基于改进沙猫群算法优化CNN-BiLSTM的热负荷预测  

Optimised CNN-BiLSTM for heat load prediction based on improved sand cat swarm algorithm

作　　者：王耀辉 薛贵军 赵广昊 WANG Yaohui;XUE Guijun;ZHAO Guanghao(School of Electrical Engineering,North China University of Science and Technology,Tangshan 063200,China;Intelligent Instrument Factory of North China University of Science and Technology,Tangshan 063000,China)

机构地区：[1]华北理工大学电气工程学院,河北唐山063200 [2]华北理工大学智能仪器厂,河北唐山063000

出　　处：《现代电子技术》2024年第14期20-29,共10页Modern Electronics Technique

摘　　要：针对传统热负荷预测精度不高、无法满足换热站及热网优化调控需求的问题,提出一种VMD-ISCSO-CNN-BiLSTM的热负荷预测模型。首先,利用变分模态分解(VMD)对原始供热负荷数据进行降噪处理,降低数据的不稳定性;其次,由K-means算法改进种群初始化,由演变机制改进寻优能力和由变异机制改进跳出局部最优能力,利用改进沙猫群算法(ISCSO)对卷积神经网络、双向长短期记忆神经网络(CNN-BiLSTM)超参数进行寻优,建立热负荷预测模型;最后通过实例进行分析。结果表明,数据降噪后模型预测精度更高,R^(2)提升1.1%;由ISCSO优化的模型比其他算法优化的模型预测效果更好,拟合度达到99.4%;VMD-ISCSO-CNN-BiLSTM的组合预测模型相较于单一模型,RMSE降低18.5%,MAE降低13.8%,R^(2)提升15.8%,并有更好的拟合度,泛化性强,满足工程实际要求。In allusion to the problems that the traditional heat load prediction accuracy is not high enough to meet the demand of heat exchange station and heat network optimisation and regulation,a VMD-ISCSO-CNN-BiLSTM heat load prediction model is proposed.The variational mode decomposition(VMD)is used to denoise the original heating load data and reduce its instability.The K-means algorithm is used to improve population initialization,the evolutionary mechanism is used to improve optimization ability,the mutation mechanism is used to improve the ability to jump out of local optima,and the improved sand cat swarm algorithm(ISCSO)is used to optimize the hyperparameters of convolutional neural networks and bidirectional long short-term memory neural networks(CNN-BiLSTM),so as to establish the heat load prediction model.The model is analysed by examples.The results show that the model prediction accuracy is higher after data noise reduction,and the R^(2)is improved by 1.1%.The model optimised by ISCSO is better than the models optimised by other algorithms,with a fit of 99.4%.In comparison with the single model,the combined prediction model of VMD-ISCSO-CNN-BiLSTM has a lower RMSE by 18.5%,a lower MAE by 13.8%,and a higher R^(2) by 15.8%.It has better goodness of fit and strong generalization,which meets the actual requirements of the project.

关 键 词：热负荷预测 卷积神经网络 双向长短期记忆神经网络 改进沙猫群算法 变分模态分解(VMD) K-MEANS算法 演变机制 变异机制 

分 类 号：TN911.23-34[电子电信—通信与信息系统] TP391.9[电子电信—信息与通信工程]