基于能耗预测的空调测试任务调度问题研究  

Research on Air Conditioning Test Task Scheduling Problem Based on Energy Consumption Prediction

作　　者：张恪 赖信君 黎展滔[1] 林深和 陈庆新[1] 毛宁[1] 李鑫 Zhang Ke;Lai Xinjun;Li Zhantao;Lin Shenhe;Chen Qingxin;Mao Ning;Li Xin(College of Electro-Mechanical Engineering,Guangdong University of Technology,Guangzhou 510006,China;Zhuhai Urban Construction Smart Energy Co.,Ltd.,Zhuhai,Guangdong 519070,China)

机构地区：[1]广东工业大学机电工程学院,广州510006 [2]珠海城建智慧能源有限公司,广东珠海519070

出　　处：《机电工程技术》2021年第10期56-63,共8页Mechanical & Electrical Engineering Technology

基　　金：国家自然科学基金面上项目(编号:51675107);广东省自然科学基金项目(编号:2021A1515012015);广东工业大学青年百人计划科研启动经费项目(编号:220413637)。

摘　　要：新空调投放到市场前需针对不同环境工况进行大量焓差实验,带来极大的能源消耗。针对焓差实验测试企业,利用机器学习模型进行能耗预测,以最小化电费为优化目标,构建分时电价政策下的空调测试任务调度问题的混合整数规划模型。针对调整工况的时间、耗电率与测试任务顺序相关且无法用常规分布函数描述的情况,构建一种基于随机森林回归的预测模型。然后,针对预测模型和数学规划分阶段求解思路的不足,提出一种基于随机森林与遗传算法的RF-GA混合算法求解思路,该思路采用并行处理方式提高了算法求解效率。通过企业实际测试数据将随机森林预测模型与其他机器学习算法进行对比,验证了该预测模型的有效性,并设计仿真实验产生各种规模的测试算例,将RF-GA算法及Inver-over算子与多种算法及算子进行对比,实验结果表明该RF-GA算法在求解效果和效率方面具有较优的表现。将该方法应用于企业实际测试,可大幅降低能源消耗及电费支出。Before the new air conditioner is put on the market,a large number of enthalpy difference experiments need to be carried out under different environmental conditions,resulting in great energy consumption.Aiming at the enthalpy difference experimental test enterprises,the machine learning model was used to predict energy consumption,and the mixed integer programming model of air conditioning test task scheduling problem under the trace operate unit price policy was constructed with the optimization goal of minimizing electricity charge.Aiming at the situation that the time and power consumption rate of adjusting working conditions were related to the sequence of test tasks and can not be described by conventional distribution function,a prediction model based on random forest regression was constructed.Aiming at the shortcomings of the phased solution of prediction model and mathematical programming,an RF-GA hybrid algorithm based on random forest and genetic algorithm was proposed,which adopted parallel processing to improve the efficiency of the algorithm.Through the actual test data of enterprises,the random forest prediction model was compared with other machine learning algorithms to verify the effectiveness of the prediction model.Simulation experiments were designed to generate test examples of various scales.The RF-GA algorithm and inver-over operator were compared with various algorithms and operators.The experimental results show that the RF-GA algorithm has better performance in solving effect and efficiency.Applying the method to the actual test of enterprises can greatly reduce energy consumption and electricity expenditure.

关 键 词：空调测试 序相关准备时间 机器学习 遗传算法 分时电价 

分 类 号：TB69[一般工业技术—制冷工程]