CN111881176A - Anomaly detection method for marine nuclear power based on logical distance characterization of safe operation domain - Google Patents

Abstract

The invention belongs to the technical field of nuclear power system abnormality detection, and discloses a marine nuclear power abnormality detection method based on a logical distance to represent a safe operation domain. The operating condition sample library is used to select characteristic parameters that characterize the operating state; build a logical distance calculation function; conduct training learning, simulation calculation, or test and verification of collected abnormal data samples through historical operating data to determine the logical distance threshold for judging system abnormality, that is, the safety domain Attribution threshold, online detection of system anomalies or retrieval of anomalies in the historical operating data of the system. The invention constructs an abnormality detection model directly from the operating state data of the target nuclear power plant, which is close to the actual system, has stronger pertinence and applicability than other algorithms from the perspective of mechanism, and has the advantages of strong interpretability, detection strength and adjustment.

Description

Anomaly detection method for marine nuclear power based on logical distance characterization of safe operation domain

technical field

The invention belongs to the technical field of abnormal detection of nuclear power systems, and in particular relates to a method for abnormal detection of marine nuclear power based on logical distance representation of a safe operation domain.

Background technique

At present, the automatic detection technology of abnormal state of nuclear power system is the front-end entry program module of system fault diagnosis technology, and it is also the servo technology module that assists operators on duty. The existing domestic and foreign research on abnormal detection and fault diagnosis technology of nuclear power system can be roughly divided into two categories: mechanism analysis method and data-driven method.

1) Mechanism analysis anomaly detection method. The mechanism analysis method developed earlier and is more popular. It is the traditional mainstream method of abnormal detection and fault diagnosis of nuclear power system. It is divided into simulation method and expert rule set-oriented method. The simulation method refers to the construction of a full-scale simulation model system of numerical calculation that is basically the same as that of the nuclear power system based on the structural characteristics of the nuclear power plant, using conservation equations and empirical formulas in physics and thermal engineering, and performing abnormal and fault analysis on the model system. and accident demonstrations to obtain rules and experience to guide the abnormal detection and fault diagnosis of the actual system. Its application has been extended to nuclear power system design verification and nuclear power operator training, and has become the mainstream engineering design verification in the nuclear power field. However, when applied to anomaly detection and fault diagnosis, this method has the problem of being difficult to synchronize with the actual system; the expert rule set method refers to the comprehensive use of physics, thermal equations and expert experience to set accident identification procedures, formulate Accident handling procedures are used to detect abnormality of the servo system, diagnose system failures, and guide the technical methods of accident handling. It is an abnormality detection and fault diagnosis method directly acting on nuclear power plants. It is also a commonly used nuclear power plant at home and abroad. However, its anomaly detection and accident identification procedures are complex, require a large number of human intervention, it is difficult to ensure its completeness, and it is easy to induce human errors, and its diagnosis and disposal process is either fixed and rigid and cannot be changed, or it is ambiguous and difficult to understand. Therefore, in recent years, many domestic and foreign experts have put forward the research on symptom-based or state-oriented disposal procedures, which divides the judgment and disposal of accidents into multiple steps, which improves the adaptability of diagnosis and reduces the risk of misjudgment. There are still shortcomings and defects in the automatic detection of system anomalies, and the nuclear power system has a wide variety of faults and faults. It is difficult to cover all fault types through enumeration, and it is easy to miss detection.

2) Data-driven anomaly detection method. In recent years, with the continuous progress of information technology, the degree of informatization of nuclear power system has been continuously improved. A large number of nuclear power operation data have been recorded in complete and detailed, and data-driven anomaly detection methods have gradually emerged. It is favored and reflects its unique advantages, such as strong online ability, close to actual installation, and high accuracy.

The data-driven anomaly detection and fault diagnosis methods are mainly divided into two categories: one is the fault diagnosis method based on the calculation data of the nuclear power plant full-scale simulation model system. The nuclear power plant full-scale simulation model system can be used to calculate different working conditions. Different levels of typical system faults and accident data, build a nuclear power plant abnormality and accident data spectrum database, and then use machine learning methods to train an accident identification and diagnosis model for abnormality detection and fault diagnosis. Known as the "blacklist method", this method utilizes the effective combination of database technology and artificial intelligence technology to solve the difficult problem of online support for the simulation method in the mechanism analysis method. However, this method relies too much on the accuracy of the full-scale simulation model. , and there is an error between the simulation model and the actual nuclear power plant. This error has not been quantitatively calibrated during the operation of the system, and the confidence interval has not been determined. The error is likely to be enlarged under abnormal and accident conditions, or even not Therefore, there are many uncertain factors in this method, and further research is needed. This direction has attracted a large number of domestic scholars to study.

The second type of data-driven method is to directly mine the actual operating data of the nuclear power system and obtain the rules for abnormal detection and diagnosis. This case studies the automatic detection technology of system abnormality according to this idea, and proposes an abnormality detection algorithm.

Based on the analysis of the actual operation data of the nuclear power system, the system abnormality detection technology is studied. The usual research idea is to sort out the abnormality, event, and accident data that have occurred in the system, perform feature extraction, and then construct a fault identification algorithm or model according to the fault characteristics. Fault diagnosis, however, the nuclear power system is a highly reliable system, and its probability of occurrence of anomalies, events and accidents is extremely low, there are very few abnormal conditions in historical operation data, and the type distribution is scattered and random, while accident data There are almost no samples. The small number of samples, wide distribution and inconspicuous features make the research idea of automatic detection technology of system anomalies based on fault characteristics into a bottleneck.

Since most of the historical operation data is the normal operation data of common working conditions, the idea is to learn from the "whitelist" concept in the field of network security, and try to use the normal operation data of the system to construct the safe operation domain of the system to detect system anomalies. A multi-dimensional vector data space composed of operating state parameters is established. When the parameter model representing the operating state of the system exceeds the data space, the system is considered abnormal. Otherwise, the system is considered to be operating normally. detection rate.

Using normal operation data to study and construct a safe operation domain method to detect nuclear power system anomalies is a novel idea, and requires continuous and massive actual operation data of the nuclear power system as support. In the field of nuclear power, research institutions and nuclear power operation units are often independent of each other. , therefore, few relevant works have been published in the nuclear power industry.

