A specialized multi-model adaptive network-based fuzzy inference system approach for optimizing latent heat storage systems from sparse experimental data

Abstract

While experimental analysis of thermal energy storage systems yields crucial data, the process is frequently constrained by significant time investment and a limited scope. To overcome these limitations, this study proposes a specialized divide-and-conquer multi-model framework to analyze and optimize the performance of a U-tube heat exchanger integrated with a phase change material (PCM). Instead of a single, complex global model, specialized Adaptive Network-Based Fuzzy Inference System (ANFIS) models were developed for discrete experimental conditions using sparse legacy data. This architectural partitioning effectively transforms historical datasets into high-fidelity digital twins capable of capturing nuanced, localized thermal dynamics. A rigorous repeated hold-out validation demonstrated outstanding predictive accuracy, with mean R-squared (R2) values consistently exceeding 0.92. The validated models were then employed to perform optimization analyses, demonstrating that operational parameters significantly impact charging durations, allowing for substantial performance improvements that are impractical to achieve through direct experimentation alone. Ultimately, this study highlights that specialized ANFIS models provide a robust and agile framework for the comprehensive analysis and rapid enhancement of complex thermal systems.