Research on parameter prediction and control model of intelligent manufacturing process based on adaptive optimization algorithm

In this paper, an Adaptive Optimization-based Predictive Control Model (AOCM) based on adaptive optimization algorithm is studied to solve the coupling problem between parameter dynamic characteristics and control requirements in intelligent manufacturing process. Traditional methods are difficult to deal with complex working conditions in high-dimensional nonlinear systems. Therefore, a closed-loop control model combining data-driven and adaptive optimization technology is proposed in this study. Through the improved adaptive optimization algorithm and deep learning prediction module, the dynamic modeling of high-dimensional nonlinear system is realized, the time-series dependence and cross-dimensional coupling relationship between parameters are captured, and the adaptive control strategy is generated according to the real-time prediction results and multi-objective constraints. In the data preprocessing stage, the mixed filling model based on GAN-Transformer is used to deal with the missing and noise problems in sensor data, and the features are extracted by improved wavelet packet decomposition method. The dynamic parameter prediction model is based on the improved Transformer architecture and adaptive attention mechanism, which enhances the modeling ability of the interaction between parameters. In the adaptive optimization control module, the improved constrained multi-objective NSGA-III algorithm combined with dynamic penalty function is used to optimize the quality deviation, energy consumption and the change rate of control quantity. In addition, the model realizes online learning and dynamic updating through sliding window incremental learning combined with Bayesian hyperparametric optimization. The experimental results show that AOPCM is superior to traditional methods in prediction accuracy, response speed and stability. Compared with LSTM, TCN and Prophet models, AOPCM significantly reduces the prediction errors of mold temperature, injection pressure and cooling rate. In terms of control performance, the response time of AOPCM is shortened by 38%, the energy consumption is reduced by 24%, and the qualified rate is increased by 2.4 percentage points. AOPCM shows better robustness when the parameters change suddenly, the control variance is reduced by 41%, the recovery time is shortened by 44% and the cold start error is reduced by 58%.