Through the above analysis, the existing problems and defects of the existing technology are: when the existing marine nuclear power system online detection technology is used, the fault samples are sparse and incomplete.

The difficulty of solving the above problems and defects is as follows:

Different from onshore nuclear facilities, marine nuclear power system has the characteristics of complex mission profile, changeable operating conditions, small space for ship nuclear power system, relatively simple safety facilities, etc., and its ability to handle accidents is relatively weak. There is also a greater reliance on the operator's experience judgment and on-site disposal. These factors increase the operational safety risk of the marine nuclear power system, and also add additional mental pressure and workload to the operator.

The significance of solving the above problems and defects is:

Safe, intelligent and humanized operation support technical means are needed to share the pressure of operators and reduce human errors caused by fatigue. System anomaly detection technology is a key technology in the nuclear power operation support technology system. If its technology is mature and intelligent, it can more effectively assist operators on duty, reduce operator work pressure, reduce human errors, and avoid safety risks.

SUMMARY OF THE INVENTION

In view of the problems existing in the prior art, the present invention provides a marine nuclear power anomaly detection method based on the logical distance characterizing the safe operation domain.

The present invention is implemented in the following way: a method for detecting anomalies in marine nuclear power based on a logical distance representing a safe operating domain, the method for detecting anomalies in marine nuclear power based on a logical distance representing a safe operating domain includes:

Step 1: Retrieve data samples of typical operating conditions of the system by analyzing the historical operating data of the nuclear power system;

Step 2: Build a sample library of standard operating conditions, and select characteristic parameters that characterize the operating state;

Step 3: Build a logical distance calculation function between the standard operating condition sample and the system operating state data to be detected;

Step 4: Determine the logical distance threshold for judging system anomalies, that is, the security domain attribution threshold, to detect system anomalies or retrieve anomalies in the system historical operation data.

Further, in step 2, the standard operating condition system includes: steady-state operating conditions, switching conditions, and start/stop conditions;

Each operating condition is an operating mode, corresponding to an operating domain and a standard operating condition;

The steady-state operating conditions include travel 1, travel 2, travel 3... travel N working conditions;

The switching conditions include switching from traveling M to traveling L, where M≠L and M, L∈(1, 2, ┅, N;

The start/stop conditions include cold start, namely physical start, hot start, normal shutdown, and emergency shutdown.

Further, in step 2, the method for constructing the standard operating condition system further includes: selecting a single-modal standard operating condition by combining two strategies of simulation test and data analysis;

The simulation test includes the use of a full-scale simulator provided with the nuclear power system to take the operating state of the system as a standard state according to the operating parameters of typical operating conditions;

The data analysis includes calculating the logical distance between all the operating data samples under a certain operating condition in the acquired samples based on the historical operating data of the system, and finding out the one operating data sample with the smallest sum of distances from all other operating data samples. For example, as a standard operating condition sample representing the operating condition, that is, the operating mode, or calculating the logical center of the data sample set in the operating area as the standard operating condition.

Further, in step 2, the characteristic parameters that characterize the operating state include: stack right inlet temperature, stack left inlet temperature, stack right outlet temperature, stack left outlet temperature, right loop flow, left loop flow, pressure of the pressure stabilizer, Pressure regulator temperature, regulator water level, purification ion exchanger inlet temperature, secondary circuit load power, nuclear power, purified water temperature, purified water flow, 1# main steam pressure, 2# main steam pressure, 1# steam generator water level , 2# steam generator water level, 1# steam generator feed water flow, 2# steam generator feed water flow, 1# steam generator steam flow, 2# steam generator steam flow and stern shaft speed, a total of 23 characteristic parameters.

Further, in step 3, the method for constructing the logical distance calculation function includes:

The logical distance calculation function is used to quantitatively analyze the metric function of the logical relationship between different stable operating states of the system; including the logical distance calculation function of the steady-state power operating condition and the logic of the startup, shutdown and operating conditions switching transient process distance calculation function;

(1) Calculation function of logical distance for steady-state power operating conditions: The weighted Euclidean distance algorithm is used to calculate the logical distance between the steady states of the system, as follows:

During steady-state power operation, the two multidimensional vectors for calculating the logical distance are:

A=[a1,a2,a3,a4,...,a23]B=[b1,b2,b3,...,b23]

The corresponding weight vector is:

C=[c1,c2,c3,c4,...,c23]

c1+c2+,…,+c23=100

The defined weighted Euclidean distance calculation function is:

(2) Calculation function of logical distance for the transient process of startup, shutdown and switching of operating conditions:

The eigenvectors used to calculate the operating state logical distance are expressed as:

A=[a1, a2, a3, a4, ..., a23, a1', a2', a3', ..., ak'], k<23

B=[b1, b2, b3, b4,..., b23, b1', b2', b3',..., bk'], k<23

The corresponding weight vector is:

C=[c1, c2, c3, c4, ..., c23, c1', c2', c3', ..., ck'], k<23

c1+c2+…, c23=100

c1'+c2'+c3',...,+ck'=100

For the logical distance of the transient operation process, it is defined as two calculation functions of the characteristic parameter logical distance and the characteristic parameter change rate, that is, the differential logical distance, as shown below:

Another object of the present invention is to provide a marine nuclear power anomaly detection system based on a logical distance to characterize a safe operation domain by implementing the method for detecting a marine nuclear power anomaly based on a logical distance to characterize a safe operating domain. Marine nuclear power anomaly detection systems in the operational domain include:

Building modules, including operating state characterization vector building unit, nuclear power system standard operating condition system building unit, logical distance function building unit, threshold determination unit, numerical test verification unit; used for operating state characterization vector building, nuclear power system System construction of standard operating conditions, logical distance function construction unit, threshold determination and numerical test verification;

The test and experiment module is used for online detection of system anomalies or retrieval of anomalies in the historical operating data of the system.

