A smart system for power grid maintenance operation scheduling and operation ticket generation based on fault prediction

Abstract

A smart system for power grid maintenance operation scheduling and operation ticket generation based on fault prediction, belonging to the field of power maintenance, is constructed. The system comprises a data acquisition interface, a fault prediction module, a maintenance scheduling module, an operation ticket generation module, and an interactive front-end. It uses recurrent neural networks for sequence prediction; establishes a power grid maintenance scheduling method based on tree-structured dynamic programming; and employs a two-ticket generation method based on deep reinforcement learning. It uses a neural network-based sequence prediction model to accurately predict future faults at nodes using a large amount of historical data; it uses a dynamic programming algorithm to schedule maintenance for both past and future faults, and automatically generates corresponding operation tickets and work tickets; completing fault repair or elimination at the optimal economic cost. It can be widely used in power system operation and maintenance management and on-site maintenance.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of power maintenance and safety management, and particularly relates to an intelligent system for power grid maintenance operation arrangement and operation ticket generation based on fault prediction. BACKGROUND

[0002] Power supply interruption has a very negative impact on the production and life of the region.

[0003] Therefore, how to accurately predict potential problems in the power system, and how to efficiently repair and eliminate the faults that have occurred and have not occurred, have important research value.

[0004] In the process of repairing and eliminating the faults that have occurred and have not occurred, the related steps and methods of operation ticket generation are inevitably involved.

[0005] The operation ticket refers to the written basis for electrical operation in the power system, including dispatching instruction ticket and substation operation ticket.

[0006] The operation ticket is generally filled out by the operator according to the order of the value length, the workshop responsible person or the power station dispatcher; before filling out the operation ticket, the operation task must be clear, the operation mode must be clear, the equipment state must be clear, and the operation ticket must be filled out according to the actual situation on site.

[0007] The invention patent with the authorized announcement date of April 2, 2019 and the authorized announcement number of CN 105446749 B discloses a method and device for generating an operation ticket, which determines the components to be operated according to the operation requirements; finds the component parameters corresponding to the components to be operated in the logic table; and calls the operation ticket generation script to generate the operation ticket according to the component parameters; wherein the logic table stores the logic sequence corresponding to the operation ticket generation script; since each component parameter is independent in this process, even if one or several component parameters need to be modified, the corresponding operation ticket generation script will also change accordingly, so the modification amount is very small, and the entire operation ticket does not need to be modified. This technical solution can avoid the modification of the operation ticket when the currently stored operation ticket does not correspond to the actual operation requirements, and improves the work efficiency.

[0008] However, the existing operation ticket generation strategy usually uses a template matching method to generate an operation ticket (including using a reinforcement learning mode to optimize the generation of an operation ticket, and scoring according to a preset rule), and only carries out remedial maintenance after the node fault occurs, which cannot accurately predict the time when the node will fail; and the maintenance scheduling is only related to the order of the fault occurrence, and the planning is poor, and the fault cannot be repaired or eliminated at the minimum economic cost. SUMMARY

[0009] The technical problems to be solved by the present application are to provide an intelligent system for power grid maintenance operation arrangement and operation ticket generation based on fault prediction.

[0010] The technical solution of the present application is to provide an intelligent system for power grid maintenance operation arrangement and operation ticket generation based on fault prediction, characterized by:

[0011] 1) An intelligent system for power grid maintenance operation arrangement and operation ticket generation is constructed, which includes a data acquisition interface, a fault prediction module, a maintenance scheduling module, an operation ticket generation module, and an interactive front end.

[0012] 2) The fault prediction module uses recurrent neural networks for sequence prediction.

[0013] 3) For a rooted tree-like power grid structure, the maintenance scheduling module establishes a tree-like dynamic programming-based power grid maintenance scheduling method.

[0014] 4) The operation ticket generation module uses a two-ticket generation method based on deep reinforcement learning.

[0015] 5) A neural network-based sequence prediction model is used to accurately predict future faults of nodes using a large amount of historical data. A dynamic programming algorithm is used to arrange maintenance schedules for faults that have occurred and have not occurred, and to automatically generate corresponding operation tickets and work tickets. The economic cost of fault repair or elimination is optimized.

[0016] 2. The intelligent system for power grid maintenance operation arrangement and operation ticket generation based on fault prediction according to claim 1, characterized in that the data acquisition interface comprises:

[0017] 1) SCADA interface:

[0018] The data analysis module of the system is connected to the SCADA system, and the running state values of each device in the power grid are obtained at fixed intervals, and data analysis is performed to predict the state values and assist decision-making.

