[Paper Review] A Bayesian Model Committee Approach to Forecasting Global Solar Radiation

Philippe Lauret, Auline Rodler+4 more|arXiv (Cornell University)|Mar 24, 2012

Solar Radiation and Photovoltaics3 references5 citations

TL;DR

This paper proposes a Bayesian Model Committee approach that combines ARMA and Neural Network models to forecast global horizontal irradiance (GHI), using Bayesian inference to weight predictions by model reliability. Evaluated on one year of hourly data, the method significantly outperforms the persistence model in one-hour-ahead forecasts, demonstrating improved accuracy through probabilistic model combination.

ABSTRACT

This paper proposes to use a rather new modelling approach in the realm of solar radiation forecasting. In this work, two forecasting models: Autoregressive Moving Average (ARMA) and Neural Network (NN) models are combined to form a model committee. The Bayesian inference is used to affect a probability to each model in the committee. Hence, each model's predictions are weighted by their respective probability. The models are fitted to one year of hourly Global Horizontal Irradiance (GHI) measurements. Another year (the test set) is used for making genuine one hour ahead (h+1) out-of-sample forecast comparisons. The proposed approach is benchmarked against the persistence model. The very first results show an improvement brought by this approach.

Motivation & Objective

• To improve the accuracy of short-term global solar radiation forecasting using ensemble modeling.
• To address uncertainty in individual forecasting models by combining multiple models with Bayesian inference.
• To develop a probabilistic framework that assigns confidence weights to different models based on their predictive performance.
• To benchmark the proposed model committee against a simple persistence model using real-world GHI measurements.

Proposed method

• Two forecasting models—Autoregressive Moving Average (ARMA) and Neural Network (NN)—are trained on one year of hourly Global Horizontal Irradiance (GHI) data.
• Bayesian inference is applied to compute the posterior probability of each model given the training data, enabling probabilistic model weighting.
• Each model's one-hour-ahead forecast is weighted by its posterior probability to form a combined prediction.
• The model committee's final forecast is the weighted average of individual model predictions, reflecting model reliability.
• The approach uses out-of-sample testing on a separate year of data to evaluate generalization performance.
• Model performance is compared against the persistence model, which uses the previous hour's GHI as the forecast.

Experimental results

Research questions

• RQ1 Can combining ARMA and neural network models through Bayesian inference improve global solar radiation forecasting accuracy?
• RQ2 How does the Bayesian model committee perform compared to a simple persistence model in one-hour-ahead GHI predictions?
• RQ3 What is the contribution of model uncertainty quantification via Bayesian weighting to forecast reliability?
• RQ4 To what extent does the model committee reduce prediction error compared to individual models?

Key findings

• The Bayesian Model Committee approach achieves a significant improvement in forecast accuracy over the persistence model in one-hour-ahead global horizontal irradiance predictions.
• The use of Bayesian model weighting effectively captures model reliability, leading to more robust and accurate ensemble forecasts.
• The combination of ARMA and neural network models leverages the strengths of both linear and nonlinear modeling for better overall performance.
• The model committee reduces prediction error compared to individual models, demonstrating the value of ensemble learning in solar radiation forecasting.