Further, the building blocks include:

The operating state characterization vector construction unit is used to characterize the characteristic parameters of the operating state of the system, and construct the feature vector of the operating state of the system;

The nuclear power system standard operating condition system construction unit is used to construct the system standard operating condition system and determine the single-modal standard operating condition;

The logical distance function construction unit is used to respectively construct the logical distance calculation function of the steady-state power operating condition and the logical distance calculation function of the transient operation process;

The threshold determination unit is used to determine the security domain attribution threshold through the training of data samples of normal operation of the system, and verify the security domain attribution threshold through simulation calculation or inspection of the collected abnormal data samples;

Numerical test verification and verification unit is used to perform numerical test verification of abnormality detection in steady state conditions and numerical test verification of abnormality detection in changing conditions.

Another object of the present invention is to provide a computer device, the computer device includes a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the following steps :

By analyzing the historical operating data of the nuclear power system, the data samples of the typical operating conditions of the system are retrieved;

Build a sample library of standard operating conditions, and select characteristic parameters that characterize the operating state;

Build a logical distance calculation function between the standard operating condition sample and the system operating state data to be detected;

The logical distance threshold for judging system anomalies, that is, the security domain attribution threshold, is determined through training learning, simulation calculation, or the inspection and verification of abnormal data samples collected through historical operation data, to detect system anomalies or retrieve anomalies in system historical operation data.

Another object of the present invention is to provide a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, causes the processor to perform the following steps:

By analyzing the historical operating data of the nuclear power system, the data samples of the typical operating conditions of the system are retrieved;

Build a sample library of standard operating conditions, and select characteristic parameters that characterize the operating state;

Build a logical distance calculation function between the standard operating condition sample and the system operating state data to be detected;

Through historical operation data for training and learning, simulation calculation or the inspection and verification of collected abnormal data samples, the logical distance threshold for judging system abnormality, that is, the security domain attribution threshold, is used to detect system abnormality or retrieve abnormal conditions in system historical operation data.

Combined with all the above-mentioned technical solutions, the advantages and positive effects possessed by the present invention are:

Starting from the historical operation data of the nuclear power system and based on the "white list" concept in the network security field, the invention proposes and constructs a data-driven nuclear power system safe operation domain characterization algorithm for assisting the operator to detect system abnormalities online.

Based on the data-driven concept, the invention starts from the normal operation data of the system and combines the characteristics of the marine reactor to construct an anomaly detection method based on the system safe operation domain, and designs the system safe operation domain characterization function based on the logical distance calculation, which is used for The abnormality of the marine nuclear power system is monitored online. Finally, the algorithm is verified by numerical experiments with the common operating conditions of the system as the object. The results show that the designed algorithm can effectively detect system anomalies and faults, with good reliability and interpretability, and its detection strength can be adjusted.

The invention constructs an anomaly detection model directly from the operating state data of the target nuclear power plant, which is close to the actual system, is an anomaly detection method customized for the system, and has stronger pertinence and applicability than other algorithms from the perspective of mechanism.

The invention starts from normal operation data to construct a system safe operation domain to detect system abnormality, and effectively solves the bottleneck problem of high reliability of nuclear power system, few abnormality and fault data samples, and inability to extract fault features.

The present invention selects 23 parameters to characterize the operating state of the system through expert experience and numerical analysis, which can well reflect the system characteristics under different operating conditions, and is divided by numerical verification.

The present invention is based on the logical distance-based safe operating domain characterization function, and expresses the subtle differences between different operating states through explicit formulas. The detection strength of the algorithm can be adjusted linearly to increase or decrease the sensitivity to anomaly and fault detection.

The invention is verified by numerical experiments, and an abnormality detection algorithm is designed, which can detect the small abnormality of the system, the missed detection rate is extremely low, and the detection speed is fast. Obvious signs of failure can be found within 1 minute, which is more sensitive than other algorithms.

The detection algorithm of the present invention presents different characteristics in the detection and representation results for different anomalies and faults, provides an information space for further identifying the types of anomalies and faults, and is superior to the neural network data-driven anomaly detection algorithm

The algorithm of the invention can be extended and applied to the abnormal detection of the operating state of the nuclear power plant unit.

The technical effect or experimental effect of the comparison. Figures 10 and 11 of the present invention show different types of faults, and the detection results show different characteristics.

The invention selects the numerical analysis basis of 23 state characterization technical parameters.

Comparing Figures 10, 11, 9 and Figure 12 (fault diagnosis results based on neural network) in the present invention, it can be seen that the abnormal detection and fault diagnosis algorithm of neural network can only output a classification number that matches the most similar, and cannot reflect the fault. The gradual change process has poor interpretability; while the present invention can reflect the degree of failure and the similarity of different faults, as well as the gradual change process of the fault, or the occasional time, and the interpretability is strong.

Compared with traditional anomaly detection methods based on fault feature analysis, this case has a wide detection range and is not limited by the scarcity of fault samples. As shown in Figure 12, traditional anomaly detection can only detect a known classic single type of fault, or perform system anomaly detection for a single parameter, and usually needs to use a simulation simulator to calculate or supplement the fault training samples, while the simulation system is similar to There are inherent calculation errors in the actual system, which will affect the effectiveness of the diagnosis method; compared with Figures 7, 8, and 9, this case builds a safe operation domain based on the actual system, which is close to reality, and can detect various types of anomalies at the same time, and can detect unknown anomalies. The detection range is expanded and the missed detection rate is reduced. At the same time, the detection intensity can be adjusted through the threshold value according to the requirements, and the practical problem of rare and uneven distribution of fault samples can be effectively avoided.

Description of drawings

In order to explain the technical solutions of the embodiments of the present application more clearly, the following will briefly introduce the drawings that need to be used in the embodiments of the present application. Obviously, the drawings described below are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative effort.

FIG. 1 is a flowchart of a method for detecting anomalies in marine nuclear power based on a logical distance representing a safe operating domain provided by an embodiment of the present invention.

2 is a schematic structural diagram of a marine nuclear power anomaly detection system based on a logical distance to characterize a safe operating domain provided by an embodiment of the present invention;

In the figure: 1. Building module; 2. Test and experiment module; 11. Operating state representation vector building unit; 12. Nuclear power system standard operating condition system building unit; 13. Logical distance function building unit; 14. Threshold determination unit ; 15. Numerical test verification unit.

FIG. 3 is an architecture diagram of a marine nuclear power anomaly detection system based on a logical distance representation of a safe operating domain provided by an embodiment of the present invention.

FIG. 4 is a schematic diagram of a logical relationship between a safe operating domain and operating state data provided by an embodiment of the present invention.

