JP7649730B2 - Anomaly detection device, anomaly detection system, and anomaly detection method - Google Patents

Description

The present invention relates to an anomaly detection device, an anomaly detection system, and an anomaly detection method.

Conventionally, abnormalities in a system have been detected by using a measuring instrument to measure physical quantities to evaluate the normality of the system and comparing the measured values with predetermined reference values.

For example, in a high-voltage power system, an abnormality such as an overload of the system load or a short circuit in the electrical circuit is detected, and a circuit breaker or other device is operated to protect the system. To do this, the load current flowing through the circuit is measured by a current meter, and the measurement signal is input to a protective relay. When the current based on the electrical signal output from the current meter exceeds a predetermined reference value, an abnormality such as an overload or short circuit is detected.

In substation equipment for DC railways, to protect the power supply system from overload currents that exceed the allowable load or excessive currents that occur during short circuits, the current flowing through the power supply system is constantly measured by a current meter, and system abnormalities such as overload currents and short circuit currents are detected and the DC circuit breaker is controlled.

Overload current in a DC power supply circuit is characterized by a gradual increase in current that exceeds the allowable current of the load, while short-circuit current is characterized by the current flowing through the circuit breaker installed at the substation increasing in a short period of time because the fault point is near the substation. DC circuit breakers may use DC selective circuit breakers that distinguish between short-circuit current and overload current and cut them off to reduce the load on the system by detecting short-circuit current early and cutting it off.

For example, there is a current change rate type DC selective circuit breaker that distinguishes between short circuit current and overload current based on the magnitude of the time rate of change (di/dt) of the feeding current. A current change rate type DC selective circuit breaker inputs a measurement signal measured by a current meter. If the di/dt based on the measurement signal is greater than a reference value, the current change rate type DC selective circuit breaker operates with a current smaller than the specified current for interrupting overload current, and if it is equal to or less than the reference value, it operates with a current larger than the specified current for interrupting overload current.

In Patent Document 1, the current I(t) flowing through a feeder line is measured by a current meter, and the measurement signal is digitized and input to a microcomputer. A transfer function of the current I(t) and its change amount ΔI is found, ΔI is calculated from the digitized I(t) that satisfies the transfer function, and the current I(t) is cut off if ΔI is greater than a reference value ΔIr. The characteristics of ΔI corresponding to di/dt can be derived from the transfer function, making it possible to distinguish between overload current and short-circuit current and cut them off.

Japanese Unexamined Patent Publication No. 63-53133

In a method such as a protective relay in a high-voltage power system, where a physical quantity is measured with a measuring instrument to determine the normality of the system and an abnormality is detected when the measured value exceeds a predetermined reference value, there is a risk of false detection of an abnormality when high-level noise is input to the abnormality detection circuit.

In a method for detecting anomalies by measuring the time change in a physical quantity such as current, as in Patent Document 1, when high-frequency noise occurs, the differential signal becomes large, increasing the possibility of falsely detecting an abnormality. An abnormality detection system is required to detect abnormalities by distinguishing between noise and abnormalities and not falsely detecting an abnormality when noise occurs. At the same time, for system abnormalities that must be detected early, such as short-circuit currents in DC power supply systems, it is required to shorten the delay time from occurrence to detection.

The present invention has been made in consideration of the above, and has as its object to improve the speed, detection accuracy, and reliability of anomaly detection in a target system based on the measurement results of physical quantities measured in the target system.

In order to solve the above-mentioned problems, one aspect of the present invention is an anomaly detection device that detects an anomaly in a target system based on a physical quantity measured in the target system, and is characterized by having a reading unit that reads quantized time series data of the physical quantity measured in the target system and converts the read quantized data into time series data that can be processed by calculations, an analysis unit that analyzes a predetermined number of the most recent time series data from the time series data and calculates a statistical amount and an error in the statistical amount based on the predetermined number of the most recent time series data, and a detection unit that detects an anomaly in the target system based on the statistical amount when the degree of variation in the statistical amount calculated by the analysis unit satisfies a predetermined condition.

According to one aspect of the present invention, for example, it is possible to improve the speed, detection accuracy, and reliability of anomaly detection in a target system based on the measurement results of physical quantities measured in the target system.

FIG. 1 is a diagram showing the configuration of an abnormality detection device according to an embodiment. 4 is a timing chart showing the timing of operation start and operation completion of measuring instrument output, AD conversion, digital signal reading, analysis, and abnormality detection in the abnormality detection device according to the embodiment. 5A to 5C are diagrams for explaining a method for calculating an average and a standard deviation of time-series data by an analysis unit according to the embodiment. FIG. 13 shows the calculation results of the mean, standard deviation, and error of time-series data of physical quantities. 1A and 1B are diagrams for explaining an anomaly evaluation method (part 1) based on the average and standard deviation of time-series data of physical quantities. 11A and 11B are diagrams for explaining an anomaly evaluation method (part 2) based on the average and standard deviation of time-series data of physical quantities. 11A and 11B are diagrams for explaining an anomaly evaluation method (part 3) based on the average and standard deviation of time-series data of physical quantities. 1 is a diagram for explaining a method for calculating a time rate of change of time series data of a physical quantity and its standard deviation. FIG. 13 is a graph showing calculation results of the time rate of change and the standard deviation of the time rate of change of time series data of a physical quantity. 1 is a diagram for explaining an anomaly evaluation method based on the time rate of change and standard deviation of time-series data of a physical quantity. 11A and 11B are diagrams for explaining an anomaly evaluation method using the average and standard deviation of time-series data of a physical quantity, and the time rate of change and standard deviation of the time-series data of the physical quantity. FIG. 1 is a diagram showing an overall configuration of a DC feeding circuit according to a first embodiment. FIG. 2 is a diagram showing the configuration of a fault current detection circuit according to the first embodiment. FIG. 4 is a diagram showing the state of an overload current and a short-circuit current in the DC feeding circuit according to the first embodiment. 5A and 5B are diagrams for explaining a method for detecting an overload current and a short-circuit current according to the first embodiment. 5A and 5B are diagrams for explaining a method for detecting an overload current and a short-circuit current according to the first embodiment. FIG. 13 is a graph showing the results of one-dimensional regression analysis of a simulation of noise due to damped vibration. FIG. 13 is a graph showing the results of one-dimensional regression analysis of a simulation of a short circuit current. FIG. 11 shows the comparison results of noise due to damped oscillation and simulation of short circuit current. FIG. 11 is a diagram showing the configuration of a fault current detection circuit according to a second embodiment. FIG. 13 is a diagram showing the configuration of a DC feeding circuit according to a second embodiment and a state in which a short circuit occurs. FIG. 11 is a diagram showing the configuration of a capacitor charge/discharge circuit using a capacitor capacitance inspection system according to a third embodiment.

The following describes embodiments and examples of the present invention with reference to the drawings. Note that the embodiments and examples described below, including the drawings, are merely examples and do not limit the invention according to the claims. Furthermore, not all of the elements and combinations thereof described in the embodiments and examples are necessarily essential to the solution of the invention. Furthermore, illustrations and descriptions of elements that are essential to the configuration of the invention but are well known may be omitted.

In the following description, the same reference numerals will be used to designate configurations that overlap with previously described embodiments or examples, and descriptions will be omitted, with the focus on the differences.

In the following description, a processor that executes a program or hardware logic that realizes functions equivalent to a program performs predetermined processing while appropriately using a memory unit and/or an interface, etc., to realize various functional units.

The term "processor" may refer to one or more. A processor is typically a microprocessor such as a CPU (Central Processing Unit), but may also be other types of processors such as a GPU (Graphics Processing Unit). A processor may also be a single-core or multi-core processor. A processor may also be a processor in the broader sense, such as a hardware circuit that performs all or part of the processing (for example, an FPGA (Field-Programmable Gate Array) or an ASIC (Application Specific Integrated Circuit)).

In the following explanation, for example, the notation "y￣" is equivalent to the notation with ￣ written directly above the y.

In the following explanation, the equal sign in determining whether the object to be determined is larger or smaller than the threshold value may be used in either range. For example, whether the threshold value Th is set to "if A≧Th, then B; if A<Th, then C" or "if A>Th, then B; if A≦Th, then B" is an item that can be changed as appropriate.

[Embodiment 1]
1 is a diagram showing the configuration of an abnormality detection device S according to an embodiment. The abnormality detection device S in this embodiment includes a measuring instrument 3, an AD (Analog-Digital) conversion unit 4, a digital signal reading unit 5, an analysis unit 6, and a detection unit 7.

The measuring instrument 3 measures a physical quantity 2 in the system 1 being measured to evaluate the normal or abnormal state of the system 1 being the target of abnormality detection. The AD conversion unit 4 converts the analog signal, which is the measurement signal measured by the measuring instrument 3, into a digital signal (quantized data). The digital signal reading unit 5 reads the digital signal converted by the AD conversion unit 4 and converts it into data that can be processed by calculations.

