TWI425226B - Method and system for fault detection, identification and location in high-voltage power transmission networks - Google Patents

Description

Method and system for fault detection, identification and location of high voltage transmission and distribution network

The invention relates to a high voltage transmission and distribution network, in particular to a method and a system for fault detection, identification and positioning of a high voltage transmission and distribution network.

The power quality of transmission lines is now seen as a major factor in enabling power plants to achieve continuous power supply to end users. Since the fail-safe system can avoid economic losses, the fail-safe system is fundamental and important for high-voltage transmission and distribution networks. The fail-safe system processes the voltage or current signal and determines if a fault has occurred. In the event of a fault, the fail-safe system determines the type of fault and the location of the fault and determines the action required to remove the fault in the power line system. A conventional transmission line system protection method for relays monitors the distortion of voltage and current signals in the time domain and the frequency domain.

The approach approach adopted by the protection system can usually be divided into three types, which are based on quantitative model-based approaches, qualitative model-based approaches and data-centric approaches. -driven approaches). Based on the quantitative mode method and the quality model based method, it has a good effect in simulation research. However, in the case where the voltage and current signals of the fault have noise, the above two methods are not effective. Therefore, measuring noise and system noise is a key factor in the performance of these two methods.

In practical applications, the data-centric approach is more flexible in implementation and can be used for fault detection and fault classification for different systems. Therefore, the data-centric approach can make the fault-protection system more resistant to noise. Robust. The data-centric approach can be, for example, artificial intelligence, so the data-centric approach is applicable to different systems with unknown levels of noise and system noise.

A failure generated on a power transmission line may cause an interruption in power supply, resulting in deterioration of two main indices of power continuous service. The two main indices are the system average interruption frequency index and the system average interruption time index. In order to reduce these two main indices, various methods have been proposed to detect faults, classify faults or locate fault locations. When a fault occurs, the fail-safe system must classify the fault type and identify the problem phase.

The characteristics of the fault type can be known by the processing of a converter or a filter such as a Fourier converter, a Kalman filter, a fractal converter or a wavelet. (wavelet) converters, etc. However, the number of types of faults is large, so traditional fail-safe systems may not be able to protect various fault types at the same time. Simply put, the traditional fault protection system may not be able to complete the detection of fault events, the determination of the fault type and the location of the fault location within a short time after the fault occurs.

Embodiments of the present invention provide a method for fault detection, identification, and location of a high voltage transmission and distribution network, and the method includes the following steps. First, a multi-level wavelet transform is performed on a voltage signal and a current signal in one cycle of the fault, and a plurality of high-frequency signals are generated accordingly. Then, Principal Component Analysis (PCA) is performed on a plurality of high frequency signals, and at least one signal characteristic value is generated accordingly, and then the signal characteristics are classified using a Support Vector Machine (SVM). Value to get the type of fault. Then, the fault location is detected using an Adaptive Structure Neural Networks (ASNN) corresponding to the fault type.

The embodiment of the invention further provides a system for fault detection, identification and location of a high voltage transmission and distribution network, wherein the system for detecting, identifying and locating the fault of the high voltage transmission and distribution network comprises at least one measuring unit and a fault diagnosis device. Waveform recorder. The measuring unit is electrically connected to the high-voltage transmission and distribution network through the bus bar connection to obtain voltage and current signals of three phases. The fault diagnosis device receives the voltage and current signals of the three phases obtained by the measurement unit, and the fault diagnosis device performs fault detection, classification and positioning. The waveform recorder is used to record the voltage and current signals of the three phases when the fault occurs.

In summary, the method, system and system for detecting, identifying, and locating a high voltage transmission and distribution network provided by the embodiments of the present invention can detect a fault, determine a fault category, and estimate a fault location. In addition, the method for detecting, identifying and locating the fault of the high-voltage transmission and distribution network and the system after the fault occurs, the time for completing the fault detection, classification and positioning is much shorter than the critical fault clearing time (critical fault clearing time). ).

The detailed description of the present invention and the accompanying drawings are to be understood by the claims The scope is subject to any restrictions.

[Embodiment of Method for Fault Detection, Identification and Location of High Voltage Transmission and Distribution Network]

Please refer to FIG. 1. FIG. 1 is a flow chart of a method for detecting, identifying, and locating a fault in a high voltage transmission and distribution network according to an embodiment of the present invention. The method of fault detection, identification and location of the high-voltage transmission and distribution network can be implemented by using a hardware circuit or by using a computer device and a software.

First, in step S1, a Negative Sequence Component (NSC) is used to detect whether the high voltage transmission and distribution network is faulty. If no fault occurs, step S1 is repeated. If a fault occurs, step S2 is performed. Then, in step S2, the voltage signal and the current signal in one cycle of the fault are multi-level wavelet converted, and a plurality of high-frequency signals are generated accordingly.

