Modified Auto Regressive Technique for Univariate Time Series Prediction of Solar Irradiance

TL;DR

This paper proposes a Modified Auto Regressive model combined with neural networks to improve the accuracy of short-term solar irradiance prediction, addressing the stochastic nature of solar energy due to weather variability.

Contribution

It introduces a novel Modified Auto Regressive approach integrated with neural networks for enhanced solar irradiance forecasting, outperforming existing models.

Findings

Modified Auto Regressive model achieves lowest prediction error.

Model performs well across multiple time horizons.

Neural network integration improves forecast robustness.

Abstract

The integration of renewable resources has increased in power generation as a means to reduce the fossil fuel usage and mitigate its adverse effects on the environment. However, renewables like solar energy are stochastic in nature due to its high dependency on weather patterns. This uncertainty vastly diminishes the benefit of solar panel integration and increases the operating costs due to larger energy reserve requirement. To address this issue, a Modified Auto Regressive model, a Convolutional Neural Network and a Long Short Term Memory neural network that can accurately predict the solar irradiance are proposed. The proposed techniques are compared against each other by means of multiple error metrics of validation. The Modified Auto Regressive model has a mean absolute percentage error of 14.2%, 19.9% and 22.4% for 10 minute, 30 minute and 1 hour prediction horizons. Therefore, the Modified Auto Regressive model is proposed as the most robust method, assimilating the state of the art neural networks for the solar forecasting problem.