Research On Intelligent Heating Load Forecasting Based On Machine Learning

With the gradual improvement of residents’quality of life,the people’s expectations for winter heating have gradually changed from meeting the basic heating needs to pursuing more comfortable heating effects,but due to the late start of central heating in China,the degree of system intelligence is relatively weak,and the heat supply is often caused by uneven heating,resulting in serious waste of energy,so choosing a suitable heating load prediction model has become the primary problem to improve heating efficiency and achieve energy conservation and emission reduction.At the same time,with the development of Internet technology and the progress of science and technology,machine learning methods with strong fitting ability have been confirmed to be able to complete the heating load prediction well,providing effective support for heating enterprises to achieve smart heating,but China’s winter heating area heating cycle is longer,a single prediction model is not necessarily applicable to all heating stages,so this paper is based on machine learning to establish a heating load prediction model and verify in turn in the three heating stages of the heating cycle,which is conducive to heating enterprises to regulate the heating pipe network more scientifically.In this project,we collected the operation data of the heat exchange station in a residential area of Changchun City from 2020 to 2021 in daily units,extracted the meteorological parameters of the same period from the China Meteorological Data Network,and preliminarily judged the input parameters of the model after qualitative analysis of the influencing factors of the heating load,and then used the Pearson correlation coefficient method to finally determine the outdoor temperature,return water supply temperature,return water supply pressure and heating load of the first three days as the model input amount,and the heating load of the last seven days as the model output.Taking MATLAB＿R2021a as the platform,the most potential BP neural network and support vector regression in machine learning are selected as the prediction algorithm,considering the influence of the two algorithm parameters on the model performance,the BP neural network prediction model based on the genetic algorithm and the support vector regression prediction model based on grid search and genetic algorithm are further established,respectively,and the mean square error(MSE),the fitting coefficient(R~2)and the mean absolute percentage error(MAPE)are used as the evaluation indicators of the model.Verify the predictive power of the model over three heating phases.The results show that the BP neural network model using the tanh activation function and optimized by genetic algorithms is more advantageous in the early cold period and the support vector regression model using the RBF kernel function,the BP neural network model using the sigmoid and tanh activation function in the severe cold period is more accurate,and the support vector regression model using the linear kernel function and the RBF kernel function and optimized by mesh search is better in the late cold period.Finally,through the analysis of the heating system regulation mode,it is decided that the above advantage model will be used to predict the water supply temperature of the heat exchange station,which provides specific heating parameters for the heating enterprise to adjust the heating pipe network.