Cluster power grid dispatching method and system

Abstract

The invention discloses a sub-cluster power grid dispatching method and system, and belongs to the technical field of power grid dispatching, and the method comprises the steps: obtaining a local target of a cluster based on cluster benefits, and carrying out the local training of an intelligent agent corresponding to the cluster according to the local target and the historical data of the cluster, and obtaining a process intelligent agent; acquiring a global target based on the comprehensive benefit of the power grid, and acquiring the final contribution of the local target to the global target by using a partially observable Markov decision process; obtaining a global reward function according to the global target, and carrying out global training on the process agent according to the final contribution and the global reward function to obtain an optimized agent; and constructing a global scheduling strategy updating rule through a PPO algorithm, obtaining a global scheduling strategy through an optimization agent under the constraint of the global scheduling strategy updating rule, and scheduling the power grid through the global scheduling strategy. The technical problem that local and global scheduling requirements are difficult to consider in the prior art is solved.

Description

Technical Field

[0001] This invention relates to the field of power grid dispatching technology, specifically to a method and system for dispatching power grids in clusters. Background Technology

[0002] With the large-scale grid connection of distributed renewable energy, energy storage devices, and flexible loads, county-level power grids exhibit characteristics such as strong source-load fluctuations, high resource heterogeneity, and diversified edge control. Traditional centralized power grid dispatching architectures are gradually revealing bottlenecks such as slow response and long control paths in the face of uneven spatial and temporal distribution of resources, increasing real-time requirements, and enhanced autonomous control capabilities. To overcome these bottlenecks, existing technologies introduce reinforcement learning to configure independent agents for each cluster in the power grid. These agents generate independent local dispatching strategies to dispatch the corresponding clusters. However, since agents can only acquire and process local observation data within their jurisdiction and lack the ability to perceive the overall operating status of the power grid, conflicts arise between the local decisions of each agent and the overall decisions of the power grid. Therefore, configuring independent agents for each cluster makes it difficult to simultaneously address both local and global dispatching needs. Summary of the Invention

[0003] To address the technical problem that existing technologies struggle to simultaneously meet both local and global scheduling needs, this invention provides a clustered power grid scheduling method and system. By combining contribution mapping of partially observable Markov decision processes with security constraints of near-end policy optimization algorithms, a learning mechanism is constructed that aligns with global objectives under local information while ensuring the safety of scheduling strategies. Ultimately, this achieves an overall improvement in cluster benefits and the safe and stable operation of the power grid. This solves the technical problem of existing technologies failing to simultaneously meet both local and global scheduling requirements.

[0004] To address the aforementioned technical problems, this invention provides a method for scheduling power grids in clusters, comprising the following steps: Based on the cluster effect, the local target of the cluster is obtained, and the agent corresponding to the cluster is locally trained based on the local target and the historical data of the cluster to obtain the process agent; The global objective is obtained based on the comprehensive benefits of the power grid, and the final contribution of local objectives to the global objective is obtained using a partially observable Markov decision process. Obtain the global reward function based on the global objective, and perform global training on the process agent based on the final contribution and the global reward function to obtain the optimized agent; A global scheduling strategy update rule is constructed using the PPO algorithm. Under the constraint of the global scheduling strategy update rule, the global scheduling strategy is obtained by optimizing the agent, and the power grid is scheduled using the global scheduling strategy.

[0005] By adopting the above technical solution, the present invention has the following advantages: Considering that agents can only acquire and process local observation data within their jurisdiction and lack the ability to perceive the overall operating status of the power grid, while partially observable Markov decision processes (MODs) possess partially observable characteristics, this approach can accurately map the local objectives of the cluster to contributions to the global objectives of the power grid. This overcomes the information barrier that local agents struggle to accurately observe the global picture, and the global system struggles to accurately perceive local influences. By globally training the process agent using the final contribution and global reward function, the agent can accurately learn the impact of changes in local objectives on the global objective. Guided by the global reward function, the optimization direction of the local objective aligns with the optimization direction of the global objective, thus balancing local and global scheduling needs. This solves the technical problem of existing technologies failing to balance local and global scheduling requirements, thereby improving both cluster efficiency and the overall efficiency of the power grid. By constructing global scheduling policy update rules through the PPO algorithm, the acquisition of global scheduling policy is constrained, thereby preventing the global scheduling policy from exceeding the equipment safety limit or system stability margin. By solving the technical problem of the lack of scheduling policy update constraint mechanism in the existing technology, the safe and stable operation of the power grid is guaranteed. By combining partially observable Markov decision processes with the PPO algorithm, the safe and stable operation of the power grid is ensured, while the scheduling efficiency of the power grid cluster and the overall operational benefits of the power grid are also improved.

[0006] Preferably, the step of obtaining the process agent by locally training the agent corresponding to the cluster based on the historical data of the local target and the cluster includes: The agent corresponding to the cluster is obtained based on reinforcement learning, and the local reward function is obtained based on the local objective. The next state of the cluster from the historical data of the cluster is input into the local reward function to obtain the first reward value. The local training dataset is obtained based on the current state of the cluster, the historical local scheduling strategy, the next state of the cluster, and the first reward value from the historical data of the cluster. The local value function of the agent is trained using a local training dataset to obtain an updated agent. The current state of the cluster is input into the updated agent to obtain a second reward value. If the second reward value meets a preset condition, the updated agent is used as the process agent. Otherwise, the local value function of the updated agent is trained using the local training dataset to obtain the process agent.

[0007] In this scheme, the agent's local value function is trained using the cluster's current state, historical local scheduling strategies, next state, and first reward value from historical data. This allows the agent to learn the pattern of "what kind of next state change will be triggered after taking a specific local scheduling strategy in a certain current state, and the corresponding local reward." Based on the learned pattern, the agent can accurately obtain local rewards and autonomously select the local scheduling strategy that maximizes the local reward. This enables the agent to independently generate local scheduling strategies that optimize local objectives, thereby improving the adaptability of local scheduling strategies to cluster benefits.

