Investigation on the long short-term memory-based models for rural heating load prediction in Northeast China

Accurate and rapid prediction of rural heating load holds considerable significance for planning heating schemes and control strategies. However, the complex variation characteristics pose a significant challenge to it. Herein, based on the long-term monitoring of the indoor and outdoor parameters of 7 typical rural buildings in Northeast China, long short-term memory, convolutional neural network-long short-term memory, and sparrow search algorithm-convolutional neural network-long short-term memory models were comparatively and progressively investigated to explore their performances of the rural heating load prediction. The results indicate that input variables, including indoor and outdoor temperature, solar radiation intensity, carbon dioxide concentration and wind speed have endowed the models with optimal performances. Furthermore, it revealed that the adoption of convolutional neural network has reduced the standard deviation of the prediction relative errors by 29.53 %, indicating that spatial features of the rural heating data were crucial. Furthermore, by applying the sparrow search algorithm to the determination of the hyperparameters, root mean square error, mean absolute percentage error and standard deviation were further reduced by 17.46 %, 24.09 %, and 30.5 %, respectively. Therefore, the sparrow search algorithm-convolutional neural network-long short-term memory model was proven to be effective for rural heating load rapid prediction.