FIG. 5 is a schematic diagram of the distribution of key parameters of a traveling sample data part according to an embodiment of the present invention.

FIG. 6 is a schematic diagram of the detection and verification of an occasional anomaly in progress provided by an embodiment of the present invention.

FIG. 7 is a schematic diagram of a traveling-destructive damage fault detection verification provided by an embodiment of the present invention.

FIG. 8 is a schematic diagram of the distribution of logical distances of state parameters in a fast-rising load condition provided by an embodiment of the present invention.

FIG. 9 is a schematic diagram of anomaly detection and verification of a system under a fast-rising load condition provided by an embodiment of the present invention.

FIG. 10 shows different types of faults provided by the embodiment of the present invention, and the detection results show different characteristic diagram 1.

FIG. 11 shows different types of faults provided by an embodiment of the present invention, and the detection results show different characteristics, as shown in FIG. 2 .

FIG. 12 is a diagram of tracking and anomaly detection of a single parameter using a traditional neural network model provided by an embodiment of the present invention.

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, but not to limit the present invention.

In view of the problems existing in the prior art, the present invention provides a method for detecting anomalies in marine nuclear power based on the logical distance characterizing the safe operation domain. The present invention is described in detail below with reference to the accompanying drawings.

As shown in FIG. 1 , the method for detecting anomalies in marine nuclear power based on a logical distance characterizing a safe operating domain provided by an embodiment of the present invention includes:

S101, by analyzing the historical operating data of the nuclear power system, retrieve data samples of typical operating conditions of the system;

S102, build a standard operating condition sample library, and select characteristic parameters that characterize the operating state;

S103, constructing a logical distance calculation function between the standard operating condition sample and the system operating state data to be detected;

S104: Determine the logical distance threshold for judging system abnormality, namely the security domain attribution threshold, by performing training learning, simulation calculation, or testing and verification of abnormal data samples collected through historical operating data, and detecting system abnormality or retrieving abnormal conditions in the historical operating data of the system.

In step S102, the standard operating condition system provided by the embodiment of the present invention includes: steady-state operating conditions, switching operating conditions, and start/stop operating conditions;

Each operating condition is an operating mode, corresponding to an operating domain and a standard operating condition;

The steady-state operating conditions include travel 1, travel 2, travel 3... travel N working conditions;

The switching conditions include switching from traveling M to traveling L, where M≠L and M, L∈(1, 2, ┅, N;

The start/stop conditions include cold start, physical start, hot start, normal shutdown, and emergency shutdown.

In step S102, the method for constructing a standard operating condition system provided by the embodiment of the present invention further includes: selecting a single-modal standard operating condition by combining two strategies of simulation test and data analysis;

The simulation test includes using a full-scale simulator matched with the nuclear power system to take the operating state of the system as a standard state according to the operating parameters of typical operating conditions;

The data analysis includes calculating the logical distance between all the operating data samples under a certain operating condition in the acquired samples based on the historical operating data of the system, and finding out the one operating data sample with the smallest sum of distances from all other operating data samples. For example, as a standard operating condition sample representing the operating condition, that is, the operating mode, or calculating the logical center of the data sample set in the operating area as the standard operating condition.

In step S102, the characteristic parameters that characterize the operating state provided by the embodiment of the present invention include: stack right inlet temperature, stack left inlet temperature, stack right outlet temperature, stack left outlet temperature, right loop flow, left loop flow, and pressure of the regulator. , pressure regulator temperature, pressure regulator water level, purification ion exchanger inlet temperature, secondary circuit load power, nuclear power, purified water temperature, purified water flow, 1# main steam pressure, 2# main steam pressure, 1# steam generation There are a total of 23 characteristic parameters including the water level of the steam generator, the water level of the 2# steam generator, the feed water flow of the 1# steam generator, the feed water flow of the 2# steam generator, the steam flow of the 1# steam generator, the steam flow of the 2# steam generator and the speed of the stern shaft. .

In step S103, the method for constructing a logical distance calculation function provided by the embodiment of the present invention includes:

The logical distance calculation function is used to quantitatively analyze the metric function of the logical relationship between different stable operating states of the system; including the logical distance calculation function of the steady-state power operating condition and the logic of the startup, shutdown and operating conditions switching transient process distance calculation function;

(1) Calculation function of logical distance for steady-state power operating conditions: The weighted Euclidean distance algorithm is used to calculate the logical distance between the steady states of the system, as follows:

During steady-state power operation, the two multidimensional vectors for calculating the logical distance are:

A=[a1,a2,a3,a4,...,a23]B=[b1,b2,b3,...,b23]

The corresponding weight vector is:

C=[c1,c2,c3,c4,...,c23]

c1+c2+,…,+c23=100

The defined weighted Euclidean distance calculation function is:

(2) Calculation function of logical distance for the transient process of startup, shutdown and switching of operating conditions:

The eigenvectors used to calculate the operating state logical distance are expressed as:

A=[a1, a2, a3, a4, ..., a23, a1', a2', a3', ..., ak'], k<23

B=[b1, b2, b3, b4,..., b23, b1', b2', b3',..., bk'], k<23

The corresponding weight vector is:

C=[c1, c2, c3, c4, ..., c23, c1', c2', c3', ..., ck'], k<23

c1+c2+…, c23=100

c1'+c2'+c3',...,+ck'=100

For the logical distance of the transient operation process, it is defined as two calculation functions of the characteristic parameter logical distance and the characteristic parameter change rate, that is, the differential logical distance, as shown below:

As shown in FIG. 2-FIG. 3, the marine nuclear power anomaly detection system based on the logical distance characterizing the safe operation domain provided by the embodiment of the present invention includes:

The building module 1 includes an operating state characterization vector building unit 11, a nuclear power system standard operating condition system building unit 12, a logical distance function building unit 13, a threshold value determining unit 14, and a numerical test verification unit 15; used for operating state Characterization vector construction, nuclear power system standard operating condition system construction, logical distance function construction unit, threshold determination and numerical test verification;

Test and experiment module 2 is used for online detection of system anomalies or retrieval of anomalies in the historical operating data of the system.