[0008] Preferably, the step of obtaining the global objective based on the comprehensive benefits of the power grid includes: The overall benefits of the power grid are obtained by considering its energy absorption efficiency, voltage deviation penalties, and regulation costs, and the overall objectives are derived from these benefits.

[0009] Preferably, the method of using a partially observable Markov decision process to obtain the final contribution of the local objective to the global objective includes: The initial contribution of local targets to the global target is obtained based on historical experience from historical data of the power grid, and the initial probability of achieving local targets is obtained based on historical experience. Execute any historical global scheduling strategy from the historical data of the power grid to obtain the observation results. Based on the observation results, use Bayesian rules to correct the initial achievement probability to obtain the first corrected achievement probability. The initial contribution is corrected based on the first corrected achievement probability to obtain a corrected contribution. If the corrected contribution meets the preset requirements, it is taken as the final contribution. Otherwise, the difference between the achievement probability and the direction of achievement probability correction are obtained based on the initial achievement probability and the first corrected achievement probability. The difference between the contribution and the direction of contribution correction are obtained based on the initial contribution and the corrected contribution. The ratio between the difference between the achievement probability and the contribution difference is obtained. The corrected achievement probability value is obtained based on the preset requirements and the ratio. The process achievement probability is obtained based on the corrected achievement probability value, the first corrected achievement probability, and the correlation coefficient representing the correlation between the correction trend of the achievement probability correction direction and the direction of contribution correction. The first corrected achievement probability is taken as the initial achievement probability. The initial achievement probability is corrected using Bayesian rules to obtain the second corrected achievement probability. The average achievement probability is obtained based on the second corrected achievement probability and the process achievement probability. The final contribution is obtained based on the average achievement probability.

[0010] In this scheme, local states are key scenario variables influencing the contribution of local targets to the global target. By obtaining the initial achievement probability of the local state, the contribution is linked to the probability of the cluster state. Bayesian rules are used to correct the initial achievement probability based on observation results to obtain a corrected achievement probability. The initial contribution is then corrected using the corrected achievement probability that fits the cluster scenario, resulting in a corrected contribution. This allows the corrected contribution to be deeply matched with the power grid scenario. Based on this, when the corrected contribution meets preset requirements, it indicates that the corrected contribution accurately reflects the true contribution of the local target to the global target in the power grid scenario, thereby improving the scenario adaptability of the final contribution. When the corrected contribution does not meet the preset requirements, the average achievement probability of the process achievement probability and the second corrected achievement probability is obtained. The final contribution is obtained based on the average achievement probability, thus considering the inherent linkage between achievement probability and contribution, compensating for the lack of correlation in Bayesian rules, and improving the efficiency of obtaining the final contribution. By accurately mapping the local target of the cluster to its contribution to the global power grid target, the information barrier of local agents being unable to accurately observe the global situation and the global situation being unable to accurately perceive the local impact is overcome.

[0011] Preferably, the step of globally training the process agent based on the final contribution and the global reward function to obtain an optimized agent includes: The global target value is obtained based on the final contribution and the local target value. The global target value is then input into the global reward function to obtain the third reward value. The global training dataset is obtained based on the third reward value and the local target value. The global value function of the process agent is trained using a global training dataset to obtain an optimized agent.

[0012] In this scheme, a global training dataset is obtained based on the third reward value and the local target value. The global value function of the process agent is trained using the global training dataset, enabling the process agent to learn the pattern of "the global reward obtained after taking a certain local target value". Based on the learned pattern, the process agent can accurately obtain the global reward and thus autonomously select the global scheduling strategy that maximizes the global reward. This allows the process agent to subsequently obtain the global scheduling strategy in the direction of maximizing the global reward, thereby improving the adaptability of the global scheduling strategy to the comprehensive benefits of the power grid.

[0013] Preferably, the step of constructing the global scheduling policy update rule using the PPO algorithm includes: Using a preset tolerance as a shearing constraint, the global scheduling switching rate is sheared according to the shearing function of the PPO algorithm to obtain the sheared scheduling switching rate. The smaller value between the global scheduling switching rate and the sheared scheduling switching rate is used with the comprehensive benefits of the power grid to obtain the scheduling strategy optimization function. The global scheduling strategy update rule is obtained based on the scheduling strategy optimization function.

[0014] Preferably, obtaining the global scheduling policy update rule based on the scheduling policy optimization function includes: The first scheduling strategy constraint is obtained based on the scheduling strategy optimization function, the power grid state change constraint is obtained based on the power grid security requirements, the second scheduling strategy constraint is obtained based on the power grid state change constraint, and the global scheduling strategy update rule is obtained based on the first scheduling strategy constraint and the second scheduling strategy constraint.

[0015] Preferably, the step of obtaining the global scheduling policy by optimizing the agent under the constraint of the global scheduling policy update rule includes: Candidate local scheduling strategies for the cluster are obtained by optimizing the local value function of the intelligent agent. Under the constraint of the global scheduling strategy update rule, candidate global scheduling strategies for the power grid are obtained based on the global value function of the optimizing intelligent agent. The candidate local target value corresponding to the candidate local scheduling strategy of the cluster is compared with the desired local target value corresponding to the candidate global scheduling strategy of the power grid. If the comparison is successful, the candidate local scheduling strategy that has been successfully compared is adopted as the global scheduling strategy. Otherwise, the local value function or global value function of the optimizing intelligent agent is modified based on the difference between the candidate local target value and the desired local target value, thereby obtaining the global scheduling strategy.

[0016] In this scheme, the candidate global scheduling strategy specifically refers to the desired local target value in order to achieve the global goal. The candidate local target value corresponding to the candidate local scheduling strategy is compared with the desired local target value. The candidate local scheduling strategy that successfully matches the target value is adopted as the global scheduling strategy. This approach not only takes into account both local and global scheduling needs but also ensures the scheduling efficiency of the cluster.

