An enhanced CNN-LSTM based multi-stage framework for PV and load short-term forecasting: DSO scenarios

Abstract

The importance of accurate forecasting in the electric sector has grown due to the increasing demand and adoption of high volume of Renewable Energy Sources (RES). Short-term forecasting (STF) using deep learning methods has shown potential for improving forecasting accuracy. However, the accuracy of these methods can be further enhanced by combining them to generate a hybrid model, selecting appropriate input features, generating new features, and optimizing model parameters. This paper proposes a novel multi-stage framework for PV and load STF that employs feature generation, feature selection, and optimal hyperparameter tuning preprocessing techniques. An enhanced hybrid CNN-LSTM deep learning model architecture is developed in the final stage of the proposed framework. The framework is assessed and compared to other leading-edge approaches across different DSO scenarios, including multiple single-phase residential loads, three-phase feeders, and secondary substation, demonstrating a significant reduction in forecasting errors.

Keywords

• CNN-LSTM
• Feature generation
• PV short-term forecasting
• Load short-term forecasting
• Distribution system operator