Next, in step S3, principal component analysis is performed on the high frequency signal, and at least one signal feature value is generated accordingly. Further, in step S4, the signal vector values are classified using the support vector machine to obtain the fault type. Then, in step S5, the fault location is detected using an adaptive structural neural network.

In high-voltage transmission and distribution networks, the distortion of voltage and current signals is mainly due to faults. A fault may cause a transient phenomenon and cause a three-phase imbalance between the voltage signal and the current signal. Negative sequence voltage signals and negative sequence current signals are sensitive to faults and misconnections, whether in balanced or unbalanced three-phase systems.

If the three-phase voltage and the three-phase current are expressed as V _abc and I _abc , where V _abc = ( V _a , V _b , V _c ) and I _abc = ( I _a , I _b , I _c ), and a to c respectively Represents different phases. The components of the three symmetrical voltage and current signals can be organized into V ₀₁₂ =( V ₀ , V ₁ , V ₂ ) and I ₀₁₂ =( I ₀ , I ₁ , I ₂ ), where 0, 1 and 2 represent zero respectively. Positive and negative sequence components. The negative sequence components of the voltage signal ( V _{2, a} , V _{2, b} , V _{2, c} ) and the negative sequence components of the current signal ( I _{2 , a} , I _{2. b} , I _{2, c} ) are used as faults index.

However, in an unbalanced three-phase power transmission and distribution system, the negative sequence voltage and the negative sequence current may have considerable values because of different system impedances and different load distributions. In order to identify the non-zero voltage negative sequence component V ₂ and the current negative sequence component I ₂ , the voltage negative sequence component V ₂ and the current negative sequence component I ₂ are differentiated from the time t , as in the following equation,

Then, in order to reduce the probability of false alarms, the differential voltage component of the partial differential will be V ₂ / t and current negative sequence component I ₂ / t is used for the cyclotron integral to obtain a joint fault indicator. The comprehensive fault indication is more reliable for detecting fault events, and the comprehensive fault indication can be expressed as equation (1).

Where D ( t ) is the contact fault indication measured at time t , and R (τ) can be a triangular area function R ( t )=(1-100 t )‧( H ( t )- H ( t -0.01 )), where H ( t ) is a step function.

Please refer to FIG. 2. FIG. 2 is a waveform diagram of a parameter R ( t ) of a contact failure indication according to an embodiment of the present invention. R ( t ) is used as a function of the integrated fault indication D ( t ) as a cyclotron integral. For example, it can be defined that when the integrated fault indication D ( t ) is greater than a threshold (for example, 1), it indicates that a fault has been detected. In the present embodiment, under the condition of R ( t ) shown in FIG. 2, if the voltage is negatively sequenced in a continuous time of 0.01 second. V ₂ / t and current negative sequence component I ₂ / If the sum of the whirling integrals of t is greater than 1, it can be considered that a fault has been detected.

In addition, the parameter R ( t ) of the integrated fault indication D ( t ) can be designed to adjust the sensitivity of detecting faults. It is worth mentioning that the method for detecting, identifying and locating the fault of the high-voltage transmission and distribution network of this embodiment uses the comprehensive fault indication D ( t ) to judge the occurrence of the fault, so that the frequency deviation of the voltage or current signal of the fault can be avoided. Misjudgment with amplitude variations, where frequency deviation and amplitude variation are caused by changes in operating parameters in the high voltage transmission and distribution network.

Please refer to FIG. 3A to FIG. 3E . FIG. 3A and FIG. 3B are waveform diagrams of a voltage signal and a voltage negative sequence component respectively according to an embodiment of the present invention. 3C and 3D are waveform diagrams of a current signal and a current negative sequence component, respectively, according to an embodiment of the present invention. FIG. 3E is a waveform diagram of a contact fault indication according to an embodiment of the present invention. In this embodiment, the fault occurs when the time t is about 1 second, and is triggered by a position of 0.95 pu (per unit) in the busbar from which the phasor measurement unit (PMU) is installed. Three-phase ground fault.

In FIGS. 3A and 3C, the voltage and current signals of phase a are indicated by broken lines, the voltage and current signals of phase b are indicated by solid lines, and the voltage and current signals of phase c are indicated by dotted lines. As can be seen from FIG. 3A and FIG. 3C, the fault is successfully detected when the time t is 1.00034 seconds, and the voltage negative sequence component is detected. Negative sequence component Also increase at the same time (please refer to FIG. 3B and FIG. 3D correspondingly). As can be seen from FIG. 3B and FIG. 3D, the voltage negative sequence component Negative sequence component It is unstable when the fault occurs, making the voltage negative sequence component Negative sequence component Cannot be directly used as an indication of a fault. The integrated fault indication D ( t ) can be used as an effective trigger for fault classification and fault location estimation.