[0017] Preferred options also include: Obtain feedback results of power grid scheduling based on the global scheduling strategy, determine whether the feedback results meet the triggering conditions, and if so, update the optimization agent based on the feedback results to obtain the optimal agent. Obtain the optimal global scheduling strategy through the optimal agent, and schedule the power grid through the optimal global scheduling strategy.

[0018] This invention also provides a clustered power grid dispatching system, applicable to the aforementioned clustered power grid dispatching method, comprising: The local training module is used to obtain local targets of the cluster based on cluster benefits, and to obtain process agents by locally training the agents corresponding to the cluster based on the local targets and the historical data of the cluster. The final contribution acquisition module is used to obtain the global objective based on the comprehensive benefits of the power grid, and to obtain the final contribution of local objectives to the global objective using a partially observable Markov decision process. The global training module is used to obtain the global reward function based on the global objective, and to perform global training on the process agent based on the final contribution and the global reward function to obtain the optimized agent. The scheduling module is used to construct global scheduling policy update rules through the PPO algorithm. Under the constraints of the global scheduling policy update rules, the module optimizes the agent to obtain the global scheduling policy and then schedules the power grid using the global scheduling policy.

[0019] By adopting the above technical solution, the present invention has the following advantages: Considering that agents can only acquire and process local observation data within their jurisdiction and lack the ability to perceive the overall operating status of the power grid, while partially observable Markov decision processes (MODs) possess partially observable characteristics, this approach can accurately map the local objectives of the cluster to contributions to the global objectives of the power grid. This overcomes the information barrier that local agents struggle to accurately observe the global picture, and the global system struggles to accurately perceive local influences. By globally training the process agent using the final contribution and global reward function, the agent can accurately learn the impact of changes in local objectives on the global objective. Guided by the global reward function, the optimization direction of the local objective aligns with the optimization direction of the global objective, thus balancing local and global scheduling needs. This solves the technical problem of existing technologies failing to balance local and global scheduling requirements, thereby improving both cluster efficiency and the overall efficiency of the power grid. Specifically, candidate local scheduling strategies for the cluster are obtained by optimizing the local value function of the agent, and candidate global scheduling strategies for the power grid are obtained based on the global value function of the agent. The candidate global scheduling strategy specifically refers to the desired local target value in order to achieve the global objective. The candidate local target value corresponding to the candidate local scheduling strategy is compared with the desired local target value. The candidate local scheduling strategy that successfully matches is adopted as the global scheduling strategy. This approach not only considers both local and global scheduling needs but also ensures the scheduling efficiency of the cluster. By constructing global scheduling policy update rules through the PPO algorithm, the acquisition of global scheduling policy is constrained, thereby preventing the global scheduling policy from exceeding the equipment safety limit or system stability margin. By solving the technical problem of the lack of scheduling policy update constraint mechanism in the existing technology, the safe and stable operation of the power grid is guaranteed. By combining partially observable Markov decision processes with the PPO algorithm, the safe and stable operation of the power grid is ensured, while the scheduling efficiency of the power grid cluster and the overall operational benefits of the power grid are also improved. Attached Figure Description

[0020] Other features, objects, and advantages of the invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. The drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings.

[0021] Figure 1 This is a flowchart illustrating a clustered power grid scheduling method according to the present invention. Figure 2 This is a flowchart illustrating the generation process of the final contribution in the sub-cluster power grid scheduling method of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only one preferred embodiment of this invention and are only used to explain this invention. They do not limit the scope of protection of this invention. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0023] Before discussing the exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe the operations (or steps) as sequential processes, many of the operations (or steps) can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but it may also have additional steps not included in the figures; the process may correspond to a method, function, procedure, subroutine, subroutine, etc.

[0024] Example 1: like Figure 1 As shown, a method for scheduling power grids by cluster includes the following steps: S1: Obtain the local target of the cluster based on the cluster benefits, and perform local training on the agent corresponding to the cluster based on the local target and the historical data of the cluster to obtain the process agent.

[0025] Specifically, the step of obtaining the process agent by locally training the agent corresponding to the cluster based on the historical data of the local target and the cluster includes: The agent corresponding to the cluster is obtained based on reinforcement learning, and the local reward function is obtained based on the local objective. The next state of the cluster from the historical data of the cluster is input into the local reward function to obtain the first reward value. The local training dataset is obtained based on the current state of the cluster, the historical local scheduling strategy, the next state of the cluster, and the first reward value from the historical data of the cluster. The local value function of the agent is trained using a local training dataset to obtain an updated agent. The current state of the cluster is input into the updated agent to obtain a second reward value. If the second reward value meets a preset condition, the updated agent is used as the process agent. Otherwise, the local value function of the updated agent is trained using the local training dataset to obtain the process agent.

[0026] In this embodiment, the local objectives specifically include maximizing cluster scheduling efficiency and maximizing cluster security. A local reward function is obtained by weighted summation of the local objectives, where security has a greater weight than scheduling efficiency. The calculation method for the sub-items in the local reward function is as follows: if all line load rates are ≤80% and the absolute value of the bus voltage deviation is ≤5%, the security reward is 10; for each line with a load rate exceeding 80% or an absolute value of voltage deviation exceeding 5%, 3 points are deducted, with a minimum of -5. If the scheduling instruction execution delay is ≤30 seconds, the scheduling efficiency reward is 8; for every 10-second delay, 2 points are deducted, with a minimum of 0. The current and next states of the cluster include active power load data, photovoltaic output, energy storage state of charge, node voltage, line load rate, and bus voltage.

[0027] The local value function is a predictor of the agent's cumulative future rewards. It measures the effectiveness of a local scheduling strategy in a given local state, categorized by the potential local reward from the current state. The agent's local value function is trained on a local training dataset, allowing it to learn the pattern of "what kind of next state change will occur after adopting a specific local scheduling strategy in a given current state, and the corresponding reward feedback." If, based on the learned pattern, the second reward value obtained by updating the agent with the current cluster state is greater than or equal to a preset value, the second reward value meets the preset condition. Otherwise, it indicates that the agent's learning of the pattern is insufficient or biased, and the local value function is further trained. Through training, the agent can subsequently adopt local scheduling strategies that maximize local rewards, thereby improving the adaptability of local scheduling strategies to cluster benefits.

