Intelligently-controlled alternating-current power grid reactive power coordination system and method

Abstract

The invention relates to the technical field of electric power system intelligent control, and discloses an intelligent control AC power grid reactive power coordination system and method, and the system comprises a data collection and communication module which is used for collecting the voltage, current, active load, reactive load and distributed energy power generation data of each node in a power grid, and uploading the data to a central control center through a communication network; and the deep reinforcement learning control center generates a reactive power scheduling strategy by using a deep reinforcement learning algorithm based on the power grid state data provided by the data acquisition and communication module. A dynamic reactive power scheduling strategy is generated by introducing a deep reinforcement learning algorithm and utilizing self-learning and self-adaptive capabilities of the deep reinforcement learning algorithm, so that intelligent decision making can be carried out according to real-time data, the limitation that a traditional method depends on a fixed rule is avoided, scheduling is more flexible and efficient, and the scheduling efficiency is improved. And the operation efficiency of the system under load fluctuation and distributed energy access is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent control of electric power systems, and in particular to an intelligently controlled AC power grid reactive power coordination system and method. Background Art

[0002] The rapid development of power systems, especially the integration of renewable energy, has challenged the operating environment and load fluctuations faced by traditional power grids. Effective scheduling of reactive power plays a vital role in ensuring grid voltage stability, improving grid operation efficiency, and reducing reactive power losses. However, existing reactive power scheduling methods mostly rely on fixed rules or static scheduling based on optimization algorithms.

[0003] In traditional power grid dispatching, reactive power dispatching schemes use manual adjustments based on preset rules and models or simple linear optimization algorithms, which have limitations: Traditional reactive power dispatching has a slow response speed to load changes and distributed energy fluctuations, making it difficult to quickly adjust the grid status in real time, resulting in increased voltage fluctuations and reactive power transmission losses; Traditional methods rely on fixed-rule scheduling, lack dynamic learning and adaptive capabilities, and are unable to make flexible decisions based on real-time data, making it difficult to cope with large fluctuations in load and power generation, reducing the scheduling efficiency of reactive power. The existing reactive power dispatching method responds immediately based on the current grid status, lacks the ability to predict future load fluctuation trends, and cannot formulate an optimized reactive power dispatching plan in advance, resulting in low grid operation efficiency; With the expansion of the power grid, especially the large-scale access of distributed energy, the burden on the centralized central control system has gradually increased. It is unable to efficiently handle large-scale, multi-node real-time reactive power scheduling problems, which has reduced the system's response speed and made it difficult to meet the high requirements of modern smart grids for real-time scheduling.

[0004] Therefore, those skilled in the art provide an intelligently controlled AC power grid reactive power coordination system and method to solve the above problems. Summary of the invention

[0005] In view of the deficiencies in the prior art, the present invention provides an intelligently controlled AC power grid reactive power coordination system and method to solve the problems raised in the above background technology.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: an intelligently controlled AC power grid reactive power coordination system, comprising: Data acquisition and communication module, used to collect voltage, current, active load, reactive load and distributed energy generation data of each node in the power grid, and upload it to the central control center through the communication network; The deep reinforcement learning control center is used to generate reactive power dispatch strategies using deep reinforcement learning algorithms based on the voltage, current, active load, reactive load and distributed energy generation data of each node in the power grid provided by the data acquisition and communication module; The reactive power compensation equipment coordination control module is used to control the reactive power compensation equipment according to the reactive power dispatching strategy generated by the deep reinforcement learning control center; The edge computing and distributed collaboration module is deployed at the distributed energy access point, responsible for reactive power regulation of the local power grid and working in collaboration with the central control center; The feedback and self-learning mechanism module is used to collect feedback data on the operation status of the power grid.

