Power grid security protection method and system combined with deep reinforcement learning

Abstract

The present invention relates to the technical field of power grid security protection, and specifically includes a power grid security protection method and system combined with deep reinforcement learning, including: for N generators in a target area power grid, real-time operating parameters are collected, a power grid topology diagram is constructed, N distributed control units are divided, and a star topology connection is adopted to form a centralized control network; a deep reinforcement learning environment is designed, the optimal control variables of each generator are selected in an action space, an optimal control instruction set is formed, and the optimal control instruction set is sent to the distributed control units for execution in real time, so as to solve the technical problems that the relay protection device can only rely on preset rules, response speed and adaptability are limited on the basis of the normal operation of the relay protection device, so as to achieve dynamic optimization of the power grid operation state, accurately select the optimal control variable combination for each generator in the preset action space, thereby quickly forming a comprehensive and targeted optimal control instruction set, and improving the technical effect of the safety and stability level of the power grid.

Description

Technical Field

[0001] The present invention relates to the technical field related to power grid security protection, and in particular to a power grid security protection method and system combined with deep reinforcement learning. Background Art

[0002] Relay protection is the most basic safety protection measure in the power grid. It is used to detect faults in the power grid and quickly remove the faulty part to prevent the fault from spreading to the entire power grid. In order to ensure continuous power supply, the power grid design usually includes multiple power supply points and multiple transmission paths. Once the main power supply or main line has problems, the backup power supply and channel can be immediately switched and put into use to ensure that important loads are not interrupted. However, relay protection devices usually rely on preset thresholds and fixed rules and cannot flexibly respond to complex changes in the operating status of the power grid.

[0003] In summary, there are technical problems in the prior art that the relay protection device can only rely on preset rules and has limited response speed and adaptability. Summary of the invention

[0004] This application aims to solve the technical problems in the prior art that relay protection devices can only rely on preset rules and have limited response speed and adaptability by providing a power grid security protection method and system combined with deep reinforcement learning.

[0005] In view of the above problems, the technical solution to implement this application is:

[0006] On the one hand, the present application provides a power grid security protection method combined with deep reinforcement learning, wherein the method comprises: based on N generators in a target area power grid, obtaining real-time operating data of each generator, wherein the real-time operating data includes generator operating parameters and generator operating status;

[0007] Based on the target regional power grid, a power grid structure topology map is generated, and according to the power grid structure topology map, each of the N generators is used as a primary central node, and each node, line, transformer and load point in the target regional power grid is compared to obtain N distributed control units;

[0008] Based on a star topology, the balancing protection unit is used as a secondary central node, and the N distributed control units are connected to obtain a centralized control network;

[0009] A deep reinforcement learning environment is designed to continuously learn protection strategies under different working conditions based on the centralized control network to obtain a safety protection model of the target regional power grid, and the action space corresponding to the N distributed control units is used as an adjustable control variable cluster of the N generators of the target regional power grid;

[0010] The real-time operation data is imported into the safety protection model, the optimal control variable is selected for each generator in the action space, the optimal control instruction set of the target area power grid is obtained, and it is sent to the N distributed control units for execution in real time.

[0011] On the other hand, the present application provides a power grid security protection system combined with deep reinforcement learning, wherein the system includes: an operation data acquisition module, which is used to acquire real-time operation data of each generator based on N generators in the target area power grid, and the real-time operation data includes generator operation parameters and generator operation status;

[0012] A control unit setting module is used to generate a power grid structure topology map based on the target area power grid, and according to the power grid structure topology map, take each of the N generators as a primary central node, and compare each node, line, transformer and load point in the target area power grid to obtain N distributed control units;

[0013] A control network setting module is used to connect the N distributed control units based on a star topology structure with a balancing protection unit as a secondary central node to obtain a centralized control network;

[0014] A protection strategy learning module is used to design a deep reinforcement learning environment, and based on the centralized control network, continuously learns the protection strategies under different working conditions to obtain a safety protection model of the target area power grid, and the action space corresponding to the N distributed control units is used as an adjustable control variable cluster of the N generators of the target area power grid;

[0015] An execution module is used to import the real-time operation data into the safety protection model, select the optimal control variable for each generator in the action space, obtain the optimal control instruction set of the target area power grid, and send it to the N distributed control units for execution in real time.