The building block 1 provided by the embodiment of the present invention includes:

The operating state characterizing vector construction unit 11 is used to characterize the characteristic parameters of the operating state of the system, and construct the characteristic vector of the operating state of the system;

The nuclear power system standard operating condition system construction unit 12 is used to construct the system standard operating condition system and determine the single-modal standard operating condition;

The logical distance function construction unit 13 is used to respectively construct the logical distance calculation function of the steady-state power operating condition and the logical distance calculation function of the transient operation process;

Threshold determination unit 14 is used to determine the security domain attribution threshold through the training of data samples of normal operation of the system, and verify the security domain attribution threshold through simulation calculation or inspection of collected abnormal data samples;

The numerical test verification and verification unit 15 is used to perform numerical test verification of abnormality detection in steady-state working conditions and numerical test verification of abnormality detection of state in changing working conditions.

The technical solutions of the present invention will be further described below in conjunction with specific embodiments.

Example:

Aiming at the problem of rare and incomplete fault samples in the online detection technology of marine nuclear power system, based on the data-driven concept, starting from the normal operation data of the system, and combining the characteristics of marine reactors, an anomaly detection algorithm based on the system's safe operation domain is designed. , and based on the logical distance calculation, the system safety operating domain characterization function is designed to monitor the abnormality of the marine nuclear power system on-line. The results show that the designed algorithm can effectively detect system anomalies and faults, with good reliability and interpretability, and its detection strength can be adjusted.

The algorithm system design mainly includes five main contents: operating state representation vector design, nuclear power system standard operating condition system design, logical distance function design, threshold determination algorithm design, and numerical test verification and verification.

The basic steps are to analyze the historical operating data of the nuclear power system, retrieve data samples of typical operating conditions of the system, build a sample library of standard operating conditions, and select characteristic parameters that characterize the operating state. Between the data, a logical distance calculation function is constructed, and the historical operating data is used for training and learning, and finally the logical distance threshold for determining the abnormality of the system is determined, which is used to detect the abnormality of the system online or retrieve the abnormality in the historical operating data of the system.

1. Construction of eigenvectors that characterize the operating state of the nuclear power system

As an energy system with security risks, the nuclear power system has high reliability requirements for its operation and maintenance. Therefore, a large number of monitoring parameters are recorded online during its operation, and all parameters are used as characteristic parameters to characterize the operating state of the system. It can retain the most feature information of the running state, but on the one hand, the large number of monitoring parameters makes the design detection algorithm a huge amount of calculation, and it is difficult to ensure the real-time performance of anomaly detection; on the other hand, the importance weights of a large number of weakly correlated parameters are not easy to determine. , which will affect the stability of the test results. Therefore, it is necessary to select the parameters so as to obtain strongly correlated parameters that characterize the operating state of the system. Taking into account the reliability redundancy design, the coupling relationship between parameters, the correlation between parameters and the system operating state, the availability and completeness of data, and then through mechanism analysis and data test screening, the following 23 parameters were selected as The characteristic parameters that characterize the operating state of the system constitute the characteristic vector of the operating state of the system.

Table 1 Characterization parameters of the operating state of the nuclear power system

2. Construction of system standard operating conditions

1) Construction of standard operating condition system

Due to the mobility requirements of ships, the marine nuclear power system needs to sail at different speeds, so it is necessary to design multiple steady-state operating conditions; at the same time, it will often switch between various operating conditions, adding many normal conversion conditions; Its working principle and design characteristics, its start-up and shutdown process lasts a long time, it is an abnormal and fault-prone operating condition, and it is quite different from power operation, and it also needs to be listed as a special operating condition. The above actual situation As a result, the standard operating conditions of the system are multi-modal, so it is necessary to build a system of standard operating conditions to support the algorithm design. The operating conditions of the marine nuclear power system are shown in the following table, and each operating condition is an operating model. state, corresponding to an operating domain and a standard operating condition.

Table 2 Nuclear power system operating condition system

2) Determination strategy for single-modal standard operating conditions

The nuclear power system is a dynamic balance system composed of several physical processes of nuclear physics, heat transfer, thermal hydraulics, and mechanical motion coupled with each other. Even under the same operating condition, the state characteristic parameters cannot be the same, but the logical distance between the operating states of the nuclear power system under a single operating condition is relatively close, which usually constitutes an operating area, and the ideal standard operating condition selected should be at the logical center of this operating area. In the selection of single-modal standard operating conditions, a combination of two strategies is mainly adopted. The first method is the simulation test, which uses the full-scale simulator of the nuclear power system to take the operating state of the system according to the operating parameters of typical operating conditions as the standard state. ; The second method is data analysis, which is based on the historical operating data of the system, calculates the logical distance between all operating data samples under a certain operating condition in the sample, and finds out the smallest sum of distances from all other operating data samples. An instance of an operating data sample of , as a standard operating condition sample characterizing the operating condition (operating mode), or it can directly calculate the logical center of the data sample set in the operating area as a standard operating condition, and check the system operating condition The schematic diagram of the relationship between the running domain and the running state data samples is shown in Figure 4.

3. Design of characterization function based on logical distance

In the above process, the data analysis method is used to determine the standard operating conditions, and the logical distance between different operating data samples needs to be calculated. When determining the boundary of the safe operating domain or identifying whether the sample is safe in the following steps, the current sample or support vector sample needs to be calculated. From the logical distance between the standard operating conditions, it can be seen that the design of the logical distance calculation function is the key link in the algorithm design. It is necessary to design a logical distance calculation function with good problem applicability and robustness to realize the anomaly detection algorithm.

1) The logical distance calculation function of the steady-state power operating condition. The designed logical distance calculation function is a measure function used to quantitatively analyze the logical relationship between different stable operating states of the system. The physical quantity characterizing the running state is a multi-dimensional vector composed of 23 selected state characteristic parameters, and a calculation function to calculate the logical distance between two 23-dimensional vectors needs to be constructed. Because the importance of 23 characteristic parameters is not the same when identifying the operating state of the system, each parameter needs to be weighted during calculation, and the selection of weights needs to be comprehensively determined based on operating experience and numerical experiments. Based on the above analysis, with Euclidean distance as the original design, a weighted Euclidean distance algorithm is defined to calculate the logical distance between system steady states, as follows.