[0007] Preferably, the deep reinforcement learning control center optimizes the reactive power dispatching strategy based on a Markov decision process model, and the state space S, action space A and reward function R of the Markov decision process are defined as follows: State space S: includes the voltage of each node in the power grid , reactive power , Active load , reactive load , Active power of distributed energy , reactive power and N represents the number of nodes in the system; , Action space A: includes control instructions for reactive power compensation equipment; , in, Represents the reactive power regulation of capacitors at different nodes, represents the reactive power regulation of the static VAR compensator at different nodes, Represents the reactive power regulation of the static synchronous compensator at different nodes Reward Function :It consists of voltage stability reward, reactive power loss minimization reward and equipment usage balance reward: , in, , , , Among them, among them, represents the penalty term for voltage deviation, represents the voltage of the ith node, is the desired voltage level of the system, the total number of N nodes, represents the penalty term for line loss, is the reactive power loss on the ith node or line, is the penalty term for device operation, is the reactive power adjustment of the jth device, and M represents the total number of adjustable devices in the system.

[0008] Preferably, the deep reinforcement learning control center uses a proximal strategy optimization algorithm to schedule the reactive power scheduling strategy, and the objective function of the proximal strategy optimization algorithm is: , in, Indicates the parameters To maximize, It means solving the expected value at time t. is the probability ratio of the current strategy, Used to measure the current action Relative to the state It indicates how good the action is compared to the average strategy and helps determine whether the probability of the action should be increased or decreased. This function is used to limit The value range of is a hyperparameter of the clipping range.

[0009] Preferably, the reactive power compensation equipment coordination control module includes a capacitor bank control unit, a static VAR compensator control unit, and a static synchronous compensator control unit: The capacitor bank control unit adjusts the reactive power output through switching operations. The output of the capacitor bank changes as follows: , and meet the start-stop frequency constraints , in, represents the reactive power output of the capacitor at time t, Indicates that the capacitor is The reactive power output, Indicates the reactive power adjustment of the capacitor.

[0010] Preferably, the static VAR compensator control unit dynamically adjusts the reactive power output of the static VAR compensator, and the adjustment amount is: , Meet the output range limit , in, represents the reactive power output of the static VAR compensator at time t, Indicates that the static VAR compensator is The reactive power output, Indicates the reactive power adjustment of the static VAR compensator; The static synchronous compensator control unit adjusts the reactive power output of the static synchronous compensator, and the adjustment amount is: , in, represents the reactive power output of the static synchronous compensator at time t, Indicates that the static synchronous compensator is The reactive power output, Represents the reactive power adjustment of the static synchronous compensator, Satisfy output range and response time constraints.

[0011] Preferably, the edge computing and distributed collaboration modules are deployed at key nodes of the power grid and distributed energy access points, which can optimize local reactive power output in real time and work in collaboration with the central control system; The local optimization objectives of the edge computing and distributed collaboration modules are: , Among them, among them, represents the voltage of the ith local node, represents the voltage level expected by the system, represents the total number of local nodes, represents the weight coefficient, represents the reactive power adjustment of the jth device, Indicates the total number of locally adjustable devices.

[0012] Preferably, the system performs short-term prediction of future load through a long short-term memory network, and the prediction model is: , in, It is for the future moment The predicted value of active power, represents the function of the long short-term memory network, Represents a series of historical active power data from time t−k to the current time t.

[0013] Preferably, the feedback and self-learning mechanism module includes: Feedback data collection unit, used to collect real-time operation status data after system scheduling; A self-learning algorithm unit, which is used to update the parameters of the deep reinforcement learning model based on the collected data through the proximal policy optimization algorithm and the gradient descent method; Historical data storage and analysis unit, used to store and analyze historical operation data, predict future loads based on the long-term and short-term memory network model, and provide a basis for long-term trend optimization; The scheduling strategy optimization unit is used to adjust the weight coefficient of the reward function according to the feedback data and update the reactive scheduling strategy.

[0014] Preferably, the feedback data acquisition unit includes a voltage deviation acquisition formula; The voltage deviation acquisition formula is: , in, is the current node voltage, is the target voltage.