The high frequency signal generated in step S2 includes first order and second The order wavelet detail signal, and may even include higher order wavelet detail signals. Wavelet transform is applied to the analysis of the unsteady and aperiodic signals in the time domain and the frequency domain, and the signal characteristics are extracted accordingly.

In this embodiment, the wavelet transform may be a multi-level continuous wavelet transform (ML-CWT), and a conversion formula of a multi-level continuous wavelet transform may be expressed as a formula (2). among them Represents the wavelet coefficients obtained by multi-level continuous wavelet transform, ψ (.) is the mother wavelet, and a and b are the real numbers representing the scaling and conversion parameters, respectively. The scaling parameter a can be used to determine the oscillation frequency and the length of the wavelet, while the conversion parameter b is used to control the position of the translation. Wavelet coefficient A measurement similarity between the proportional and translational wavelets corresponding to the voltage signal V _a ( t ) is provided.

It should be noted that multi-level continuous wavelet transform is only one of the embodiments for implementing wavelet transform, however, the present invention is not limited thereto. In practical applications, multi-level continuous wavelet transform can be implemented by using multi-level discrete wavelet transform (Multi-level DWT, ML-DWT). According to this, the multi-level continuous wavelet transform of equation (2) Can be discretized into equation (3),

The multi-level discrete wavelet transform is used to recursively decompose the three-phase voltage signal V _abc and the three-phase current signal I _abc to obtain a corresponding wavelet approximation coefficient and wavelet detailed coefficient. This process can also be called multi-analytic wavelet analysis (multi- Resolution analysis, MRA). The wavelet approximation coefficient and the wavelet detailed coefficient represent the low frequency signal and the high frequency signal of the three-phase voltage signal V _abc and the three-phase current signal I _abc , respectively.

Please refer to FIG. 4. FIG. 4 is a schematic diagram of a Mallat-type discrete wavelet decomposition process of multiple analytical wavelet analysis according to an embodiment of the present invention. The multi-level discrete wavelet transform of equation (3) can be decomposed by the Marley algorithm. In this embodiment, the first-order and second-order wavelet detailed coefficients can be expressed as ( And, in step S3, the first order and second order wavelet detail signals are taken for principal component analysis.

And fault classification accuracy selection mother wavelet [Psi] (.) Is related, in the present embodiment, Daubechies (of Daubechies) Db4 wavelet wavelet multi-stratum has been selected as discrete wavelet transform (ML-DWT) the mother wavelet. The Db4 wavelet of the Daubechies wavelet has been proven to be the most effective mother wavelet for fault analysis of transmission and distribution systems, as in the reference, "AH Osman and OP Malik, "Protection of parallel transmission lines using wavelet transform, IEEE Trans. Power Deliv ., Vol. 19, no. 1, pp. 49-55, 2004.". For example, if the sampling frequency is 3840 Hz (the frequency of the power system is 60 Hz, 64 times per cycle is sampled) to measure the voltage and current, the frequency of the first-order wavelet detailed coefficient and the second-order wavelet detailed coefficient are respectively 960- 1920Hz and 480-960Hz. In addition, through the above two first-order and second-order wavelet coefficients in the high frequency band, information on the transient phenomenon of the fault can be obtained.

Since the signal measured from the actual high-voltage transmission and distribution network is full of redundant information, and this redundant information is also present in the converted wavelet coefficients. In this embodiment, the main components of the high frequency signal are a first order wavelet detail signal and a second order wavelet detail signal. After the faulty voltage and current signals pass the multi-level discrete wavelet transform (ML-DWT), the first-order wavelet detail signal and the second-order wavelet detail signal of the voltage and current signals can be expressed as the following vectors.

Where C _a , C _b and C _c represent the high frequency characteristics of the phases a to c of the transmission line, respectively. In this embodiment, the number of samples per cycle is 64, so C _a , C _b and C _c respectively contain 64 vectors ( k =1 to 64). The average of C _a , C _b and C _c is formula (7) and formula (8),

, where x { a , b , c }, y { d 1, d 2}.

Next, C _a , C _b and C _{c are} subjected to mean-centered to obtain equation (9).

Where x { a , b , c }, A wavelet detail coefficient that represents the sum of the voltage of the phase x of the transmission line and the current signal. In order to find Each column of the largest possible projection vector e , find a set of orthogonal vectors e _x ( k ) such that the value of the following formula is the largest,

And the orthonormality limit is ( l ) e _x ( k )= δ _x ( l,k ), where x { a , b , c }, δ _x ( l, k ) is an identity matrix of size l × k . The vector e _x ( k ) and the scalar λ _x ( k ) are covariances Cov _x = The eigenvectors and eigenvalues.

In order to obtain the vector e _x ( k ) and the scalar λ _x ( k ), it is assumed that d ( k ) and μ ( k ) are the eigenvectors and eigenvalues of the co-variation, ie d ( k )= μ _k d _k . will d ( k )= μ _k d _k multiplied by both sides And become Therefore, the first M -1 eigenvector e _x ( k ) of the covariance number Cov _x and the eigenvalue λ _x ( k ) can be d ( k ) and μ _k are obtained, where x { a , b , c }.