[0016] In summary, one or more technical solutions provided in this application realize the collection of real-time operating parameters and status data for N generators in the target area power grid, construct a power grid topology diagram, and divide N distributed control units, each unit centered around a generator; a star topology is used to form a centralized control network; a deep reinforcement learning environment is designed to learn safety protection strategies for different working conditions, and generate a protection model containing a set of adjustable variables of the generator; operating data is received in real time, the optimal control variables of each generator are selected in the action space, an optimal control instruction set is formed, and it is sent to the distributed control units for execution in real time, thereby achieving the technical effect of dynamically optimizing the operating state of the power grid, accurately selecting the optimal control variable combination for each generator in the pre-set action space, thereby quickly forming a comprehensive and targeted optimal control instruction set, and improving the safety and stability level of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A possible flow chart of a power grid security protection method combined with deep reinforcement learning is provided for this application;

[0018] Figure 2 A possible flow chart of the control strategy for dynamically adjusting FACTS devices in a power grid security protection method combined with deep reinforcement learning is provided for this application;

[0019] Figure 3 A possible structural diagram of a power grid security protection system combined with deep reinforcement learning is provided for this application.

[0020] Description of reference numerals: operation data acquisition module M100, control unit setting module M200, control network setting module M300, protection strategy learning module M400, execution module M500. DETAILED DESCRIPTION

[0021] Embodiment 1

[0022] The present application is described in detail below in conjunction with the accompanying drawings. Figure 1 As shown, the present application provides a power grid security protection method combined with deep reinforcement learning, wherein the method includes:

[0023] Step-1: Based on N generators in the target area power grid, obtain the real-time operation data of each generator, wherein the real-time operation data includes generator operation parameters and generator operation status;

[0024] Step-2: Based on the target area power grid, a power grid structure topology diagram is generated. According to the power grid structure topology diagram, each of the N generators is used as a primary central node, and each node, line, transformer and load point in the target area power grid is compared to obtain N distributed control units;

[0025] In the scenario of large-scale renewable energy grid connection and high proportion of DC transmission, most power outage accidents are characterized by: peak load, successive tripping of lines after overload, large fluctuations in interval power flow, voltage drop and even collapse, and dispatching failure to handle them in time for some reason. Although the relay protection device works normally, it may have insufficient adaptability and control decision lag when dealing with complex and dynamically changing power system conditions.

[0026] Based on this, in the target area power grid, sensors and monitoring equipment are deployed at key parts of each generator, such as rotor speed, voltage, current, power output, temperature, oil pressure and other key operating parameters; connected to the data acquisition system to ensure that various operating status information of the generator can be recorded in real time, including but not limited to operating mode, load conditions, fault alarm signals; the collected data is transmitted in real time through a communication network (such as optical fiber, wireless, etc.) to the power grid safety protection system combined with deep reinforcement learning.

[0027] Based on the actual physical layout and electrical connection relationship of the power grid, a detailed power grid structure topology diagram is created, covering the location and electrical connection of all nodes, lines, transformers and load points. The power grid topology is analyzed to clarify the location and function of each generator in the power grid structure topology diagram.

[0028] According to the grid structure topology diagram, with each generator as the core, the surrounding related nodes, lines, transformers and related loads are included in an independent control area, and a distributed control unit is set up in each area. The distributed control unit is responsible for monitoring and controlling the operation of all power facilities in the area, especially the operating status and control strategy of the core generator. Any one of the N distributed control units combines the generator operation data obtained in real time to adjust the generator operation parameters according to the overall operation requirements of the grid and local conditions to ensure the safe and stable operation of the grid.

[0029] Step-3: Based on the star topology, the balancing protection unit is used as the secondary central node, and the N distributed control units are connected to obtain a centralized control network;

[0030] Step-4: Design a deep reinforcement learning environment, based on the centralized control network, continuously learn the protection strategies under different working conditions, and obtain the safety protection model of the target area power grid. The action space corresponding to the N distributed control units is used as the adjustable control variable cluster of the N generators in the target area power grid;

[0031] Step-5: Import the real-time operation data into the safety protection model, select the optimal control variables for each generator in the action space, obtain the optimal control instruction set of the target area power grid, and send it to the N distributed control units for execution in real time.