[0015] An intelligently controlled AC power grid reactive power coordination method comprises the following steps: Step 1: Collect the voltage of each node in the power grid through the data acquisition and communication module , reactive power , Active load , reactive load , Active power of distributed energy and reactive power , and upload the data to the central control center through the communication network; Step 2: Preprocess the collected data and convert it into the state space of the Markov decision process ; Step 3: The central control center generates a dispatching strategy for reactive power compensation equipment based on the status data through a deep reinforcement learning algorithm; Step 4: Control the reactive power compensation equipment according to the generated dispatching strategy; Step 5: The edge computing and distributed collaboration modules perform local optimization on the distributed energy access points; Step 6: Through the feedback and self-learning mechanism module, feedback data such as voltage deviation and reactive power loss are collected in real time.

[0016] The present invention provides an intelligently controlled AC power grid reactive power coordination system and method, which has the following beneficial effects: 1. The present invention introduces a deep reinforcement learning algorithm and utilizes its self-learning and adaptive capabilities to generate a dynamic reactive power dispatching strategy, which can then make intelligent decisions based on real-time data, avoiding the limitations of traditional methods that rely on fixed rules, making dispatching more flexible and efficient, and improving the operating efficiency of the system under load fluctuations and distributed energy access.

[0017] 2. The present invention realizes the ability to adjust the dispatching strategy in real time according to the grid feedback data through the feedback and self-learning mechanism module. The self-learning mechanism enables the system to adapt to different operating environments and gradually optimize the future dispatching strategy by adjusting the parameters of the deep reinforcement learning model. Its dynamic optimization capability significantly improves the robustness of the system in dealing with complex and changing environments, especially under the situation of rapid load changes and distributed energy fluctuations, and can quickly recover and adjust.

[0018] 3. The present invention introduces edge computing and distributed collaborative modules to perform local reactive power regulation at distributed energy access points, reducing dependence on the central control system. The introduction of edge computing greatly improves the response speed of the system, especially when the local power grid load fluctuates. The edge nodes can autonomously adjust the reactive power. The distributed collaborative optimization technology is relatively rare in the prior art and is a significant improvement in the system in terms of rapid response and real-time scheduling.

[0019] 4. The present invention uses a long short-term memory network model to predict future loads, which can predict the load change trend of the power grid and the power generation fluctuation of distributed energy in advance, and perform scheduling optimization based on this. Its forward-looking scheduling method improves the accuracy of current scheduling, provides a basis for the long-term optimization of the system, and thus improves the operating efficiency and stability of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a system framework diagram of the present invention; Figure 2 It is a schematic diagram of the feedback and self-learning mechanism module of the present invention; Figure 3 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0021] In order to make the technical personnel in the technical field understand the scheme of the present invention, the technical scheme in the embodiment of the present invention will be clearly and completely described below in combination with the drawings in the embodiment of the present invention. Obviously, the described embodiment is a partial embodiment of the present invention, not a complete embodiment. Based on the embodiment of the present invention, other embodiments obtained by ordinary technicians in the field without creative work should fall within the scope of protection of the present invention.

[0022] The present invention is described in detail below in conjunction with the accompanying drawings: Example: Please see attached Figure 1 -Attached Figure 3 The embodiment of the present invention provides an intelligently controlled AC power grid reactive power coordination system, comprising: Data acquisition and communication module, used to collect voltage, current, active load, reactive load and distributed energy generation data of each node in the power grid, and upload it to the central control center through the communication network; The deep reinforcement learning control center is used to generate reactive power dispatch strategies using deep reinforcement learning algorithms based on the voltage, current, active load, reactive load and distributed energy generation data of each node in the power grid provided by the data acquisition and communication module; The reactive power compensation equipment coordination control module is used to control the reactive power compensation equipment according to the reactive power dispatching strategy generated by the deep reinforcement learning control center; The edge computing and distributed collaboration module is deployed at the distributed energy access point, responsible for reactive power regulation of the local power grid and working in collaboration with the central control center; The feedback and self-learning mechanism module is used to collect feedback data on the operation status of the power grid.