[0019] 2) Middle station:

[0020] The data information required in the operation ticket generation module needs the support of the middle station system. By calling the corresponding API interface in the middle station system, relevant data can be obtained and analyzed to generate relevant operation sequences for related tasks.

[0021] 3.The intelligent system for power grid maintenance operation arrangement and operation ticket generation based on fault prediction according to claim 1, characterized in that the fault prediction module predicts the state values of all devices in the power grid in a future period of time according to the operation state data and interaction mode of the devices in the existing power grid, thereby predicting the operation condition of the entire power grid and providing decision information for subsequent operation arrangement of the power station.

[0022] 4.The intelligent system for power grid maintenance operation arrangement and operation ticket generation based on fault prediction according to claim 3, characterized in that the fault prediction module is divided into two parts.

[0023] The first part designs a deep learning algorithm model for time series prediction, and fits the model through massive historical data, so that the model has the ability to obtain time series feature information.

[0024] The second part is mainly divided into three sub-modules: the first sub-module is connected with SCADA to obtain real-time operation data of all devices in the power grid in multiple dimensions; the second sub-module preprocesses the obtained data to remove meaningless values and performs operations including normalization of part of the data; the third sub-module obtains the feature data of the preprocessed power grid devices, inputs the data into the deep learning algorithm model for time series prediction, predicts the future condition of the devices, and transmits the prediction to the operation generation module to assist device operation decision.

[0025] 5.The intelligent system for power grid maintenance operation arrangement and operation ticket generation based on fault prediction according to claim 1, characterized in that the maintenance scheduling module is used to generate a minimum cost operation scheme sequence; first, the module calls the API interface of the middle station to obtain corresponding data information, including node information and time information of fault or maintenance, and the module can solve the best operation strategy through a mathematical model established based on the dynamic programming idea, so that the cost is minimized, and then the generated minimum cost strategy sequence is sent to the operation ticket generation module based on the template to generate corresponding operation instruction tickets.

[0026] 6.The intelligent system for power grid maintenance operation arrangement and operation ticket generation based on fault prediction according to claim 1, characterized in that the operation ticket generation module analyzes the minimum cost strategy sequence generated and transmitted by the minimum cost operation generation module, and through the analysis, obtains relevant station name and operation category information and strategy information of how to open and close the switch; then, the module retrieves the operation instruction ticket template of the operation category stored in the knowledge base according to the operation category information, and then generates the operation instruction ticket in combination with the relevant station name; finally, the operation instruction ticket information is transmitted to the front end for display.

[0027] The operation category includes fault or maintenance.

[0028] 7.The fault prediction based power grid maintenance operation arrangement and work order generation intelligent system according to claim 6, characterized in that the interactive front end comprises a front end web visualization display page.

[0029] The interactive front end respectively interacts with the user in the relevant page according to the data information intelligently generated by the power grid data, the fault prediction module and the work order generation module.

[0030] 8.The fault prediction based power grid maintenance operation arrangement and work order generation intelligent system according to claim 1, characterized in that the fault prediction module adopts a power grid fault prediction method based on a recurrent graph neural network, abstracts the power grid into a graph structure, abstracts the power distribution location into a node and the line into an edge, and then uses a recurrent neural network for sequence prediction; the input, state and output of the recurrent neural network are all tensor graph structures, and the calculation of the state change is changed from full connection to graph convolution, so that the structure information of the power grid can be effectively utilized to make more accurate prediction.

[0031] 9.The fault prediction based power grid maintenance operation arrangement and work order generation intelligent system according to claim 1, characterized in that the maintenance scheduling module adopts a power grid maintenance scheduling method based on tree-shaped dynamic programming, adopts a dynamic programming algorithm for the root tree-shaped power grid structure, assumes the power outage loss of the power grid node, obtains the expected fault time from the fault prediction, optimizes layer by layer from the leaf node to the root node, designs the most reasonable power outage maintenance scheme, and minimizes the economic loss caused by power outage.

[0032] 10.The fault prediction based power grid maintenance operation arrangement and work order generation intelligent system according to claim 1, characterized in that the work order generation module accepts the scheduling task as input, generates a plurality of operation order combinations with high probability through operation order generation, selects and modifies the operation order combinations by the user and actually uses the operation order combinations, finally, performs feedback training according to the structure selected and modified by the user, performs iterative reinforcement learning according to the operation feedback of the user, and optimizes the operation order generation model.

