A Fault Prediction Method for Power Grid Dispatching Control System

Abstract

The present invention relates to the field of power systems, and in particular to a method for predicting faults in a power grid dispatching and control system. A method based on event chain combination is used to construct a knowledge graph with reasoning function for known fault events in the dispatching and control system. A digital algorithm based on fuzzy clustering is used to vectorize the acquired alarm signal. The similarity of discrete variables and continuous variables between nodes in the knowledge graph after vectorization and real-time alarm signals is calculated to determine similar event chains, and the terminal node of the search event chain is used as the fault point to realize fault prediction. The present invention can timely detect faults before they occur, effectively reduce or avoid the occurrence of faults, and ensure the safe operation of the dispatching and control system.

Description

Technical Field

[0001] The present invention relates to the field of electric power systems, and in particular to a fault prediction method for a power grid dispatching and control system. Background Art

[0002] As an important technical means for power grid enterprises to control power grid operation and dispatch production management, the power grid dispatching control system is a key infrastructure to ensure the safe operation of the power grid. The dispatching control system integrates applications such as secondary equipment monitoring, online safety analysis, network security management, and load regulation. In recent years, with the expansion of the scale of power grids and the increasing number of interconnected objects, the architecture of the dispatching control system has also developed from traditional centralized interconnection to distributed unified interconnection. The complex structure of the dispatching control system consists of software and hardware equipment and model data distributed in the main station, channel and substation. Once part of the system fails, it will seriously affect the operation monitoring and accident handling of the power grid, and supervision and operation safety will face severe challenges. Therefore, the detection of dispatching control system faults is crucial to the operation of the power system.

[0003] At present, there are three methods for fault tracking in dispatching control systems that have been studied in the literature and applied to practical projects: case-based methods, code matching methods, and rule-based methods. The information processed by case-based or code-based fault location algorithms usually faces the dimensionality curse due to various warning signals. The increase in the number of components will bring a large amount of data encoding, making it very time-consuming to search the code book. Summary of the invention

[0004] The purpose of the present invention is to provide a fault prediction method for a power grid dispatching control system, which can timely discover faults before they occur, effectively reduce or avoid the occurrence of faults, and ensure the safe operation of the dispatching control system.

[0005] In order to solve the above technical problems, the technical solution of the present invention is: a fault prediction method for a power grid dispatching control system, comprising the following steps:

[0006] Step 1: Construct an event chain based on known historical fault alarm events in the dispatching control system, and combine the event chains to form a knowledge graph network;

[0007] Step 2: Process the real-time alarm signal in the initial process of the power grid dispatching control system fault, vectorize the real-time alarm signal containing the object, fault type, and related applications, and obtain a discrete-continuous vector as a variable for subsequent similarity calculation;

[0008] Step 3: Vectorize the alarm signal of the node object in the knowledge graph network according to the vectorization method in step 2;

[0009] Step 4: Calculate the similarity of discrete variables and continuous variables between nodes in the vectorized knowledge graph network and real-time event alarm signals. The event chain consists of multiple events. Calculate the event chain similarity based on the event similarity, and determine the similar event chain in the knowledge graph network based on the maximum similarity.

[0010] Step 5: Search the knowledge graph network to find the terminal node of the similar event chain, which is the fault point, and complete the fault prediction.

[0011] Preferably, in step 1 and step 2, before quantizing the alarm signal, the collected alarm signal is first formatted and the information is standardized according to the component type-event-state mode.

[0012] Preferably, in step 1, the control system is divided into hardware, software and network types, and the alarm signal originates from the control system; in step 2, a digital algorithm based on fuzzy clustering is used to simplify the alarm signal into a formatted vector.

