Research On Short-term Heat Load Prediction Of Central Heating System Based On Deep Learning

Centralized heating is the most important way to heat in northern areas in winter,but there is still a mismatch between heat supply and demand in centralized heating systems.In order to reduce energy waste while meeting the comfort needs of heat users.accurate heat load forecasting for centralized heating systems is crucial.Short-term heat load prediction can assist energy systems to plan and allocate heat rationally,which is a prerequisite for safe,stable and efficient operation of centralized heating systems,and helps to realize demand-based heat supply and energy saving,with significant economic and environmental benefits.The rise of deep learning has brought new techniques and methods for the research of heat load prediction.In this paper,we address the problem of low accuracy of current heat load prediction models,and conduct a detailed study on the improvement of short-term heat load prediction methods based on convolutional neural networks and bidirectional long-and short-term memory networks in deep learning and combined with attention mechanism.In this paper,the Pearson correlation coefficient is used to analyze the relevant influencing factors of heat load,and six factors with a large degree of influence on heat load are obtained:outdoor temperature,relative humidity,solar radiation intensity and heat load in the first,second and third moments,and they are identified as the input parameters of the prediction model.A short-term heat load prediction model based on convolutional neural network-bi-directional long and short-term memory(CNN-BiLSTM)is constructed by using convolutional neural network to extract the potential relationships between heat load data to form feature vectors,and a bi-directional long and short-term memory network to perform long-term memory of information from heat load history sequences,which fully combines the advantages of a single model,and introduces the spatially dependent characteristics of the heat supply network into the heat The spatial dependence of the heat supply network is introduced into the heat load prediction model,and the influencing factors of heat load are considered comprehensively from both time and space dimensions.Detailed comparisons and analyses with the single model LSTM and BiLSTM are conducted through experimental simulations,and the experimental results demonstrate the efficiency and superiority of the combined prediction model.In order to further improve the prediction accuracy and prediction stability of the CNN-BiLSTM model,a combined prediction model based on CNN-BiLSTM-AM is proposed by introducing an attention mechanism to achieve adaptive estimation of the feature weights of each influencing factor of heat load,and it is found through the arithmetic simulation that,compared with the CNN-BiLSTM model,the CNN-BiLSTM-AM The RMSE,MAE and MAPE of the model are smaller and the R2 is larger,which indicates that the model has higher prediction accuracy and stability and better prediction effect,and the introduction of attention mechanism helps to improve the prediction performance of the model.It is also verified that the combined prediction model proposed in this paper is more conducive to achieving on-demand heat supply and energy saving and carbon reduction than a single prediction model,and can better guide the energy-saving and efficient operation of central heating systems.