[0033] Compared with the prior art, the application has the following advantages:

[0034] 1.A power grid fault prediction method based on a recurrent graph neural network is disclosed, compared with a single-point fault prediction method, the technical scheme of the application can more effectively utilize the correlation information between power grid nodes to make more accurate and timely fault prediction.

[0035] 2. Disclosed is a power grid maintenance scheduling method based on tree-shaped dynamic programming, compared with the traditional single-task-oriented maintenance scheduling method, the technical scheme can better reduce the economic loss caused by maintenance power outage under the premise of ensuring safety.

[0036] 3. Disclosed is a two-ticket generation method based on deep reinforcement learning, compared with the existing template or rule-based method, the technical scheme can effectively utilize user online operation for feedback iteration, and continuously optimize the generation effect.

[0037] 4. The technical scheme of the present application combines the above multiple functions into a system, which can directly complete all work from fault prediction to operation ticket generation. BRIEF DESCRIPTION OF DRAWINGS

[0038] Figure 1 is a schematic diagram of the overall structure of the system of the present application;

[0039] Figure 2 is a schematic diagram of the power grid fault prediction method based on the recurrent graph neural network of the present application;

[0040] Figure 3 is a schematic diagram of the power grid maintenance scheduling method based on the tree-shaped dynamic programming of the present application;

[0041] Figure 4 is a schematic diagram of the two-ticket generation method based on deep reinforcement learning of the present application. DETAILED DESCRIPTION

[0042] The present application will be further described below in conjunction with the drawings.

[0043] I. Overall structure:

[0044] The overall structure of the present application is shown in Figure 1

[0045] 1. Data acquisition interface:

[0046] a. SCADA:

[0047] SCADA (Supervisory Control And Data Acquisition) system, i.e. data acquisition and monitoring control system.

[0048] SCADA system is a computer-based DCS and power automation monitoring system.

[0049] ​In the power system, the SCADA system is the most widely used, and the technology is the most mature. It plays an important role in the remote control system, can monitor and control the running equipment in the field, to realize data acquisition, equipment control, measurement, parameter adjustment and various signal alarm and other functions. The data analysis module of the technical scheme is connected with the SCADA system, the running state value of each device in the power grid is obtained at fixed interval time, and data analysis is carried out, the state value is predicted, and decision is assisted.

[0050] b. Middle station:

[0051] The middle station system is a data platform for obtaining information, displaying power grid state and manually printing operation tickets in the power grid. The data information required in the minimum cost operation generation module of the technical scheme needs the support of the middle station system, and the minimum cost operation generation module can obtain related data and analyze by calling the corresponding API interface in the middle station system, so as to generate related operation sequence for related task.

[0052] 2. Fault prediction module:

[0053] The power grid data and state prediction module mainly aims to predict the state value of all devices in the power grid in a period of time in the future according to the running state data and interaction mode of each device in the existing power grid, so as to predict the running condition of the whole power grid. Provide decision information for subsequent operation arrangement of power station. The power grid data and state prediction module mainly includes two parts.

[0054] The first part is to design a deep learning algorithm model of time series prediction, and fit the model through a large amount of historical data, so that the model has the ability to obtain time series characteristic information.

[0055] The second part mainly includes three submodules. The first submodule is to connect with SCADA to obtain real-time running data of all devices in the power grid in multiple dimensions. The second submodule is to preprocess the obtained data, remove meaningless values and normalize part of the data and other operations. The third submodule obtains the characteristic data of the preprocessed power grid equipment, inputs the time series prediction deep learning algorithm model, predicts the future state of the equipment, and transmits the prediction to the operation generation module to assist the equipment operation decision.

[0056] 3. Maintenance scheduling module:

[0057] This module is primarily used to generate a sequence of minimum-cost operation schemes. First, it calls the platform's API to obtain corresponding data, including node information and timing information for faults or maintenance. Then, using a mathematical model based on dynamic programming, it solves for the optimal operation strategy to minimize costs. Finally, it sends the generated minimum-cost strategy sequence to the template-based operation ticket generation module to generate corresponding operation instruction tickets.

[0058] 4. Operation Ticket Generation Module:

[0059] This module parses the minimum-cost strategy sequence generated by the minimum-cost operation generation module. Parsing reveals the relevant site name, operation category (fault or maintenance) information, and the strategy information for opening and closing the switch. Then, based on the operation category information, the module retrieves the operation instruction ticket template for that operation category from the knowledge base, and then combines it with the relevant site name to generate an operation instruction ticket. Finally, the operation instruction ticket information is sent to the front end for display.