Need to pay attention to, in order to let d ( k ) is the same as e _x ( k ), d ( k ) has to be normalized. Mutation corresponding to a total number of non-zero eigenvalues of Cov _x normalized feature vector generation substrate, a substrate formed of this subspace such that the voltage and current of the detail wavelet most signal transmission line having only a small error. The feature vector e _x ( k ) may be arranged in a decreasing manner according to the feature value λ _x ( k ) of the feature vector. The feature vector with the largest eigenvalue can be used to reflect the largest amount of variation in the wavelet detail signal, since the aligned eigenvalues are exponentially decremented such that approximately ninety percent of the change is in the front percent Within five to ten percent of the dimensions.

By calculating Ω _x =[υ _{x . 1} ,υ _{x . 2} ,...,υ _{x . M '} ], where υ _{xk '} = ( k ') ( k '), k '=1, 2,..., M ', the wavelet detail signal can be projected in the vector space of the M '(<<64) dimension. υ _{xk '} is the k'th coordinate of the wavelet detail signal in the new vector space, and υ _{xk '} is also the transmission line at phase x ( x { a , b , c }) The main components of the voltage and current signals. In short, in step S3, the signal characteristic value υ _{xk '} is obtained according to the high frequency features C _a , C _b and C _c of the phases a , b , c of the transmission line, where x { a , b , c }.

The first two dimensions of the signal characteristic value υ _{xk '} make the wavelet and detail signals of the voltage and current signals at each phase of the transmission line ( =[υ _{a ,1} ,υ _{a ,2} ], =[υ _{b . 1} ,υ _{b . 2} ] =[υ _{c .1} , υ _{c .2} ]) Projected in a feature plane. In this embodiment, the support vector machine described in step S4 is used to classify the principal component values of the first two dimensions of the feature space, and classify different types of faults accordingly. It should be noted, however, that the present invention is not limited thereto. The support vector machine may also classify the principal component values of the first two or more dimensions of the feature space, and classify different types of faults accordingly. The support vector machine is a general learning algorithm with a single output and can handle linear and nonlinear separation models by theoretical results of statistical theory. Therefore, the support vector machine (SVM) can classify newly generated faults based on different types of faults obtained by classifying the training samples.

Please refer to FIG. 5. FIG. 5 is a schematic diagram of a fault type according to an embodiment of the present invention. In FIG. 5, the resistance R _f represents a fault resistance, the resistance R _g represents a ground resistance, and the resistance HIF represents a high impedance fault resistance. The types of faults in this embodiment can be classified into the following six categories: (a) single-phase ground fault, "phase a ground", "phase b ground" and "phase c ground"; (b) double phase ground fault , "phase and b ground", "phase b and c ground" and "phase a and c ground"; (c) three-phase ground fault, "phase a , b , c ground"; (d) between two phases Faults, "phase a and phase b ", "phase b and phase c " and "phase a and phase c "; (e) three-phase fault, "fault between phases a , b , c "; and (f) High-impedance fault (HIF).

In order to correctly classify the above fault types, a set of six support vector machines (SVM#1 to SVM#6) and six support vector machines (SVM#1 to SVM#6) correspond to six adaptive structural neural networks. Adaptive Structure Neural Networks (ASNN) are integrated to classify faults and locations. The six support vector machines (SVM#1 to SVM#6) correspond to the aforementioned six failure types (a) to (f), respectively. The support vector machine generates the feature space and the decision boundary according to the training sample, and according to the position of the signal feature value corresponding to the fault in the feature space, the decision boundary is matched to classify the signal feature value. Thereby, the fault can be classified using the support vector machine. In addition, the manner in which the support vector machine generates the feature space and determines the boundary based on the training sample is as follows.

Give a set of known knowledge bases =[υ _x , y _x ], where x { a , b , c }, and y _x is the label associated with υ _x , which is the known fault category. The resolution function of the support vector machine (SVM) can be expressed as , where i =1, 2,..., N , N represents the number of hyperplanes, Is the weight vector, and b _i is the offset term. Each hyperplane divides the feature space into two parts, and the regression function f _i ( The positive and negative signs are used to indicate that the feature point is on one side of the hyperplane.

A nonlinear classifier can be used to classify the nonlinear transient phenomena of the fault signal, but at the same time a nonlinear transfer function φ is used to draw the feature points from the input space to the feature space. The resolution function can be ,among them . In this case, the decision boundary of the sorter can be written as . This transformation adds flexibility to the decision boundary so that nonlinearly determined boundaries can be achieved.