[0023] Benefits of data acquisition and communication module Real-time and accurate data collection of voltage, current, active load, reactive load and distributed energy generation data of each node in the power grid, providing basic data support for the reactive power dispatch of the system, and the data can be quickly transmitted to the central control center, so that the system can respond promptly according to the latest power grid status, effectively ensuring the real-time and reliability of data transmission, and ensuring the accuracy of subsequent dispatch decisions; Benefits of Deep Reinforcement Learning Control Center The system has self-learning and adaptive capabilities, and can dynamically generate the optimal reactive power dispatch strategy according to the real-time status of the power grid. Compared with the traditional static dispatch method, deep reinforcement learning can adjust the strategy in real time, avoid relying on fixed rules, and make the dispatch more flexible and efficient, especially when the load fluctuates frequently or distributed energy is connected, it can better optimize the operation of the power grid and improve the overall operation efficiency; Benefits of the coordinated control module for reactive power compensation equipment: It can realize intelligent control of reactive power compensation equipment, accurately adjust reactive power output according to the dispatching strategy generated by deep reinforcement learning, and then improve the voltage stability of the power grid, reduce reactive power transmission loss, and effectively improve the quality of power. In addition, the intelligent coordinated control of equipment can balance the use of equipment, extend the service life of equipment, and reduce maintenance costs; The benefits of edge computing and distributed collaborative modules improve the response speed of the system, especially at the distributed energy access point. The edge computing node can quickly process the reactive power regulation needs of the local power grid, reduce the pressure on the central control system, realize global and local reactive power dispatch optimization, and effectively improve the system's regulation ability and dispatch efficiency when the local load fluctuates, so that the system can respond to dynamic changes in the power grid faster and more effectively; The benefit of the feedback and self-learning mechanism module is that it provides a basis for dynamic optimization of the deep reinforcement learning model, continuously updates the dispatch strategy according to the feedback data, improves the accuracy and adaptability of the dispatch, and enables the system to have self-optimization and self-adaptation capabilities. When the load changes sharply and distributed energy fluctuates, the system can adjust quickly to improve the robustness and stability of the power grid operation.

[0024] The deep reinforcement learning control center optimizes the reactive power dispatch strategy based on the Markov decision process model. The state space S, action space A and reward function R of the Markov decision process are defined as follows: State space S: includes the voltage of each node in the power grid , reactive power , Active load , reactive load , Active power of distributed energy , reactive power and N represents the number of nodes in the system; , Action space A: includes control instructions for reactive power compensation equipment; , in, Represents the reactive power regulation of capacitors at different nodes, represents the reactive power regulation of the static VAR compensator at different nodes, Represents the reactive power regulation of the static synchronous compensator at different nodes Reward Function :It consists of voltage stability reward, reactive power loss minimization reward and equipment usage balance reward: , in, , , , Among them, among them, represents the penalty term for voltage deviation, represents the voltage of the ith node, is the desired voltage level of the system, the total number of N nodes, represents the penalty term for line loss, is the reactive power loss on the ith node or line, is the penalty term for device operation, is the reactive power adjustment of the jth device, and M represents the total number of adjustable devices in the system.

[0025] The state space defines the current operating conditions of the power grid. By comprehensively collecting various types of data in the power grid, the system can evaluate the real-time operating conditions of the power grid and generate reactive power dispatch strategies based on this. The state space, as the input of the deep reinforcement learning model, directly affects the accuracy of the strategy and the system's ability to perceive complex power grid dynamics. The action space defines the control output of the system, that is, the specific adjustment operation of the reactive compensation equipment. By generating appropriate control instructions, the system can dynamically adjust the state of the reactive compensation equipment to ensure the voltage stability and reactive power balance of the power grid. The determination of each action directly affects the operation quality of the power grid. Especially when the load fluctuates or the output of distributed energy changes, the control instructions in the action space can respond to changes quickly. The reward function is used to guide the training of the deep reinforcement learning model and measure the quality of the current strategy. By optimizing the value of the reward function, the system can generate a better dispatching strategy. The three indicators of voltage stability, minimization of reactive power loss and balance of equipment use cover the main operating objectives of the power grid, ensuring that the system optimizes the dispatching efficiency and equipment usage while meeting the operating requirements.