The analysis unit 6 analyzes the time series data, including the latest data among the data converted from the digital signals read by the digital signal reading unit 5, based on statistical methods, and calculates the estimated values such as the average and time rate of change of the time series data, the error of the estimated values, and the degree of variation of the time series data (standard deviation, etc.). The detection unit 7 detects an abnormality in the system 1 based on the estimated values calculated by the analysis unit 6, the error of the estimated values, and the degree of variation of the time series data.

The digital signal reading unit 5, the analysis unit 6, and the detection unit 7 may be configured as an anomaly detection device using a processor such as a microcomputer or FPGA. The anomaly detection device may be configured using one processor or multiple processors. When the anomaly detection device is configured using multiple processors, the digital signal reading unit 5, the analysis unit 6, and the detection unit 7 are appropriately integrated and distributed and placed in each processor. The anomaly detection device and the AD conversion unit 4 may also be included in the configuration of an anomaly detection system. The anomaly detection system may further include a measuring instrument 3.

Figure 2 is a timing chart showing the timing of the output of the measuring instrument 3, AD conversion, digital signal reading, analysis, and the start and end of anomaly detection operations in the anomaly detection device S according to the embodiment.

The following describes an example of short circuit and overload current detection when the system 1 being the target of anomaly detection is a power system or a DC power supply system. In this case, the physical quantity 2 is, for example, a current flowing in an electric circuit. The measuring instrument 3 is a current transformer, Rogowski coil, Hall CT (Current Transformer), etc. for measuring current. The digital signal reading unit 5, analysis unit 6, and detection unit 7 are processors capable of arithmetic processing of digital data and other processing.

In FIG. 2, the rising edge of the pulse marks the start of execution, and the falling edge marks the completion of the operation. The output 8 of the measuring device 3 is continuously input to the AD conversion unit 4, and is therefore always ON. The operation 9 of the AD conversion unit converts an analog signal into a digital signal at a predetermined cycle T that ensures real-time performance, and therefore the start and completion of the conversion process are repeated at a predetermined cycle T.

Each operation (operation 10, operation 11, and operation 12, respectively) of the digital signal reading unit 5, the analysis unit 6, and the detection unit 7 is completed by the start of the next operation of the AD conversion unit 4 (i.e., by the input of the next digital signal). Operation 10 of the digital signal reading unit 5 is approximately synchronized with operation 9 of the AD conversion unit. Operation 11 of the analysis unit 6 starts approximately at the same time as operation 10 is completed. Operation 12 of the detection unit 7 starts approximately at the same time as operation 10 is completed, and is completed by the start of the next operation 9. The total operation time of operations 10, 11, and 12 in one period is set within the predetermined period T so that the digital signal reading unit 5, the analysis unit 6, and the detection unit 7 repeatedly start and complete in the predetermined period T. In this way, abnormalities are always evaluated in real time taking into account the latest data, making it possible to detect abnormalities early.

The following describes in detail the analysis unit 6 of the anomaly detection device S, which analyzes the time-series data of the digital signal based on statistical methods to calculate estimated values such as the average and time rate of change, as well as the error in the estimated values and the degree of variation in the time-series data, and the detection unit 7, which evaluates and detects anomalies in the system 1 based on the results of the calculations by the analysis unit 6.

Figure 3 is a diagram for explaining a method for calculating the mean y^ and standard deviation σy of time series data by the analysis unit 6 according to the embodiment. In Figure 3, the horizontal axis represents time x, and the vertical axis represents data y of physical quantity 2. For example, when physical quantity 2 is a current, the unit is amperes (A).

For the time series data of physical quantity 2, the average y￣ and the degree of variation, i.e., the standard deviation σy, of N pieces of data is calculated. For simplicity, Figure 3 shows the case where the number of data N = 3, and in the first calculation, the average and standard deviation are calculated for x = [x1, x2, x3] and y = [y1, y2, y3]. In the second calculation, the average and standard deviation are calculated for x = [x2, x3, x4] and y = [y2, y3, y4], and in the third calculation, the average and standard deviation are calculated for x = [x3, x4, x5] and y = [y3, y4, y5].

In this way, every time new data is input, the oldest data used in the previous calculation is discarded using the FIFO (First In First Out) method, and a new calculation is performed including the new data. For example, in the second calculation in which y4 is input, the oldest data x1 and y1 are discarded, and the latest data x4 and y4 are included, and the average and standard deviation are found for the combination of x = [x2, x3, x4] and y = [y2, y3, y4]. Every time data is input, the calculation of the average and standard deviation is repeated for the combination including new x and y. The average y and standard deviation σy are calculated from equations (1) and (2).

Here, y is the average of y, j is the number of calculations, and σy is the standard deviation of y. In formula (2), the standard deviation is calculated using unbiased variance, but normal variance may be used in which N-1 in 1/(N-1) is replaced with N. Since j is the number of calculations, for example, in an explanation using FIG. 3, N=3, and in the second calculation, i=j=2, j+N-1=4, and the average and standard deviation of data from i=2 to 4 (combination of x=[x2, x3, x4], y=[y2, y3, y4]) are calculated. By calculating the average y and standard deviation σy, it is possible to evaluate how much the data varies with respect to the average, so that, for example, if the measuring instrument 3 has a 1% error, it is possible to estimate that noise has occurred if the standard deviation of the time-series data acquired by the measuring instrument 3 is greater than 1% of the average. In this way, the detection accuracy can be improved by using the standard deviation as an index of anomaly detection. In addition, formula (3) may be substituted for the standard deviation to be used as an index of anomaly detection.

In the above, the standard deviation σy has been described as an index for detecting an anomaly, but the standard deviation σy may be used as an index for error. The standard deviation σy can be calculated using equation (4). In Fig. 4 and subsequent figures, the method for evaluating an anomaly will be described using the standard deviation σy, but the standard deviation σy can also be used instead.

Figure 4 shows the calculation results of the mean y￣, standard deviation σy, and error y￣error for the time series data of physical quantity 2. The time x increases by 1, and the physical quantity data y varies between 2 and 4. As in the example of Figure 3, the number of data N = 3. For example, as shown in the row i = 1 of the table in Figure 3, the mean y￣ for x = [x1, x2, x3] and y = [y1, y2, y3] is 3.0, the standard deviation σy is 0.12, and the error is 4.1%. Also, as shown in the row i = 5, the mean y￣ for x = [x5, x6, x7] and y = [y5, y6, y7] is 3.2, the standard deviation σy is 0.41, and the error is 13.1%, which means that the standard deviation σy and error y￣error are larger than those for x = [x1, x2, x3] and y = [y1, y2, y3]. This is because the data range for y = [y5, y6, y7] is 2.6 to 3.6, which is a larger variation in the data than the data range for y = [y1, y2, y3], which is 2.9 to 3.2. In this way, the standard deviation σy and error y￣error become larger according to the variation in the data.

Figure 5 is a diagram for explaining an anomaly evaluation method (part 1) based on the average y￣ and standard deviation σy of time-series data of physical quantity 2. In Figure 5, the horizontal axis is the average y￣ and the vertical axis is the standard deviation σy. Figure 5 shows an example of a system in which an anomaly is detected when physical quantity 2 is larger than a predetermined reference value. The reference value Cty￣ of physical quantity 2 is shown by a dotted line on the horizontal axis. In Figure 5, when the standard deviation σy is larger than expected, the reliability of the estimated value is low and it is not determined to be abnormal, so the reference value Ctσy of the standard deviation σy is also shown on the vertical axis, and the shaded area is the anomaly detection area.

The reference value Ctσy of the standard deviation σy may be, for example, a value estimated from the measurement error of the measuring instrument 3, or may be determined from the expected fluctuation range of the physical quantity 2 as a system. In either case, for example, when unexpected noise enters the anomaly detection device S, the standard deviation σy becomes large and an abnormality is not evaluated even if the estimation accuracy of the mean y￣ exceeds the reference value Cty￣; on the other hand, when the standard deviation σy is small, the estimation accuracy of the mean y￣ increases, and an abnormality is evaluated when the mean y￣ exceeds the reference value Cty￣. In this way, the accuracy and reliability of anomaly detection can be improved by evaluating anomalies taking the standard deviation into account.

Figure 6 is a diagram for explaining an anomaly evaluation method (part 2) based on the average y ̂ and standard deviation σy of time-series data of physical quantity 2. Figure 6 shows an example of a system in which an anomaly is detected when physical quantity 2 is smaller than a predetermined reference value Cty ̂. In Figure 6, similar to the example shown in Figure 5, when the standard deviation σy is larger than expected, the reliability of the estimated value is low and it is not determined to be abnormal, so the vertical axis shows the reference value Ctσy of the standard deviation σy, and the shaded area is the anomaly detection area.

