| At present,traditional industrial production enterprises are facing the trials of the era of transformation and upgrading,the tasks of pollution prevention and energy saving and consumption reduction are arduous.As a high energy consuming industry,the industrial output and power consumption of non-ferrous metal industry are closely related to parameter regulation in the production process.In actual production,technicians adjust parameters based on the experience accumulated in the past.Due to the differences in personnel experience and the complexity of production,this method is difficult to make scientific research and timely adjustment to the current production situation of equipment,so it has a certain lag.As information technology advances,machine learning algorithms provide more solutions for industrial production.In view of the complex relationship between process parameters and the difficulty of enterprise production control,this thesis uses the accumulated aluminum electrolytic process data of enterprises to conduct data analysis through machine learning algorithm modeling,selects the key technical indicators,establishes the problem optimization model,and combines the intelligent optimization algorithm to optimize the production process parameters.It provides decision-making basis for enterprise technicians to adjust parameters.The following constitutes the primary research for this thesis:(1)In order to solve the problem that it is difficult to determine the optimization objective of process parameters,a process parameter prediction model based on TPE-Light GBM was established in this thesis.The preprocessed industrial production data is used as the input data of the Light GBM model for training.The independent adjustment of Light GBM parameters was realized through TPE Bayesian optimization and Optuna.The experimental results demonstrate that the prediction model established in this thesis has a minimal prediction error and a high accuracy,which lays the model groundwork for the ensuing optimization algorithm to be used for the optimization search.(2)Improvements are made to solve the problems of low random initialization quality and slow convergence speed of the marine predator algorithm.The thesis adopts adopts the chaotic opposition method to initialize the population,and introduces the adaptive t distribution and chaotic search strategy to enhance the search ability of the original algorithm.The experimental results show that compared with the standard marine predator algorithm,the improved algorithm has higher optimization accuracy and stability,and can effectively optimize the process parameters.(3)Establish a process parameter optimization model to achieve the best parameter range matching.Through the trained process parameter prediction model,the objective function of the optimization problem is established.The improved marine predator algorithm is used to solve the problem,and multiple sets of parameter point sets corresponding to the optimal production index are output.After Uniform Manifold Approximation and Projection dimension processing,range division is realized through variational Bayesian Gaussian Mixture clustering.(4)Combined with the above algorithm research,the production process parameter optimization system is designed.The system realizes the functions of data visualization analysis,data preprocessing,model prediction and parameter optimization,which can provide guidance for production technicians. |
