基于生成对抗网络的Q学习能耗预测方法

Q Learning Energy Consumption Prediction Algorithms Based on Generative Adversarial Networks

针对大型公共建筑之中实际能耗数据缺乏、能耗预测精度低的问题,提出了一种方法——基于生成对抗网络的Q学习能耗预测算法(Reinforcement Learning Algorithm Based on Generative Adversarial Networks,GAN_RL)。该算法首先将能耗数据转化成时间标记的数据,同时利用生成对抗网络生成部分建筑能耗数据,并将其加入原始能耗数据之中,将前几个时辰的能耗数据通过环境状态进行建模,结合Q学习方法预测后续能耗。该算法采用结合生成对抗网络与Q学习的方法解决了能耗数据不足的问题。实验表明,该算法能有效地预测建筑能耗,预测精度高。

Aiming at the problem of lack of actual energy consumption data and low accuracy of energy consumption prediction in large public buildings,a method based on Q-learning algorithm based on generative adversarial networks(GAN_RL)is proposed.Firstly,the energy consumption data is transformed into time tagged data.At the same time,part of the building energy consump⁃tion data is generated by the generation of countermeasures network and added to the original energy consumption data.The energy consumption data of the first few hours are modeled through the environmental state,and the subsequent energy consumption is pre⁃dicted by combining the Q-learning method.In this algorithm,the problem of energy consumption data shortage is solved by com⁃bining the generation of countermeasure network and Q-learning.Experimental results show that the algorithm can effectively pre⁃dict building energy consumption with high accuracy.