Analysis of Solar Power Generation Forecasting Using Machine Learning Techniques

Abstract

Accurate solar power generation forecasting is critical for efficient grid integration and renewable energy management. This paper presents a comparative analysis of machine learning techniques for predicting solar power output using weather and historical generation data. We evaluate the performance of Naive Bayes and Artificial Neural Network (ANN) models trained on a curated dataset containing temperature, humidity, wind speed, and solar irradiance features. Our methodology emphasizes robust data preprocessing, including outlier removal, missing value imputation, and normalization, to enhance model reliability. Experimental results demonstrate that the ANN model achieves superior accuracy (RMSE: 0.18, R²: 0.92) compared to Naive Bayes (RMSE: 0.32, R²: 0.81) in day-ahead forecasts, attributed to its ability to capture non-linear relationships in solar irradiation patterns. The study also highlights the critical role of feature selection, with solar irradiance and temperature identified as the most influential predictors. These findings provide actionable insights for energy operators seeking to optimize forecasting systems for grid stability and renewable energy utilization

Introduction

The increasing use of solar energy necessitates accurate forecasting due to the intermittent and weather-dependent nature of solar power, which challenges grid stability and energy management. Traditional forecasting methods struggle with the complex, nonlinear relationship between weather and solar generation, leading to growing interest in machine learning (ML) approaches.

Literature Survey:
Recent research highlights three main areas in ML-based solar forecasting:

• ML Techniques: Hybrid and ensemble models outperform traditional algorithms, with studies showing improved accuracy using methods like Markov-switch models and deep learning (ANN).

• Feature Engineering: Key predictors include solar irradiance and temperature, with emerging use of humidity, wind speed, and panel-level data to improve fault detection and accuracy.

• Real-time Deployment: Advanced systems integrate batch and real-time processing with IoT monitoring, achieving low latency forecasts suitable for grid operations.

Critical gaps include limited comparative studies between simpler (Naive Bayes) and deep learning models, scarcity of high-resolution public datasets, and underexplored feature selection trade-offs.

Modeling and Analysis:
This study compares Gaussian Naive Bayes and a 3-layer ANN for solar forecasting using historical and weather data. Feature selection identified solar irradiance (42%) and temperature (31%) as most influential. The ANN significantly outperformed Naive Bayes (RMSE 0.18 vs. 0.32 kW/m², R² 0.92 vs. 0.81), especially in volatile weather, though with longer training times. Results demonstrate ANN’s ability to capture complex weather-power dynamics critical for grid reliability.

Related Work:
Traditional persistence and statistical models have been supplemented or replaced by ML methods like support vector machines, clustering-regression hybrids, and neural networks (MLP, RBF). AI-based approaches better capture nonlinear weather and power generation relationships. Challenges remain in data availability, modeling diffuse irradiance variability, and scaling models for large, diverse PV installations.

Proposed Future Work:
Plans include developing a hybrid ANN-LSTM model to improve temporal accuracy, especially during dawn/dusk transitions, and integrating satellite and sky camera imagery for multi-modal forecasting. Optimization for edge devices aims to enable real-time deployment with low latency and memory footprint, while transfer learning will help adapt models to various geographic locations.

Conclusion

This study demonstrated the effectiveness of machine learning techniques for solar power generation forecasting, with a focus on comparing the performance of Naive Bayes and Artificial Neural Network (ANN) models. The ANN model consistently outperformed Naive Bayes, achieving superior accuracy with an RMSE of 0.18 kW/m² and an R² score of 0.92, owing to its ability to capture complex, non-linear relationships between weather variables and solar power output. Key features such as solar irradiance and temperature were identified as the most influential predictors, reinforcing their importance in forecasting models. While the ANN required longer training times, its robustness in handling volatile weather conditions makes it a viable solution for real-world applications where prediction accuracy is critical. The proposed preprocessing pipeline—incorporating outlier removal, missing value imputation, and normalization—further enhanced model reliability. Future work will explore hybrid models and real-time hardware integration to improve computational efficiency. These advancements can significantly benefit energy grid operators by enabling more stable renewable energy integration and reducing reliance on fossil fuel backups. This research contributes to the global transition toward sustainable energy systems, aligning with SDG 7 (Affordable and Clean Energy).

References

[1] A. Khan, R. Bhatnagar, and V. B. Lobo, \"A comparative study on solar power forecasting using ensemble learning,\" Proc. 4th Int. Conf. Trends Electron. Inform., 2020, pp. 401-408, doi: 10.1109/ICOEI48184.2020.9142937 [2] Z. Li and S. M. Rahman, \"Day-ahead prediction of solar irradiance with Markov-switch models,\" IEEE Trans. Sustain. Energy, vol. 7, no. 3, pp. 1421-1429, Jul. 2016, doi: 10.1109/TSTE.2016.2538268. [3] B. Andò et al., \"Sentinella: Smart monitoring of photovoltaic systems at panel level,\" IEEE Trans. Instrum. Meas., vol. 66, no. 6, pp. 1615-1623, Jun. 2017, doi: 10.1109/TIM.2017.2669942 [4] C. Ventura and G. M. Tina, \"Utility-scale PV plant monitoring for fault detection,\" Renew. Energy, vol. 127, pp. 102-113, Nov. 2018, doi: 10.1016/j.renene.2018.04.053.. [5] A. G. Phadke and P. Wall, \"Improving power system protection using wide-area monitoring,\" IEEE Power Energy Mag., vol. 16, no. 5, pp. 52-61, Sep. 2018, doi: 10.1109/MPE.2018.2840123 [6] A. Chikh and A. Chandra, \"Optimal MPPT algorithm for PV systems with climatic estimation,\" IEEE Trans. Sustain. Energy, vol. 8, no. 2, pp. 514-525, Apr. 2017, doi: 10.1109/TSTE.2016.2606663. [7] S. Yagli et al., \"Automatic hourly solar forecasting using machine learning models,\" Renew. Sustain. Energy Rev., vol. 105, pp. 487-498, May 2019, doi: 10.1016/j.rser.2019.02.006

Copyright

Copyright © 2025 Natesh Khadse, Prathemesh Nawale, Raj Mankar, Ashwin Pawar, Prof. Kalyani Akhade. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.