Source Load Forecasting of Customer-side Multi-energy System Based Feature Engineering

Abstract

Existing research mainly focuses on the single forecasting target and algorithm optimization, rarely mentions how to extract features for forecasting. For the source and load data of the customer-side multi-energy system, the influencing factors are complex and coupled. So, the quality of features gradually becomes the bottleneck that limits its forecasting accuracy. In this regard, based on EDA technology, this paper proposes a source and load forecasting method of customer-side multi-energy system based on feature engineering. First, combined with domain knowledge, perform a more systematic and complete feature analysis for the original observation data. Through this process, we can extract time features, statistical features and combined features. Subsequently, according to the characteristics of the data, the corresponding algorithm is selected to build a forecasting model. Finally, conduct two experiments based on source-side data and load-side data respectively, which proved that this approach can significantly improve the forecasting accuracy.