Power Load Forecasting Based On Couple Prophet-PSO-GPR Combined Model

With the economic and social development of countries around the world,the power load has increased sharply,and the optimization and improvement of its security and supply capacity has attracted great attention.Power shortage will not only lead to economic problems,but also social and political issues related to national energy security,economic and social development and people’s livelihood and well-being.Conversely,oversupply leads to wasted energy and resources.Therefore,how to accurately predict the future power load and formulate a reasonable power supply plan under the premise of ensuring supply is the key to reducing power generation costs,ensuring stable power supply,and avoiding waste.Firstly,the experimental data of power load in Belgium from 2018.01.01-2022.09.30was collected,which was composed of 1733 pieces of data sampled every 15 minutes,and then preprocessed the unbalanced data.Taking the collected data as the empirical analysis object,machine learning methods are used to deeply explore the impact of power load at different times of the day on the average load of the next day.This method not only has high accuracy,but also does not require complex power load model parameters,so it is widely used in the field of power load forecasting.Secondly,based on 1733 power load data,the Prophet model is used to model a load prediction model,and then the power load forecast value and residual value are obtained.Gaussian process regression(GPR)analysis is then performed on the power load residual values and the model is corrected,during which particle swarm optimization(PSO)is used to search for the optimal hyperparameters.By superimposing the predicted value of the Prophet model with the residual prediction value,the power load forecast value of the model is determined.Finally,a coupled Prophet-PSO-GPR power load forecasting model is obtained.Finally,the coefficient of determination(R~2),root mean square error(RMSE),mean absolute percentage error(MAPE)and normalized root mean square error(NRMSE)are used as indexes to verify that the Prophet-PSO-GPR model is more effective than the support vector regression(SVR),BP neural network,particle swarm optimization gaussian process regression(PSO-GPR)model,and Prophet model.The results show that the R~2 of the Prophet-PSO-GPR prediction model is 0.9290,RMSE is 248.5688,MAPE is 1.8428%,and NRMSE is 0.0594,which indicates that the Prophet-PSO-GPR prediction model can better reflect the change trend of power load and effectively improve the accuracy of power load prediction,which means that the model has high feasibility and practicality.