[0060] 5. Interactive Front-End:

[0061] The front-end web visualization display page will interactively display the data information intelligently generated by the power grid data and status prediction module and the template-based operation ticket generation module on the relevant pages. The details are as follows.

[0062] For displaying data information in the power grid data and status prediction module, the front-end page provides text boxes for users to input three parameters: the name of the power plant / station, its type, and the name of the equipment. The front-end then packages these three parameters into JSON format and sends it to the back-end via an HTTP request. Upon receiving the request, the back-end parses and extracts the three parameters, then calls the power grid data and status prediction module. This module packages the previously obtained real-time prediction values ​​along with historical data values ​​in response to the front-end request. The front-end then parses and extracts the data from the back-end response and displays it to the user in the form of a line chart. Furthermore, different colors are used to display historical and predicted data for easier and clearer comparison by the user.

[0063] For displaying data information from the template-based operation ticket generation module, a polling request is sent to the backend. When the backend module responds to the request and returns the relevant operation instruction ticket information it generated, the frontend receives the response, analyzes and extracts the data, and then displays it on the page.

[0064] II. A Power Grid Fault Prediction Method Based on Cyclic Graph Neural Networks:

[0065] like Figure 2 As shown in the figure, the technical solution of the present invention abstracts the power grid into a graph structure, the location of power distribution and transformation is abstracted into nodes, and the lines are abstracted into edges.

[0066] Then, a recurrent neural network is used for sequence prediction.

[0067] However, the technical solution of this invention differs from existing single-point prediction methods. Its input, state, and output are all tensor graph structures, and the calculation of state changes is changed from fully connected to graph convolution, thereby enabling this method to effectively utilize power grid structure information for more accurate prediction.

[0068] III. Power Grid Maintenance and Dispatch Method Based on Tree-based Dynamic Programming:

[0069] like Figure 3 As shown, this invention designs a dynamic programming algorithm for rooted tree-structured power grids. The algorithm assumes the power outage losses at grid nodes and obtains the expected fault time from fault prediction. By optimizing layer by layer from leaf nodes to root nodes, it designs the most reasonable power outage maintenance plan, minimizing the economic losses caused by power outages.

[0070] IV. Two-vote generation method based on deep reinforcement learning:

[0071] like Figure 4 As shown, the operation ticket generation system of the present invention accepts the aforementioned scheduling task as input, generates multiple combinations of operation tickets with high probabilities, and allows the user to select, modify, and actually use them. Finally, the system performs feedback training based on the user's selected and modified structures, and iteratively performs reinforcement learning based on the user's operation feedback to optimize the operation ticket generation model.

[0072] In summary, the technical solutions of this invention disclose a power grid fault prediction method based on cyclic graph neural networks, a power grid maintenance scheduling method based on tree dynamic programming, and a two-ticket generation method based on deep reinforcement learning.

[0073] The technical solution of this invention adopts a power grid fault prediction method based on cyclic graph neural networks, which can more effectively utilize the correlation information between power grid nodes to make more accurate and timely fault prediction. Under the premise of ensuring safety, it can better reduce the economic losses caused by power outages due to maintenance. It can effectively utilize online user operations for feedback iteration and continuously optimize the generation effect. By combining the above multiple functions into a system, it can directly complete all the work from fault prediction to operation ticket generation.

[0074] This invention can be widely used in the operation and maintenance management and on-site repair of power systems.

