基于改进型BP神经网络的大型CCHP系统负荷预测研究  

Load forecasting of large CCHP system based on improved BP neural network

作　　者：顾兆雄 黄志坚 谢鸣 林萌雅 齐家业 高文忠[2] Gu Zhaoxiong;Huang Zhijian;Xie Ming;Lin Mengya;QI Jiaye;Gao Wenzhong(Shanghai Electric Power Company Zhabei Power Plant of State Grid,Shanghai,China)

机构地区：[1]国网上海市电力公司闸北发电厂 [2]上海海事大学

出　　处：《暖通空调》2021年第1期128-133,127,共7页Heating Ventilating & Air Conditioning

摘　　要：为了解决BP神经网络在预测空调负荷时存在的学习速度慢、维数灾难、容易陷入局部收敛及无法保证全局收敛最优解等问题,首先采用Spearman秩相关系数分析冷负荷的主要影响因素,确定了动态冷负荷预测模型的输入参数,然后构建复合遗传算法的改进型GA-BP神经网络预测模型,并分别利用BP和GA-BP神经网络模型对位于上海的某大型区域CCHP系统进行了冷负荷预测。结果显示:利用Spearman秩相关系数分析.可缩短模型训练时间,规避维数灾难,提高预测精度;相对BP神经网络预测结果,GA-BP神经网络避免了局部收敛,明显提高了模型的预测精度。BP neural network has problems such as slow learning speed, dimensionality disaster, easy to fall into local convergence, and unable to guarantee the best global convergence. In order to solve above problems, applies Spearman rank correlation coefficient firstly to analyse the main influencing factors of cooling load, so as to determine the input parameters of a dynamic cooling load prediction model, and then develops an improved BP neural network prediction model of the combined genetic algorithm. Finally, uses BP and GA-BP neural network models to predict the cooling load of a large district CCHP system located in Shanghai respectively. The results show that Spearman rank correlation coefficient analysis can shorten the training time, avoid dimensionality disaster and improve the prediction accuracy. Compared with the prediction results of BP neural network, GA-BP neural network avoids local convergence and obviously improves the prediction accuracy of the model.

关 键 词：CCHP系统 负荷预测 GA-BP神经网络 Spearman秩相关系数 影响因素 

分 类 号：TU831[建筑科学—供热、供燃气、通风及空调工程] TP183[自动化与计算机技术—控制理论与控制工程]