[0028] S2: Obtain the global objective based on the comprehensive benefits of the power grid, and use a partially observable Markov decision process to obtain the final contribution of local objectives to the global objective.

[0029] The method of obtaining the overall objective based on the comprehensive benefits of the power grid includes: The overall benefits of the power grid are obtained by considering its energy absorption efficiency, voltage deviation penalties, and regulation costs, and the overall objectives are derived from these benefits.

[0030] In this embodiment, the expression for the global objective, i.e., the joint objective function, is: ; in, This represents the policy network parameters, and the global scheduling policy is learned by optimizing these parameters. This indicates that the policy network parameters are... In this context, the global scheduling strategy refers to the total benefit that the global scheduling action aims to achieve throughout the entire decision-making process. This represents the expectation operation. Indicates the global scheduling cycle. This represents the policy probability of the global scheduling policy. This indicates the global scheduling strategy, such as load transfer amount and energy storage start / stop signals. Indicates the current state of the power grid. This indicates that the overall benefits of the power grid, i.e., the comprehensive benefits corresponding to the global dispatch strategy, are taken into account. , , , These represent the empirical weights of the power grid's energy absorption efficiency, voltage deviation penalty, and regulation cost, respectively, and can be dynamically adjusted. This indicates the increased local consumption of renewable energy after dispatching. This indicates the reduction in node voltage fluctuation after scheduling. This represents the costs incurred in dispatching, such as energy storage conversion losses and equipment start-up and shutdown expenses. Indicates the current state of the power grid. The following global scheduling strategy is adopted. The overall benefits obtained.

[0031] In some embodiments, such as Figure 2 As shown, the method of using a partially observable Markov decision process to obtain the final contribution of the local objective to the global objective includes: S2a: Obtain the initial contribution of local targets to the global target based on historical experience from historical data of the power grid, and obtain the initial probability of local state achieving local target based on historical experience; S2b: Execute any historical global scheduling strategy from the historical data of the power grid to obtain the observation results. Based on the observation results, use Bayesian rules to correct the initial achievement probability and obtain the first corrected achievement probability. S2c: Based on the first corrected achievement probability, the initial contribution is corrected to obtain the corrected contribution. If the corrected contribution meets the preset requirements, the corrected contribution is taken as the final contribution. Otherwise, based on the initial achievement probability and the first corrected achievement probability, the difference in achievement probability and the direction of achievement probability correction are obtained. Based on the initial contribution and the corrected contribution, the difference in contribution and the direction of contribution correction are obtained. The ratio of the difference in achievement probability to the difference in contribution is obtained. Based on the preset requirements and the ratio, the achievement probability correction value is obtained. Based on the achievement probability correction value, the first corrected achievement probability, and the correlation coefficient representing the correlation between the direction of achievement probability correction and the direction of contribution correction, the process achievement probability is obtained. The first corrected achievement probability is taken as the initial achievement probability. The initial achievement probability is corrected using Bayesian rules to obtain the second corrected achievement probability. Based on the second corrected achievement probability and the process achievement probability, the average achievement probability is obtained. Based on the average achievement probability, the final contribution is obtained.

[0032] Partially observable Markov decision processes are a mathematical framework for describing sequential decision-making by agents in environments with incomplete state information. In this embodiment, the final contribution of local objectives to global objectives is obtained through partially observable Markov decision processes, overcoming the information barrier between the cluster and the power grid where local agents cannot accurately observe the global situation and the global situation cannot accurately perceive local influences.

