大型集中供热系统全网热用户供热负荷预测研究

Heating load forecasting of large-scaledistrict heating system

提出一种遗传算法(GA)与BP神经网络相结合的BP-GA模型,以解决BP神经网络算法容易陷入局部最优的问题,用于换热站供热负荷预测;进一步基于典型住宅和公建用户历史运行数据发展住宅和公建用户的通用负荷预测模型,以提高全网热用户供热负荷预测效率。将上述模型与算法应用于西安市某大型集中供热系统,结果表明:对典型住宅及公建用户训练过程的平均供热负荷预测绝对百分比误差为8.56%和8.78%;对94%的非典型用户预测误差小于15%。证明该模型能够更加高效地对大型集中供热系统全网热用户供热负荷进行预测。

This paper proposes a novel approach,the BP-GA model,which effectively mitigates the inherent challenge of the BP neural network algorithm’s vulnerability to local optima,by the coupling of the genetic algorithm(GA)and the BP neural network,and the present model is employed for the heating load forecasting of substations.More importantly,the generalized load forecasting models for both typical residential and public construction user are developed based on their historical operational data to enhance the efficiency of heating load predictions for all the users in the entire network.Employing the aforementioned models and algorithms for the heating load prediction of a large-scale district heating system in Xi'an,the results reveal that the average absolute percentage errors during the training process for typical residential and public construction users are 8.56%and 8.78%,respectively;The prediction error of heating load for 94%of atypical users is less than 15%.The above results serve a compeling evidence that the present model is capable of predicting the heating load for all the substations of the large-scale districtheatingsystem more efficiently.