Parameters priority analysis for improving radio frequency heating uniformity in agricultural products drying based on machine learning

Abstract

Drying is essential to reduce postharvest agricultural product losses to ensure food security. Traditional drying methods like hot air drying suffer from high energy consumption and low product quality. Radio frequency (RF) drying has emerged as a promising alternative to traditional methods due to its volumetric heating and deep penetration advantages. However, non-uniform heating of RF technology remains a critical challenge restricting further commercialization. Previous approaches (such as finite element software) in improving RF heating uniformity have limitations. Therefore, a parameter priority analysis method combining literature data and machine learning was proposed to improve RF heating uniformity in agricultural products drying. A total of 356 data points were extracted from 18 selected literature on RF drying via a systematic search (2014-2024) of databases, including Web of Science and Google Scholar, constructing a dataset after applying screening criteria. Twelve machine learning models were used for analysis, with superior models selected based on model performance. The gradient boosting models were selected for parameter priority analysis. Parameter priority analysis identified moisture content and material (agricultural products) thickness as key parameters (thickness dominated in fruits & vegetables, moisture content in nuts) influencing RF heating uniformity. Across heating methods (RF, Hot air-assisted RF, Vacuum-RF), moisture content remained the most influential, with electrode gap as the primary parameter for RF and Hot air-assisted RF, and moisture content dominating in Vacuum-RF. The developed parameter priority analysis method can be further expanded to improve RF heating uniformity in different application scenarios (such as thawing and pasteurization).