[0026] The deep reinforcement learning control center uses the proximal strategy optimization algorithm to schedule the reactive power scheduling strategy. The objective function of the proximal strategy optimization algorithm is: , in, Indicates the parameters To maximize, It means solving the expected value at time t. is the probability ratio of the current strategy, Used to measure the current action Relative to the state It indicates how good the action is compared to the average strategy and helps determine whether the probability of the action should be increased or decreased. This function is used to limit The value range of is a hyperparameter of the clipping range.

[0027] Ratio of current policy to old policy Control the update range of the current strategy relative to the old strategy to ensure the stability of the strategy update; Advantage function Used to measure the pros and cons of an action relative to other actions, helping the strategy determine the optimal action; The pruning mechanism limits the range of policy updates, prevents policy changes, and ensures the smoothness of updates; While maintaining learning efficiency, control the stability of updates to ensure that strategy divergence does not occur.

[0028] The coordinated control module of reactive power compensation equipment includes a capacitor bank control unit, a static reactive power compensator control unit, and a static synchronous compensator control unit: The capacitor bank control unit adjusts the reactive power output through switching operations. The output of the capacitor bank changes as follows: , and meet the start-stop frequency constraints , in, represents the reactive power output of the capacitor at time t, Indicates that the capacitor is The reactive power output, Indicates the reactive power adjustment of the capacitor.

[0029] Preferably, the static VAR compensator control unit dynamically adjusts the reactive power output of the static VAR compensator, and the adjustment amount is: , Meet the output range limit , in, represents the reactive power output of the static VAR compensator at time t, Indicates that the static VAR compensator is The reactive power output, Indicates the reactive power adjustment of the static VAR compensator; The static synchronous compensator control unit adjusts the reactive power output of the static synchronous compensator. The adjustment amount is: , in, represents the reactive power output of the static synchronous compensator at time t, Indicates that the static synchronous compensator is The reactive power output, Represents the reactive power adjustment of the static synchronous compensator, Satisfy output range and response time constraints.

[0030] Edge computing and distributed collaboration modules are deployed at key nodes of the power grid and distributed energy access points, which can optimize local reactive power output in real time and work in collaboration with the central control system. The local optimization goals of edge computing and distributed collaboration modules are: , Among them, among them, represents the voltage of the ith local node, represents the voltage level expected by the system, represents the total number of local nodes, represents the weight coefficient, represents the reactive power adjustment of the jth device, Indicates the total number of locally adjustable devices.

[0031] The voltage deviation term is used to minimize the voltage deviation of each node in the local power grid. By minimizing the difference between the actual voltage and the reference voltage, the system can ensure the voltage stability of the local power grid, which is particularly important when the load fluctuates and distributed energy is connected. Minimizing the voltage deviation helps maintain the overall voltage level of the power grid and prevent the voltage from being too high or too low. The function of the reactive power compensation equipment usage balance item is to balance the usage frequency of the reactive power compensation equipment to prevent a certain equipment from being frequently used, resulting in wear or reduced efficiency; Optimize the overall usage frequency of reactive power compensation equipment to prevent over-reliance on a certain device, thereby extending the life of the equipment, reducing the maintenance cost of the equipment, and ensuring balanced use of the equipment; The role of the overall optimization objective is to adjust the voltage of each node in the local power grid to the reference voltage, ensure the voltage stability of the local power grid, avoid the impact of excessive voltage deviation on the power grid, and on the basis of ensuring voltage stability, through the penalty factor Balance the use of reactive power compensation equipment, avoid excessive use of equipment, extend the service life of equipment, and maintain the overall operating efficiency of reactive power compensation equipment; Edge computing nodes can make autonomous decisions locally to ensure the voltage stability of the power grid, optimize the use of reactive power compensation equipment, and avoid overuse or underuse of equipment. At the same time, this local optimization works in conjunction with the central control system to form a distributed optimized smart grid dispatching system.

[0032] The system uses a long short-term memory network to make short-term predictions of future loads. The prediction model is: , in, It is for the future moment The predicted value of active power, represents the function of the long short-term memory network, Represents a series of historical active power data from time t−k to the current time t.