Figure 7 is a diagram for explaining an anomaly evaluation method (part 3) based on the average, standard deviation y^ and standard deviation σy of time-series data of physical quantity 2. In Figure 7, a system is shown as an example in which a reference value is set in advance for the extent to which the standard deviation σy can vary under normal circumstances. When the standard deviation σy exceeds the reference value Ctσy, it is evaluated and detected as an anomaly, so the shaded area is set as the anomaly determination area.

In this way, in the anomaly detection device S of this embodiment, the physical quantity 2 is measured by the measuring instrument 3, and the measured analog signal is converted to a digital signal for processing, so that the anomaly detection conditions can be easily changed as shown in Figures 5, 6, and 7 to suit the system 1 for which anomaly detection is to be performed. That is, the characteristic curves of the standard deviation y￣ and the standard deviation σy can be obtained, and the area above or below the characteristic curve can be used as the anomaly detection condition. Also, although the vertical axis in Figures 5, 6, and 7 is the standard deviation σy, the error y￣error in equation (3) or the standard deviation σy￣ in equation (4) can be used instead of the standard deviation σy.

Figure 8 is a diagram for explaining a method of calculating the time change rate a and its standard deviation σa of the time series data of physical quantity 2. In Figure 3, the horizontal axis is time x, and the vertical axis is data y of physical quantity 2. For example, when physical quantity 2 is current, the unit is ampere (A). For the time series data of physical quantity 2, the time change rate a and its standard deviation σa are calculated for N pieces of data. For simplicity, Figure 8 shows the case where the number of data N = 3, and in the first calculation, the time change rate a and its standard deviation σa are calculated for x = [x1, x2, x3], y = [y1, y2, y3]. In the second calculation, the time change rate a and its standard deviation σa are calculated for x = [x2, x3, x4], y = [y2, y3, y4], and in the third calculation, the time change rate a and its standard deviation σa are calculated for x = [x3, x4, x5], y = [y3, y4, y5].

In this way, every time new data is input, the oldest data used in the previous calculation is discarded by FIFO (First In First Out) and a new calculation is performed including the new data. For example, in the second calculation in which y4 is input, the oldest data x1 and y1 are discarded and the latest data x4 and y4 are included, and the time change rate a and its standard deviation σa are obtained for the combination of x = [x2, x3, x4] and y = [y2, y3, y4]. Every time data is input, the calculation of the time change rate a and its standard deviation σa is repeated for the combination including new x and y. The time change rate a and the standard deviation σa are calculated, for example, using regression analysis. The regression equation with the slope (i.e., the time change rate) as a and the intercept as b is y = ax + b. The total error W between yi and the regression equation is expressed by Equation (5). N is the number of data and j is the number of calculations.

For example, in the case of N=3 and the second calculation, i=j=2, j+N-1=4, the time change rate a and standard deviation σa are calculated for data from i=2 to 4. By finding a and b when the value obtained by partially differentiating equation (5) with respect to a or b is zero, that is, δW/δa=0 and δW/δb=0, it is possible to calculate the values of slope a and intercept b that minimize the total error W. a and b are expressed as equations (6) and (7), respectively.

After deriving a, b may be obtained from the linear regression equation y = ax + b. In reality, these estimated values a and b are estimated values just like the average y, so the standard deviation σa (error of a) and the standard deviation σb (error of b) can be calculated. When the data varies widely, that is, when noise occurs, the standard deviations σa and σb become large. In this case, as shown in equations (8), (9), and (10), σa and σb are derived after the residual sum of squares ^Sy2 of y is found.

The error aerror of the time rate of change a may be defined as shown in equation (11).

In the above, a method for calculating the time rate of change using the linear regression equation y = ax + b was explained, but if the time change of physical quantity 2 in the target system 1 can be expressed as an exponential function (for example, physical quantity 2 = αexp(-βt), where α and β are coefficients and t is time), regression analysis can be performed using that function as the regression equation to calculate the time rate of change and standard deviation.

Figure 9 shows the calculation results of the time rate of change a and the standard deviation σa of the time rate of change a of the time series data of physical quantity 2. The time x increases by 1, and the physical quantity data y varies between 2 and 10. As with the example in Figure 8, the number of data N = 3. For example, as shown in the row i = 1 in the table in Figure 3, the time rate of change a of x = [x1, x2, x3], y = [y1, y2, y3] is 1, the standard deviation σa is 0, and the error aerror is 0%. This is because y always increases by 1 every time x increases by 1, and the time rate of change a can be calculated without error.

On the other hand, as shown in the row i=4, the time change rate a for x=[x4, x5, x6] and y=[y4, y5, y6] is -0.05, the standard deviation σa is 0.144, and the error aerror is 28.868%. This is because if the time change rate a is calculated from the two points y4=3.5 and y5=3.2, it will be negative, but if the time change rate a is calculated from y5 and y6 because y6=3.5, it will be positive. Therefore, the time change rate a calculated from the three points y4, y5, and y6 is both positive and negative, and the standard deviation σa and error aerror will be large. When detecting an abnormality using the time change rate a, the estimated accuracy of the time change rate a is verified based on the magnitude of the standard deviation σa and error aerror, and the abnormality is evaluated.

Figure 10 is a diagram for explaining an anomaly evaluation method based on the time rate of change a and standard deviation σa of the time series data of physical quantity 2. In Figure 10, the horizontal axis represents the time rate of change a, and the vertical axis represents the standard deviation σa. Figure 10 shows an example of a system in which an anomaly is detected when the time rate of change a of physical quantity 2 is greater than a predetermined reference value. The reference value Cta of the time rate of change a of physical quantity 2 is indicated by a dotted line on the horizontal axis. In Figure 10, when the standard deviation σa is greater than expected, the reliability of the estimated value is low and it is not determined to be an anomaly, so the reference value Ctσa of the standard deviation σa is also indicated on the vertical axis, and the shaded area is designated as the anomaly detection area.

The reference value Ctσa of the standard deviation σa may be, for example, a value expected from the measurement error of the measuring instrument 3, or may be determined from the expected fluctuation range of the time rate of change a of the physical quantity 2 in the system 1. In either case, for example, when unexpected noise enters the anomaly detection device S, the standard deviation σa becomes larger than the reference value Ctσa, and the estimation accuracy of the time rate of change a decreases, so even if the time rate of change a exceeds the reference value Cta, it is not evaluated as an abnormality. However, when the standard deviation σa is equal to or less than the reference value Cta, the estimation accuracy of the time rate of change a increases, so if the time rate of change a exceeds the reference value Cta, it is evaluated as an abnormality. In this way, the accuracy and reliability of the anomaly evaluation can be improved.

As in the abnormality evaluation method using the average y and standard deviation σy of the physical quantity 2 shown in Figures 5, 6, and 7, even when the time change rate a and its standard deviation σa obtained by regression analysis are used, the abnormality detection region may be determined according to the target system. For example, if the time change rate a of the physical quantity 2 is determined to be abnormal when it is smaller than a predetermined reference value Cta, it is evaluated as abnormal when the time change rate a of the physical quantity 2 calculated in the analysis unit 6 is smaller than the reference value Cta and the standard deviation σa is equal to or smaller than the reference value Ctσa. Alternatively, if the time change rate a of the physical quantity 2 is determined to be abnormal when it is larger than the predetermined reference value Cta, it is evaluated as abnormal when the time change rate a of the physical quantity 2 is larger than the reference value Cta and the standard deviation σa is equal to or smaller than the reference value Ctσa. Alternatively, in a system 1 in which the extent to which the standard deviation σa can vary is known in advance, it is determined as abnormal when the standard deviation σa exceeds the reference value Ctσa.

In this way, in the anomaly detection device S of this embodiment, the physical quantity 2 is measured by the measuring instrument 3, and the measured analog signal is converted to a digital signal for statistical processing, so that the anomaly detection conditions can be easily changed to match the system 1 that is the target of anomaly detection. A characteristic curve of the time change rate a and standard deviation σa may be obtained, and the area above or below this characteristic curve may be used as the anomaly detection condition. The vertical axis may also be the error aerror in equation (10).

