Optimized Controller Design Using Hybrid Real-Time Model Identification with LSTM-Based Adaptive Control

Most of the processes with various dynamic characteristics can be reduced to the Second Order Plus Time Delay (SOPTD) model by using the model reduction method. We propose a novel hybrid approach that combines Long Short-Term Memory (LSTM)-based real-time model identification with Genetic Algorithms to enhance the Smith predictor control structure. This method compensates for the delay time of the SOPTD model while minimizing the Integral Time Absolute Error performance index. Our approach integrates an optimally adaptive Proportional–Integral–Derivative (PID) controller design algorithm that estimates the coefficients of the SOPTD model in the Smith Predictor control structure and adjusts the PID controller parameters dynamically. The method is improved through a combination of numerical calculation, Genetic Algorithms, and LSTM networks, showing approximately 15% better performance compared to conventional methods. The system demonstrates significant improvements in both performance metrics and resource utilization, including a 40% reduction in execution time and enhanced resource efficiency. Simulation results show that the proposed scheme exhibits improved adaptability to disturbances and process variations, with faster response times and reduced overshoots compared to traditional methods. The steady-state response of the higher-order model and the reduced model shows perfect matching for the unit feedback input.