When the steady-state power is running, the two multi-dimensional vectors that need to calculate the logical distance are as follows:

A=[a1,a2,a3,a4,...,a23]B=[b1,b2,b3,...,b23]

The corresponding weight vector is:

C=[c1,c2,c3,c4,...,c23]

c1+c2+,…,+c23=100

The defined weighted Euclidean distance calculation function is:

2) For the transient process of startup, shutdown and operating condition switching, because the characteristic parameters are in a one-way change, the operating state is a process-oriented space-time region, and it is difficult to reflect the system’s behavior simply by the characteristic values of the characteristic parameters at a single moment. Dynamic change process, therefore, to calculate the logical distance under this kind of working conditions, not only its characterization value, but also the difference value of its time series should be considered. First, it is not necessary to introduce the difference values of all the feature parameters into the feature vector, but only some key parameters can be selected to extract the difference values, which need to be analyzed according to the actual problem.

Therefore, during transient operation, the eigenvector used to calculate the logical distance of the operating state is expressed as:

A=[a1, a2, a3, a4, ..., a23, a1', a2', a3', ..., ak'], k<23

B=[b1, b2, b3, b4,..., b23, b1', b2', b3',..., bk'], k<23

The corresponding weight vector is:

C=[c1, c2, c3, c4, ..., c23, c1', c2', c3', ..., ck'], k<23

c1+c2+…, c23=100

c1'+c2'+c3',...,+ck'=100

Regarding the logical distance of the transient operation process, it needs to be defined as two calculation functions: the logical distance of the characteristic parameter and the logical distance of the rate of change of the characteristic parameter (difference), as shown below:

k<23

4. Determining the algorithm system for the characterization of the safe operation domain and the attribution threshold

For the safe operating domain characterization algorithm of a single operating condition of the system, it includes three elements: standard operating conditions and their eigenvectors, a logical distance calculation function and a safe domain belonging threshold. The safe domain belonging threshold is after the design of the first two elements is completed. , which is determined through the training of data samples of normal operation of the system, and needs to be verified through simulation calculation or the inspection and verification of abnormal data samples collected before it can be finally determined.

Corresponding to the nuclear power system standard operating condition system, the nuclear power system safe operating domain characterization algorithm is also an algorithm system composed of different modes, as shown in the following table.

Table 3 Standard operating condition system and logical distance calculation function

5. Algorithm numerical test verification

In order to verify the effectiveness of the algorithm, the historical operating data of a marine nuclear power plant was selected as the experimental object, and numerical tests were carried out for the commonly used traveling-steady-state operating conditions and the fast-rising load conditions with fast operating transient rhythm.

(1) Numerical test verification of abnormal detection under steady state conditions

1) Numerical test setting

a. Data sample structure

First, use the actual system operation data to test the design algorithm, mainly to verify the algorithm's ability to detect random and occasional deviations, disturbances, and anomalies under steady-state conditions, and to test the training samples and test samples of the anomaly detection algorithm under steady-state conditions. The travel-state data of the object nuclear power plant in different burnup stages are extracted to construct, the sampling window width of each stage is 5000 seconds, a total of 10 stages of data samples are selected, and the time-domain curves of some key characteristic parameters of the selected data samples are As shown in Figure 5.

From the above characteristic parameter distribution, it can be seen that the operating state parameters of a specific working condition exhibit obvious regional aggregation phenomenon, and a typical safe operating area can be constructed.

Secondly, for the system offset and fault caused by equipment damage, because the actual system cannot obtain complete relevant test samples, the test samples are obtained through the simulation machine corresponding to the actual system, and the set equipment damage abnormality and fault test sample types Including micro-breaks in the heat transfer tubes of the steam generator, micro-breaks at the cold end of the left loop of the reactor, stuck open safety valves, breaks in the secondary-circuit steam pipes, and uncontrolled extraction of a single control rod, covering the reactor core, primary and secondary loops , main and auxiliary systems and weak links.

b. Determination of standard operating conditions and allocation of parameter weights

Take the average value of all sample data values of each characteristic parameter that characterizes the operational state of the system as the standard value, and use it as the component to obtain the characteristic vector of the standard operating condition of march one; , determine the distribution weights of each component parameter when calculating the logical distance, as shown in the following table, because the design algorithm mainly detects the abnormality and failure of the nuclear power primary circuit system, so the primary circuit and the parameters directly related to the primary circuit have higher weights.

Table 4. Standard operating conditions and parameter weight assignment

2) Analysis of numerical test results

a. The verification results of the algorithm's ability to detect random and occasional deviations, disturbances, and anomalies under steady-state conditions using actual system data are shown in Figure 6 below.

It can be seen from the figure that during normal operation, the logical distance between the actual system operating state and the standard operating condition is uniformly distributed in the time domain, and presents a typical band-like aggregation phenomenon. After an occasional abnormality or failure occurs, its value shows a step increase, and the abnormal symptoms are obvious. As long as an appropriate threshold is selected according to the actual demand, an online abnormality detection algorithm with strong sensitivity and linear adjustment can be obtained. Through manual inspection of the detected abnormal detection samples, the true existence of the abnormality or deviation is confirmed; the manual inspection of the remaining normal detection samples with no abnormality detected, the results are all normal, as shown in Figure 7. . Therefore, the method has excellent performance both from the perspective of false detection rate and missed detection rate.

b. Using the destructive damage data samples obtained by the full-scale simulation computer calculation, the verification effect of the anomaly detection algorithm is shown in Figure 7 below. From the data analysis of full-scale simulation and calculation operation data, and calculating the logical distance between it and the standard operating conditions, it is not difficult to find that its value presents typical periodic high-frequency interference in the time domain, and its stability is lower than the actual operation data. Therefore, from the perspective of stability, it is reasonable to believe that the application effect of the anomaly detection method based on the safe operation domain for the actual system is better than that of the simulator.

In order to verify the detection ability of the algorithm for destructive damage faults, five types of destructive fault detection numerical tests are set up. From the analysis of the test results, the designed safe operation domain algorithm and the defined logical distance characterization technology can be obtained for the five types of faults. Obvious signs of failure.