FIG. 11 is a diagram for explaining an anomaly evaluation method using the average y and standard deviation σy of the time series data of physical quantity 2, and the time change rate a and standard deviation σa of the time series data of physical quantity 2. In FIG. 11, the horizontal axis is the average y and the vertical axis is the time change rate a, and the reference values are indicated by dotted lines. As indicated by diagonal lines in FIG. 11, the anomaly detection region is defined as a region in which the average y is equal to or greater than the reference value Cty and the time change a of physical quantity 2 is equal to or greater than the reference value Cta. In this case, if at least one of the standard deviation σy of the time series data and the standard deviation σa of the time change a is greater than the respective reference values Ctσy and Ctσa, the anomaly is not evaluated as being abnormal. In other words, the following are the anomaly detection conditions.
(y￣>Cty￣)∧(σy<Ctσy)∧(a>Cta)∧(σa<Ctσa)
...(12)

The method in Figure 11 is effective for a system 1 in which the abnormal mode depends on either the average y￣ or the time rate of change a of physical quantity 2. The abnormality detection region of system 1 is determined using the average y￣, standard deviation σy (or error y￣error), time rate of change a, and standard deviation σa of the time rate of change a (or error aerror of the time rate of change a).

11, the abnormality detection region is defined as the region where the average y￣ of physical quantity 2 is equal to or greater than the reference value Cty￣ and the time change a of physical quantity 2 is equal to or greater than the reference value Cta, but since the abnormality evaluation region differs depending on the target system 1 and the physical quantity 2 being measured, the abnormality detection region may be changed according to the characteristics of the system 1. For example, the abnormality detection region may be defined as the region where the average y￣ of physical quantity 2 is equal to or less than the reference value Cty￣ and the time change a of physical quantity 2 is equal to or less than the reference value Cta. In a system 1 in which the fluctuation range of the standard deviation σy of the time series data and the standard deviation σa of the time change rate a is known in advance, an abnormality may be detected when the standard deviation σy or the standard deviation σa exceeds the respective reference values Cty￣ and Ctσa.

In Example 1, an example is described in which the abnormality detection device S and the abnormality detection method described in the embodiment are applied to a railway DC power supply system. Note that "cl" in the figure of Example 1 indicates the closed state of the circuit breaker.

Figure 12 is a diagram showing the overall configuration of the DC power supply circuit 1S according to the first embodiment. The DC power supply circuit 1S includes a substation 13 (substation A) which is a first substation facility and a substation 14 (substation B) which is a second substation facility adjacent to the substation 13. The substation is also called a substation system. The substation 13 constitutes a substation system that receives AC power from the power grid and supplies DC power to the power supply line. The substation 13 includes an AC circuit breaker 15, a transformer 16, a rectifier 17, a DC circuit breaker 18, a DC circuit breaker 19, a DC circuit breaker 20, and a DC circuit breaker 21. The substation 13 converts the AC power received from the power grid into DC power via the transformer 16 and the rectifier 17, and supplies the power to the power supply line.

Substation 14 is equipped with AC circuit breaker 22, transformer 23, rectifier 24, DC circuit breaker 25, DC circuit breaker 26, DC circuit breaker 27, and DC circuit breaker 28. Similar to substation 13, substation 14 converts AC power received from the power grid into DC power via transformer 23 and rectifier 24, and supplies the power to the feeder line.

Under normal conditions, DC circuit breaker 18, DC circuit breaker 19, DC circuit breaker 20, DC circuit breaker 21, DC circuit breaker 25, DC circuit breaker 26, DC circuit breaker 27, DC circuit breaker 28, AC circuit breaker 15, and AC circuit breaker 22 are in the closed state. DC circuit breaker 18 is connected to section A of the upward overhead line 29. DC circuit breaker 21 and DC circuit breaker 25 are connected to section B of the upward overhead line 29. DC circuit breaker 28 is connected to section C of the upward overhead line 29. DC circuit breaker 19 is connected to section F of the downward overhead line 30. DC circuit breaker 20 and DC circuit breaker 26 are connected to section E of the downward overhead line 30. DC circuit breaker 27 is connected to section D of the downward overhead line 30.

When train 31 is traveling in section B, DC power is supplied from substations 13 and 14 via up-going overhead lines 29. The DC power supplied to train 31 returns to substations 13 and 14 via rails 32. The paths of the current supplied from substations 13 and 14 to the train are as shown in current paths 33 and 34.

In FIG. 12, for example, in the substation 13, the current meter 35 measures the current downstream of the DC circuit breaker 21 and inputs the analog signal to the fault current detection circuit 36. The fault current detection circuit 36 is a DC selective interruption device. The fault current detection circuit 36 evaluates abnormalities such as overload current and short circuit current based on the input analog signal. When a fault current occurs, the fault current detection circuit 36 detects the abnormality and opens the DC circuit breaker 21 to interrupt the current. Although not shown, the DC circuit breaker 18, the DC circuit breaker 19, and the DC circuit breaker 20 are also each provided with a current meter and a fault current detection circuit. When a fault current occurs, each fault current detection circuit detects the abnormality and controls the corresponding DC circuit breaker 18, the DC circuit breaker 19, and the DC circuit breaker 20.

Similarly, in the substation 14, the current meter 37 measures the current downstream of the DC circuit breaker 25 and inputs the analog signal to the fault current detection circuit 38. The fault current detection circuit 38 is a DC selective interruption device. The fault current detection circuit 38 evaluates anomalies such as short circuit current and load current based on the input analog signal. When a fault current occurs, the fault current detection circuit 38 detects the abnormality and opens the DC circuit breaker 25 to interrupt the current. Although not shown, the DC circuit breaker 26, the DC circuit breaker 27, and the DC circuit breaker 28 are also each provided with a current meter and a fault current detection circuit. When a fault current occurs, each fault current detection circuit detects the abnormality and controls the corresponding DC circuit breaker 26, the DC circuit breaker 27, and the DC circuit breaker 28, respectively.

Figure 13 is a diagram showing the configuration of the fault current detection circuit 36 according to the first embodiment. The fault current detection circuit 38 has the same configuration as the fault current detection circuit 36. An analog signal measured by a current meter 35 is input to the fault current detection circuit 36. The fault current detection circuit 36 has the same configuration as the abnormality detection device S shown in Figure 1 of the embodiment, and is configured to include an AD conversion unit 4, a digital signal reading unit 5, an analysis unit 6, and a detection unit 7. Furthermore, the fault current detection circuit 36 is configured to include a breaker control unit 7a.

The analysis unit 6 in this embodiment calculates the average I of the time series data of the current I, the standard deviation σI of the time series data of the current I, the time rate of change dI/dt of the time series data of the current I and its standard deviation σ _dt/dI . These calculations are performed by the methods described in Fig. 3 and Fig. 8 of the embodiment. When the detection unit 7 detects an abnormality, the circuit breaker control unit 7a transmits an opening command to the DC circuit breaker 21 to control the DC circuit breaker 21.

The fault current detection circuit 36 may be configured with a processor such as a microcomputer or FPGA. The fault current detection circuit 36 may be configured with one processor or multiple processors. When the fault current detection circuit 36 is configured with multiple processors, the AD conversion unit 4, the digital signal reading unit 5, the analysis unit 6, the detection unit 7, and the circuit breaker control unit 7a are appropriately integrated and distributed and arranged in each processor. For example, the AD conversion unit 4 may be arranged in one processor, and the digital signal reading unit 5, the analysis unit 6, the detection unit 7, and the circuit breaker control unit 7a may be arranged in one processor. Alternatively, the AD conversion unit 4 may be arranged in one processor, the digital signal reading unit 5, the analysis unit 6, and the detection unit 7 may be arranged in one processor, and the circuit breaker control unit 7a may be arranged in one processor.

14 is a diagram showing the state of the overload current and the short circuit current in the DC power supply circuit 1S according to the first embodiment. In FIG. 14, the horizontal axis represents time t, and the vertical axis represents the current I. In the DC power supply system, when the load current becomes larger than the allowable current or when a short circuit occurs, the occurrence of a fault current, i.e., an abnormality in the power supply system, is detected and the DC circuit breaker is controlled. As shown in FIG. 14, the overload current 39 is characterized by gradually increasing and becoming larger than the allowable current CtI of the load, and the short circuit current 40 is characterized by the current flowing through the circuit breaker arranged at the substation increasing in a short time (the time change rate dI/dt is greater than a predetermined value) as shown in the time change rate 40a, since the fault point is near the substation in the example of FIG. 14. The fault current detection circuit 36 is required to selectively detect the overload current 39 and the short circuit current 40, not to operate unnecessarily when noise occurs, and to detect the abnormality of the overload and the short circuit early. From the above, the anomaly detection device and anomaly detection method shown in the embodiment can be applied to a DC power supply system.

15A and 15B are diagrams for explaining a method for detecting an overload current and a short circuit current according to the first embodiment. In FIG. 15A, the horizontal axis indicates the average I of the time series data of the current I, and the vertical axis indicates the time rate of change dI/dt of the time series data of the current I. In FIG. 15A, the reference values of the current I are I1 and I2, and the reference value of the time rate of change dI/dt of the current I is A1. In FIG. 15B, the horizontal axis indicates the standard deviation σI of the time series data of the current I, and the vertical axis indicates the standard deviation σ _dI/ dt of the time rate of change dI/dt of the time series data of the current I. In FIG. 15B, the reference values of the current I are I1 and I2, and the reference value of the time rate of change dI/dt of the current I is A1.

