Research on an LSTM-PID Temperature Control Method With Adaptive Parameter Tuning Incorporating Load Forecasting

Temperature control systems subjected to random disturbances present significant challenges in maintaining stability and precision. Conventional fixed-parameter proportional–integral–derivative (PID) controllers often suffer from low accuracy, overshoot, and oscillations. Although adaptive PID controllers with tunable parameters can enhance robustness, achieving stringent performance requirements in high-precision temperature regulation remains difficult. To address these limitations, this work proposes an adaptive temperature control strategy that integrates load prediction with a long short-term memory (LSTM)-based PID controller. A load prediction module, implemented using an LSTM neural network, is first developed to model thermal inertia and forecast the control output in real time based on ambient temperature, target temperature, and the current operating state. The predicted output is then refined by the LSTM-PID controller to perform fine-scale temperature regulation. A PLC-based freezer temperature control platform is established for experimental validation. Comparative results demonstrate that the proposed method achieves accurate load prediction, significantly reduces the control range required by the LSTM-PID, and enables dynamic PID parameter tuning for stable regulation with minimal overshoot. When regular PID approaches fail in model-free, strongly disturbed, and high-demand scenarios, the proposed framework provides a novel and effective control solution.