The detection algorithm is aimed at different types of faults, and there are differences in the fault characteristics represented by the logical distance. For example, the pressure boundary of the primary circuit is damaged. The logical distance after high-frequency filtering gradually deviates from the normal working condition and increases linearly. , its time domain value is presented as a straight line, and the slope of the straight line can be used to characterize the degree of failure (crack size), that is, its slope can be used to define the abnormal detection threshold; when a reactive accident occurs, the logical distance shows a step-by-step expansion Phenomenon; when the secondary circuit steam pipeline breaks, the system heat sink is abnormal, the logical distance will deviate from the normal area as a whole, and the deviation will gradually increase with the loss of the secondary circuit working medium. The latter two can be directly detected in the time domain. threshold.

The test results also show that when the algorithm detects destructive faults, different types of faults will show different characterization phenomena in the logical distance, which can be used as the basis for the identification and diagnosis of fault types in the next link of abnormal detection. In the follow-up research Targeted in-depth analysis will be carried out in the work.

(2) Numerical test verification of abnormal state detection in changing conditions

1) Numerical test setting

For the verification of the algorithm under variable working conditions, the system selects the rapid load-rising process with drastic parameter changes as the object working condition, and the state parameters and their distribution weights still select the same scheme as the steady-state working condition. When calculating the logical distance of the state parameter, the selected standard operating condition is the standard operating condition corresponding to the initial operating condition before the power increase, that is, a standard operating condition; when calculating the logical distance of the parameter change rate, set the calculation parameter change The time interval of the rate (difference) is 1 second, and the selected standard condition component is the median of all sample difference values of the corresponding feature parameters in the training sample, as shown in the following formula.

a'i criterion=(MAX(b'i)+Min(b'i))/2,i∈(1,2,...,23)

2) Analysis of numerical test results

From the historical operation database of the actual system, the typical data samples of the running-to-fast-rising load are extracted as the training samples. First, according to the designed state parameter logical distance calculation algorithm, the logical distance dAB between the training sample and the running-standard operating condition is calculated, The logical distance of the state parameter characterization value of each training sample is obtained as shown in Figure 8. The step change at (300 to 400)*0.125 seconds in the figure is caused by the main coolant pump being switched from low speed to high speed. It can be seen from the figure that the rapid load increase condition, with the continuous increase of the system load, the logical distance from the standard operating condition continues to increase, and it has obvious regional aggregation performance, which can form an effective safe operation domain, indicating that based on The anomaly detection algorithm in the safe operation domain is feasible, but the system anomaly is determined directly by a parameter of the state parameter characterization value logical distance, and the information is not sufficient, and some auxiliary parameters are needed to reach the conclusion.

The operating state of the system during the rapid load escalation process is not only related to the characteristic values of the characteristic parameters, but more information actually exists in the rate of change of the characteristic parameters, so it may be easier to retrieve the fault data from the differential data of the characteristic parameters, as shown in Figure 9. The distribution of the logical distance d'AB between the characteristic parameter score value and the standard score value, and the fault state characteristics, it can be seen from the figure that in the differential logical distance space, as long as the appropriate threshold is set, it can be easily retrieved A system exception occurred. The verification shows that by selecting appropriate characterization parameters, the design algorithm can effectively detect the abnormality and faults that occur under the transient operating conditions of the system.

Figures 10 and 11 of the present invention show different types of faults, and the detection results show different characteristics.

The invention selects the numerical analysis basis of 23 state characterization technical parameters.

Comparing Figures 10, 11, 9 and Figure 12 (fault diagnosis results based on neural network) in the present invention, it can be seen that the abnormal detection and fault diagnosis algorithm of neural network can only output a classification number that matches the most similar, and cannot reflect the fault. The gradual change process has poor interpretability; while the present invention can reflect the degree of failure and the similarity of different faults, as well as the gradual change process of the fault, or the occasional time, and the interpretability is strong.

Compared with traditional anomaly detection methods based on fault feature analysis, this case has a wide detection range and is not limited by the scarcity of fault samples. As shown in Figure 12, traditional anomaly detection can only detect a known classic single type of fault, or perform system anomaly detection for a single parameter, and usually needs to use a simulation simulator to calculate or supplement the fault training samples, while the simulation system is similar to There are inherent calculation errors in the actual system, which will affect the effectiveness of the diagnosis method; compared with Figures 7, 8, and 9, this case builds a safe operation domain based on the actual system, which is close to reality, and can detect various types of anomalies at the same time, and can detect unknown anomalies. The detection range is expanded and the missed detection rate is reduced. At the same time, the detection intensity can be adjusted through the threshold value according to the requirements, and the practical problem of rare and uneven distribution of fault samples can be effectively avoided.

In the description of the present invention, unless otherwise stated, "plurality" means two or more; the terms "upper", "lower", "left", "right", "inner", "outer" The orientation or positional relationship indicated by , "front end", "rear end", "head", "tail", etc. are based on the orientation or positional relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, not An indication or implication that the referred device or element must have a particular orientation, be constructed and operate in a particular orientation, is not to be construed as a limitation of the invention. Furthermore, the terms "first," "second," "third," etc. are used for descriptive purposes only and should not be construed to indicate or imply relative importance.

The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited to this. Any person skilled in the art is within the technical scope disclosed by the present invention, and all within the spirit and principle of the present invention Any modifications, equivalent replacements and improvements made within the scope of the present invention should be included within the protection scope of the present invention.

Claims (9)

