Research On Heat Load Forecasting Of Centralized Heating System Based On Meteorological Factors

As the main influencing factors,meteorological factors have great research value.The random forest and GBDT prediction algorithm based on machine learning can effectively solve the heating load forecasting problem.This paper mainly discusses the heating load forecast based on meteorological factors.The work including the importance analysis of meteorological factors,the correlation analysis between meteorological factors and load,the load forecasting based on random forest algorithm and on GBDT model,in order to solve the non-parallel problem of GBDT,the stochastic gradient ascending algorithm is introduced,which is based on the ieda of the forest algorithm concept.In the fourth chapter,the popular dropout technique against dropover is introduced,and it is merged with GBDT to form the DGBDT algorithm,which makes the prediction accuracy of the model further improved.The implementation of random forest and GBDT algorithms in this paper are all work under the Spark platform,which makes the training speed of the model significantly improved.Finally,the influence of meteorological factors on heating load,the effect of number of samples and the prediction bias and parameter comparison of different algorithms are compared and analyzed in the experiment.The experiment further proves the effectiveness of DGBDT prediction.In the study of the relationship between meteorological factors and load,the traditional statistical learning method is used,and the attribute importance analysis is carried out in the third chapter using the random forest model.Finally,the date attribute,meteorological factor,and yesterday load are determined as the predictive attributes.In the load forecasting research based on machine learning algorithm,model construction of random forest,GBDT and DGBDT are carried out.DGBDT is a model obtained by regularization processing on the basis of GBDT and merged with Dropout.The model is in experiment and can further improve the prediction accuracy of GBDT,at the same time,it can better resist over-fitting ability.In the Spark construction process,the parallelization processing of the above algorithm is realized,and the operation efficiency of the model is improved.Finally,the influence of attributes such as meteorological factors on the heating load and the prediction accuracy of each model are analyzed and compared in the experimental link.