首页|基于LSTM与DDPG的空调能耗优化控制策略

基于LSTM与DDPG的空调能耗优化控制策略

扫码查看
我国建筑能耗快速增长,能源供需矛盾严重.针对空调系统高能耗低能效的问题,论文提出了一种基于长短期记忆网络(LSTM)与深度确定性策略梯度算法(DDPG)的空调能耗优化控制策略.将空调能耗优化问题抽象为强化学习问题,建立空调制冷系统马尔科夫决策过程模型,采用DDPG算法对空调控制策略进行优化.为进一步提升模型训练效率,我们提出基于LSTM算法训练能耗预测模型模拟与智能体交互的环境,解决无模型强化学习方法在线训练收敛时间长的问题.最后,以某工厂工业空调制冷系统为研究对象进行对比实验,实验结果表明,采用论文提出的控制策略模型训练收敛速度快,并且可以在保证系统正常运行的前提下有效降低运行能耗.
Optimal Control Strategy of Air Conditioning Energy Consumption Based on LSTM and DDPG
Building energy consumption is growing rapidly,and the contradiction between energy supply and demand is seri-ous.Aiming at the problem of high energy consumption and low energy efficiency of air conditioning system,this paper proposes an air conditioning control strategy based on LSTM and DDPG.This paper abstracts the energy consumption optimization problem of air conditioning into reinforcement learning problem,establishes the markov decision process model of air conditioning,and uses DDPG algorithm to optimize the control strategy of air conditioning.In order to further improve the model training efficiency,this pa-per proposes a training energy consumption prediction model based on LSTM algorithm to simulate the interaction environment with an agent,and solves the problem of long convergence time for online training without model reinforcement learning method.Finally,a factory air conditioning system is taken as the research object for the experiment,the experimental results show that the algorithm proposed in this paper has fast convergence speed,and can effectively reduce the operation energy consumption.

reinforcement learningneural networksenergy optimizationLSTM-DDPG

王涛、于泽沛、时斌、赵永俊、尹鹏、张思哲

展开 >

中国石油大学(华东)青岛 266555

青岛海尔空调电子有限公司 青岛 266100

强化学习 神经网络 能耗优化 LSTM-DDPG

2024

计算机与数字工程
中国船舶重工集团公司第七0九研究所

计算机与数字工程

CSTPCD
影响因子:0.355
ISSN:1672-9722
年,卷(期):2024.52(11)
  • 1