A Novel Multi-Step Ahead Solar Power Prediction Scheme Based on Transformer Structure

Abstract

Photovoltaic (PV) power generation inherently possesses uncertainty and issusceptible to significant short-term fluctuations, posing a notable risk topower grid stability. To address this challenge, accurate solar irradianceprediction emerges as a viable solution to mitigate power intermittency. Inparticular, the complexity increases when considering multistep predictionas opposed to single-step prediction. Consequently, the pursuit of effectivemulti-step prediction methods becomes a pressing and essential research endeavor. This paper introduces a novel approach for multi-step solar prediction (MSSP) model, founded upon the transformer framework. This modeladeptly captures prolonged dependencies within solar data, thus accommodating trend variations. The MSSP model innovatively integrates a distillingoperation and a generative decoder. These additions serve to reduce error propagation, construct replicas, and enhance model generalization androbustness. Additionally, rigorous experimentation involving real solar datavalidates the efficacy of the MSSP model, Further experiments on the MSSP’sapplication in the Denmark’s electricity market reveal that it significantlyenhances profitability, indicating its potential for diverse applications. Comparative analyses against several existing methods underscore its superiorityin terms of prediction accuracy and stability, particularly for long-term multistep prediction scenarios.

Keywords: PV power Prediction, Multi-step ahead prediction, Deep learning, Transformer