[0013] Preferably, in step 4, a similarity combination measurement algorithm is used to calculate the similarity between events. The similarity measurement of a single event is to calculate the discrete variables and the continuous variables respectively, and finally combine the calculation results of the discrete variables and the continuous variables;

[0014] The calculation of discrete variables in the event vector is as follows:

[0015]

[0016] Where: D i and D j denote discrete variables in events i and j, respectively, Represents the similarity of the discrete variables of events i and j. If the discrete variables have the same value, then The value of is 1, otherwise it is 0;

[0017] The Euclidean distance calculation is used for continuous variables between events as follows:

[0018]

[0019] Where: X i,k and X j,k are the kth value of the continuous variable in events i and j respectively, n is the length of the continuous variable for events i and j, The similarity of the continuous variable representing events i and j;

[0020] The calculation for combining discrete and continuous variables is as follows:

[0021]

[0022] Where: represents the similarity of the discrete variables of events i and j, as shown in formula (1), w represents the weight factor of the discrete variable similarity, The similarity of the continuous variable representing events i and j is shown in formula (2), S ij It represents the similarity between events i and j, as shown in formula (3).

[0023] Preferably, the event chain similarity is obtained by averaging the sum of the similarities of multiple events; after determining the event chain with the largest similarity value, the terminal node of the event chain with the largest similarity value in the knowledge graph network is the fault to be predicted.

[0024] Compared with the prior art, the present invention has the following beneficial effects:

[0025] The present invention is a fault prediction method for a power grid dispatching and control system based on event similarity and knowledge graph reasoning. A large number of real-time alarm signals will be obtained during the initial process of a fault in a power grid dispatching and control system. The present invention uses these alarm signals to match known fault event chains for fault prediction. The present invention constructs a knowledge graph network based on historical fault information, effectively simplifies the alarm signal into a vector that can be calculated by a computer, proposes a similarity calculation method for a combination of discrete-continuous variables, calculates the similarity of discrete variables and continuous variables between nodes in the knowledge graph after vectorization and real-time alarm signals, determines similar event chains, and searches for fault nodes in the knowledge graph based on similar event chains to achieve fault prediction;

[0026] The present invention uses the early warning signals of fault development to predict subsequent faults before the fault occurs, helping the maintenance personnel of the dispatching control system to prevent the occurrence of faults, and guiding the power grid dispatching control operation staff to use early warning signals to predict subsequent faults, timely discover faults and take corresponding countermeasures, thereby reducing or avoiding the occurrence of faults to a certain extent. Compared with traditional fault detection methods, the method of the present invention can timely discover faults before they occur, ensuring the safe operation of the dispatching control system. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a flow chart of the method of the present invention;

[0028] Figure 2 A distribution diagram of the number of fault event chains in an embodiment of the present invention;

[0029] Figure 3 A schematic diagram of the similarity between hardware, software and network single events in an embodiment of the present invention;

[0030] Figure 4 Schematic diagram of the relationship between event chain length and prediction accuracy in an embodiment of the present invention. DETAILED DESCRIPTION

[0031] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0032] Please refer to Figure 1 The present invention provides a fault prediction method for a power grid dispatching control system, the steps comprising:

[0033] Step 1: According to the importance of system applications in the power grid dispatching and control system, the system is divided into three parts, namely hardware (H), software (S), and network (N). The event chain consists of these three parts and eventually causes a certain fault; the event chain is constructed based on the known historical fault alarm events in the dispatching and control system, and the event chains are combined to form a knowledge graph network with reasoning function;

[0034] Due to the distributed design of the components in the power grid dispatching and control system, and the collected alarm information is also multi-source data, in order to facilitate vectorization and calculation in the following steps, the collected information is first formatted and standardized according to the component type-event-state mode;

[0035] Step 2: A large number of real-time alarm signals will be obtained during the initial process of the power grid dispatching and control system fault. First, the real-time alarm signals in the initial process of the power grid dispatching and control system fault are processed, and the real-time alarm signals containing objects, fault types, and related applications are vectorized in a unified format to obtain a discrete-continuous vector as a variable for subsequent similarity calculation; after the system is divided into three parts: hardware (H), software (S), and network (N), all alarm signals come from here. A digital algorithm based on fuzzy clustering is used to simplify the alarm signal into a formatted vector. The important considerations in hardware alarms are the location and the importance of the warning, the software alarms consider the scope and impact of the warning, and the network alarms are the links and layers to which they are associated. The common information is the correlation and the node. According to this requirement, the simplified information is shown in Table 1:

[0036] Table 1: Formatted alarm information D is a discrete variable, C is a continuous variable

[0037]

[0038] Step 3: Vectorize the alarm signal of the node object in the knowledge graph network according to the vectorization method in step 2;

[0039] Step 4: Calculate the similarity of discrete variables and continuous variables between the known real-time event alarm signals after vectorization and the event alarm signals in each corresponding knowledge graph network. The event chain is composed of multiple events. The event chain similarity is calculated based on the event similarity. The similarity of the event chain is obtained by the average value of the sum of the similarities of multiple events. The similar event chain in the knowledge graph network is determined by the maximum similarity.