As shown by the diagonal lines in Figure 15A, the fault current judgment region is divided into region Re1 and region Re2. Region Re1 is the region used to detect short-circuit current, and is the region where a fault is judged to exist when the current is greater than reference value I1 and the time rate of change of current I is greater than reference value A1. Region Re2 is the region used to detect overload current, and is the region where a fault is judged to exist when the current is greater than reference value I2 and the time rate of change of current I is less than reference value A1.

In the fault current detection circuit 36, the analysis unit 6 calculates the average I of the time series data of current I, the standard deviation σI of the time series data of current I, the time rate of change dI/dt of the time series data of current I and its standard deviation σ _dI/dt by statistical methods, and detects an overload current or a short circuit current based on these calculation results. For example, the average I of the time series data of current I is obtained from equation (1), the standard deviation σI of the time series data of current I from equation (2), the time rate of change dI/dt of the time series data of current I from equation (6), and the standard deviation σ _dI/dt of the time rate of change of the time series data of current I from equation (9). In FIG. 15A, a reference value CtσI of the standard deviation σI in the time-series data of current I and a reference value CtσdI _{/ dt} of the standard deviation σdI/dt of the time change rate are determined, and if at least one of the standard deviation σI and the standard deviation σdI _/dt exceeds the reference value shown in FIG. 15B, even if the conditions shown in FIG. 15A fall within the regions Re1 and Re2, the current is not evaluated as a fault current (i.e., an overload current and a short circuit current) and is not detected.

That is, the detection condition for the short-circuit current corresponds to the region Re1 and is expressed by the formula (13).
(I￣＞I1)∧(σI＜CtσI)∧(dI/dt＞A1)∧(σ _dI/dt <Ctσ _dI/dt )
...(13)

Moreover, the detection condition for the overload current corresponds to the region Re2 and is expressed by the formula (14).
(I￣＞I2)∧(σI＜CtI)∧(dI/dt＜A1)∧(σ _dI/dt <Ctσ _dI/dt )
...(14)

In the above, the standard deviation σI and the standard deviation σ _dI/dt are used as indexes of error in the calculation results. However, the error I error of the average I based on equation (3) and the error dI/dterror of the time rate of change dI/dt based on equation (11) may also be used as indexes of error.

Fig. 16A is a diagram showing the results of one-dimensional regression analysis of the simulation of noise due to damped vibration. Fig. 16B is a diagram showing the results of one-dimensional regression analysis of the simulation of short circuit current. Fig. 16C is a diagram showing the results of comparison between the simulation of noise due to damped vibration and the short circuit current. Using the results of regression analysis of the simulation of noise and short circuit current in Figs. 16A and 16B, the effectiveness of the anomaly detection method in this embodiment will be explained. Fig. 16A shows a waveform 16a of time series data in which noise is simulated by damped vibration. Fig. 16B shows a waveform 16b of time series data in which a short circuit current rising at 3 x ¹⁰⁶ A/s is simulated.

For the short-circuit current waveform in Figure 16B, the measuring instrument was assumed to have an error of ±2%, and the error was calculated using pseudorandom numbers as y = 3 x 10 ⁶ x ±2%. For comparison, it was necessary to perform regression analysis with the same number of data and time width, so regression analysis was performed on data points with the same time width ΔT ((0-T) seconds), as shown in Figures 16A and 16B. The number of data within ΔT in Figures 16A and 16B was also the same. The results are shown in Figure 16C.

As shown in Fig. 16C, the time rate of change dI/dt of the current I was 3.12 x ¹⁰⁶ A/s for the damped vibration noise and 2.96 x ¹⁰⁶ A/s for the short circuit current. The standard deviation σ dI/dt of the time rate of change dI _/dt of the current I was 0.53 x ¹⁰⁶ A/s for the damped vibration noise and 0.0059 x ¹⁰⁶ A/s for the short circuit current. The error σ _dI/dt error of the time rate of change dI/dt of the current I was 17% for the damped vibration noise and 0.2% for the short circuit current. The time rate of change dI/dt of the damped vibration noise and the short circuit current are almost the same, but there is a large difference in the standard deviation σ _dI/dt . This shows that according to this embodiment, the standard deviation σ _dI/dt is used to prevent unnecessary detection of an abnormality when noise occurs.

Furthermore, when detecting the time rate of change dI/dt using an analog differentiation circuit, for example, if the measured current signal has a measurement error of ±2% as shown in FIG. 16B, the measurement result of the differentiation signal is expected to have an error of about ±4%, twice the measurement error of the current. However, by using regression analysis, which is a statistical method, as in this embodiment, the error σ _dI/dt error of the time rate of change dI/dt of the time rate of change can be suppressed to 0.2% as shown in FIG. 16C. Therefore, it can be seen that this embodiment improves the accuracy of detecting anomalies. The error σ _dI/dt error was calculated using formula (11).

In Example 2, a method for estimating a fault point when a short circuit current occurs between two adjacent substations is described using the fault current detection circuits 36 and 38 described in Example 1.

FIG. 17 is a diagram showing the configuration of a fault current detection circuit 41 according to the second embodiment. In the fault current detection circuit 41 according to the second embodiment, an analog signal measured by a current meter 35 is input, similarly to the fault current detection circuit 36 described in FIG. 13. The fault current detection circuit 41 includes an AD conversion unit 4, a digital signal reading unit 5, an analysis unit 6, a detection unit 7, and a fault point analysis unit 42. The fault point analysis unit 42 uses the average I of the time series data of the current I calculated by the analysis unit 6, the standard deviation σI of the time series data of the current I, the time rate of change dI/dt of the time series data of the current I and its standard deviation σ _dI/dt to calculate the resistance R and inductance L of the electric path formed from the substation via the fault point 44 (FIG. 18), and transmits them to the remote slave station 47 (FIG. 18). The remote slave station 47 is a computer or the like.

The fault current detection circuit 41 may be configured with a processor such as a microcomputer or FPGA. The fault current detection circuit 41 may be configured with one processor or multiple processors. When the fault current detection circuit 41 is configured with multiple processors, the AD conversion unit 4, the digital signal reading unit 5, the analysis unit 6, the detection unit 7, and the fault point analysis unit 42 are appropriately integrated and distributed and placed in each processor.

Figure 18 is a diagram showing a DC power supply circuit 2S according to the second embodiment and the state when a short circuit occurs. The DC power supply circuit 2S according to the second embodiment is configured to include a substation 13 (substation A) which is a first substation facility and a substation 14 (substation B) which is a second substation facility adjacent to the substation 13, similar to Figure 12 in the first embodiment. In Figure 18, a case where a short circuit occurs between section B of the up-going overhead line 29 and the rail 32 is considered, and for simplicity, the DC circuit breaker 18, DC circuit breaker 19, DC circuit breaker 20, DC circuit breaker 26, DC circuit breaker 27, and DC circuit breaker 28 shown in Figure 12 are not shown.

In FIG. 18, in the substation 13, the current meter 35 measures the current downstream of the DC circuit breaker 21 and inputs the analog signal to the fault current detection circuit 41 described in FIG. 17. Similarly in the substation 14, the current meter 37 measures the current downstream of the DC circuit breaker 25 and inputs the analog signal to the fault current detection circuit 43. The configuration of the fault current detection circuit 43 is the same as that of the fault current detection circuit 41.

If a short circuit occurs between section B of the up-going overhead line 29 and the rail 32, the resistance at the fault point 44 becomes zero, resulting in a current path 45 that flows from the AC input of the substation 13 to the fault point 44 via the DC circuit breaker 21, and a current path 46 that flows from the AC input of the substation 14 to the fault point 44 via the DC circuit breaker 25.

18, the resistance of the upstream overhead line 29 from the substation 13 to the fault point 44 is _R11 , the inductance is _L11 , and the resistance of the rail 32 from the fault point 44 to the substation A13 _{is R12} , and the inductance is _L12 . The resistance _R1 of the resistor in the current path 45 is _R1 = _R11 + _R12 , and the inductance _L1 in the current path 45 is _L1 = _L11 + _L12 . Similarly, the resistance of the upstream overhead line 29 from the substation 14 to the fault point 44 is _R21 , the inductance is _L21 , and the resistance of the rail 32 from the fault point 44 to the substation 14 is _R22 , and the inductance is _L22 . The resistance R ₂ in the current path 46 is expressed as R ₂ =R ₂₁ +R ₂₂ , and the inductance L ₂ in the current path 46 is expressed as L ₂ =L ₂₁ +L ₂₂ .

