Voltage regulation method and system based on evolutionary learning and deep reinforcement learning

Abstract

The invention discloses a voltage regulation method and system based on evolutionary learning and deep reinforcement learning, and belongs to the field of artificial intelligence and control system cross technology, and the method comprises the steps: obtaining an environment state detected in real time, inputting the environment state into a trained strategy network, and obtaining a voltage regulation strategy; voltage regulation resources are called according to the voltage regulation strategy to complete voltage regulation; the strategy network is trained through the following method: multi-stage progressive multi-node deep reinforcement learning training is performed on the strategy network, evolutionary learning is applied in each stage of training, and the number of the trained strategy networks is doubled through intersection between the trained strategy networks. Performing mutation operation on the trained policy networks in the next stage of performing the interlace operation until the number of the trained policy networks reaches a preset target; each node corresponds to one policy network; the method is suitable for multi-node power distribution network collaborative voltage regulation, promotes the diversity of the network training process, and has strong expansibility.

Description

technical field [0001] The invention relates to a voltage regulation method and system based on evolutionary learning and deep reinforcement learning, and belongs to the cross technical field of artificial intelligence and control systems. Background technique [0002] With the proposal of my country's dual carbon goals and the rapid development of new energy power generation technology, the high proportion of distributed new energy access to the distribution network has brought great challenges to the safe operation of the power grid; the intermittent, random and The imbalance between power generation and load can easily cause the problem of voltage crossing of feeder nodes. The existing distribution network voltage control has the characteristics of local adjustment and dependence on external equipment. Therefore, how to effectively coordinate the active and reactive output of high-proportion new energy sources has important practical significance for grid voltage regulation...

AI Technical Summary

Problems solved by technology

[0002]With the proposal of China's dual carbon targets and the rapid development of new energy power generation technology, the high proportion of distributed new energy access to the distribution network has brought great benefits to the safe operation of the power grid. Big challenge; the intermittent and random nature of distributed new energy output and the imbalance between power generation and load can easily cause the problem of voltage crossing of feeder nodes. The existing distribution network voltage control has the characteristics of local adjustment and dependence on external equipment. With the exhaustion of external regulation resources, it is gradually unable to meet the voltage regulation requirements of the distribution network under the high proportion of decentralized new energy access. Therefore, how to effectively coordinate the high proportion of new energy active and reactive power output has important practical significance for the grid voltage regulation.

[0003]Currently based on traditional distribution network voltage regulation methods, there are mathematical optimization method, intelligent optimization method, stochastic optimization method and model prediction method, etc. Although these methods can However, these methods generally have problems such as large amount of calculation, easy to fall into local optimum, heavy reliance on forecast data, and difficulty in realizing online control. Moreover, the existing distribution network communication infrastructure is poor, and information interaction is limited to some Observable nodes. At the same time, the traditional voltage regulation method has high requirements on the accuracy of the model, which is difficult to apply to the development trend of new forms of high-proportion new energy generation connected to the distribution network; based on the shortcomings of the above traditional voltage regulation methods and the limitations of artificial intelligence technology Development, the latest research uses data-driven deep reinforcement learning methods, including Deep Deterministic Policy Gradient (DDPG), Multi-Agent Deep Deterministic Policy Gradient (MADDPG), Multi-Agent Constrained Soft Actor-Critic (MACSAC) and other methods, in multiple Remarkable results have been achieved in node coordinated voltage regulation. However, these methods have obvious limitations and cannot be applied to large-scale new energy power generation node coordinated voltage regulation scenarios.

Images