Multi-energy system optimal cooperative operation method based on intelligent agent

Abstract

The invention discloses a multi-energy system optimal cooperative operation method based on an intelligent agent, and relates to the technical field of multi-energy systems; the method includes forming a target network through a pi network and a Q network, and setting a cycle period T; inputting a group of historical data into a target network, determining an action through a pi network, calculating an observation state and a return at a next moment through a multi-energy system physical model, updating parameters of the pi network and a Q network through an r value, and completing offline learning of a DDPG algorithm after T times of circulation; using the observation device to obtain observation data of the DDPG algorithm, and inputting the observation data into a DDPG algorithm which completes offline learning to obtain a decision action, thus completing real-time self-optimization operation of the multi-energy system. According to the invention, the problem that complex modeling needs to be carried out on the coupling relation between physical models in a traditional mathematical algorithm is solved, the action space of a general machine learning algorithm is expanded, and the decision can be closer to the optimal decision.

Description

technical field [0001] The invention relates to the technical field of multi-energy systems, in particular to an intelligent agent-based optimal cooperative operation method for multi-energy systems. Background technique [0002] The multi-energy system integrates various energy forms such as heat, electricity, and gas, and can realize the mutual conversion and complementary utilization of various energy forms. However, due to the uncertainty of load and renewable energy output, as well as the complex energy coupling relationship, the economic operation of multi-energy systems faces major challenges. [0003] The existing multi-energy system optimal scheduling is divided into day-ahead scheduling and real-time scheduling, in which day-ahead scheduling cannot dynamically respond to new energy output and load fluctuations, and it is difficult to obtain optimal scheduling effects. For real-time scheduling, the scheduling method based on model predictive control is generally us...

AI Technical Summary

Problems solved by technology

The DQN algorithm solves the problem of Q value storage by adding a neural network, eliminates the curse of dimensionality, and realizes the expansion to high-dimensional space, but it can only output discrete actions, which loses a lot of action space, resulting in an increase in scheduling costs. Unable to get optimal scheduling

Images