Biologically Based Intelligent Multi-Objective Optimization for Automatically Deriving Explainable Rule Set for PV...

Area of Science:

• Renewable Energy Systems
• Environmental Science
• Machine Learning Applications

Background:

• Antarctic research stations need dependable, low-carbon power sources.
• Extreme polar conditions pose significant challenges for energy generation.

Purpose of the Study:

• To develop and validate an interpretable, multi-objective framework for photovoltaic (PV) power systems in Antarctica.
• To create a synchronized PV-meteorological data set for polar conditions.
• To compare the performance of different PV module types.

Main Methods:

• Compiled a high-resolution (30s, 1min, 5min) PV-meteorological time-series data set on Horseshoe Island.
• Developed a modified SPEA-2 algorithm to optimize precision and recall for rule extraction.
• Compared rule-based models with benchmark machine learning models (kNN, SVM).
• Evaluated performance using precision, recall, F1 score, balanced accuracy, and MCC.

Main Results:

• The proposed rule-based approach achieved competitive predictive performance, interpretability, and stability.
• Explainable AI (XAI) models demonstrated 92.3% precision and 89.7% recall.
• Performance remained robust across different PV panel types and sampling intervals.

Conclusions:

• The study provides a novel, explainable AI methodology for PV systems in harsh, high-latitude environments.
• The findings support the design and operation of reliable PV systems for polar research stations.
• A valuable high-resolution data set for Antarctic PV research was created.