When a short circuit current is detected, the fault current detection circuit 41 calculates the resistance _R1 and the inductance _L1 , and the fault current detection circuit 43 calculates the resistance _R2 and the inductance _L2 , and transmits them to a remote slave station 47 such as a remote monitoring and control facility. The remote slave station 47 calculates the location of the fault point based on the received resistance _R1 and inductance _L1 , and resistance _R2 and inductance _L2 . The calculation method will be described in detail below.

18, the current path 45 is a series LR circuit. If the sending voltage of the substation 13 is V0 (V) and the current flowing through the current path 45 is I, the following formula (15) is established. For example, V0 may be a constant such as 1500 (V), or the sending voltage may be measured by a voltage meter (not shown).

As described with reference to Fig. 17, in the fault current detection circuit 41, the fault point analysis unit 42 records the average I of the time series data of the current I calculated by the analysis unit 6, the standard deviation σI of the time series data of the current I, the time rate of change dI/dt of the time series data of the current I and its standard deviation σdI/dt. The fault point analysis unit 42 also records the sending voltage V0 measured by a voltage meter (not shown). If the time when the short circuit current is detected is t0 and the time ΔT seconds after that is t1 = t0 + ΔT, the following formulas (16) and (17) are established from formula (15).

The resistance R1 and the inductance L1 are determined from the simultaneous equations (13) and (14).

In this case, the average I of the time series data of current I and the time rate of change dI/dt of the time series data of current I are used such that the calculation results of the standard deviation σI of the time series data of current I and the standard deviation σdI _/dt of the time rate of change of the time series data of current I are smaller than the respective reference values. By doing so, _R1 and _L1 can be obtained with high accuracy.

In this way, the results calculated by the analysis unit 6 of the fault current detection circuit 41 are recorded by the fault point analysis unit 42, and the resistance and inductance of the electric circuit at the substation and the fault point can be derived by using the time series data of the calculation results and equations (18) and (19). The resistance _R2 and inductance _L2 of the current path 46 can also be calculated using equations (18) and (19). Since the resistance and inductance are proportional to the length of the electric circuit, if the distance _DAB between the substation 13 and the substation 14 is known, the remote substation 47 can use _R1 and _R2 to derive the distance D _A from the substation 13 to the fault point 44 and the distance D _B from the substation 14 to the fault point 44 as shown in equations (20) and (21). For example, _R1 : _R2 = D _A :D _B = D _A :(D _AB - _{D A} ) = (D _AB - _{D B} ):D _B.

Furthermore, distances D _A and D _B can be derived using L ₁ and L ₂ as shown in equations (22) and (23).

In addition, in a DC power feeding system consisting of a series circuit of inductance L and resistance R, the resistance and inductance may be obtained from equation (24) which expresses the current I at the time t when a short circuit occurs when the output voltage E from the substation is E.

As can be seen from equation (24), in a DC LR circuit, the current I approaches E/R as time passes. Regression analysis is performed using equation (25) as a regression curve to calculate the regression coefficients c and d. The resistance and inductance are calculated from these values and the voltage E, and the distance from the substation to the fault point can be calculated from equations (20) and (21), or (21) and (22).

Note that it is not necessary for the remote substation 47 to calculate the fault point 44. For example, when a short circuit current occurs, the fault point analysis unit 42 of the fault current detection circuit 41 or 43 that detected the short circuit current communicates with the fault point analysis unit 42 of the fault current detection circuit 41 or 43 of the adjacent substation to receive the resistance and inductance values. The fault point analysis unit 42 of the fault current detection circuit 41 or 43 that detected the short circuit current may calculate the fault point using the resistance and inductance received from the fault point analysis unit 42 of the fault current detection circuit 41 or 43 of the adjacent substation and the resistance and inductance calculated by the device itself.

In Example 3, a capacitor capacity inspection system and a capacitor capacity inspection method are described that apply the anomaly detection device S and the anomaly detection method described in the embodiment. Example 3 is an example of a case where multiple physical quantities are measured in the target system.

Figure 19 is a diagram showing the configuration of a capacitor charge/discharge circuit 3S using a capacitor capacity inspection system according to the third embodiment. The capacitor charge/discharge circuit 3S shown in Figure 19 includes a DC power source 48, a charge switch 49, a charge resistor 50, a capacitor 51, a discharge resistor 52, a discharge switch 53, a charge/discharge control device 54, a diode 55, a current meter 56, a voltage meter 57, and a capacitor capacity inspection unit 58.

The charge switch 49 and the discharge switch 53 use contact relays or non-contact semiconductor switches. When charging, the charge/discharge control device 54 turns off the discharge switch 53 and then turns on the charge switch 49 to start charging, and when discharging, it turns off the charge switch 49 to stop charging, and then turns on the discharge switch 53 to start discharging.

Diode 55 is arranged to control the current flow from capacitor 51 to discharge resistor 52 during discharge. Ammeter 56 measures the current flowing through capacitor 51. Voltage meter 57 measures the voltage of capacitor 51. Ammeter 56 inputs an analog signal of the measured current to capacitor capacitance inspection unit 58. Voltage meter 57 inputs an analog signal of the measured voltage to capacitor capacitance inspection unit 58.

The capacitor capacity inspection unit 58 is composed of an AD conversion unit 4a, a digital signal reading unit 5a, and an analysis unit 6a. The AD conversion unit 4a converts the analog signal of the measured current into a digital signal. The digital signal reading unit 5a reads the digital signal converted by the AD conversion unit 4a and converts it into data so that it can be processed. The analysis unit 6a analyzes the time series data, including the latest data among the data converted from the digital signal read by the digital signal reading unit 5a, based on statistical methods, and calculates the estimated values such as the average and time change rate of the time series data, the error of the estimated values, and the degree of variation of the time series data (standard deviation, etc.).

The capacitor capacity inspection unit 58 is composed of an AD conversion unit 4b, a digital signal reading unit 5b, and an analysis unit 6b. The AD conversion unit 4b converts the analog signal of the measured voltage into a digital signal. The digital signal reading unit 5b reads the digital signal converted by the AD conversion unit 4b and converts it into data so that it can be processed. The analysis unit 6b analyzes the time series data, including the latest data among the data converted from the digital signal read by the digital signal reading unit 5a, based on statistical methods, and calculates the estimated values such as the average and time rate of change of the time series data, the error of the estimated values, and the degree of variation of the time series data (standard deviation, etc.).

The capacitor capacity inspection unit 58 also includes a capacitor capacity calculation unit (detection unit) 59 that estimates the capacity of the capacitor 51 based on the calculation results of the analysis units 6a and 6b. The capacitor capacity calculation unit 59 transmits the calculated estimated value of the capacitor capacity to the management unit 60.

The management unit 60 compares the capacitor capacity management value with the estimated capacitor capacity calculated by the capacitor capacity calculation unit 59, and issues an alarm indicating an abnormality if the value deviates from the management value. The management unit 60 also records the estimated capacitor capacity received from the capacitor capacity calculation unit 59.

In FIG. 19, the management unit 60 is illustrated outside the capacitor capacity inspection unit 58, but it may be disposed inside the capacitor capacity inspection unit 58. The evaluation and inspection of the capacity of the capacitor 51 is performed when the capacitor 51 is charged or discharged, and is executed in response to a command from the charge/discharge control device 54.

The analysis unit 6a calculates the average I of the time series data of the current I and the standard deviation σI of the time series data of the current I. The analysis unit 6b calculates the time rate of change dV/dt and its standard deviation σ _dV/dt of the time series data of the voltage V. The average I and the standard deviation σI are calculated by the method shown in Fig. 3, and the time rate of change dV/dt and the standard deviation σ _dV/dt are calculated by the method shown in Fig. 8.

Alternatively, a model function expected for the charging circuit can be created, and the model function can be used as a regression equation to perform regression analysis. In either case, the calculation is performed using statistical methods.

The following describes how the capacitor capacitance is calculated by the capacitor capacitance calculation unit 59. If the capacitance of the capacitor 51 is C, the charge stored in the capacitor is Q, and the capacitor voltage is V, then equation (26) holds.

When equation (26) is differentiated with respect to time, equation (27) is obtained.

Since the change in charge over time is a current I, equation (27) can be transformed into equation (28).

The capacitance of the capacitor can be calculated by substituting the average I of the time series data of the current I calculated by the analysis unit 6a and the time rate of change dV/dt of the time series data of the voltage V calculated by the analysis unit 6b into equation (28). At this time, the accuracy of the calculation of the capacitance of the capacitor can be improved by using data when the standard deviation σI of the time series data of the current I and the standard deviation σ _dV/dt of the time rate of change of the time series data of the voltage V are smaller than predetermined reference values.

The capacitor capacity inspection unit 58 and the management unit 60 may be configured with a processor such as a microcomputer or FPGA. The capacitor capacity inspection unit 58 may be configured with one processor or multiple processors. When the capacitor capacity inspection unit 58 is configured with multiple processors, the AD conversion units 4a, 4b, the digital signal reading units 5a, 5b, the analysis units 6a, 6b, the capacitor capacity calculation unit 59, and the management unit 60 are appropriately integrated and distributed and placed in each processor.