[0032] The N distributed control units are organized in a star topology, with each distributed control unit as a leaf node and the balancing protection unit as a central node. At the same time, it is necessary to ensure that each distributed control unit can communicate with the central node in real time and report information such as the generator operating status and line load status in their respective areas to the central node; in addition, a reasonable communication protocol is designed to ensure the timeliness, integrity and security of data transmission.

[0033] The set of adjustable control variables (such as output, frequency, voltage, etc.) of N generators is defined as the action space of N distributed control units; each control unit can select an appropriate generator control strategy in the corresponding action space according to the current grid operation status.

[0034] Create a simulation environment that simulates the power grid in the target area, which can receive real-time operating data from the actual power grid or historical data as input; set different operating scenarios to simulate various possible power grid anomalies and disturbance events; use deep reinforcement learning algorithms (such as DQN, A3C, DDPG, etc.) to train the intelligent agent to learn the optimal protection strategy in the simulation environment, and continuously update its policy network in the process of interaction with the environment by trying different action combinations, and gradually form an effective safety protection model for various operating conditions.

[0035] The real-time generator operation data is imported into the trained safety protection model. The safety protection model calculates the optimal control variables for each generator according to the current power grid status, and sends the optimal control instruction set to the corresponding distributed control unit. Each control unit executes the control operation of the corresponding generator to achieve dynamic optimization control of the target area power grid.

[0036] In addition, after implementing the control instructions, the actual power grid response data can be collected and fed back into the deep reinforcement learning environment and safety protection model for further correction and optimization of model performance. The above process is repeated continuously so that the safety protection model can continue to learn and improve in actual applications, thereby improving the ability to ensure the safe operation of the power grid under complex working conditions.

[0037] Furthermore, based on the centralized control network, the protection strategies under different working conditions are continuously learned to obtain the security protection model of the target area power grid. Step-4 includes:

[0038] In the deep reinforcement learning environment, determining key operating status indicators of the target area power grid;

[0039] Convert the key operating status indicators into a state vector that can be perceived and processed, and construct a state-action space mapping;

[0040] Based on the state-action space mapping, an action set is designed, wherein the action set includes effective control dimensions corresponding to different operating states of N generators in a target area power grid, and the effective control dimensions are used to construct an action space for functional characteristic configuration.

[0041] Collect and analyze the actual operating data of the power grid in the target area, identify the operating status indicators that are critical to the safety and stability of the power grid, such as generator output, voltage level, frequency, line flow, transformer load, load change rate, grid structure parameters, etc., standardize the key operating status indicators, and convert them into state vectors that can be processed by deep reinforcement learning algorithms. The state vector should be able to fully reflect the current operating status of the power grid so that the model can make accurate decisions.

[0042] According to the physical characteristics and operation mechanism of the power grid, a state-action space mapping relationship is designed. In the mapping, each state vector corresponds to a set of possible actions. The action represents the adjustment of control variables of N generators and other equipment in the power grid, such as generator regulation action, reactive power compensation control, transformer tap position change, etc. For example, generator regulation action: for N generators, the action set may include increase or decrease adjustment of generator active power output, compensation and release of reactive power, speed or frequency control, etc. These actions help to maintain the stability of power grid frequency and qualified voltage.

[0043] Based on the functional characteristics of the power grid control equipment, an action set is defined in detail, which includes effective control dimensions that can be taken for each generator under different operating conditions, such as the active output adjustment range, reactive output adjustment range, equipment start and stop status, etc. of the generator; the effective control dimensions of all generators are integrated to construct a complete action space for the functional characteristics configuration of the target area power grid. Then, based on the action space, a deep reinforcement learning algorithm (such as DQN, DDPG, A2C, etc.) is used to perform model training in a deep reinforcement learning environment. Through continuous trial and error learning, the optimal action strategy is found under different working conditions, thereby obtaining a safety protection model suitable for the target area power grid.

[0044] Furthermore, based on the state-action space mapping, an action set is designed, wherein the action set includes effective control dimensions corresponding to different operating states of N generators in the target area power grid. The method of the present application also includes:

[0045] Locating key load equipment in the power grid structure topology diagram;

[0046] Perform load classification scheduling based on the key load equipment to obtain the scheduling priority of the key load equipment;

[0047] Load action management is performed through the key load equipment and the scheduling priority.