[0033] By accurately predicting the changing trend of load, the power grid dispatching system can arrange the operation of reactive power compensation equipment more reasonably to ensure the stable operation of the power grid under high or low load conditions; The long short-term memory network model is particularly suitable for processing and predicting sequence data. It can capture the long-term dependencies in historical load data and overcome the gradient vanishing problem that traditional neural networks are prone to encounter when processing time series. The model effectively saves and updates information through memory units and gating mechanisms, ensuring that when performing load forecasting, the system can take into account the impact of past load changes on future loads, thereby improving the accuracy of forecasts. By accurately predicting future loads, the system can formulate reactive power dispatch strategies in advance to cope with load changes. The output of the prediction model can directly affect the dispatch strategy generated by the deep reinforcement learning control center, making it more forward-looking and accurate. Before the load peak arrives, the system can adjust the output of reactive power compensation equipment in advance to avoid voltage fluctuations and low power supply caused by sudden load increase, thereby improving the operating efficiency and reliability of the power grid; The long short-term memory network model is trained based on historical load data. The prediction results generated by the model are used to promote the intelligence of the decision-making process. By combining historical data with real-time predictions, the system can achieve dynamic adjustment and continuously optimize reactive power scheduling, thereby improving the flexibility and adaptability of the overall power grid.

[0034] The feedback and self-learning mechanism modules include: Feedback data collection unit, used to collect real-time operation status data after system scheduling; A self-learning algorithm unit, which is used to update the parameters of the deep reinforcement learning model based on the collected data through the proximal policy optimization algorithm and the gradient descent method; Historical data storage and analysis unit, used to store and analyze historical operation data, predict future loads based on the long-term and short-term memory network model, and provide a basis for long-term trend optimization; The scheduling strategy optimization unit is used to adjust the weight coefficient of the reward function according to the feedback data and update the reactive scheduling strategy.

[0035] Benefits of the Feedback Data Acquisition Unit The system can obtain real-time feedback, promptly detect the implementation effect of the dispatch strategy, help identify potential problems and adjust demand, make the operation status of the power grid more transparent, and improve the response speed and flexibility of the system; The benefits of the self-learning algorithm unit make the system adaptive and can optimize the dispatch strategy according to the real-time operation status of the power grid and historical experience. Its self-learning ability ensures that the system continues to improve in a complex and changing environment, thereby improving the efficiency and accuracy of dispatch; The benefits of the historical data storage and analysis unit are to predict future loads and help the system identify long-term trends and patterns. The process enhances the system's use of historical data and provides a more scientific basis for power grid dispatching. The benefits of the dispatch strategy optimization unit enable the system to balance multiple objectives such as voltage stability, minimization of reactive power losses and equipment utilization efficiency, and flexibly adjust strategies based on real-time feedback to ensure the best results in the dispatch process.

[0036] The feedback data acquisition unit includes a voltage deviation acquisition formula; Voltage deviation acquisition formula: , in, is the current node voltage, is the target voltage.

[0037] The function of the voltage deviation acquisition formula is to calculate the deviation of the voltage of each node in the power grid relative to the target voltage. By quantifying the voltage deviation, the system can monitor in real time whether the voltage is within a reasonable range. The calculation of voltage deviation enables the system to quickly identify voltage fluctuations and take timely measures to adjust them. Benefits of the voltage deviation collection formula The system can monitor voltage stability in real time and identify potential voltage problems. The real-time data of voltage deviation provides feedback to the deep reinforcement learning control center, so that the dispatching strategy can be adjusted quickly to avoid voltage being too high or too low. The reactive power loss calculation formula calculates the reactive power transmission loss of the lines in the power grid. By evaluating the total reactive power loss, the system can quantify the energy loss caused by reactive power transmission, providing important data support for subsequent dispatch optimization; The benefits of the reactive power loss calculation formula are that the system can clearly understand the transmission loss of reactive power in the power grid, evaluate the performance of the power grid, identify the lines with high reactive power loss, and take corresponding measures to optimize scheduling, reduce unnecessary energy consumption, and improve the overall power grid operation efficiency, so as to give priority to the lines with higher reactive power loss in scheduling and ensure the economy and safety of the power grid.

