供水管网实时优化调度深度自注意力强化学习框架

Deep self-attention reinforcement learning framework for real-time optimal scheduling of water distribution network

水泵能耗为水务企业运营中的主要耗能环节,精细化管理水泵运行状况对于水务企业实现绿色低碳发展至关重要。提出了一个基于自注意力机制的深度强化学习框架,能够应对供水管网中需水量的动态变化,提供水泵实时调度方案。通过深度强化学习对具有随机需水量的供水管网环境进行训练,引入了自注意力机制,用于学习隐含的状态重要性,有利于智能体提取状态信息。使用案例管网和启发式算法来验证所提出的模型的有效性。结果表明,基于自注意力机制的深度强化学习,在供水管网水泵实时调度中具有明显优势,能够在短时间内获得优于启发式算法的调度结果,为供水管网实时优化调度提供新的解决思路。

Pump energy consumption is the main energy-consuming part of the daily operation of water enterprises.Fine operation management of pumps is crucial for water enterprises to achieve green and low-carbon development.A deep reinforcement learning framework based on selfattention mechanism is proposed,which can cope with the dynamic change of water demand in the water distribution network and provide a real-time scheduling scheme for pumps.The environment of the water distribution network with random water demand is trained through deep reinforcement learning,and the self-attention mechanism is introduced to learn the implicit state importance,which is conducive to the extraction of state information by agents.Use case network and heuristic algorithm to verify the effectiveness of the proposed model.The results show that the deep reinforcement learning based on the self-attention mechanism has obvious advantages in the real-time scheduling of pumps of the water distribution network,and can obtain better scheduling results in a short time compared to the heuristic algorithm.This framework provides a new solution for the real-time optimal scheduling of water distribution network.