Solar Irradiance Forecast Using Naïve Bayes Classifier Based on Publicly Available Weather Forecasting Variables

Abstract

This paper develops an approach for two-day-ahead global horizontal irradiance (GHI) forecast using the naïve Bayes classifier (NB). Based on publicly available weather forecasting information about temperature, relative humidity, dew point, and sky coverage, they are used as a training set in NB classification with hourly resolution. To reduce having two times with the same GHI affecting the classification in the proposed model, two characteristics of the GHI under different weather conditions are considered: The daylight variation and diurnal cycle. More importantly, NB’s independence assumption-based on simple Bayes’ theorem makes the process speed faster and less constrained than other classification algorithms. The forecast performance is verified with several error criteria from established analytical practices using relevant statistics. Moreover, commonly used forecasting error criteria are discussed. This NB model shows improved results regarding error criteria and a good agreement for a clear day that satisfies the guideline for the evaluation of two-days-ahead forecast, when compared with other recent techniques.

Keywords

• global horizontal irradiance
• naïve Bayes classification
• diurnal variation
• kernel density estimation