Furthermore, assume that the weight vector can use training samples ( Linear combination to represent . Then, the decision boundary can be expressed as . In the feature space (in this embodiment, the feature space is a feature plane), the decision boundary can be written as . A notation in which the variable α _x is a term is a dual representation called a decision boundary.

The value of b _i can be used to solve the regularized risk function (Eq. (10)),

Where the penalty risk function is limited to " With ξ _x In the condition of 0". ξ _x represents the slack variable, and the idling allows the feature point to be at the boundary of the hyperplane (0 ξ _x 1, known as the limit error), or the feature points are misclassified (ξ _x 1), and C is the penalty parameter.

By the method of Lagrange multipliers, the equation (11) with the representation of the variable α _x can be obtained from the equation (10).

Where equation (11) is limited to " "Under conditions.

Please refer to FIG. 6A and FIG. 6B. FIG. 6A and FIG. 6B are respectively a plan view of a phase ground fault and a high impedance fault according to an embodiment of the present invention. The broken lines in Figs. 6A and 6B are decision boundaries corresponding to the phases a , b , and c , respectively. Each training sample produces a point on Figures 6A and 6B based on the signal characteristic values of the training samples. The location indicated next to each point is the location of the fault on the transmission line estimated by the adaptive neural network. 0 pu and 1 pu represent the distance from the busbar to the point where the fault occurred. For example, 0 pu is 0 km and 1 pu is 1000 km.

Please refer to FIG. 7. FIG. 7 is a flow chart showing the sub-steps of detecting the location of a fault event using an adaptive structural neural network according to an embodiment of the present invention. Step S5 in Fig. 1 may further include the sub-steps described below. First, in sub-step S51, a construction algorithm for adaptive structural neural network neuron weights is performed to obtain an adaptive structural neural network. Then, in sub-step S52, the principal components of the fault voltage and current signals are input to the adaptive structural neural network. Next, in sub-step S53, the principal components of the fault voltage and current signals are analyzed using an adaptive structural neural network to obtain the fault location.

Please refer to the same figure and FIG. 7 and FIG. 8. FIG. 8 is a structural diagram of an adaptive structural neural network according to an embodiment of the present invention. The adaptive structural neural network (ASNN) obtained in step S51 is as shown in FIG. 8. The adaptive structural neural network includes an input layer, a hidden layer and an output layer.

The data input nodes X ₁ , X ₂ ... X _{p of the} input layer input the wavelet transformed principal component when a fault occurs. In this embodiment, the principal component is a first order wavelet detail signal and a second order wavelet detail signal. The neuron nodes N ₁ , N ₂ ... N _p in the hidden layer are used to link the data input nodes or the neuron nodes N ₁ , N ₂ ... N _n outside the node to the input node-neuron node weight W _{X (1 , 2... p ), (1, 2... n )} , neuron node-neuron node weight W _{(1, 2... n ). (1, 2... n )} and neuron node-output node weight W _{Y (1 , 2... n ), (1, 2... q )} Estimate the fault location. The above input node-neuron node weights W _{X (1, 2... p ), (1, 2... n )} and neuron node-output node weights W _{Y (1, 2... n ), (1, 2... q It} is obtained through the adaptive structure-based neural network neuron weight reconstruction algorithm, and the algorithm can make the architecture of the adaptive structure-like neural network meet the environmental requirements.

Please refer to FIG. 9. FIG. 9 is a flowchart of a constructing algorithm for the weight of the neural network of the adaptive structure-like neural network according to an embodiment of the present invention. The step of obtaining an adaptive structural neural network (step S51) includes the following sub-steps. First, in step S511, the number of neurons of the hidden layer, the weight, and a plurality of neuron connections are randomly generated. Then, in step S512, the error rate of the output is calculated. Further, in step S513, it is judged whether or not the error rate of the output is acceptable. If the error rate of the output can be received, the construction of the adaptive structure-like neural network neuron weights is completed, and the subsequent step S52 is performed. If the error rate of the output is not acceptable, step S514 is performed.

In step S514, a plurality of neuron node-output node weights are adjusted. Then, in step S515, a plurality of input node-neuron node weights are adjusted. Then, in step S516, a plurality of neuron node-neuron node weights are adjusted. Next, in step S517, at least one new neuron is established or at least one neuron is deleted. Then, in step S518, at least one of the connections between the plurality of neurons is deleted or at least one neuron connection is established. Then, step S512 is performed again, that is, the error rate of the output is calculated. Steps S512 to S518 are repeated in this manner until it is judged in step S513 that the error rate of the output can be accepted, and the construction of the adaptive structure-like neural network neuron weight is completed, and the subsequent step S52 is performed.

It should be noted that the construction of adaptive structural neural network neuron weights is the training of adaptive structural neural networks without the need to reset the adaptive structural neural network. Because power companies may not be able to afford the risk and cost of shutting down the power grid detection system for too long, such an adaptive structural neural network to detect fault locations can meet the needs of current power companies.