[0033] In this embodiment, taking the power grid of a core area of ​​a prefecture-level city as an example, the power grid includes two key clusters: a commercial center cluster encompassing three 110kV substations and eight 10kV feeders, serving five large commercial complexes with a peak load of 80MW; and a residential community cluster encompassing two 110kV substations and twelve 10kV feeders, serving twelve residential communities with a peak load of 60MW. The overall goal is to optimize the overall efficiency of the power grid. The quantitative indicators for the overall goal are: power supply reliability ≥ 99.99%, comprehensive network loss rate ≤ 3.5%, and renewable energy absorption rate ≥ 92%. The local goals are as follows: for the commercial center cluster, the highest dispatch efficiency (load adjustment response delay ≤ 25 seconds) and the highest security (line load rate ≤ 82%, absolute value of bus voltage deviation ≤ 4%); for the residential community cluster, the highest dispatch efficiency (energy storage charging and discharging response delay ≤ 30 seconds) and the highest security (line load rate ≤ 80%, absolute value of bus voltage deviation ≤ 5%). Local states are defined as key scenario variables affecting contribution. For commercial center clusters, the local states are real-time load fluctuation amplitude and feeder load balance; for residential community clusters, the local states are energy storage SOC level and peak-valley load difference. The initial contribution is obtained by statistically analyzing the improvement of the global target when the cluster achieves its local target, using the formula: Initial Contribution = Single Indicator Improvement Amplitude × Indicator Weight. If the weights of power supply reliability, network loss rate, and absorption rate are 0.4, 0.3, and 0.3 respectively, when the commercial center cluster achieves its local target, the power supply reliability increases by 0.01% and the network loss rate decreases by 0.2%, then the initial contribution of the commercial center cluster is C1 = 0.01% × 0.4 + 0.2% × 0.3 = 0.01. When the residential community cluster achieves its local target, the renewable energy absorption rate increases by 1.5% and the network loss rate decreases by 0.1%, then the initial contribution of the residential community cluster is C2 = 1.5% × 0.3 + 0.1% × 0.3 = 0.0048. The initial achievement probability is determined based on the frequency with which local states support the achievement of local targets in historical data. If, out of 1000 events, the local conditions of "load fluctuation amplitude ≤ 10% and feeder load balance ≥ 0.8" support the achievement of the local target in 650 events, then the initial probability of achieving "load fluctuation amplitude ≤ 10% and feeder load balance ≥ 0.8" is 65%. If, out of 1000 events, the local conditions of "energy storage SOC ≥ 50% and peak-valley load difference ≤ 40MW" support the achievement of the local target in 700 events, then the initial probability of achieving "energy storage SOC ≥ 50% and peak-valley load difference ≤ 40MW" is 70%.After selecting and executing the historical global scheduling strategy, the observation results are collected through the cluster monitoring system. The observation results for the commercial center cluster are "load fluctuation amplitude of 8%, feeder load balance of 0.85 (both meet the support status), scheduling delay of 22 seconds, and load rate of 78% (local target achieved)". The observation results for the residential community cluster are "energy storage SOC of 62%, peak-valley load difference of 35MW (both meet the support status), charging and discharging delay of 28 seconds, and load rate of 76% (local target achieved)". The initial achievement probability is corrected using Bayesian rules as follows: Define the observed event O as a local state meeting the supporting conditions and the local goal being achieved. If P(O|local state meets supporting conditions) = 98% and P(O|local state does not meet supporting conditions) = 15%, then the first corrected achievement probability P(local state meets supporting conditions|O) = [P(O|local state meets supporting conditions) × initial achievement probability] / [P(O|local state meets supporting conditions) × initial achievement probability + P(O|local state does not meet supporting conditions) × (1 - initial achievement probability)]. Therefore, the first corrected achievement probability of the local state in the business center cluster is P1 = (0.98 × 0.65). The first correction probability of the local state in the residential community cluster is P2 = (0.98 × 0.7) / (0.98 × 0.7 + 0.15 × 0.3) ≈ 0.94. The correction contribution is obtained by calculating the correction contribution = initial contribution × (first correction probability / initial probability). Therefore, the correction contribution of the local target in the commercial center cluster is C11 = 0.01 × (0.92 / 0.65) ≈ 0.0142, and the correction contribution of the local target in the residential community cluster is C21 = 0.0048 × (0.94 / 0.7) ≈ 0.0065. It is understandable that when the absolute value of the deviation between the correction contribution and the initial contribution is less than or equal to 0.005, it indicates that the correction contribution meets the preset requirements. In this embodiment, by obtaining the initial achievement probability of the local state, the contribution is linked to the probability of the cluster state. The initial achievement probability is then corrected using Bayesian rules based on the observation results to obtain a first corrected achievement probability. This first corrected achievement probability, tailored to the cluster scenario, is used to correct the initial contribution, resulting in a corrected contribution. This allows the corrected contribution to be deeply matched with the power grid scenario. Furthermore, when the corrected contribution meets preset requirements, it indicates that the corrected contribution accurately reflects the true contribution of the local target to the global target in the power grid scenario, thereby improving the scenario adaptability of the final contribution. By mapping the local target of the cluster to the quantified contribution of the global power grid target, the information barrier of local agents being unable to accurately observe the global situation and the global situation being unable to accurately perceive the local impact is overcome.

[0034] In this embodiment, the correlation coefficient is 1 when the achievement probability correction direction is consistent with the contribution correction direction, and -1 when they are inconsistent. If the contribution increases with an increase in achievement probability or decreases with a decrease in achievement probability, it indicates that the achievement probability correction direction and contribution correction direction are consistent; otherwise, it indicates they are inconsistent. Obtaining the achievement probability correction value according to preset requirements and proportions specifically refers to: obtaining an initial difference based on preset requirements and initial contribution, and obtaining the achievement probability correction value based on the initial difference and proportion. Obtaining the process achievement probability based on the achievement probability correction value, the first corrected achievement probability, and the correlation coefficient representing the correlation between the achievement probability correction direction and the contribution correction direction specifically refers to: Process achievement probability = Correlation coefficient × Achievement probability correction value + First corrected achievement probability. Obtaining the final contribution based on the average achievement probability includes: using the corrected contribution as the initial contribution, using the average achievement probability as the first corrected achievement probability, and iteratively executing S2c. In this embodiment, the average achievement probability of the process achievement probability and the second modified achievement probability is obtained, and the final contribution is obtained based on the average achievement probability. This takes into account the intrinsic linkage between achievement probability and contribution, thereby making up for the lack of correlation in Bayesian rules. It significantly reduces the iterative convergence time of obtaining the final contribution by only relying on Bayesian rules to modify the achievement probability. It also avoids the one-sidedness caused by relying solely on observation. Thus, it improves the efficiency of obtaining the final contribution and also improves the accuracy of the final contribution.

[0035] S3: Obtain the global reward function based on the global objective, and perform global training on the process agent based on the final contribution and the global reward function to obtain the optimized agent.

[0036] The step of globally training the process agent based on the final contribution and the global reward function to obtain an optimized agent includes: The global target value is obtained based on the final contribution and the local target value. The global target value is then input into the global reward function to obtain the third reward value. The global training dataset is obtained based on the third reward value and the local target value. The global value function of the process agent is trained using a global training dataset to obtain an optimized agent.

[0037] In this embodiment, the global value function is a predictor of the agent's future cumulative rewards, used to measure the merits of adopting a certain global scheduling strategy in a given global state. The criterion for merit is how much global reward can be obtained from the current global state. The global value function of the process agent is trained based on the third reward value and the local target value, enabling the process agent to learn the pattern of "the global reward obtained after adopting a certain local target value." Through training, the process agent can subsequently adopt global scheduling strategies in the direction that maximizes the global reward value, thereby improving the adaptability of the global scheduling strategy to the overall benefits of the power grid.

[0038] S4: Construct a global scheduling policy update rule through the PPO algorithm. Under the constraint of the global scheduling policy update rule, obtain the global scheduling policy by optimizing the agent, and schedule the power grid through the global scheduling policy.

