Thermal Load Forecasting Based on PSO-SVR

Heat load forecasting has a guiding role for large time constant central heating systems. It is difficult to meet the forecast demand simply relying on outdoor temperature-heat load linear modeling. It is also important to make qualitative and quantitative analysis in combination with detailed meteorological conditions and time period factors. Based on the Particle Swarm Optimization Support Vector Machine (PSO-SVR) nonlinear prediction model, the global optimization function of PSO algorithm for historical data is used to optimize the parameters of SVM, and the output value of thermal load prediction is obtained through the magic input variable which is Pre-processed weather and date data. The average absolute error of the test data obtained from the measured samples of a heat supply enterprise in dalian is 4.6%, which is significantly improved compared to linear prediction and SVR prediction methods. And it can play a good guiding role in the scientific and refined quantitative management of the heating production process.