[0040] In step 4, a similarity combination measurement algorithm is used to calculate the similarity between events. After obtaining the formatted vector of the alarm signal, a new calculation method is proposed for the similarity calculation in a single alarm event. For discrete variables in the event, a similarity measurement method is used in which the same is 1 and the different is 0. For continuous variables in the event, the Euclidean distance formula is used as the similarity measurement method after standardization. Finally, the weight factor is added to the calculated similarity of the discrete variable and the similarity combination of the continuous variable is used as the similarity of a single event. The coefficient of the discrete variable is used as the weight factor and optimized using historical data in the offline training stage;

[0041] The calculation of discrete variables in the event vector is as follows:

[0042]

[0043] Where: D i and D j denote discrete variables in events i and j, respectively, Represents the similarity of the discrete variables of events i and j. If the discrete variables have the same value, then The value of is 1, otherwise it is 0;

[0044] The Euclidean distance calculation is used for continuous variables between events as follows:

[0045]

[0046] Where: X i,k and X j,k are the kth value of the continuous variable in events i and j respectively, n is the length of the continuous variable for events i and j, The similarity of the continuous variable representing events i and j;

[0047] The calculation for combining discrete and continuous variables is as follows:

[0048]

[0049] Where: represents the similarity of the discrete variables of events i and j, as shown in formula (1), w represents the weight factor of the discrete variable similarity, The similarity of the continuous variable representing events i and j is shown in formula (2), Sij It represents the similarity between events i and j, as shown in formula (3).

[0050] Step 5: Search the knowledge graph network to find the terminal node of the similar event chain, which is the fault point, and complete the fault prediction. Use the same method to handle other alarm events. The knowledge graph network stores the connection between all alarm signals and faults. The different fault nodes divide the graph into multiple chains. After determining the event chain with the largest similarity value, its terminal node is the fault to be predicted.

[0051] A specific embodiment is provided below:

[0052] In order to verify that the present invention is a fault prediction method for power grid dispatching control system based on event similarity and knowledge graph reasoning, this embodiment is verified in a typical urban power grid dispatching center consisting of more than 1,000 devices. Since the actual data is not enough to build a knowledge graph, a simulation case is added. The logical connection of hardware, software and network devices adopts a deterministic model, while the warning and fault connection adopts a probabilistic model. Table 2 lists the key parameters of the model:

[0053] Table 2: Formatted alarm information

[0054] U is a uniformly distributed variable N is a normally distributed variable C is a constant

[0055]

[0056] According to Table 2, a Monte Carlo simulation case was generated, simulating 100,000 event chains. In this simulation, it was defined that 3 warnings from the same category would trigger a fault. Figure 2 The summarized fault chain is shown.

[0057] from Figure 2 It can be seen that 80% of the event chains have a length of 9-20 before a serious failure occurs.

[0058] According to the method of calculating the similarity of a single event in step 2, the similarity measure between single events can be obtained, such as Figure 3 As shown, the test verifies the similarity between single events of three types: hardware, software, and network; Figure 3 In the figure, the gray box shows the similarity of events of the same type. For example, H(1,1) is two events with lower storage device capacity. The white box shows the similarity of events of different types. It can be clearly observed that the similarity value in the gray box is greater than the similarity value in the white box.