[0039] The construction of global scheduling policy update rules using the PPO algorithm includes: Using a preset tolerance as a shearing constraint, the global scheduling switching rate is sheared according to the shearing function of the PPO algorithm to obtain the sheared scheduling switching rate. The smaller value between the global scheduling switching rate and the sheared scheduling switching rate is used with the comprehensive benefits of the power grid to obtain the scheduling strategy optimization function. The global scheduling strategy update rule is obtained based on the scheduling strategy optimization function.

[0040] The core objective of the Proximal Policy Optimization (PPO) algorithm is to efficiently and stably optimize the agent's decision-making policy. By limiting the step size of policy updates, the PPO algorithm ensures that the difference between the new and old policies is not too large, thereby preventing the global scheduling policy from exceeding equipment safety limits or system stability margins. By addressing the technical problem of the lack of scheduling policy update constraint mechanisms in existing technologies, it ensures the safe and stable operation of the power grid. The specific expression of the scheduling policy optimization function is: ; in, Indicates the policy network parameters The optimized value of the scheduling strategy is as follows. Indicates the policy network parameters Global scheduling switching rate under the following conditions Represents the shearing function. This indicates a preset tolerance, which, through the constraint of the shear function, prevents frequency fluctuations and voltage disturbances caused by oscillations in the global scheduling strategy. , This indicates the old global scheduling strategy under the current state of the power grid. Select global scheduling strategy The probability, This indicates the new global scheduling strategy in the current state of the power grid. Select global scheduling strategy The probability of.

[0041] In some embodiments, obtaining the global scheduling policy update rule according to the scheduling policy optimization function includes: The first scheduling strategy constraint is obtained based on the scheduling strategy optimization function, the power grid state change constraint is obtained based on the power grid security requirements, the second scheduling strategy constraint is obtained based on the power grid state change constraint, and the global scheduling strategy update rule is obtained based on the first scheduling strategy constraint and the second scheduling strategy constraint.

[0042] In this embodiment, the first scheduling strategy constraint specifically refers to the range of the global scheduling switching rate. Furthermore, the global scheduling strategy is always updated around the global objective. The second scheduling strategy constraints specifically include: frequency fluctuation not exceeding 50MHz, voltage offset rate not exceeding 5%, and load switching within any cluster not exceeding 10%. If any threshold is violated, the global scheduling strategy is rejected, and the system rolls back to the previous stable global scheduling strategy. By constructing a two-layer global scheduling strategy filtering mechanism, on the one hand, the clip function limits the update amplitude of the global scheduling strategy to avoid oscillations; on the other hand, thresholds for physical indicators such as frequency, voltage, and load switching are set as a safety supervisor to perform post-hoc filtering and rollback judgment on the scheduling results. If any indicator is detected to exceed the safety domain during the scheduling process, the protection mode is automatically triggered and the scheduling authority is downgraded, thereby ensuring the safe and stable operation of the power grid.

[0043] In another embodiment, the step of obtaining the global scheduling policy by optimizing the agent under the constraint of the global scheduling policy update rule includes: Candidate local scheduling strategies for the cluster are obtained by optimizing the local value function of the intelligent agent. Under the constraint of the global scheduling strategy update rule, candidate global scheduling strategies for the power grid are obtained based on the global value function of the optimizing intelligent agent. The candidate local target value corresponding to the candidate local scheduling strategy of the cluster is compared with the desired local target value corresponding to the candidate global scheduling strategy of the power grid. If the comparison is successful, the candidate local scheduling strategy that has been successfully compared is adopted as the global scheduling strategy. Otherwise, the local value function or global value function of the optimizing intelligent agent is modified based on the difference between the candidate local target value and the desired local target value, thereby obtaining the global scheduling strategy.

[0044] In this embodiment, the candidate global scheduling strategy specifically refers to the desired local target value in order to achieve the global objective. The candidate local target value corresponding to the candidate local scheduling strategy is compared with the desired local target value. The candidate local scheduling strategy that successfully matches is adopted as the global scheduling strategy. This approach not only considers both local and global scheduling needs but also ensures the scheduling efficiency of the cluster. Obtaining the global scheduling strategy under the constraints of the global scheduling strategy update rules also guarantees the safety of the power grid.

[0045] Also includes: Obtain feedback results of power grid scheduling based on the global scheduling strategy, determine whether the feedback results meet the triggering conditions, and if so, update the optimization agent based on the feedback results to obtain the optimal agent. Obtain the optimal global scheduling strategy through the optimal agent, and schedule the power grid through the optimal global scheduling strategy.

[0046] The feedback results specifically refer to the difference in active power output, the offset of energy storage SOC, and the voltage deviation rate. The triggering conditions include: 1. Within three consecutive scheduling cycles. 1. The advantage function declines by more than 30%; 2. The energy storage SOC trend reverses for two consecutive cycles (i.e., repeated charging and discharging); 3. The photovoltaic curtailment rate in a single region exceeds a set threshold (e.g., 30%). Once the triggering conditions are met, the feedback results are sent back to the cluster scheduling layer. After being summarized and encoded by the local model, they are sent to the central control layer. If the current scheduling strategy causes a significant decline in the target benefit function, it enters the rapid retraining mode. The central control layer loads the historical training model and supplements it with the most recent feedback data to fine-tune the strategy, completing a rapid strategy repair process. This feedback mechanism can also trigger the main model to refit or only correct and optimize the agent strategy. In addition, it can call historical 30-day data for sample expansion training and use a short window model (e.g., 10 iterations) to converge quickly, which is suitable for seasonal or event-driven scenario changes. Through real-time feedback of strategy operation effects and rapid retraining mechanism, and based on key operation indicators (e.g., decline in advantage function, repeated SOC, surge in curtailment rate) for anomaly identification, it automatically triggers local strategy fine-tuning or historical model refitting, forming an online adaptive scheduling closed loop of strategy-execution-feedback-relearning. It also effectively solves the problem of slow response of traditional global scheduling strategies to seasonal and event-driven sudden changes, and greatly improves the robustness and evolution capability of the system in long-term operation.

