A hybrid ANN–AHP–GIS framework with dimensionality reduction and uncertainty quantification for solar site selection in Southern India

Document Type

Article

Publication Title

Energy Conversion and Management X

Abstract

This study presents a novel hybrid framework for assessing solar energy feasibility across nineteen sites in Southern India by combining artificial neural networks (ANN) and the analytic hierarchy process (AHP). Using a 40:60 weighting, the model integrates expert-driven AHP and data-driven ANN scores, demonstrating 85 % ranking stability across different settings, indicating a robust and reliable site prioritization that remains consistent despite input variability through Monte Carlo simulations. Nine spatial criteria, including solar irradiation (4–7 kW/m2), land cost variability (±12 %), grid proximity, unused land, land slope, land area, ecological impact, population density, and future energy demand, are incorporated into actionable suitability maps using geographic information systems (GIS). Principal component analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE) diminish dimensionality, encapsulating 94 % of data variance, thereby facilitating the simplification of intricate criteria for enhanced interpretability without substantial information loss and uncovering latent patterns in site suitability. Robust concordance among scoring systems is validated by Spearman, Pearson, and Kendall correlation analyses (Pearson > 0.99). The framework also includes uncertainty quantification, modeling variance in input data (e.g., ±5% solar irradiation) and ANN prediction uncertainty (±0.03), producing 95 % confidence intervals for site rankings. Among the top-ranked sites are Vizag, Guntur, and Srikakulam. The hybrid technique enhances classification accuracy by 22 % compared to individual models. Three-dimensional scatter plots, heat maps, and radar charts, among other visualization methods, illustrate the tradeoffs between land cost, environmental impact, and infrastructural accessibility. The fully automated MATLAB framework offers policymakers a swift, reproducible, and scalable decision-support tool for efficient, transparent, and risk-informed solar site selection aligned with national energy objectives.

DOI

10.1016/j.ecmx.2025.101280

Publication Date

10-1-2025

This document is currently not available here.

Share

COinS