[0059] Finally, the fault warning is predicted. Since the similarity calculation only considers the alarm vector and cannot directly calculate the similarity of the event chain, the terminal fault node is used to verify the similarity. In the event chain simulated above, 8 typical fault event chains are artificially selected as the benchmark. Figure 3 From the similarity measurement results, we can observe that the similarity values ​​of the same type of event similarity and the same terminal fault are all greater than 0.5. The same type of event is represented by a gray box, so the conclusion uses a similarity value of 0.5 as the threshold. If the similarity value > 0.5, it is judged to be the same fault. In the simulation cases of 8 typical fault event chains, the similarity values ​​of the events are calculated, and the number of samples greater than the threshold is obtained. After statistics, the proportion of the same terminal fault in these samples can also be obtained. The results are shown in Table 3:

[0060] Table 3: Event chain similarity

[0061]

[0062] In Benchmark 1 of Table 3, 2578 represents the number of selected simulation case samples that are greater than the selected threshold after similarity calculation; 2372 represents the percentage of the same terminal failures in the above samples.

[0063] The similarity calculation is verified by the high percentage of identical terminal failures, which means that in this test case, the failure can be accurately predicted based on the event chain similarity. Since it is defined in the simulation case that three warnings from the same category will trigger a failure, the event length is set to 3 to 10 for calculation. Figure 4 As can be seen from the figure, as the length of the event chain increases, the accuracy of the prediction improves.

[0064] To sum up, combined with the test results of this implementation example, the method of the present invention can use real-time alarm signals to more accurately and efficiently realize fault prediction of the power grid dispatching control system. The fault type can be known in advance according to the alarm signal, and maintenance personnel can solve the problem in advance to ensure safe and stable operation of the power grid.

[0065] The parts not involved in the present invention are the same as the prior art or are implemented by using the prior art.

[0066] The above contents are further detailed descriptions of the present invention in combination with specific implementation methods, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A fault prediction method for a power grid dispatching control system, It is characterized in that the steps include: Step 1: Construct an event chain based on known historical fault alarm events in the dispatching control system, and combine the event chains to form a knowledge graph network; Step 2: Process the real-time alarm signal in the initial process of the power grid dispatching control system fault, vectorize the real-time alarm signal containing the object, fault type, and related applications, and obtain a discrete-continuous vector as a variable for subsequent similarity calculation; Step 3: Vectorize the alarm signal of the node object in the knowledge graph network according to the vectorization method in step 2; Step 4: Calculate the similarity of discrete variables and continuous variables between nodes in the vectorized knowledge graph network and real-time event alarm signals. The event chain consists of multiple events. Calculate the event chain similarity based on the event similarity, and determine the similar event chain in the knowledge graph network based on the maximum similarity. Step 5: Search the knowledge graph network to find the terminal node of the similar event chain, which is the fault point, and complete the fault prediction; In step 4, a similarity combination measurement algorithm is used to calculate the similarity between events. The similarity measurement of a single event is to calculate the discrete variables and the continuous variables respectively, and finally combine the calculation results of the discrete variables and the continuous variables; The calculation of discrete variables in the event vector is as follows: Where: D i and D j denote discrete variables in events i and j, respectively, Represents the similarity of the discrete variables of events i and j. If the discrete variables have the same value, then The value of is 1, otherwise it is 0; The Euclidean distance calculation is used for continuous variables between events as follows: Where: X i,k and X j,k are the kth value of the continuous variable in events i and j respectively, n is the length of the continuous variable for events i and j, The similarity of the continuous variable representing events i and j; The calculation for combining discrete and continuous variables is as follows: Where: represents the similarity of the discrete variables of events i and j, as shown in formula (1), w represents the weight factor of the discrete variable similarity, The similarity of the continuous variable representing events i and j is shown in formula (2), S ij represents the similarity between events i and j, as shown in formula (3); The event chain similarity is obtained by averaging the sum of the similarities of multiple events; after determining the event chain with the largest similarity value, the terminal node of the event chain with the largest similarity value in the knowledge graph network is the fault to be predicted.

2. The fault prediction method of the power grid dispatching control system according to claim 1, Features: In the step 1 and the step 2, before the alarm signal is quantized, the collected alarm signal is first formatted and the information is standardized according to the mode of component type-event-state.

3. The fault prediction method of the power grid dispatching control system according to claim 1, Features: In step 1, the control system is divided into hardware, software and network types, and the alarm signal comes from the control system; in step 2, a digital algorithm based on fuzzy clustering is used to simplify the alarm signal into a formatted vector.

Images