Hybrid Grey Wolf Optimization and ANN Model for Solar Panel Efficiency Prediction

Abstract

Solar power generation is inherently difficult to forecast because output is strongly influenced by continuously changing environmental conditions such as irradiance, temperature, humidity, and cloud cover, and because field sensor readings are frequently noisy. Conventional machine learning approaches such as Random Forest combined with Lasso-based feature selection perform reasonably well under stable weather but struggle to adapt when sky conditions change abruptly, since their feature sets are fixed and their capacity to model nonlinear relationships is limited. This paper proposes a hybrid forecasting framework that couples Grey Wolf Optimization (GWO) with an Artificial Neural Network (ANN) to overcome these limitations. GWO is employed as an adaptive, metaheuristic feature-selection mechanism that dynamically identifies the most influential parameters — irradiance, temperature, humidity, voltage, current, panel age, soiling ratio, and module temperature — from large meteorological datasets, while the ANN learns the nonlinear mapping between the optimized feature subset and solar power output through backpropagation. The resulting GWO-ANN model reduces feature redundancy, accelerates convergence, and improves robustness to sensor noise relative to static feature-selection methods. Evaluation of the training and validation loss/MAE curves demonstrates stable convergence and consistent error reduction across training epochs, and the trained model was verified through a real-time weather-simulation and manual-input interface. The proposed hybrid framework offers a scalable, adaptive solution suitable for real-time photovoltaic monitoring and smart-grid energy-management applications. Keywords—Grey Wolf Optimization (GWO), Artificial Neural Network (ANN), Solar Power Forecasting, Photovoltaic (PV) Systems, Feature Selection, Renewable Energy, Smart Grid, Metaheuristic Optimization.