[0048] Analyze the grid structure topology map, identify key load equipment that has a significant impact on the stable operation of the grid, such as high-priority power supply points such as large industrial users and data centers, and clearly mark the locations of these key load equipment and their associated lines on the topology map.

[0049] According to factors such as the importance and sensitivity of the load, a set of load classification standards is formulated to divide key load equipment into different levels, such as primary, secondary, and tertiary loads. For each key load equipment, its dispatch priority under various working conditions is calculated according to the above classification standards. This can be obtained through a pre-set weight formula or historical data analysis to ensure that in emergency situations or power grid instability, priority is given to continuous power supply to high-level loads.

[0050] Combined with the dispatch priority of key load equipment, load-side operation dimensions are added to the action set, such as demand response measures and orderly power consumption strategies, and the specific form of the action is designed. For example, the intelligent agent is allowed to decide whether to adjust the load, when to adjust it, and the adjustment range according to the operating status of the power grid and the priority of key load equipment, including reducing some non-critical loads, transferring loads, starting backup power supplies or energy storage facilities, and other operations.

[0051] The dispatching actions of key load equipment are incorporated into the original action set of generator control to form a more refined comprehensive action set, so that the safety protection model can take into account the dynamic management of the load side while considering the operating status of the generator; in a deep reinforcement learning environment, the expanded action set is used for strategy training, so that the intelligent agent can not only control the generator according to the operating status of the power grid, but also perform reasonable load action management according to the priority of key load equipment, thereby realizing a more comprehensive and efficient power grid safety protection strategy.

[0052] Furthermore, before locating the key load equipment in the power grid structure topology diagram, the method of the present application further includes:

[0053] Locating the tap position of the transformer in the power grid structure topology diagram;

[0054] According to the tap positions of the transformers, the power flow distribution in the target area power grid is optimized.

[0055] On the one hand, global voltage adjustment and power flow optimization are required, including marking the locations of all transformers in the power grid structure topology diagram, and recording the number of taps, gear settings and current actual tap positions of each transformer in detail, which will help to analyze and optimize the operation status of the transformer in the system in the future. Through the SCADA system (Supervisory Control And Data Acquisition) or other real-time monitoring equipment, the actual operation status data of each transformer tap, including the current tap gear, voltage level and other information, can be obtained in real time or periodically.

[0056] According to the known grid structure topology and transformer tap status data, power system analysis software (such as PSASP, PSS / E, etc.) is used to perform power flow calculations to analyze the current power flow distribution of the target area power grid; based on the power flow calculation results, the voltage deviation in the power grid, the direction and size of the line power flow, and other issues are analyzed, and the transformer taps are optimized and adjusted in a targeted manner to improve the grid voltage quality, balance the line power flow distribution, and improve the grid operation efficiency. For example, if the voltage in a certain area is too high or too low, the tap position of the relevant transformer can be adjusted to increase or decrease the voltage level in the area; if the power flow of a certain line is too large, the power flow within the power grid can be redistributed by adjusting the tap positions of the transformers at both ends.

[0057] In addition, after optimizing the transformer tap settings, the power flow calculation can be performed again to compare the grid operation status before and after the optimization to verify the effectiveness of the optimization measures. If the optimization results fail to achieve the expected goals or new problems are found, the tap positions are further adjusted according to the actual conditions until the grid operation status reaches a satisfactory state. Only after completing the above-mentioned positioning, analysis and optimization of the transformer taps can the reasonable distribution of the grid power flow be ensured. At this time, the key load equipment is further located in the grid structure topology diagram to ensure that the overall operation of the grid is in a relatively optimized state when load management is considered.

[0058] Furthermore, if Figure 2 As shown, the present application method also includes:

[0059] Based on the local network structure characteristics and dynamic load changes, FACTS equipment is used to obtain the charge and discharge status and energy storage boundary of each energy storage unit under different operating conditions;

[0060] According to the charge and discharge status and energy storage boundary of each energy storage unit, the node voltage in the power grid structure topology diagram is evaluated for stability, and the node stability component is obtained;

[0061] The node stability component is used as the local optimization constraint parameter of the power flow distribution, and the control strategy of the FACTS device is dynamically adjusted in combination with the deep reinforcement learning environment.

[0062] On the one hand, local voltage adjustment and power flow optimization are required. Based on the local network structure characteristics and dynamic load changes, FACTS (Flexible AC Transmission Systems) equipment is used to optimize the grid stability, including:

[0063] Real-time monitoring of local grid parameters at the location of each FACTS device, including line flow, node voltage, frequency and other important indicators. At the same time, real-time data of each energy storage unit is collected and integrated, including key information such as energy storage capacity, current charge and discharge status, remaining available storage energy, charge and discharge rate, etc.

[0064] Based on the collected energy storage unit data, the maximum charge capacity, minimum discharge capacity and corresponding energy storage boundary conditions of each energy storage unit under the current and predicted future operating conditions are analyzed, the health status, cycle life and other factors of the energy storage unit are calculated, and the long-term stability and safety of its charging and discharging capabilities are comprehensively considered.

[0065] Combined with the grid structure topology diagram, for each node, the power system stability theory and algorithm are used, combined with the charging and discharging status and energy storage boundary of the energy storage unit, to evaluate the stability of the node voltage under various operating conditions, determine the stability component of the node voltage, that is, the node's ability to withstand disturbances, and the regulation margin required to maintain voltage stability under extreme conditions.

[0066] The node stability component obtained previously is used as a constraint condition for local optimization and added to the objective function or constraint set of deep reinforcement learning to ensure that the voltage stability of the power grid nodes is fully guaranteed while optimizing the control strategy of the FACTS equipment.

[0067] The safety protection model is trained using a deep reinforcement learning algorithm to autonomously learn and explore the optimal FACTS device control strategy in a complex and ever-changing power grid environment. The safety protection model continuously iterates and optimizes the decision-making process. According to the actual operating status of the power grid, the characteristics of the local network structure and dynamic load changes, the parameters of the FACTS device are adjusted in a timely manner (such as injecting reactive power, changing impedance, etc.) to achieve efficient distribution of power grid trends and effective control of node voltages, thereby achieving collaborative work between FACTS devices and energy storage units under a variety of operating conditions, effectively enhancing the stability of the power grid and improving the quality and reliability of power supply.

[0068] Furthermore, in locating the key load equipment in the power grid structure topology diagram, the method of the present application further includes:

[0069] Obtaining abnormal operating condition records of the target area power grid;

[0070] Based on the key load equipment, a correlation search is performed in the abnormal operating condition record to extract key abnormal information;

[0071] A safety action is set based on the key abnormal information, and the safety action is added to the action space. The safety protection equipment for executing the safety action includes a circuit breaker and a busbar tie switch.

[0072] Collect abnormal operating condition records of the target area power grid in the past period of time from the power grid monitoring system or historical database, including but not limited to fault events, overload conditions, voltage drops, and frequency anomalies. According to the location and attributes of key load equipment, perform associated searches in the abnormal operating condition records to find abnormal events directly or indirectly related to these key loads, conduct in-depth analysis of the retrieved abnormal events, and extract key abnormal information, such as the time, location, cause, scope of impact, duration, and degree of impact on key loads.

[0073] According to the extracted key abnormal information, combined with the power grid protection principles and technical specifications, the corresponding safety action strategy is designed. The safety actions may include but are not limited to: the opening and closing operation of the circuit breaker (when an abnormality occurs, the fault area is quickly isolated by the circuit breaker to protect the critical load from being affected), and the switching operation of the busbar tie switch (the power supply path is switched by the busbar tie switch to ensure that the critical load obtains a stable power supply);

[0074] Add safety actions set based on key abnormal information to the existing action space, expand the decision-making scope of the safety protection model in the deep reinforcement learning environment, update the action space mapping, ensure that the newly added safety actions are closely related to the protection strategies of key load equipment, and can be effectively implemented in actual operations.

[0075] The actual application process also includes, in a deep reinforcement learning environment, using the extended action space that includes key load equipment protection actions to train the safety protection model, and learning to accurately perform safety actions under different abnormal conditions. During actual operation, the safety protection model directs line circuit breakers, busbar tie switches and other safety protection devices to perform corresponding safety actions based on real-time monitoring data and learned strategies, so as to improve the safety and stability of the power grid in the target area.

[0076] Furthermore, the safety protection equipment for performing the safety action includes a circuit breaker and a busbar tie switch, and the method of the present application further includes:

[0077] A relay protection device connected to the target area power grid;

[0078] Downloading the equipment operation log of the relay protection device, analyzing the response time of the equipment operation log, and extracting the action logic rules of the relay protection device;

[0079] Based on the action logic rules, work in coordination with the safety protection device.

[0080] The relay protection device of the target area power grid is physically connected to the data acquisition and supervisory control system (SCADA) or other monitoring platform of the central control system through a communication interface or field bus to ensure that data can be read and transmitted in real time. Log in to the management system of the relay protection device or download the equipment operation log of the relay protection device through a remote access interface. The equipment operation log usually includes detailed information such as protection action time, triggering conditions, and protection action category.

[0081] The downloaded operation logs of relay protection devices are collated and analyzed, with a focus on the response time and action logic of each protection action, including the judgment logic, delay setting, and specific protection measures implemented when the protection device detects a fault or abnormal condition. Based on the analysis results, the action rules of the relay protection device under different working conditions and its role in power grid protection are determined.

[0082] According to the log analysis results, the action logic rules of the relay protection device are extracted and summarized, and it is clear which fault conditions different types of protection actions correspond to, as well as the specific order and timing of the execution of the protection actions. The action logic rules are combined with the control strategies of safety protection equipment such as circuit breakers and busbar interconnection switches to develop specific plans for collaborative work. For example, when the relay protection device detects a fault and decides to trip a circuit breaker, the switching operation of the adjacent busbar interconnection switches can be preset to be synchronously triggered to quickly isolate the fault area and ensure normal power supply to the non-fault area.

[0083] In addition, it is necessary to upload the formulated collaborative working strategy between relay protection devices and safety protection equipment to the central control system, and make corresponding programming or configuration changes to the control logic of the protection equipment; conduct simulation tests in a safe experimental environment to verify the correctness and effectiveness of the collaborative working strategy under different working conditions, and make necessary adjustments and optimizations based on the test results. After sufficient verification, this collaborative working strategy will be officially enabled in actual power grid operation to ensure that when facing abnormal working conditions, the relay protection device and safety protection equipment can quickly and accurately coordinate the execution of protection actions, thereby maximizing the safe and stable operation of the power grid.

[0084] In summary, the beneficial effects of the embodiments of the present application are:

[0085] 1. Based on a large amount of grid operation data and simulation environment, it can autonomously learn and adjust the optimal control strategy under different working conditions to improve the accuracy and timeliness of protection actions.

[0086] 2. By constructing a state-action space mapping, the optimal control decision can be quickly found in the complex grid state space to adapt to the dynamic changes of the grid, including factors such as new energy grid connection and load fluctuations.

[0087] 3. Deep reinforcement learning methods can integrate global information and realize coordinated optimization control of multiple elements such as generators, transformers, lines, load equipment, energy storage systems, and FACTS devices, thereby improving the overall safety, stability, and operating efficiency of the power grid.

[0088] 4. Based on the local network structure characteristics and dynamic load changes, FACTS devices are used to obtain the charge and discharge status and energy storage boundary of each energy storage unit under different operating conditions; according to the charge and discharge status and energy storage boundary of each energy storage unit, the node voltage in the power grid structure topology diagram is evaluated for stability and the node stability component is obtained; the node stability component is used as the local optimization constraint parameter of the power flow distribution, combined with the deep reinforcement learning environment, to dynamically adjust the control strategy of the FACTS device. The node stability component obtained previously is used as a constraint condition for local optimization and added to the objective function or constraint set of deep reinforcement learning to ensure that while optimizing the control strategy of the FACTS device, the voltage stability of the power grid node is fully guaranteed, thereby effectively enhancing the stability of the power grid and improving the quality and reliability of power supply.

[0089] Embodiment 2

[0090] Based on the same inventive concept as the power grid security protection method combined with deep reinforcement learning in the aforementioned embodiment, Figure 3 As shown, the embodiment of the present application provides a power grid security protection system combined with deep reinforcement learning, wherein the system includes:

[0091] The operation data acquisition module M100 is used to acquire the real-time operation data of each generator based on the N generators in the target area power grid, wherein the real-time operation data includes the generator operation parameters and the generator operation status;

[0092] The control unit setting module M200 is used to generate a power grid structure topology map based on the target area power grid, and obtain N distributed control units according to the power grid structure topology map, taking each of the N generators as a primary central node and comparing each node, line, transformer and load point in the target area power grid;

[0093] The control network setting module M300 is used to connect the N distributed control units based on a star topology structure with the balancing protection unit as the secondary central node to obtain a centralized control network;

[0094] The protection strategy learning module M400 is used to design a deep reinforcement learning environment, and based on the centralized control network, continuously learn the protection strategies under different working conditions to obtain the safety protection model of the target area power grid, and the action space corresponding to the N distributed control units is used as the adjustable control variable cluster of the N generators of the target area power grid;

[0095] The execution module M500 is used to import the real-time operation data into the safety protection model, select the optimal control variable for each generator in the action space, obtain the optimal control instruction set of the target area power grid, and send it to the N distributed control units for execution in real time.

[0096] Furthermore, the protection strategy learning module M400 is used to execute the following method:

[0097] In the deep reinforcement learning environment, determining key operating status indicators of the target area power grid;

[0098] Convert the key operating status indicators into a state vector that can be perceived and processed, and construct a state-action space mapping;

[0099] Based on the state-action space mapping, an action set is designed, wherein the action set includes effective control dimensions corresponding to different operating states of N generators in a target area power grid, and the effective control dimensions are used to construct an action space for functional characteristic configuration.

[0100] Furthermore, the protection strategy learning module M400 is also used to execute the following method:

[0101] Locating key load equipment in the power grid structure topology diagram;

[0102] Perform load classification scheduling based on the key load equipment to obtain the scheduling priority of the key load equipment;

[0103] Load action management is performed through the key load equipment and the scheduling priority.

[0104] Furthermore, the protection strategy learning module M400 is also used to execute the following method:

[0105] Locating the tap position of the transformer in the power grid structure topology diagram;

[0106] According to the tap positions of the transformers, the power flow distribution in the target area power grid is optimized.

[0107] Furthermore, the protection strategy learning module M400 is also used to execute the following method:

[0108] Based on the local network structure characteristics and dynamic load changes, FACTS equipment is used to obtain the charge and discharge status and energy storage boundary of each energy storage unit under different operating conditions;

[0109] According to the charge and discharge status and energy storage boundary of each energy storage unit, the node voltage in the power grid structure topology diagram is evaluated for stability, and the node stability component is obtained;

[0110] The node stability component is used as the local optimization constraint parameter of the power flow distribution, and the control strategy of the FACTS device is dynamically adjusted in combination with the deep reinforcement learning environment.

[0111] Furthermore, the protection strategy learning module M400 is also used to execute the following method:

[0112] Obtaining abnormal operating condition records of the target area power grid;

[0113] Based on the key load equipment, a correlation search is performed in the abnormal operating condition record to extract key abnormal information;

[0114] A safety action is set based on the key abnormal information, and the safety action is added to the action space. The safety protection equipment for executing the safety action includes a circuit breaker and a busbar tie switch.

[0115] Furthermore, the protection strategy learning module M400 is also used to execute the following method:

[0116] A relay protection device connected to the target area power grid;

[0117] Downloading the equipment operation log of the relay protection device, analyzing the response time of the equipment operation log, and extracting the action logic rules of the relay protection device;

[0118] Based on the action logic rules, work in coordination with the safety protection device.

[0119] In summary, any step can be stored as a computer instruction or program in an unlimited computer memory, and can be called and recognized by an unlimited computer processor, and no unnecessary restrictions are made here.

[0120] Furthermore, the above-mentioned technical scheme only reflects the preferred technical scheme of the technical scheme of the embodiment of the present application. Some changes that may be made to some parts thereof by technicians in this technical field all reflect the novel principles of the embodiment of the present application. Obviously, technicians in this field can make various changes and modifications to the present application without departing from the scope of the present application.

Claims

1. A power grid security protection method combined with deep reinforcement learning, characterized in that: The method comprises: Based on N generators in the target area power grid, real-time operation data of each generator is obtained, wherein the real-time operation data includes generator operation parameters and generator operation status; Based on the target regional power grid, a power grid structure topology map is generated, and according to the power grid structure topology map, each of the N generators is used as a primary central node, and each node, line, transformer and load point in the target regional power grid is compared to obtain N distributed control units; Based on a star topology, the balancing protection unit is used as a secondary central node, and the N distributed control units are connected to obtain a centralized control network; A deep reinforcement learning environment is designed to continuously learn protection strategies under different working conditions based on the centralized control network to obtain a safety protection model of the target regional power grid, and the action space corresponding to the N distributed control units is used as an adjustable control variable cluster of the N generators of the target regional power grid; Importing the real-time operation data into the safety protection model, selecting the optimal control variable for each generator in the action space, obtaining the optimal control instruction set of the target regional power grid, and sending it to the N distributed control units for execution in real time; A deep reinforcement learning environment is designed to continuously learn protection strategies under different working conditions based on the centralized control network to obtain a security protection model of the target area power grid. The method includes: In the deep reinforcement learning environment, determining key operating status indicators of the target area power grid; Convert the key operating status indicators into a state vector that can be perceived and processed, and construct a state-action space mapping; Based on the state-action space mapping, an action set is designed, wherein the action set includes effective control dimensions corresponding to different operating states of N generators in the target area power grid, and the effective control dimensions are used to construct an action space for functional characteristic configuration; Based on the state-action space mapping, an action set is designed, wherein the action set includes effective control dimensions corresponding to different operating states of N generators in the target area power grid, and the method further includes: Locating key load equipment in the power grid structure topology diagram; Perform load classification scheduling based on the key load equipment to obtain the scheduling priority of the key load equipment; Load action management is performed through the key load equipment and the scheduling priority.

2. The power grid security protection method combined with deep reinforcement learning as claimed in claim 1, characterized in that: Prior to locating the key load equipment in the power grid structure topology map, the method further includes: Locating the tap position of the transformer in the power grid structure topology diagram; According to the tap positions of the transformers, the power flow distribution in the target area power grid is optimized.

3. The power grid security protection method combined with deep reinforcement learning as claimed in claim 2, characterized in that: The method further comprises: Based on the local network structure characteristics and dynamic load changes, FACTS equipment is used to obtain the charge and discharge status and energy storage boundary of each energy storage unit under different operating conditions; According to the charge and discharge status and energy storage boundary of each energy storage unit, the node voltage in the power grid structure topology diagram is evaluated for stability, and the node stability component is obtained; The node stability component is used as the local optimization constraint parameter of the power flow distribution, and the control strategy of the FACTS device is dynamically adjusted in combination with the deep reinforcement learning environment.

4. The power grid security protection method combined with deep reinforcement learning as claimed in claim 1, characterized in that: Positioning key load equipment in the power grid structure topology map, the method further comprising: Obtaining abnormal operating condition records of the target area power grid; Based on the key load equipment, a correlation search is performed in the abnormal operating condition record to extract key abnormal information; A safety action is set based on the key abnormal information, and the safety action is added to the action space. The safety protection equipment for executing the safety action includes a circuit breaker and a busbar tie switch.

5. The power grid security protection method combined with deep reinforcement learning as claimed in claim 4, characterized in that: The safety protection equipment for executing the safety action includes a circuit breaker and a busbar tie switch, and the method further includes: A relay protection device connected to the target area power grid; Downloading the equipment operation log of the relay protection device, analyzing the response time of the equipment operation log, and extracting the action logic rules of the relay protection device; Based on the action logic rules, work in coordination with the safety protection device.

6. A power grid security protection system combined with deep reinforcement learning, characterized in that: A method for protecting power grid security combined with deep reinforcement learning according to any one of claims 1 to 5, the system comprising: An operation data acquisition module is used to acquire real-time operation data of each generator based on N generators in the target area power grid, wherein the real-time operation data includes generator operation parameters and generator operation status; A control unit setting module is used to generate a power grid structure topology map based on the target area power grid, and according to the power grid structure topology map, take each of the N generators as a primary central node, and compare each node, line, transformer and load point in the target area power grid to obtain N distributed control units; A control network setting module is used to connect the N distributed control units based on a star topology structure with a balancing protection unit as a secondary central node to obtain a centralized control network; A protection strategy learning module is used to design a deep reinforcement learning environment, and based on the centralized control network, continuously learns the protection strategies under different working conditions to obtain a safety protection model of the target area power grid, and the action space corresponding to the N distributed control units is used as an adjustable control variable cluster of the N generators of the target area power grid; An execution module is used to import the real-time operation data into the safety protection model, select the optimal control variable for each generator in the action space, obtain the optimal control instruction set of the target area power grid, and send it to the N distributed control units for execution in real time.

Images