Levy-CPSO-GA算法在空调冷负荷预测模型LSTM中的应用  

Application of Levy-CPSO-GA Algorithm in Air Conditioning Cooling Load Prediction Model LSTM

作　　者：后理通 张晨晨 丛意林 吴伊成 马永志[1] HOU Litong;ZHANG Chenchen;CONG Yilin;WU Yicheng;MA Yongzhi(College of Mechanical and Electrical Engineering,Qingdao University,Qingdao 266071,China)

机构地区：[1]青岛大学机电工程学院,山东青岛266071

出　　处：《青岛大学学报（工程技术版）》2023年第1期16-23,共8页Journal of Qingdao University(Engineering & Technology Edition)

基　　金：国家自然科学基金资助项目(51875296);山东省重点研发计划资助项目(2018GGX105007)。

摘　　要：针对单一算法优化空调冷负荷模型参数存在的局限性及对预测精度的需求,本文提出了基于遗传算法(genetic algorithm,GA)进化、莱维飞行(Levy)及粒子群(particle swarm optimization,PSO)优化算法的协同并行混合算法Levy-CPSO-GA。将初始种群初始化为2个同规模种群,分别按照合作机制和竞争机制并行更新,种群1采用Levy飞行产生随机新巢方式自适应初始化PSO,同时引入迭代极值,记录粒子群的信息交换;种群2按照GA更新,种群间通过适应度交流,以最优适应度更新群体,将混合算法应用于优化长短期预测模型(long short-term memory,LSTM),并将结果与各预测算法进行比较。研究结果表明,优化后的预测模型,预测精度大幅提高,与ELM相比,RMSE降低了81.1%;与LSTM模型相比,误差显著降低,RMSE降低了26.4%,在第105个预测点处,该预测模型的绝对误差为-0.6829,相比于ELM的绝对误差值-7.3135,其精度提高了90.66%,预测性能优于其他算法。该研究对准确预测冷负荷具有重要意义。Aiming at the limitations of a single algorithm to optimize the parameters of the air-conditioning cooling load model and the demand for prediction accuracy,this paper proposes a method based on genetic algorithm(GA)evolution,Levy flight(Levy)and particle swarm optimization(PSO),say,a cooperative parallel hybrid algorithm for optimization algorithms Levy-CPSO-GA.It initializes the initial population as two populations of the same scale,and updates them in parallel according to the cooperation mechanism and the competition mechanism.Population 1 uses the Levy flight to generate random new nests to adaptively initialize the PSO,and at the same time,iterative extreme values are introduced to record the information exchange of the particle swarm;Second,according to the GA update,the population is updated with the optimal fitness through fitness exchange,and the hybrid algorithm is applied to optimize the long short-term prediction model(LSTM),and the results are compared with each prediction algorithm.The research results show that the optimized prediction model,the prediction accuracy is greatly improved,and the RMSE is reduced by 81.1%compared with the ELM;compared with the LSTM model with the same network structure,the error is significantly reduced,and the RMSE is reduced by 26.4%.At the prediction point,the absolute error of the prediction model is-0.6829,compared with the absolute error of ELM-7.3135,its accuracy is improved by 90.66%,and the prediction performance is better than other algorithms.This research is of great significance for accurate prediction of cooling load.

关 键 词：GA LEVY CPSO LSTM负荷预测 协同并行优化 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论] TU831.2[自动化与计算机技术—计算机科学与技术]