Referring back to FIG. 1, through the above steps S11 to S15, the fault type and the fault location can be obtained, whereby the protection decision can be made. Protection decisions can include taking action to troubleshoot to reduce the damage caused by the failure. Protection decisions may also include recording analysis results for further analysis. All in all, the power company can customize the protection decisions for each type of failure.

[Embodiment of System for Fault Detection, Identification and Location of High Voltage Transmission and Distribution Network]

Please refer to FIG. 10. FIG. 10 is a structural diagram of a system for detecting, identifying, and locating faults in a high voltage transmission and distribution network according to an embodiment of the present invention. The high-voltage power transmission and distribution network has three phases of transmission lines and bus bars for electrically connecting the loads. For convenience of description, the high-voltage transmission and distribution network of the embodiment includes three phases ( S _a , S _b , S _c ) of the transmission line and the bus bars Bus #1 to 5, and each bus bar can be electrically connected to the load. . As shown in FIG. 10, the bus bars Bus#1~4 are electrically connected to the load respectively.

The system 1 for detecting, identifying and locating faults of a high-voltage transmission and distribution network includes measurement units 1a, 1c, 1e, fault diagnosis device 10 and waveform recording Device 11. The measuring units 1a, 1c, and 1e are electrically connected to the bus bars Bus#1, Bus#3, and Bus#5, respectively. The fault diagnosis apparatus 10 receives the voltage and current signals measured by the measurement units 1a, 1c, and 1e through a Wide Area Network (WAN). The waveform recorder 11 is connected to the fault diagnostic device 10 through a wide area network (WAN).

The measuring units 1a, 1c, 1e are used to obtain voltage and current signals of three phases ( S _a , S _b , S _c ). The fault diagnosis apparatus 10 receives the voltage and current signals of the three phases obtained by the measurement units 1a, 1c, and 1e, and the fault diagnosis apparatus 10 detects, classifies, and locates the fault. The waveform recorder 11 is used to record voltage and current signals of three phases when a fault occurs.

Please refer to FIG. 11. FIG. 11 is a block diagram of a fault diagnosis apparatus according to an embodiment of the present invention. The fault diagnosis device 10 includes a signal measurement module 12, a fault diagnosis module 13, and an image display module 14. The signal measurement module 12 includes analog to digital converters (AD converters) 15-17. The fault diagnosis module 13 includes a fault detection module (not shown), a fault classification module (not shown), and a fault location module (not shown). In this embodiment, the fault diagnosis module 13 is implemented by a Field-Programmable Gate Array (FPGA), so that the functions of the fault detection module, the fault classification module, and the fault location module are integrated. A programmatic logic gate array (FPGA), but the invention is not so limited. The functions of fault detection, fault classification and fault location can also be implemented by other hardware circuits or software programs.

The input of the signal measurement module 12 receives the voltage and current signals of the three phases of the high voltage transmission and distribution network. The output end of the signal measurement module 12 is electrically connected to the input end of the fault diagnosis module 13. The output end of the fault diagnosis module 13 is electrically connected to the image display module 14.

The analog/digital converters 15-17 of the signal measurement module 12 are used for analog/digital conversion of voltage and current signals. In the present embodiment, the analog-to-digital converters (AD converters) 15 to 17 are 14-bit analog/digital converters, but the present invention is not limited thereto. In addition, in order to facilitate the processing of the 14-bit digital signal (voltage signal or current signal) output by the analog/digital converters (15 to 17), the 14-bit digital signal can be converted into a floating-point digital signal. . Therefore, the signal measurement module 12 can further include another digital signal converter for converting the digital signal into a floating point digital signal. For example: Convert this 14-bit digital signal to a 16-bit half-precision format.

The fault diagnosis module 13 receives the digitized voltage signal and the current signal for negative sequence conversion to determine whether the fault occurs, and then performs multi-level wavelet transform, principal component analysis, and uses the support vector machine to classify the fault, and finally uses the fault. The adaptive structural neural network locates the fault.

The negative sequence conversion can be performed by a combination of an adder and a multiplier to calculate a voltage negative sequence component as in the previous embodiment. V ₂ / t and current negative sequence component I ₂ / t , which in turn produces a comprehensive fault indication D ( t ). When the contact failure indication D ( t ) is greater than the threshold value, the fault is detected, whereby the fault diagnosis module 13 performs subsequent wavelet conversion. For the description of the integrated fault indication D ( t ), reference may be made to the description of the previous embodiment.

