Research On Building Heat Load Forecasting Method Based On GRU

Accurate prediction of heat load is the premise for the efficient operation of the heating system.There are many factors affecting the building heat load,and the change of heat load brings many challenges to the stable operation of the actual heating system.Accurately and scientifically judging the trend of heating load can effectively improve the thermal comfort of heat users,improve the operating efficiency of the pipe network,and reduce waste of resources.This thesis takes the logistic office building of a university as the research object,and analyzes the thermal process and thermal load characteristics of the building.This thesis focuses on considering the correlation between heat load and indoor temperature,simulates the building through De ST-c,and determines the best room for indoor temperature measurement points according to the simulation results.The heating operation parameters of the heating season from 2021 to 2022 were collected.For the data with noise in the actual data set,the Laida criterion and the K nearest neighbor algorithm were used to process it,to eliminate gross errors and fill in the missing values in the data.Using the random forest model to analyze the influence of various meteorological factors on the building heat load,it is finally determined that the outdoor temperature,humidity,solar radiation,wind speed and wind direction are the main factors affecting the heat load change.The time series neural network model GRU network in the field of deep learning is used to predict the building heat load.Considering the characteristics of heat load and the uncertainty of weather forecast,a multi-input model that can control the room temperature at the next moment is designed based on the GRU network.linear relationship.Reasonable selection of the hyperparameters of the model.The number of layers and neurons of the GRU network is determined,and it is found that the predicted time step and the value of the model learning rate have an important impact on the accuracy of the model.The particle swarm algorithm is used to further optimize the time step and learning rate of the model.The results show that the GRU neural network optimized by the particle swarm algorithm can further improve the prediction accuracy of the model,which is better than the RNN network in prediction accuracy and training speed.outperforms LSTM neural networks.In order to further improve the research and application value,the online prediction program was written in python language by applying crawler,database and other related technologies.A control strategy for the terminal valve under the heat load prediction of GRU neural network model is proposed.Realize the automatic upload and release of data and online real-time prediction of heat load.The research results have important application value for the operation regulation of the heating system.