基于GA改进LSTM-BP神经网络的智慧楼宇用能行为预测方法

Intelligent building energy consumption behavior prediction method based on GA improved LSTM-BP neural network

针对大部分预测方法难以适用于多源异构数据的处理,且存在能源类型考虑不全面等问题,提出了基于GA改进LSTM-BP神经网络的智慧楼宇用能行为预测方法.该方法通过K-means聚类算法分析用能行为并减少用能数据规模,利用遗传算法优化长短时记忆网络(LSTM)结合反向传播神经网络(BP)的预测模型,实现对智慧楼宇的能耗预测.基于TensorFlow深度学习框架进行实验分析,结果表明所提方法在12 h及120 h内预测结果的MAE值分别为1.79 J和2.11 J,预测效果稳定并优于其他对比方法,故具有一定的应用前景.

In view of the problems that most prediction methods are difficult to apply to the processing of multi source heterogeneous data and the incomplete consideration of energy types,a prediction method of energy consumption behavior of intelligent buildings based on GA improved LSTM-BP neural network was proposed.A K-means clustering algorithm was used to analyze energy consumption behavior and reduce the scale of energy consumption data.A genetic algorithm was used to optimize the prediction model of long and short-term memory network(LSTM)combined with back propagation neural network(BP)to realize the energy consumption prediction of intelligent buildings.The experimental analysis based on TensorFlow deep learning framework shows that the MAE values of the prediction results of the proposed method in 12 h and 120 h are 1.79 J and 2.11 J,respectively,the prediction effect is stable and better than other comparison methods,and has a certain application prospect.