[0047] Example 2: This embodiment also provides a clustered power grid dispatching system, applicable to the aforementioned clustered power grid dispatching method, including: The local training module is used to obtain local targets of the cluster based on cluster benefits, and to obtain process agents by locally training the agents corresponding to the cluster based on the local targets and the historical data of the cluster. The final contribution acquisition module is used to obtain the global objective based on the comprehensive benefits of the power grid, and to obtain the final contribution of local objectives to the global objective using a partially observable Markov decision process. The global training module is used to obtain the global reward function based on the global objective, and to perform global training on the process agent based on the final contribution and the global reward function to obtain the optimized agent. The scheduling module is used to construct global scheduling policy update rules through the PPO algorithm. Under the constraints of the global scheduling policy update rules, the module optimizes the agent to obtain the global scheduling policy and then schedules the power grid using the global scheduling policy.

[0048] The final contribution of a local objective to a global objective is obtained using a partially observable Markov decision process. This includes: obtaining an initial contribution of the local objective to the global objective based on historical experience from historical power grid data; obtaining an initial probability of achieving the local objective based on the local state based on historical experience; executing any historical global scheduling strategy from the historical power grid data to obtain observation results; using Bayesian rules to correct the initial probability of achievement based on the observation results to obtain a first corrected probability of achievement; correcting the initial contribution based on the first corrected probability of achievement to obtain a corrected contribution; if the corrected contribution meets a preset requirement, the corrected contribution is taken as the final contribution; otherwise, the final contribution is determined based on the initial probability of achievement and the first corrected probability of achievement. The process involves obtaining the probability difference and correction direction of achievement probability, acquiring the contribution difference and correction direction based on the initial contribution and the corrected contribution, obtaining the ratio of the achievement probability difference to the contribution difference, obtaining the achievement probability correction value according to preset requirements and ratio, obtaining the process achievement probability based on the achievement probability correction value, the first corrected achievement probability, and the correlation coefficient representing the correlation between the achievement probability correction direction and the contribution correction direction, using the first corrected achievement probability as the initial achievement probability, and correcting the initial achievement probability using Bayesian rules to obtain the second corrected achievement probability, obtaining the average achievement probability based on the second corrected achievement probability and the process achievement probability, and obtaining the final contribution based on the average achievement probability. The process agent is then globally trained based on the final contribution and the global reward function to obtain an optimized agent, including: obtaining a global target value based on the final contribution and the local target value, inputting the global target value into the global reward function to obtain a third reward value, obtaining a global training dataset based on the third reward value and the local target value; and training the global value function of the process agent using the global training dataset to obtain an optimized agent. The global scheduling strategy update rules are constructed using the PPO algorithm, including: using a preset tolerance as a shearing constraint, shearing the global scheduling switching rate according to the shearing function of the PPO algorithm to obtain the sheared scheduling switching rate, using the smaller value between the global scheduling switching rate and the sheared scheduling switching rate and the comprehensive benefits of the power grid to obtain the scheduling strategy optimization function, and obtaining the global scheduling strategy update rules based on the scheduling strategy optimization function.

[0049] In this embodiment, considering that the agent can only acquire and process local observation data within its jurisdiction and lacks the ability to perceive the overall operating status of the power grid, while the partially observable Markov decision process has partially observable characteristics, it is possible to accurately map the local objectives of the cluster to the contributions to the global objectives of the power grid through the partially observable Markov decision process. This overcomes the information barrier that local agents cannot accurately observe the global situation and the global system cannot accurately perceive the local impact. By training the process agent globally with the final contribution and the global reward function, the agent can accurately learn the impact of changes in local objectives on the global objectives. Under the guidance of the global reward function, the optimization direction of the local objectives is aligned with the optimization direction of the global objectives, thus taking into account both local and global scheduling needs. This approach addresses the technical challenge of balancing local and global scheduling needs in existing technologies, thereby improving both cluster efficiency and overall grid efficiency. By constructing global scheduling policy update rules using the PPO algorithm, it constrains the acquisition of global scheduling policies, preventing them from exceeding equipment safety limits or system stability margins. This solves the problem of the lack of scheduling policy update constraint mechanisms in existing technologies, thus ensuring the safe and stable operation of the power grid. Furthermore, by combining partially observable Markov decision processes with the PPO algorithm, it not only ensures the safe and stable operation of the power grid but also improves cluster scheduling efficiency and overall grid operational efficiency.

[0050] The specific embodiments described above are preferred embodiments of the distributed power grid dispatching method and system of the present invention, and are not intended to limit the specific scope of the present invention. The scope of the present invention includes, but is not limited to, these specific embodiments. All equivalent changes made in accordance with the shape and structure of the present invention are within the protection scope of the present invention.

Claims

1. A method for dispatching power grids by cluster, characterized in that, Includes the following steps: Based on the cluster effect, the local target of the cluster is obtained, and the agent corresponding to the cluster is locally trained based on the local target and the historical data of the cluster to obtain the process agent; The global objective is obtained based on the comprehensive benefits of the power grid, and the final contribution of local objectives to the global objective is obtained using a partially observable Markov decision process. Obtain the global reward function based on the global objective, and perform global training on the process agent based on the final contribution and the global reward function to obtain the optimized agent; A global scheduling strategy update rule is constructed using the PPO algorithm. Under the constraint of the global scheduling strategy update rule, the global scheduling strategy is obtained by optimizing the agent, and the power grid is scheduled using the global scheduling strategy.

