Track: Machine Learning

Abstract

As the world moves towards renewable energy sources, solar energy is becoming more widely recognized. This impacts photovoltaic (PV) panel prices, which requires deeper analysis to shape strategic planning. This paper fills a gap in PV panel price forecasting by utilizing machine learning models for multivariate time series forecasting, offering a more robust, comprehensive, and adaptable approach. Based on literature review, feature analysis, and linear regression, PV panel prices factors were identified, and data was collected. The dataset consists of 26 monthly time series features covering January 2018 to July 2023. The study's methodology consists of two phases: Phase I involves applying machine learning models (CNN, KNN, Random Forest, SVR, XGBoost) to forecast prices, resulting in the selection of the Random Forest model as the most effective based on performance metrics and comprehensive evaluations. Phase II utilizes the aforementioned model, along with forecasted features from the SARIMA model, to predict future PV panel prices. Forecasts indicate a stabilized PV panel market for 36 months from July 2023, with slight price fluctuations. This stabilization hypothesis aligns with other research findings, suggesting industry agreement. As the solar industry is dynamic, regular market analyses and model calibrations are essential.Overall, the research enhances the understanding of solar energy price time series forecasting, merging data gaps, and outlining pivotal price-influencing factors, offering stakeholders direction towards a sustainable energy future. Recommendations include using a broader dataset, improving methodology, exploring more models, applying a hybrid model, and continually updating forecasting methods in alignment with industry dynamics.