Claims

1. A fault prediction-based power grid maintenance operation arrangement and operation ticket generation intelligent system, characterized by: 1) constructing a power grid maintenance operation arrangement and operation ticket generation intelligent system composed of a data acquisition interface, a fault prediction module, a maintenance scheduling module, an operation ticket generation module, and an interactive front end; 2) the fault prediction module uses a recurrent neural network for sequence prediction; The fault prediction module predicts the state values of all devices in the power grid for a period of time based on the operating state data and interaction mode of the existing power grid, thereby predicting the operating condition of the entire power grid and providing decision information for subsequent operation arrangement of the power plant; 3) for a rooted tree-like power grid structure, the maintenance scheduling module establishes a tree-like dynamic programming-based power grid maintenance scheduling method; Its maintenance scheduling module is used to generate a minimum cost operation scheme sequence: First, the maintenance scheduling module calls the API interface of the middle station to obtain the corresponding data information, including fault or maintenance node information and time information, and the maintenance scheduling module solves the optimal operation strategy through a mathematical model based on the dynamic programming idea, thereby minimizing the cost; Then, the maintenance scheduling module sends the generated minimum cost strategy sequence to the operation ticket generation module based on the template to generate the corresponding operation instruction ticket; 4) the operation ticket generation module uses a two-ticket generation method based on deep reinforcement learning; The operation ticket generation module parses the minimum cost strategy sequence generated and transmitted by the minimum cost operation generation module, and through parsing, the relevant station name and operation category information and how to open and close the switch strategy information are obtained; Then, the operation ticket generation module retrieves the operation instruction ticket template of the operation category stored in the knowledge base according to the operation category information, and then generates the operation instruction ticket in combination with the relevant station name; Finally, the operation instruction ticket information is transmitted to the front end for display; 5) using a neural network-based sequence prediction model, a large amount of historical data is used to accurately predict the future faults of nodes; using a dynamic programming algorithm, maintenance scheduling is arranged for faults that have occurred and have not occurred, and corresponding operation tickets and work tickets are automatically generated; and the fault repair or elimination is completed with the optimal economic cost.

2. The failure prediction based power grid maintenance operation scheduling and work order generation intelligent system as claimed in claim 1, characterized in that The data acquisition interface includes: 1) SCADA interface: The data analysis module of the system is connected with the SCADA system, and the running state values of each device in the power grid are obtained at fixed intervals, and data analysis is performed, the state value is predicted, and decision-making is assisted; 2) middle station: The data information required in the operation ticket generation module needs the support of the middle station system; by calling the corresponding API interface in the middle station system, the relevant data can be obtained and analyzed, thereby generating the relevant operation sequence for the relevant task.

3. The failure prediction based power grid maintenance operation scheduling and work order generation intelligent system as claimed in claim 1, characterized in that The fault prediction module is divided into two parts: First, design a deep learning algorithm model for time series prediction, and fit the model through a large amount of historical data, so that the model has the ability to obtain time series feature information; The second part is mainly divided into three sub-modules: the first submodule is to interface with SCADA to obtain real-time operation data of all devices in the power grid in multiple dimensions; the second submodule is to preprocess the obtained data to remove meaningless values and perform operations including normalization of part of the data; The third submodule obtains the feature data of the preprocessed power grid devices and inputs them into the deep learning algorithm model of time series prediction to predict the future status of the devices and transmits the prediction to the operation generation module to assist in device operation decision-making.

4. The failure prediction based power grid maintenance operation scheduling and work order generation intelligent system as claimed in claim 1, characterized in that The operation category includes fault or maintenance.

5. The failure prediction based power grid maintenance operation scheduling and work order generation intelligent system as claimed in claim 1, characterized in that The interactive front end includes a front-end web visualization display page. The interactive front end respectively displays the data information intelligently generated by the power grid data, fault prediction module and operation ticket generation module in the relevant pages to interact with the user.

6. The failure prediction based power grid maintenance operation scheduling and work order generation intelligent system as claimed in claim 1, characterized in that The fault prediction module adopts a power grid fault prediction method based on a recurrent graph neural network, abstracts the power grid into a graph structure, abstracts the distribution location into a node, and abstracts the line into an edge, and then uses a recurrent neural network for sequence prediction. The input, state and output are all tensor graph structures, and the calculation of state changes is changed from full connection to graph convolution, so that the structure information of the power grid can be effectively utilized for more accurate prediction.

7. The failure prediction based power grid maintenance operation scheduling and work order generation intelligent system as claimed in claim 1, characterized in that The maintenance scheduling module adopts a power grid maintenance scheduling method based on tree-shaped dynamic programming, adopts a dynamic programming algorithm for a rooted tree-shaped power grid structure, assumes the power outage loss of the power grid node, obtains the expected fault time from the fault prediction, optimizes layer by layer from the leaf node to the root node, designs the most reasonable power outage maintenance scheme, and minimizes the economic loss caused by power outage.

8. The failure prediction based power grid maintenance operation scheduling and work order generation intelligent system as claimed in claim 1, characterized in that The operation ticket generation module accepts the scheduling task as input, generates multiple operation ticket combinations with high probability through the operation ticket, selects and modifies the operation ticket by the user, and actually uses it; finally, according to the structure selected and modified by the user, feedback training is performed, iterative reinforcement learning is performed according to the operation feedback of the user, and the operation ticket generation model is optimized.

Images