The fault diagnosis module 13 performs wavelet transform using multi-level discrete wavelet transform (ML-DWT). Multi-level discrete wavelet transform (ML-DWT) can be calculated using the Mallat algorithm. The wavelet coefficients based on the jth level of the Mallat algorithm can be expressed by the wavelet coefficients of the ( j -1) hierarchy, such as the following equation:

Where H ( z ) and G ( z ) are wavelet low-pass filter coefficients and wavelet high-pass filter coefficients. Wavelet approximation signals and wavelet detail signals for voltages in the j- level. Wavelet approximation signal and wavelet detail signal for current in the j- level. In order to increase the speed of the operation, the above operation can be implemented using a serial-in parallel-out (SIPO) architecture.

After completing the multi-level wavelet analysis, the fault diagnosis module 13 performs principal component analysis to reduce the dimension of the wavelet coefficients generated by the multi-level discrete wavelet transform (ML-DWT). The method of principal component analysis may be the method described in the previous embodiment. The most computationally intensive part of principal component analysis is the calculation of eigenvalues and ordering.

Taking the fault diagnosis module 13 as an example of a programmable logic gate array (FPGA) implementation, the implementation of principal component analysis can use a multi-stage pipeline architecture, and the use of a large number of multi-stage pipelines can reduce the computational latency. For example, a Stratix-III FPGA with 48982 logic elements is used, where the maximum operational clock of a Stratix-III FPGA can be 76 MHz, and this operational clock benefits from a large number of pipeline stages.

Then, the fault diagnosis module 13 processes the support vector machine. The support vector machine consists of two phases, a learning phase and a fault classification phase. In the learning phase, as mentioned in the previous embodiment, in the feature space, the decision boundary can be written as . The values of the variables α _x and b _i can be calculated in advance and stored in the storage unit of the fault diagnosis module 12. In the fault classification phase, the wavelet detail signal representing the first two coordinates of the principal component value ( Describe the characteristics of the fault signal, whereby the fault can be classified according to the decision boundary.

The fault diagnostic module 13 then processes the adaptive structural neural network. For the processing steps of the adaptive structural neural network, reference may be made to the description of the previous embodiment, and details are not described herein again. It should be noted that the number of neurons and the parameters of each neuron (including weights, number of connected neurons, etc.) are also limited due to limitations of hardware resources during implementation. The number of neurons, the weights, and the plurality of neuron connections can be stored in a lookup table of a programmable logic gate array (FPGA). The number of new neurons, weights, and complex neuron connections generated during the calculation can be stored in a register of a programmable logic gate array (FPGA).

The image display module 14 is configured to display the results of fault detection, identification, and positioning. The image display module 14 can also display the waveform of the fault. The image display module 14 can be a display, such as a liquid crystal screen. In addition, the image display module 14 can also be a touch screen, so that an operator or an engineer who handles a fault event can select the fault information to be viewed through touch.

[Possible efficacy of the embodiment]

According to an embodiment of the invention, the above method and system for detecting, identifying and locating a high voltage transmission and distribution network can detect a fault, determine a fault category and estimate a fault location. In addition, the method for detecting, identifying and locating the fault of the high-voltage transmission and distribution network and the system after the fault occurs, the time for detecting, identifying and locating the fault is much shorter than the critical fault clearing time.

The above description is only an embodiment of the present invention, and is not intended to limit the scope of the invention.

S1～S5, S51～S53, S511～S518. . . Step flow

1‧‧‧System for fault detection, identification and location of high voltage transmission and distribution networks

10‧‧‧Diagnostic device

11‧‧‧ Waveform recorder

1a, 1c, 1e‧‧‧ measurement unit

12‧‧‧Signal Measurement Module

13‧‧‧Troubleshooting Module

14‧‧‧Image display module

15~17‧‧‧ Analog/Digital Converter

1 is a flow chart of a method for detecting, identifying, and locating a fault in a high voltage transmission and distribution network according to an embodiment of the present invention.

2 is a waveform diagram of a parameter R ( t ) of a contact failure indication according to an embodiment of the present invention.

3A and 3B are waveform diagrams of a voltage signal and a voltage negative sequence component, respectively, according to an embodiment of the present invention.

3C and 3D are waveform diagrams of a current signal and a current negative sequence component, respectively, according to an embodiment of the present invention.

FIG. 3E is a waveform diagram of a contact fault indication according to an embodiment of the present invention.

4 is a schematic diagram of a Mallat-type discrete wavelet decomposition process of multiple analytical wavelet analysis according to an embodiment of the present invention.

FIG. 5 is a schematic diagram of a fault type according to an embodiment of the present invention.

6A and 6B are respectively a plan view of a phase ground fault and a high impedance fault according to an embodiment of the present invention.

FIG. 7 is a schematic diagram of a location of detecting a fault event using an adaptive structural neural network according to an embodiment of the present invention.

FIG. 8 is a structural diagram of an adaptive structural neural network according to an embodiment of the present invention.

FIG. 9 is a flowchart of a construction algorithm for the weight of a neural network element of an adaptive structure based on an embodiment of the present invention.

FIG. 10 is a structural diagram of a system for detecting, identifying, and locating a fault of a high voltage transmission and distribution network according to an embodiment of the present invention.

