CNN-LSTM with Fused Residual Blocks for Heating Load Prediction

Abstract

The heating industry in northern regions accounts for a large share of energy consumption, and the heating system mainly relies on manual experience for heating load regulation, thus resulting in energy waste. Load prediction is a key part of industrial IoT technology, and heating load prediction can provide accurate predictions for a heating load to achieve effective energy utilization. Traditional load prediction algorithms have low accuracy and strict requirements for data. Therefore, this paper proposes a convolutional long-short term memory neural network (CNN-LSTM) model integrating residual blocks to predict heating load. The model uses convolutional neural network (CNN) and long and short-term memory neural network (LSTM) to extract the temporal and spatial characteristics of heating load. Furthermore, the model uses residual blocks to deepen network depth and improve network performance, and solves the problem of feature and information loss in feature extraction. The experimental results show that the CNN-LSTM model with fused residual convolution blocks has the lowest prediction error, and the average absolute percentage error of heating load prediction results of a heat exchange station in Lanzhou City reaches 0.57%.