Similarity-Based Models for Day-Ahead Solar PV Generation Forecasting

Abstract

Accurate forecasting of solar photovoltaic (PV) power for the next day plays an important role in unit commitment, economic dispatch, and storage system management. However, forecasting solar PV power in high temporal resolution such as five-minute resolution is challenging because most of PV forecasting models can only achieve the same temporal resolution as their predictors(i.e., weather variables), whose temporal resolution is usually low (i.e., hourly). To address this challenge, similarity-based forecasting models (SBFMs) are advocated in this paper to forecast PV power in high temporal resolution using low temporal resolution weather variables. To effectively generalize the model for different scenarios of available weather data, three forecasting models (i.e., basic SBFM, categorical SBFM, and hierarchical SBFM) are proposed. As a case study, the PV power generated by the solar panels on the rooftop of a commercial building is forecasted for the next day with a five-minute resolution under three different scenarios of available weather data. The leave-one-out cross-validation analysis reveals that using only two or three weather variables, the proposed SBFMs can achieve higher forecasting accuracy than several benchmark models.