基于生成对抗网络的深度Q学习能耗预测算法  

Deep Q-learning Energy Consumption Prediction Algorithm Based on Generating Adversarial Network

作　　者：刘青松 戴大东 章挺飞 张大龙 LIU Qing-song;DAI Da-dong;ZHANG Ting-fei;ZHANG Da-long(Institute of Electronic and Information Engineering,Suzhou University of Science and Technology,Suzhou 215009,China;Jiangsu Prov⁃ince Key Laboratory of Intelligent Building Enery Efficiency,Suzhou University of Science and Technology,Suzhou,215009,China;Suzhou Key Laboratory of Mobile Networking and Applied Technologies,Suzhou University of Science and Technology,Suzhou 215009,China)

机构地区：[1]苏州科技大学电子与信息工程学院,江苏苏州215009 [2]苏州科技大学江苏省建筑智慧节能重点实验室,江苏苏州215009 [3]苏州科技大学苏州市移动网络技术与应用重点实验室,江苏苏州215009

出　　处：《电脑知识与技术》2019年第11Z期69-71,共3页Computer Knowledge and Technology

摘　　要：针对基于生成对抗网络的Q学习能耗预测算法中,将传统Q学习算法,应用于大状态空间存在收敛速度慢以及非线性条件下能耗预测性能较差的问题,提出一种基于生成对抗网络的深度Q学习能耗预测算法(Deep Q-Learning Energy Consumption Prediction Algorithm Based on Generative Adversarial Networks,DGQL)。该算法引入深度神经网络,通过构建深度Q网络作为非线性函数逼近器去近似表示动作值函数,并利用深度Q网络值函数近似的方法解决传统Q学习算法在大状态空间中算法收敛速度慢的问题。实验结果表明,在引入深度Q网络值函数近似方法后,能耗预测的精度显著提高。Aiming at the problem of Q-learning energy prediction algorithm Based on generating anti-network, the traditional Q-learning algorithm is applied to the problem of slow convergence in large state space and poor performance in energy prediction under nonlinear conditions. Deep Q-Learning Energy consumption Prediction Algorithm Based on Generative Adversarial Networks(DGQL). The algorithm introduces deep neural network, constructs depth Q network as nonlinear function approximator to approximate action value function, and uses depth Q network value function approximation to solve the slow convergence of traditional Q learning algorithm in large state space. problem. The experimental results show that the accuracy of energy consumption prediction is significantly improved after the introduction of the depth Q network value function approximation method.

关 键 词：深度Q学习 生成对抗网络 建筑能耗 函数逼近器 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]