# Biologically Based Intelligent Multi-Objective Optimization for Automatically Deriving Explainable Rule Set for PV Panels Under Antarctic Climate Conditions

## TL;DR

This paper presents a new method for optimizing solar panel performance in Antarctica by using biologically inspired algorithms and high-resolution data.

## Contribution

The study introduces a high-resolution PV dataset for Antarctica and an interpretable rule-based model for solar panel optimization.

## Key findings

- The rule-based approach achieved 92.3% precision and 89.7% recall in predicting solar panel performance.
- The proposed method outperformed standard machine learning models in terms of interpretability and stability.
- The dataset captures PV output alongside meteorological variables at high temporal resolution in polar conditions.

## Abstract

Antarctic research stations require reliable low-carbon power under extreme conditions. This study compiles a synchronized PV-meteorological time-series data set on Horseshoe Island (Antarctica) at 30 s, 1 min, and 5 min resolutions and compares four PV module types (monocrystalline, polycrystalline, flexible mono, and semitransparent) under controlled field operation. Model development adopts an interpretable, multi-objective framework: a modified SPEA-2 searches rule sets on the Pareto front that jointly optimize precision and recall, yielding transparent, physically plausible decision rules for operational use. For context, benchmark machine-learning models (e.g., kNN, SVM) are evaluated on the same splits. Performance is reported with precision, recall, and complementary metrics (F1, balanced accuracy, and MCC), emphasizing class-wise behavior and robustness. Results show that the proposed rule-based approach attains competitive predictive performance while retaining interpretability and stability across panel types and sampling intervals. Contributions are threefold: (i) a high-resolution field data set coupling PV output with solar radiation, temperature, wind, and humidity in polar conditions; (ii) a Pareto-front, explainable rule-extraction methodology tailored to small-power PV; and (iii) a comparative assessment against standard ML baselines using multiple, class-aware metrics. The resulting XAI models achieved 92.3% precision and 89.7% recall. The findings inform the design and operation of PV systems for harsh, high-latitude environments.