1. a marine nuclear power anomaly detection method based on a logical distance representation of a safe operation domain, is characterized in that, the marine nuclear power anomaly detection method based on a logical distance representation of a safe operation domain comprises: By analyzing the historical operating data of the nuclear power system, the data samples of the typical operating conditions of the system are retrieved; Build a sample library of standard operating conditions, and select characteristic parameters that characterize the operating state; Build a logical distance calculation function between the standard operating condition sample and the system operating state data to be detected; The logical distance threshold for judging system anomalies, that is, the security domain attribution threshold, is determined through training learning, simulation calculation, or the inspection and verification of abnormal data samples collected through historical operation data, to detect system anomalies or retrieve anomalies in system historical operation data. 2. The marine nuclear power anomaly detection method based on the logical distance characterizing the safe operating domain according to claim 1, wherein the standard operating condition system comprises: steady-state operating Shutdown conditions; Each operating condition is an operating mode, corresponding to an operating domain and a standard operating condition; The steady-state operating conditions include travel 1, travel 2, travel 3... travel N working conditions; The switching conditions include switching from traveling M to traveling L, where M≠L and M, L∈(1, 2, ┅, N; The start/stop conditions include cold start, namely physical start, hot start, normal shutdown, and emergency shutdown. 3. The marine nuclear power anomaly detection method based on the logical distance characterizing the safe operating domain according to claim 1, wherein the standard operating condition system construction method further comprises: adopting a combination of two strategies of simulation test and data analysis method to select single-modal standard operating conditions; The simulation test includes the use of a full-scale simulator provided with the nuclear power system to take the operating state of the system as a standard state according to the operating parameters of typical operating conditions; The data analysis includes calculating the logical distance between all the operating data samples under a certain operating condition in the acquired samples based on the historical operating data of the system, and finding out the one operating data sample with the smallest sum of distances from all other operating data samples. For example, as a standard operating condition sample representing the operating condition, that is, the operating mode, or calculating the logical center of the data sample set in the operating area as the standard operating condition. 4. The method for detecting anomalies in marine nuclear power based on logical distance characterizing a safe operating domain according to claim 1, wherein the characteristic parameters characterizing the operating state comprise: stack right inlet temperature, stack left inlet temperature, stack right outlet temperature Temperature, stack left outlet temperature, right loop flow, left loop flow, regulator pressure, regulator temperature, regulator water level, purification ion exchanger inlet temperature, secondary loop load power, nuclear power, purified water temperature, purification Water flow, 1# main steam pressure, 2# main steam pressure, 1# steam generator water level, 2# steam generator water level, 1# steam generator feed water flow, 2# steam generator feed water flow, 1# steam generator There are a total of 23 characteristic parameters including steam flow, steam flow of 2# steam generator and stern shaft speed. 5. The marine nuclear power anomaly detection method based on the logical distance characterizing the safe operation domain as claimed in claim 1, wherein the logical distance calculation function construction method comprises: The logical distance calculation function is used to quantitatively analyze the metric function of the logical relationship between different stable operating states of the system; including the logical distance calculation function of the steady-state power operating condition and the logic of the startup, shutdown and operating conditions switching transient process distance calculation function; (1) Calculation function of logical distance for steady-state power operating conditions: The weighted Euclidean distance algorithm is used to calculate the logical distance between the steady states of the system, as follows: During steady-state power operation, the two multidimensional vectors for calculating the logical distance are: A = [a1, a2, a3, a4, ..., a23] B = [b1, b2, b3, ..., b23] The corresponding weight vector is: C=[c1,c2,c3,c4,...,c23] c1+c2+,…,+c23=100 The defined weighted Euclidean distance calculation function is:

(2) Calculation function of logical distance for the transient process of startup, shutdown and switching of operating conditions: The eigenvectors used to calculate the operating state logical distance are expressed as: A=[a1, a2, a3, a4, ..., a23, a1', a2', a3', ..., ak'], k<23 B=[b1, b2, b3, b4,..., b23, b1', b2', b3',..., bk'], k<23 The corresponding weight vector is: C=[c1, c2, c3, c4, ..., c23, c1', c2', c3', ..., ck'], k<23 c1+c2+…, c23=100 c1'+c2'+c3',...,+ck'=100 For the logical distance of the transient operation process, it is defined as two calculation functions of the characteristic parameter logical distance and the characteristic parameter change rate, that is, the differential logical distance, as shown below:

6. A marine nuclear power anomaly detection system based on a logical distance to characterize a safe operating domain, which implements the method for detecting marine nuclear power anomalies based on a logical distance to characterize a safe operating domain according to claims 1-5, wherein the The marine nuclear power anomaly detection system whose logical distance characterizes the safe operating domain includes: Building modules, including operating state characterization vector building unit, nuclear power system standard operating condition system building unit, logical distance function building unit, threshold determination unit, numerical test verification unit; used for operating state characterization vector building, nuclear power system System construction of standard operating conditions, logical distance function construction unit, threshold determination and numerical test verification; The test and experiment module is used for online detection of system anomalies or retrieval of anomalies in the historical operating data of the system. 7. The marine nuclear power anomaly detection system based on the logical distance representation of the safe operating domain as claimed in claim 6, wherein the building module comprises: The operating state characterization vector construction unit is used to characterize the characteristic parameters of the operating state of the system, and construct the feature vector of the operating state of the system; The nuclear power system standard operating condition system construction unit is used to construct the system standard operating condition system and determine the single-modal standard operating condition; The logical distance function construction unit is used to respectively construct the logical distance calculation function of the steady-state power operating condition and the logical distance calculation function of the transient operation process; The threshold determination unit is used to determine the security domain attribution threshold through the training of data samples of the normal operation of the system, and verify the security domain attribution threshold through simulation calculation or inspection of the collected abnormal data samples; Numerical test verification and verification unit is used to perform numerical test verification of abnormality detection in steady state conditions and numerical test verification of abnormality detection in changing conditions. 8. A computer device, characterized in that the computer device comprises a memory and a processor, the memory stores a computer program, and when the computer program is executed by the processor, the processor is caused to perform the following steps: By analyzing the historical operating data of the nuclear power system, the data samples of the typical operating conditions of the system are retrieved; Build a sample library of standard operating conditions, and select characteristic parameters that characterize the operating state; Build a logical distance calculation function between the standard operating condition sample and the system operating state data to be detected; The logical distance threshold for judging system anomalies, that is, the security domain attribution threshold, is determined through training learning, simulation calculation, or the inspection and verification of abnormal data samples collected through historical operation data, to detect system anomalies or retrieve anomalies in system historical operation data. 9. A computer-readable storage medium storing a computer program, when the computer program is executed by a processor, the processor is caused to perform the following steps: By analyzing the historical operating data of the nuclear power system, the data samples of the typical operating conditions of the system are retrieved; Build a sample library of standard operating conditions, and select characteristic parameters that characterize the operating state; Build a logical distance calculation function between the standard operating condition sample and the system operating state data to be detected; The logical distance threshold for judging system anomalies, that is, the security domain attribution threshold, is determined through training learning, simulation calculation, or the inspection and verification of abnormal data samples collected through historical operation data, to detect system anomalies or retrieve anomalies in system historical operation data.

Images