A Comparative Study on Machine Learning-Based Renewable Energy Forecasting Utilising Advanced Optimisation Algorithms: Wind, Solar and Hydroelectric Power

Purpose: This study aims to develop a machine learning framework to predict power generation from wind, solar, and hydroelectric sources using state-wise data for four Indian states for the period 2015-2026. Design/Methodology/Approach: Three models were trained with three optimisation algorithms: Support Vector Regression (SVR), Random Forest (RF), and XGBoost (XGB). We used ten optimisation algorithms: Adam, AdamW, RMSprop, Lion, NovoGrad, Stochastic Gradient Descent (SGD), Shampoo, Sophia, AdaHessian, and Lioso (a combination of Lion and Sophia). The performance was evaluated by the RMSE, MAE, R², the convergence of the training loss, and the Lift-Over-Cumulative (LOC) curve. Research Limitation: The study is constrained to three standard machine learning algorithms and state-specific renewable energy datasets for the 2015-2026 period. The external environment and economic and socio-economic factors that affect renewable energy generation were not considered. Findings: The results show that the choice of optimiser is relevant to the forecasting performance. AdaHessian obtained the lowest RMSE (0.3509) and highest R² (0.0265) among the evaluated optimisers, while Shampoo obtained the highest RMSE (0.5790) and lowest R² (−1.6516). The state- and source-specific analysis showed that SVR performed best for hydro generation in several states, while RF and XGBoost performed competitively for solar and wind generation. For hydroelectric generation in Rajasthan, RF had the lowest state-level RMSE (230.22 MU). The 2030 projection shows high solar generation for Rajasthan (34,961.19 MU) and high wind and hydro generation for Karnataka (12,612.09 MU and 9,640.53 MU, respectively). Practical Implication: The proposed framework can support energy planners, power-system operators and policymakers in renewable energy forecasting, generation scheduling, grid management and long-term capacity planning. Social Implication: State- and source-specific forecasting can also support informed renewable energy planning and management, making it more sustainable. Originality/Value: Combining training loss convergence and LOC curve analysis with traditional error metrics provides further evidence of optimisation and prediction behaviour that can support the development of optimisation-based renewable energy forecasting frameworks.

Authors

Institutions

Publication Details

Journal
African Journal Of Applied Research
Published
2026-11-18
DOI
https://doi.org/10.26437/g60eep94
Primary Topic
Energy Load and Power Forecasting
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A Comparative Study on Machine Learning-Based Renewable Energy Forecasting Utilising Advanced Optimisation Algorithms: Wind, Solar and Hydroelectric Power

N. Vaidya, S. Tayade
1 links
African Journal Of Applied Research
Energy Load and Power Forecasting
article

A Comparative Study on Machine Learning-Based Renewable Energy Forecasting Utilising Advanced Optimisation Algorithms: Wind, Solar and Hydroelectric Power

N. Vaidya, S. Tayade
article en

Abstract

Purpose: This study aims to develop a machine learning framework to predict power generation from wind, solar, and hydroelectric sources using state-wise data for four Indian states for the period 2015-2026. Design/Methodology/Approach: Three models were trained with three optimisation algorithms: Support Vector Regression (SVR), Random Forest (RF), and XGBoost (XGB). We used ten optimisation algorithms: Adam, AdamW, RMSprop, Lion, NovoGrad, Stochastic Gradient Descent (SGD), Shampoo, Sophia, AdaHessian, and Lioso (a combination of Lion and Sophia). The performance was evaluated by the RMSE, MAE, R², the convergence of the training loss, and the Lift-Over-Cumulative (LOC) curve. Research Limitation: The study is constrained to three standard machine learning algorithms and state-specific renewable energy datasets for the 2015-2026 period. The external environment and economic and socio-economic factors that affect renewable energy generation were not considered. Findings: The results show that the choice of optimiser is relevant to the forecasting performance. AdaHessian obtained the lowest RMSE (0.3509) and highest R² (0.0265) among the evaluated optimisers, while Shampoo obtained the highest RMSE (0.5790) and lowest R² (−1.6516). The state- and source-specific analysis showed that SVR performed best for hydro generation in several states, while RF and XGBoost performed competitively for solar and wind generation. For hydroelectric generation in Rajasthan, RF had the lowest state-level RMSE (230.22 MU). The 2030 projection shows high solar generation for Rajasthan (34,961.19 MU) and high wind and hydro generation for Karnataka (12,612.09 MU and 9,640.53 MU, respectively). Practical Implication: The proposed framework can support energy planners, power-system operators and policymakers in renewable energy forecasting, generation scheduling, grid management and long-term capacity planning. Social Implication: State- and source-specific forecasting can also support informed renewable energy planning and management, making it more sustainable. Originality/Value: Combining training loss convergence and LOC curve analysis with traditional error metrics provides further evidence of optimisation and prediction behaviour that can support the development of optimisation-based renewable energy forecasting frameworks.

African Journal Of Applied ResearchVol. 12(8)
Raisoni Group of Institutions (IN), G H Raisoni University (IN)
Affordable and clean energy, Climate action, Industry, innovation and infrastructure
Energy Load and Power Forecasting
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon