Mid- and Long-term Forecast of Load Peak-Valley Difference based on Random Forest and Secondary Correction

Abstract

Recent years have seen the increasing flexibility in the changes of the demand-side user load, which makes it difficult to evaluate the demand-side response. Moreover, the accountability of the evaluation of the response plan depends on the accuracy of load peak-valley difference forecast. Therefore, considering the complexity of the load peak-valley difference, we establish a mid- and long-term peak-valley prediction model based on random forest and secondary correction to evaluate the response effect. First, the binary feature combination is used to identify the optimal feature set. Secondly, the random forest model is applied to the first mid-and long-term long-term prediction of the monthly and quarterly peak-valley differences. Finally, taking into account the impact of the influencing factors of different years on the seasonal peak-valley difference, we use the support vector regression machine to obtain the fitting features of the correction factors and the load peak-valley difference, which facilitates the secondary correction of the prediction. The validity of the model proposed is verified by the residential user load data of a city in Jiangsu Province.