Near-end strategy optimization method based on deep reinforcement learning

The invention provides a near-end strategy optimization method based on deep reinforcement learning. The near-end strategy optimization method comprises the following steps: step 1, constructing a multi-agent flexible action evaluation framework based on a deep reinforcement learning algorithm; step 2, utilizing an elasticity enhancement algorithm to carry out elasticity enhancement on the multi-agent flexible action evaluation framework; and step 3, training the multi-agent flexible action evaluation framework after elasticity enhancement, and optimizing a near-end strategy by using the trained multi-agent flexible action evaluation framework after elasticity enhancement. The invention introduces a hybrid flexible action evaluation algorithm based on an intelligent agent, and is used for offline positioning, grading and online control of the parallel reactive power compensator so as to improve the voltage recovery capability of the parallel reactive power compensator. 本发明提出一种基于深度强化学习的近端策略优化方法，包括以下步骤：步骤1：基于深度