[0038] An intelligently controlled AC power grid reactive power coordination method comprises the following steps: Step 1: Collect the voltage of each node in the power grid through the data acquisition and communication module , reactive power , Active load , reactive load , Active power of distributed energy and reactive power , and upload the data to the central control center through the communication network; Step 2: Preprocess the collected data and convert it into the state space of the Markov decision process ; Step 3: The central control center generates a dispatching strategy for reactive power compensation equipment based on the status data through a deep reinforcement learning algorithm; Step 4: Control the reactive power compensation equipment according to the generated dispatching strategy; Step 5: The edge computing and distributed collaboration modules perform local optimization on the distributed energy access points; Step 6: Through the feedback and self-learning mechanism module, feedback data such as voltage deviation and reactive power loss are collected in real time.

[0039] The role of step 1 is to ensure the timeliness and accuracy of information, and the process provides real-time data support for the reactive power dispatch of the system; The role of step 2 is to provide standardized input for the deep reinforcement learning algorithm to facilitate subsequent decision making. The definition of the state space enables the system to fully and intuitively understand the current state of the power grid; The role of step 3 is to utilize the adaptive and self-learning capabilities of deep reinforcement learning. Strategy generation can be adjusted dynamically, and the best scheduling plan can be formulated based on historical data and current status, thereby improving the flexibility and accuracy of scheduling. The effect of step 4 is that the system can maintain the voltage stability of the power grid, reduce reactive power loss, improve the quality of power, and ensure the safe operation of the power grid; The role of step 5 is to quickly respond to local load changes, reduce the burden on the central control system, and improve the overall response speed and efficiency of the system; The role of step 6 is to ensure that the system can make adaptive adjustments based on the latest grid status and improve the accuracy and flexibility of future dispatch decisions.

[0040] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligently controlled AC power grid reactive power coordination system, characterized in that: include: Data acquisition and communication module, used to collect voltage, current, active load, reactive load and distributed energy generation data of each node in the power grid, and upload it to the central control center through the communication network; The deep reinforcement learning control center is used to generate reactive power dispatch strategies using deep reinforcement learning algorithms based on the voltage, current, active load, reactive load and distributed energy generation data of each node in the power grid provided by the data acquisition and communication module; The reactive power compensation equipment coordination control module is used to control the reactive power compensation equipment according to the reactive power dispatching strategy generated by the deep reinforcement learning control center; The edge computing and distributed collaboration module is deployed at the distributed energy access point, responsible for reactive power regulation of the local power grid and working in collaboration with the central control center; The feedback and self-learning mechanism module is used to collect feedback data on the operation status of the power grid.

2. The intelligently controlled AC power grid reactive power coordination system according to claim 1, characterized in that: The deep reinforcement learning control center optimizes the reactive power dispatch strategy based on the Markov decision process model. The state space S, action space A and reward function R of the Markov decision process are defined as follows: State space S: includes the voltage of each node in the power grid , reactive power , Active load , reactive load , Active power of distributed energy , reactive power and N represents the number of nodes in the system; ， Action space A: includes control instructions for reactive power compensation equipment; ， in, Represents the reactive power regulation of capacitors at different nodes, represents the reactive power regulation of the static VAR compensator at different nodes, Indicates the reactive power regulation of the static synchronous compensator at different nodes; Reward Function :It consists of voltage stability reward, reactive power loss minimization reward and equipment usage balance reward: ， in, , , ， Among them, among them, represents the penalty term for voltage deviation, represents the voltage of the ith node, is the desired voltage level of the system, the total number of N nodes, represents the penalty term for line loss, is the reactive power loss on the ith node or line, is the penalty term for device operation, is the reactive power adjustment of the jth device, and M represents the total number of adjustable devices in the system.

3. The intelligently controlled AC power grid reactive power coordination system according to claim 1, characterized in that: The deep reinforcement learning control center uses a proximal strategy optimization algorithm to schedule the reactive power scheduling strategy. The objective function of the proximal strategy optimization algorithm is: ， in, Indicates the parameters To maximize, It means solving the expected value at time t. is the probability ratio of the current strategy, Used to measure the current action Relative to the state The goodness of the average behavior. It indicates how much better the action is than the average strategy, which helps determine whether the probability of the action should be increased or decreased; This function is used to limit The value range of is a hyperparameter of the clipping range.