Other Embodiments
In relation to the above-mentioned embodiments and examples, the following embodiments are further disclosed.

(1) An anomaly detection device that detects an anomaly in a target system based on a physical quantity measured in the target system, comprising: a reading unit that reads quantized time series data of the physical quantity measured in the target system and converts the read quantized data into time series data that can be processed by calculation; an analysis unit that analyzes a predetermined number of the most recent time series data from the time series data and calculates a statistical quantity and an error in the statistical quantity based on the predetermined number of the most recent time series data; and a detection unit that detects an anomaly in the target system based on the statistical quantity when the degree of variation in the statistical quantity calculated by the analysis unit satisfies a predetermined condition.

In the above (1), a measuring instrument is used to measure physical quantities to evaluate the normality of a target system, the measured analog signals are converted into digital signals and digitized for processing, and time-series data including the digitized past data points and the latest data is used to calculate estimated values (statistics) such as the average and time rate of change, as well as their errors and the degree of variation in the time-series data, and anomalies in the system are detected based on the calculation results. Therefore, when the error in the calculated average, time rate of change, etc., and the degree of variation in the time-series data are small, the reliability of the estimated value is high, and an abnormality is detected when the estimated value reaches a reference value for determining an abnormality. When the error in the calculated average, time rate of change, etc., and the degree of variation in the time-series data are large, the reliability of the estimated value is low, and therefore an abnormality is not detected even if the estimated value reaches the reference value for an abnormality. This improves the accuracy and reliability of anomaly detection.

(2) The anomaly detection device according to (1) above, characterized in that the reading unit reads the quantized data at a predetermined cycle and converts the read quantized data into the time series data, and between the time when the reading unit reads the quantized data for a current cycle of the predetermined cycle and the time when the reading unit reads the quantized data for a next cycle of the predetermined cycle, the analysis unit analyzes the latest predetermined number of time series data including time series data converted from the quantized data for the current cycle so as to be capable of being processed, and completes a process of calculating the statistics for the current cycle and the error of the statistics for the current cycle based on the latest predetermined number of time series data, and the detection unit completes a process of detecting an anomaly in the target system based on the statistics for the current cycle when the statistics for the current cycle and the error of the statistics for the current cycle calculated by the analysis unit satisfy a predetermined condition.

In (2) above, calculation of estimates and errors such as the average value and time rate of change of time series data, and anomaly detection are completed between the reception of the latest digital signal and the input of the next digital signal, and anomalies are evaluated and detected in real time, always taking into account the latest data. Therefore, because estimates and errors, and the degree of variation in time series data are calculated and evaluated always taking into account the latest data, early anomaly detection is possible without delay.

(3) The anomaly detection device according to (1) above, wherein the statistic is the average of the latest predetermined number of time series data, the error of the statistic is the average or the standard deviation of the latest predetermined number of time series data, and the detection unit detects an anomaly in the target system based on the average when the standard deviation calculated by the analysis unit satisfies a predetermined condition.

In the above (3), when the average of a specified number of the most recent time series data or the standard deviation of a specified number of the most recent time series data satisfies a specified condition, the reliability of the average is high. Therefore, by detecting an anomaly in the target system based on the average, the accuracy and reliability of anomaly detection can be improved.

(4) The anomaly detection device according to (1) above, wherein the statistic is the time change rate of the latest predetermined number of time series data, the error of the statistic is the standard deviation of the time change rate, and the detection unit detects an anomaly in the target system based on the time change rate when the standard deviation calculated by the analysis unit satisfies a predetermined condition.

In the above (4), when the standard deviation of the time change rate of the latest specified number of time series data satisfies a specified condition, the reliability of the time change rate is high, so by detecting an anomaly in the target system based on the time change rate, the accuracy and reliability of anomaly detection can be improved.

(5) The anomaly detection device according to (1) above, wherein the statistics are the average and time rate of change of the latest predetermined number of time series data, the error of the statistics is the average or the standard deviation of the latest predetermined number of time series data and the standard deviation of the time rate of change, and the detection unit detects an anomaly in the target system based on the average and the time rate of change when the standard deviation calculated by the analysis unit satisfies a predetermined condition.

In (5) above, when the average of the latest specified number of time series data or the standard deviation of the latest specified number of time series data and the standard deviation of the time change rate of the latest specified number of time series data satisfy specified conditions, the reliability of the average and the time change rate is high, so by detecting an anomaly in the target system based on the average and the time change rate, the accuracy and reliability of anomaly detection can be improved.

(6) The anomaly detection device according to (4) or (5) above, wherein the analysis unit calculates the time change rate and the standard deviation of the time change rate by regression analysis of the latest predetermined number of time series data.

In (6) above, the accuracy of anomaly detection can be improved by calculating the time rate of change and the standard deviation of the time rate of change through regression analysis.

(7) The abnormality detection device according to (5) above, wherein the target system is a substation system that converts the voltage of power received from a power grid and supplies it to a feeder line via a circuit breaker, the physical quantity is a current flowing through the substation system, and the abnormality detection device has a circuit breaker control unit that opens the circuit breaker to cut off the supply of power to the feeder line when an abnormality in the substation system is detected by the detection unit.

According to (7) above, abnormal currents such as overload currents and short-circuit currents can be accurately detected and distinguished from noise in the power supply system, and the circuit breaker can be opened quickly and appropriately.

(8) The abnormality detection device described in (7) above, wherein the substation systems are present in multiple locations along the feeder line, and when an abnormality in the substation system is detected by the detection unit, the device has a fault point analysis unit that calculates the resistance and inductance of the electric path formed from the substation system through a fault point in the feeder line using the average of the current time series data, the current time series data or the standard deviation of the average, the time change rate of the current time series data, and the standard deviation of the time change rate, and the fault point is calculated based on the resistance and the inductance of the substation system in which an abnormality is detected, calculated by the fault point analysis unit, and the resistance and the inductance of a substation system adjacent to the substation system, calculated by the fault point analysis unit.

In the above (8), the fault point in the circuit beginning and ending at the substation system can be estimated based on the resistance and inductance of the substation system that are quickly and accurately calculated, and the fault point can be estimated quickly and with high accuracy.

(9) The anomaly detection device according to (1) or (2) above, wherein the physical quantities are multiple, the reading unit reads the time series quantized data for each of the multiple physical quantities and converts the read quantized data into time series data that can be processed by calculation, the analysis unit calculates the statistics and the errors of the statistics for each of the multiple physical quantities, and the detection unit detects an anomaly in the target system based on the statistics for each of the multiple physical quantities when the degree of variation of the statistics for each of the multiple physical quantities calculated by the analysis unit satisfies a predetermined condition.

In the above (9), when the degree of variation in the statistics for multiple physical quantities satisfies a predetermined condition, an anomaly in the target system is detected based on the statistics for multiple physical quantities. Therefore, the accuracy of anomaly detection can be improved and highly accurate estimates of various indicators can be calculated using the multiple physical quantities measured in the target system.

(10) The abnormality detection device according to (9) above, wherein the target system is a capacitor charging/discharging system having a capacitor that receives a DC voltage from a DC power supply unit, the physical quantities are a current flowing through the capacitor and a voltage of the capacitor, the reading unit reads the time series quantized data for each current flowing through the capacitor and each voltage of the capacitor and converts the read quantized data into time series data that can be processed, the analysis unit calculates the average of the current flowing through the capacitor and the error between the average, and the time rate of change of the voltage of the capacitor and the error between the time rate of change, and the detection unit calculates the capacitance of the capacitor based on the average and the time rate of change when the error between the average and the time rate of change calculated by the analysis unit satisfy a predetermined condition.

In the above (10), when the error in the average current flowing through the capacitor and the error in the time rate of change of the capacitor voltage satisfy predetermined conditions, the capacitance of the capacitor can be calculated based on the average current of the capacitor and the time rate of change of the capacitor voltage, thereby enabling accurate inspection of the capacitance of the capacitor.

(11) An anomaly detection system comprising the anomaly detection device according to any one of (1) to (10) above, and a conversion unit that converts a physical quantity measured in the target system into the time-series quantized data.

According to (11) above, by detecting anomalies in the target system, etc., starting from time-series quantized data obtained by converting physical quantities measured in the target system, noise specific to analog signals can be identified on a digital signal, so that anomalies in the target system can be detected with high accuracy.