FIG. 11 is a block diagram of a fault diagnosis module according to an embodiment of the present invention.

S1 ~ S5. . . Step flow

Claims (11)

A method for detecting, identifying and locating faults in a high-voltage transmission and distribution network includes: using a negative sequence conversion to detect whether the fault occurs; and performing a multi-level voltage signal and a current signal in one cycle of a fault Wavelet conversion, and generating a plurality of high frequency signals according to the same; performing principal component analysis on the high frequency signals, and generating at least one signal characteristic value according to the same; using a support vector machine to classify the signal characteristic values to obtain a fault type And detecting a fault location using an adaptive structural neural network corresponding to one of the fault types. The method for detecting, identifying, and locating a fault of a high-voltage power transmission and distribution network as described in claim 1, wherein the fault includes single-phase grounding, dual-phase grounding, three-phase grounding, and fault between two phases, Phase failure and high-impedance fault (HIF). The method for detecting, identifying, and locating a fault of a high-voltage power transmission and distribution network according to claim 1, wherein the high-frequency signals include a plurality of first-order wavelet detail signals and a plurality of second-order wavelet detail signals . The method for detecting, identifying, and locating a fault in a high-voltage power transmission and distribution network according to claim 1, wherein the step of using the support vector machine to classify the signal feature value comprises: generating a feature space and at least one according to the training sample Determining the boundary; according to the position of the signal characteristic value corresponding to the fault in the feature space, matching the decision boundary to classify the signal feature value, thereby obtaining the fault type. The method for detecting, identifying, and locating a high-voltage power transmission and distribution network according to claim 1, wherein the step of detecting the fault location by using the adaptive structure-like neural network corresponding to the fault type includes: Performing a construction algorithm for adaptive structural neural network neuron weights to obtain an adaptive structural neural network; inputting the principal components of the fault voltage and current signals to the adaptive structural neural network; The adaptive structural neural network analyzes the principal components of the voltage and current signals of the fault to obtain the fault location. For example, the method for detecting, identifying and locating faults of a high-voltage power transmission and distribution network as described in claim 1 of the patent application, wherein a construction algorithm for adaptive structural neural network neuron weights is performed to obtain the adaptive structure class The steps of the neural network include: randomly generating a hidden layer of neurons, weights, and a plurality of neuron connections; calculating an output error rate; determining whether the output error rate is acceptable; adjusting a plurality of neurons Node-output node weight; adjusting a plurality of input nodes-neuron node weights; adjusting a plurality of neuron nodes-neuron node weights; establishing at least one new neuron or deleting at least one neuron; and deleting between the neurons At least one of the connections or establishes at least one neuron connection. A system for fault detection, identification and location of a high voltage transmission and distribution network, comprising: at least one measuring unit electrically connected to one of the high voltage transmission and distribution networks The row is used to obtain voltage and current signals of three phases; a fault diagnosis device receives the voltage and current signals of the three phases obtained by the measuring unit, and the fault diagnosis device detects, classifies and locates a fault, wherein The fault diagnosis apparatus includes: a signal measurement module that receives voltage and current signals from three phases of the measurement unit; and a fault diagnosis module that performs a negative sequence conversion to determine whether the fault occurs, and then The three-phase voltage and current signals at the time of the fault are multi-level wavelet transform, principal component analysis, and using a support vector machine, thereby classifying the fault and then estimating the neural network by adapting one of the corresponding fault types. a location of the fault; and an image display module for displaying the type of the fault and the estimated fault location; and a waveform recorder for recording the voltage and current signals of the three phases at the time of the fault. A system for fault detection, identification, and location of a high-voltage power transmission and distribution network as described in claim 7 wherein the fault includes single-phase grounding, dual-phase grounding, three-phase grounding, and faults between two phases, Phase failure and high-impedance fault (HIF). A system for detecting, identifying, and locating a fault of a high voltage transmission and distribution network as described in claim 8 wherein the signal measurement module includes three analogies corresponding to three phases of a high voltage transmission and distribution network, respectively. /digital converter. A system for detecting, identifying, and locating a fault of a high-voltage power transmission and distribution network as described in claim 8 wherein the fault classification module performs multi-level wavelet transform on voltage and current signals of three phases, and accordingly produce a plurality of high frequency signals, the fault classification module performs principal component analysis on the high frequency signals, and generates at least one signal feature value according to the fault classification module, and the fault classification module classifies the signal feature values by using a support vector machine, thereby The fault classification. The system for detecting, identifying, and locating a fault of a high-voltage power transmission and distribution network according to claim 8 of the patent scope, wherein the fault location module performs a construction algorithm for adaptive structural neural network neuron weights to obtain An adaptive structural neural network, the fault location module uses the adaptive structural neural network to analyze the principal components of the fault voltage and current signals to obtain the fault location.