4. The intelligently controlled AC power grid reactive power coordination system according to claim 1, characterized in that: The reactive power compensation equipment coordination control module includes a capacitor bank control unit, a static VAR compensator control unit, and a static synchronous compensator control unit: The capacitor bank control unit adjusts the reactive power output through switching operations. The output of the capacitor bank changes as follows: , and meet the start-stop frequency constraints , in, represents the reactive power output of the capacitor at time t, Indicates that the capacitor is The reactive power output, Indicates the reactive power adjustment of the capacitor.

5. The intelligently controlled AC power grid reactive power coordination system according to claim 4, characterized in that: The static VAR compensator control unit dynamically adjusts the reactive power output of the static VAR compensator, and the adjustment amount is: ， Meet the output range limit , in, represents the reactive power output of the static VAR compensator at time t, Indicates that the static VAR compensator is The reactive power output, Indicates the reactive power adjustment of the static VAR compensator; The static synchronous compensator control unit adjusts the reactive power output of the static synchronous compensator, and the adjustment amount is: , in, represents the reactive power output of the static synchronous compensator at time t, Indicates that the static synchronous compensator is The reactive power output, Indicates the reactive power adjustment of the static synchronous compensator; Satisfy output range and response time constraints.

6. The intelligently controlled AC power grid reactive power coordination system according to claim 1, characterized in that: The edge computing and distributed collaboration modules are deployed at key nodes of the power grid and distributed energy access points, and can optimize local reactive power output in real time and work in collaboration with the central control system; The local optimization objectives of the edge computing and distributed collaboration modules are: ， Among them, among them, represents the voltage of the ith local node, represents the voltage level expected by the system, represents the total number of local nodes, represents the weight coefficient, represents the reactive power adjustment of the jth device, Indicates the total number of locally adjustable devices.

7. The intelligently controlled AC power grid reactive power coordination system according to claim 1, characterized in that: The system uses a long short-term memory network to make short-term predictions of future loads. The prediction model is: ， in, It is for the future moment The predicted value of active power, represents the function of the long short-term memory network, Represents a series of historical active power data from time t−k to the current time t.

8. The intelligently controlled AC power grid reactive power coordination system according to claim 1, characterized in that: The feedback and self-learning mechanism module includes: Feedback data collection unit, used to collect real-time operation status data after system scheduling; A self-learning algorithm unit, which is used to update the parameters of the deep reinforcement learning model based on the collected data through the proximal policy optimization algorithm and the gradient descent method; Historical data storage and analysis unit, used to store and analyze historical operation data, predict future loads based on the long-term and short-term memory network model, and provide a basis for long-term trend optimization; The scheduling strategy optimization unit is used to adjust the weight coefficient of the reward function according to the feedback data and update the reactive scheduling strategy.

9. The intelligently controlled AC power grid reactive power coordination system according to claim 8, characterized in that: The feedback data acquisition unit includes a voltage deviation acquisition formula; The voltage deviation acquisition formula is: , in, is the current node voltage, is the target voltage.

10. An intelligently controlled AC power grid reactive power coordination method, according to an intelligently controlled AC power grid reactive power coordination system according to any one of claims 1 to 8, characterized in that: The steps include: Step 1: Collect the voltage of each node in the power grid through the data acquisition and communication module , reactive power , Active load , reactive load , Active power of distributed energy and reactive power , and upload the data to the central control center through the communication network; Step 2: Preprocess the collected data and convert it into the state space of the Markov decision process ; Step 3: The central control center generates a dispatching strategy for reactive power compensation equipment based on the status data through a deep reinforcement learning algorithm; Step 4: Control the reactive power compensation equipment according to the generated dispatching strategy; Step 5: The edge computing and distributed collaboration modules perform local optimization on the distributed energy access points; Step 6: Through the feedback and self-learning mechanism module, feedback data such as voltage deviation and reactive power loss are collected in real time.

Images