Generative adversarial imitation learning-based dynamic economic dispatching system and method for power system

Abstract

The invention discloses a dynamic economic dispatching system and method for a power system based on generative adversarial imitation learning, and the method comprises the steps: firstly, observing the state of the power system through a generator network module, and generating a dispatching strategy through employing a reinforcement learning near-end strategy optimization algorithm; secondly, the perfect scheduling module generates a perfect scheduling strategy; thirdly, the discriminator network module compares the scheduling strategy generated by the generator network module with a perfect scheduling strategy to obtain feedback information to train generator network parameters and discriminator network parameters; and finally, the generator network module obtains a final scheduling strategy based on the generator network parameters trained by the discriminator network module in combination with the state of the power system. According to the method, the generative adversarial network is combined, the subjectivity introduced by artificially defining a reward function in deep reinforcement learning is avoided, strategy-to-strategy end-to-end learning is realized, the convergence problem of the algorithm is improved, the modeling difficulty is reduced, and the ability of the algorithm to deal with high-dimensional complex problems is enhanced.

Description

technical field [0001] The technical field of electric power system of the present invention specifically relates to a dynamic economic dispatching system and method of electric power system based on generative confrontation imitation learning. Background technique [0002] Building a new power system with a high proportion of new energy is the only way to achieve the goal of "double carbon". However, new energy power generation has the characteristics of intermittent, volatility and randomness, and the uncertainty introduced by large-scale new energy grid connection is bound to bring challenges to the power system dispatching operation. [0003] Based on new energy and load forecasting data, dynamic economic dispatching arranges the output plan of thermal power units to minimize the cost of system power generation within the dispatching cycle. Traditional methods model dynamic economic dispatch as an optimization problem with uncertainty, which can be solved by stochastic ...

AI Technical Summary

Problems solved by technology

Reinforcement learning is a necessary means to improve the effectiveness and flexibility of dynamic economic dispatch in power systems. However, reinforcement learning faces too large a space for action exploration because there is no prior knowledge guidance in the early stage of agent training, resulting in too long time for strategy convergence.

Images