Deep Learning Enhanced ANFIS-PID Control for Intelligent and Energy-efficient Wood Drying Systems

Abstract

Traditional wood drying systems suffer from significant control inaccuracies, excessive temperature fluctuations, and suboptimal energy efficiency, leading to wood defects and economic losses. This study introduces a Machine Learning–Enhanced Adaptive Network-Based Fuzzy Inference System with Proportional-Integral-Derivative control (ML-ANFIS-PID). The proposed system incorporates Long Short-Term Memory (LSTM) networks for temporal pattern recognition, Convolutional Neural Networks (CNN) for temporal feature extraction, and adaptive fuzzy inference for real-time control parameter optimization. The results of experimental validation on ayous wood show that the performance was significantly improved: 98.4% temperature prediction accuracy, 42-second rise time (8.7% faster than traditional ANFIS-PID), 0.02% overshoot (50% less than traditional PID), 58-second settling time (5% better than conventional PID), and remarkably low 0.04°C steady-state error (96% lower than traditional PID). Also, the ML-ANFIS-PID system attained 23.7% less energy use, 18.4% less drying duration, and 31.2% less defect rate with high-quality wood produced under the tested ayous wood conditions. These results demonstrated significant performance improvement compared with classical and adaptive PID-based controllers under controlled experimental conditions.