Research On Short-term Heat Load Prediction Of Heat Exchange Stations Based On Improved LSTM Neural Network

With the penetration and development of information technology and digitalization in the heating industry,"on-demand heat supply" has become the focus of research in the field of centralized heating at home and abroad.Since the real heat demand keeps changing according to the changes of different influencing factors and the control of the heating system has large inertia and lag,most of the heating enterprises adopt the regulation method of rough heating supply at present,lacking the consideration of the precise heat load demand,which leads to the phenomenon of mismatch between the heat supply of enterprises and the heat demand of users often.This study aims to get accurate short-term heat load demand value through prediction and guide the heat exchange stations according to the predicted heat supply,and front-load the regulation of the heating system in the next hour to realize heat supply according to demand.The specific technical means is to build a CEEMDAN-SE-LSTM heat exchange station short-term heat load prediction model to guide heat exchange stations by extracting the effective knowledge from the historical data of heat exchange stations through time-series data processing and deep learning methods.The main research of this thesis is as follows:(1)A short-term heat load prediction model for heat exchange stations was constructed.First of all,study the intrinsic change pattern of heat load data as well as extrinsic influencing factors,use traditional data processing methods combined with data characteristics to remove abnormal and missing values,and unify the data dimensions of different variables.Pearson correlation analysis method(Pearson)was used to analyze the correlation between heat load and its influencing factors and identify the strongly correlated factors as model inputs.Based on a large amount of heat load data and strong nonlinearity,three models of recurrent neural network(RNN),long and short-term memory neural network(LSTM),and gated recurrent unit(GRU)were used to model and predict the short-term heat load data of heat exchange stations,and the particle swarm algorithm(PSO)was introduced to optimize the parameters in the model.The mean absolute percentage error(MAPE)and root mean square error(RMSE)are used to verify the prediction accuracy,and the results show that the PSO-LSTM model is more suitable for the short-term heat load prediction of heat exchange stations.(2)The optimization model further improves the model prediction capability.For the characteristics of non-stationary short-term heat load data of heat exchange stations,the complete ensemble empirical modal decomposition with adaptive noise(CEEMDAN)method is introduced to split the heat load series into multiple relatively smooth sub-series,and the sample entropy(SE)model is used to combine the sub-series with similar entropy values into one sub-series and then use the PSO-LSTM model for prediction.The results show that the MAPE value of the optimized model is 16.4% lower than that before optimization.(3)Build a cyber-physical system(CPS).The system includes online monitoring,automatic control,remote regulation,and alarm functions,and embeds the heat load prediction model proposed in this thesis into the CPS to regulate the primary network booster pump according to the predicted heat load to realize heat exchange station on-demand supply.The experiment proves that adjusting the heat exchange station according to the predicted heat load value makes the temperature in the room more stable than the current heating mode,while improving the indoor comfort time by 30%.It is proved that the short-term heat load prediction model of the heat exchange station proposed in the thesis can achieve better control effect in the heating system.