2. The method for dispatching a power grid by sub-cluster according to claim 1, characterized in that, The process of acquiring intelligent agents by locally training the agents corresponding to the cluster based on historical data of local targets and the cluster includes: The agent corresponding to the cluster is obtained based on reinforcement learning, and the local reward function is obtained based on the local objective. The next state of the cluster from the historical data of the cluster is input into the local reward function to obtain the first reward value. The local training dataset is obtained based on the current state of the cluster, the historical local scheduling strategy, the next state of the cluster, and the first reward value from the historical data of the cluster. The local value function of the agent is trained using a local training dataset to obtain an updated agent. The current state of the cluster is input into the updated agent to obtain a second reward value. If the second reward value meets a preset condition, the updated agent is used as the process agent. Otherwise, the local value function of the updated agent is trained using the local training dataset to obtain the process agent.

3. The method for dispatching a power grid by sub-cluster according to claim 1, characterized in that, The method of obtaining the overall objective based on the comprehensive benefits of the power grid includes: The overall benefits of the power grid are obtained by considering its energy absorption efficiency, voltage deviation penalties, and regulation costs, and the overall objectives are derived from these benefits.

4. The method for dispatching a power grid by sub-cluster according to claim 1, characterized in that, The use of partially observable Markov decision processes to obtain the final contribution of local objectives to the global objective includes: The initial contribution of local targets to the global target is obtained based on historical experience from historical data of the power grid, and the initial probability of achieving local targets is obtained based on historical experience. Execute any historical global scheduling strategy from the historical data of the power grid to obtain the observation results. Based on the observation results, use Bayesian rules to correct the initial achievement probability to obtain the first corrected achievement probability. The initial contribution is corrected based on the first corrected achievement probability to obtain a corrected contribution. If the corrected contribution meets the preset requirements, it is taken as the final contribution. Otherwise, the difference between the achievement probability and the direction of achievement probability correction are obtained based on the initial achievement probability and the first corrected achievement probability. The difference between the contribution and the direction of contribution correction are obtained based on the initial contribution and the corrected contribution. The ratio between the difference between the achievement probability and the contribution difference is obtained. The corrected achievement probability value is obtained based on the preset requirements and the ratio. The process achievement probability is obtained based on the corrected achievement probability value, the first corrected achievement probability, and the correlation coefficient representing the correlation between the correction trend of the achievement probability correction direction and the direction of contribution correction. The first corrected achievement probability is taken as the initial achievement probability. The initial achievement probability is corrected using Bayesian rules to obtain the second corrected achievement probability. The average achievement probability is obtained based on the second corrected achievement probability and the process achievement probability. The final contribution is obtained based on the average achievement probability.

5. The method for dispatching a power grid by sub-cluster according to claim 1, characterized in that, The step of globally training the process agent based on the final contribution and the global reward function to obtain an optimized agent includes: The global target value is obtained based on the final contribution and the local target value. The global target value is then input into the global reward function to obtain the third reward value. The global training dataset is obtained based on the third reward value and the local target value. The global value function of the process agent is trained using a global training dataset to obtain an optimized agent.

6. The method for dispatching a power grid by sub-cluster according to claim 1, characterized in that, The construction of global scheduling policy update rules using the PPO algorithm includes: Using a preset tolerance as a shearing constraint, the global scheduling switching rate is sheared according to the shearing function of the PPO algorithm to obtain the sheared scheduling switching rate. The smaller value between the global scheduling switching rate and the sheared scheduling switching rate is used with the comprehensive benefits of the power grid to obtain the scheduling strategy optimization function. The global scheduling strategy update rule is obtained based on the scheduling strategy optimization function.

7. The method for dispatching a power grid cluster according to claim 6, characterized in that, The step of obtaining the global scheduling policy update rule based on the scheduling policy optimization function includes: The first scheduling strategy constraint is obtained based on the scheduling strategy optimization function, the power grid state change constraint is obtained based on the power grid security requirements, the second scheduling strategy constraint is obtained based on the power grid state change constraint, and the global scheduling strategy update rule is obtained based on the first scheduling strategy constraint and the second scheduling strategy constraint.

8. The method for dispatching a power grid by sub-cluster according to claim 1, characterized in that, The step of obtaining the global scheduling policy by optimizing the agent under the constraints of the global scheduling policy update rules includes: Candidate local scheduling strategies for the cluster are obtained by optimizing the local value function of the intelligent agent. Under the constraint of the global scheduling strategy update rule, candidate global scheduling strategies for the power grid are obtained based on the global value function of the optimizing intelligent agent. The candidate local target value corresponding to the candidate local scheduling strategy of the cluster is compared with the desired local target value corresponding to the candidate global scheduling strategy of the power grid. If the comparison is successful, the candidate local scheduling strategy that has been successfully compared is adopted as the global scheduling strategy. Otherwise, the local value function or global value function of the optimizing intelligent agent is modified based on the difference between the candidate local target value and the desired local target value, thereby obtaining the global scheduling strategy.

9. The method for dispatching a power grid by sub-cluster according to claim 1, characterized in that, Also includes: Obtain feedback results of power grid scheduling based on the global scheduling strategy, determine whether the feedback results meet the triggering conditions, and if so, update the optimization agent based on the feedback results to obtain the optimal agent, obtain the optimal global scheduling strategy through the optimal agent, and schedule the power grid through the optimal global scheduling strategy.

10. A sub-cluster power grid dispatching system, applicable to the sub-cluster power grid dispatching method according to any one of claims 1-9, characterized in that, include: The local training module is used to obtain local targets of the cluster based on cluster benefits, and to obtain process agents by locally training the agents corresponding to the cluster based on the local targets and the historical data of the cluster. The final contribution acquisition module is used to obtain the global objective based on the comprehensive benefits of the power grid, and to obtain the final contribution of local objectives to the global objective using a partially observable Markov decision process. The global training module is used to obtain the global reward function based on the global objective, and to perform global training on the process agent based on the final contribution and the global reward function to obtain the optimized agent. The scheduling module is used to construct global scheduling policy update rules through the PPO algorithm. Under the constraints of the global scheduling policy update rules, the module optimizes the agent to obtain the global scheduling policy and then schedules the power grid using the global scheduling policy.

Images