(12) An anomaly detection method executed by an anomaly detection device that detects an anomaly in a target system based on a physical quantity measured in the target system, the anomaly detection method comprising: a reading step of reading quantized time series data of the physical quantity measured in the target system and converting the read quantized data into time series data that can be processed by calculation; an analysis step of analyzing a predetermined number of the most recent time series data from the time series data and calculating a statistical quantity and an error in the statistical quantity based on the predetermined number of the most recent time series data; and a detection step of detecting an anomaly in the target system based on the statistical quantity when the degree of variation in the statistical quantity calculated by the analysis step satisfies a predetermined condition.

The present invention is not limited to the above-described embodiments, and includes various modified examples. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and are not necessarily limited to those having all of the configurations described. Furthermore, it is possible to replace part of the configuration of one embodiment with the configuration of another embodiment, and to add the configuration of another embodiment to the configuration of one embodiment, so long as there is no contradiction. It is also possible to add, delete, replace, integrate, or distribute part of the configuration of each embodiment. Furthermore, the configurations and processes shown in the embodiments can be distributed, integrated, or replaced as appropriate based on processing efficiency or implementation efficiency.

1: System, 1S, 2S: DC power supply circuit, 2: Physical quantity, 3: Measuring instrument, 4, 4a, 4b: AD conversion unit, 5, 5a, 5b: Digital signal reading unit, 6, 6a, 6b: Analysis unit, 7: Detection unit, 13, 14: Substation, 15: AC circuit breaker, 16: Transformer, 17: Rectifier, 18, 19, 20, 21, 25, 26, 27, 28: DC circuit breaker, 22: AC circuit breaker, 23: Transformer, 24: Rectifier, 29: Upstream overhead line, 30: Downstream overhead line, 3 1: train, 32: rail, 35, 37: current meter, 36, 38: fault current detection circuit, 41: fault current detection circuit, 42: fault point analysis unit, 43: fault current detection circuit, 47: remote slave station, 48: DC power supply, 49: charging switch, 50: charging resistor, 51: capacitor, 52: discharging resistor, 53: discharging switch, 55: diode, 56: current meter, 57: voltage meter, 59: capacitor capacity calculation unit (detection unit), 60: management unit

Claims (11)

1. An anomaly detection device for detecting an anomaly in a target system based on a physical quantity measured in the target system, comprising:
a reading unit that reads , at a predetermined cycle, quantized data of a time series of physical quantities measured in the target system and converts the read quantized data into time series data that can be processed by calculation;
an analysis unit that analyzes a predetermined number of latest time series data among the time series data at the predetermined period and calculates an estimated value based on the predetermined number of latest time series data and an error of the estimated value or a degree of variation of the predetermined number of latest time series data ;
a detection unit that determines whether the error or the degree of variation of the estimated value calculated by the analysis unit satisfies a predetermined condition at the predetermined period , and detects an abnormality of the target system based on the estimated value when the predetermined condition is satisfied ;
From when the quantized data of the current period of the predetermined cycle is read by the reading unit until when the quantized data of the next period of the predetermined cycle is read by the reading unit,
The analysis unit includes:
discarding, by a first-in, first-out method, a certain number of the oldest time-series data from the predetermined number of time-series data used when calculating the error or the degree of variation between the estimated value and the estimated value in the previous period of the predetermined period, and calculating the error or the degree of variation between the estimated value and the estimated value in the current period based on the latest predetermined number of time-series data including the time-series data converted from the quantized data by the reading unit in the current period;
The detection unit is
determining whether or not an error or a degree of variation in the estimated value for the current cycle calculated by the analysis unit satisfies a predetermined condition, and if the predetermined condition is satisfied, detecting an abnormality in the target system based on the estimated value for the current cycle . The abnormality detection device according to claim 1 ,
the estimated value is an average of the latest predetermined number of time series data;
the degree of variation is a standard deviation of the latest predetermined number of time-series data;
The anomaly detection device, wherein the detection unit detects an anomaly in the target system based on the average when the standard deviation calculated by the analysis unit satisfies a predetermined condition. The abnormality detection device according to claim 1 ,
the estimated value is a time change rate of the latest predetermined number of time-series data;
the error of the estimate is the standard deviation of the time rate of change;
The anomaly detection device, wherein the detection unit detects an anomaly in the target system based on the time change rate when the standard deviation calculated by the analysis unit satisfies a predetermined condition. The abnormality detection device according to claim 1 ,
The estimated value is an average and a time rate of change of the latest predetermined number of time series data,
the degree of variation is a standard deviation of the latest predetermined number of time-series data;
the error of the estimate is the standard deviation of the time rate of change;
the detection unit detects an anomaly in the target system based on the average and the time rate of change when the latest predetermined number of time series data and the standard deviation of the time rate of change calculated by the analysis unit satisfy a predetermined condition. The abnormality detection device according to claim 3 or 4 ,
The anomaly detection device according to claim 1, wherein the analysis unit calculates the time change rate and a standard deviation of the time change rate by performing a regression analysis on the latest predetermined number of time-series data. The abnormality detection device according to claim 4 ,
The target system is a substation system that converts the voltage of power received from a power grid and supplies the power to a feeder line via a circuit breaker,
the physical quantity is a current flowing through the power transformation system,
an abnormality detection device comprising: a circuit breaker control unit that performs control to open the circuit breaker to cut off the supply of power to the feeder when an abnormality in the power substation system is detected by the detection unit. The abnormality detection device according to claim 6 ,
The substation system is provided in plurality along the feeder line,
a fault point analysis unit that calculates, when an abnormality in the power transformation system is detected by the detection unit, a resistance and an inductance of an electric path formed from the power transformation system through a fault point in the feeder using an average of the time series data of the current, a standard deviation of the time series data of the current or the average, a time change rate of the time series data of the current, and a standard deviation of the time change rate,
An abnormality detection device characterized in that the fault point is calculated based on the resistance and the inductance of the substation system in which an abnormality has been detected, calculated by the fault point analysis unit, and the resistance and the inductance of a substation system adjacent to the substation system, calculated by the fault point analysis unit. The abnormality detection device according to claim 1 ,
The physical quantities are plural,
The reading unit reads the time-series quantized data for each of the plurality of physical quantities, and converts the read quantized data into time-series data that can be processed by arithmetic operations;
the analysis unit calculates, for each of the plurality of physical quantities, an error between the estimated value and the estimated value or a degree of variation in the estimated value ;
the detection unit detects an abnormality in the target system based on the estimated values for the plurality of physical quantities when an error or a degree of variation in the estimated values for the plurality of physical quantities calculated by the analysis unit satisfies a predetermined condition. The abnormality detection device according to claim 8 ,
the target system is a capacitor charge/discharge system having a capacitor that receives a DC voltage from a DC power supply unit,
the physical quantities being a current flowing through the capacitor and a voltage across the capacitor,
the reading unit reads the time-series quantized data for each of the current flowing through the capacitor and the voltage of the capacitor, and converts the read quantized data into time-series data that can be processed by arithmetic operations;
the analysis unit calculates an average of the current flowing through the capacitor, the degree of variation of the latest predetermined number of time-series data related to the current , and a time rate of change of the voltage of the capacitor and an error between the time rate of change,
The abnormality detection device, characterized in that the detection unit calculates the capacitance of the capacitor based on the average and the time rate of change when the degree of variation and the error of the time rate of change calculated by the analysis unit satisfy predetermined conditions. The abnormality detection device according to any one of claims 1 to 9 ,
a conversion unit that converts a physical quantity measured in the target system into the time-series quantized data. 1. An anomaly detection method executed by an anomaly detection device that detects an anomaly in a target system based on a physical quantity measured in the target system, comprising:
a reading step of reading , at a predetermined period, quantized data of a time series of physical quantities measured in the target system and converting the read quantized data into time series data that can be processed by calculation;
an analysis step of analyzing a predetermined number of latest time series data among the time series data at the predetermined period , and calculating an estimated value based on the predetermined number of latest time series data and an error of the estimated value or a degree of variation of the predetermined number of latest time series data ;
a detection step of determining , at the predetermined period, whether the error or the degree of variation of the estimated value calculated by the analysis step satisfies a predetermined condition, and detecting an abnormality of the target system based on the estimated value when the predetermined condition is satisfied;
In the reading step, from when the quantized data of the current period of the predetermined cycle is read to when the quantized data of the next period of the predetermined cycle is read,
In the analyzing step,
discarding, by a first-in, first-out method, a certain number of the oldest time-series data from the predetermined number of time-series data used when calculating the error or the degree of variation between the estimated value and the estimated value in the previous period of the predetermined period, and calculating the error or the degree of variation between the estimated value and the estimated value in the current period based on the latest predetermined number of time-series data including the time-series data converted from the quantized data in the reading step in the current period;
In the detection step,
determining whether or not an error or a degree of variation of the estimated value for the current cycle calculated by the analysis step satisfies a predetermined condition, and if the predetermined condition is satisfied, detecting an anomaly in the target system based on the estimated value for the current cycle .

Images