Research On Prediction And Optimization Of Oil Processing Based On Surrogate Model

With the development of China’s economy and the continuous improvement of people’s material living standards,people’s consumption level of edible oil continues to improve.In order to meet people’s increasing consumption demand for edible oil,the production scale of oil industry is expanding,prompting enterprises to put forward higher pursuits for production indicators such as yield,quality,energy consumption and and efficiency,so as to comply with the trend of sustainable production.By establishing the prediction model of production indicators and combined with intelligent optimization technology,the processing parameters can be accurately regulated,so as to reduce energy consumption and improving the quality and yield of edible oil.Surrogate model is an approximate mathematical model(also known as approximate model or metamodel)of the original system trained from limited experimental data.It has the advantages of low modeling cost,simple operation and high computational efficiency.The application of agent model technology to the prediction and optimization of oil processing process is an effective method,and it is also the main research topic at home and abroad.Therefore,based on the theory of surrogate model and taking oil processing as the application background,this paper intends to systematically carry out the research on the prediction and optimization methods of oil processing.The main work of this paper is summarized as follows:(1)In order to improve the robustness and accuracy of the surrogate model to adapt to varies oil processing scenarios,an ensemble of surrogate model based on the hybrid error metric(ES-HEM)is designed.In this method,the hybrid error metric combines the stability of the global error metric and the flexibility of the local error metric to improve the predictive ability of the ensemble model.At the same time,considering the influence of the difference in sample distribution near different prediction points on the local accuracy of the model,a distance criterion is introduced in the adopted PV(prediction variance)criterion to improve the flexibility of the local error metric at different prediction points.In addition,in order to improve the robustness of ES-HEM in dealing with noise data in actual oil processing engineering,regression models with the ability to filter noise are selected to construct ESHEM.Finally,the performance of ES-HEM is verified by five numerical examples and decolorization experiment of camellia oil.The results show that the proposed method has high accuracy and robustness.(2)In order to improve the global prediction accuracy of the surrogate model,a sequential sampling method based on Bootstrap prediction error is proposed.This method first uses Bootstrap resampling strategy instead of cross-validation(CV)to evaluate the prediction error of the surrogate model at the prediction point,avoiding the phenomenon that the estimated error value is too large for samples at and around the stagnation point of the prediction function;secondly,a distance criterion is introduced to prevent sample aggregation;Then,the performance of the method is verified by three numerical examples with different noise levels.The results show that the proposed method has high global prediction accuracy.Finally,the proposed sequential sampling method is applied to improve the performance of ES-HEM of the decolorization rate and peroxide value.On this basis,the sensitivity and influence trend of Camellia oil decolorization parameters are analyzed.(3)In order to solve the multi-objective optimization problem of oil processing process and improve the reliability of optimization,a multi-objective optimization method based on ensemble of surrogate and NSGA-II(non-dominated sorting genetic algorithm)is proposed.In this method,the optimization process is divided into two steps: Modeling and multiobjective optimization.In the modeling stage,the ES-HEM is selected to establish the relationship between process parameters and production indexes due to its advantages of better prediction accuracy and robustness than individual surrogate models.On this basis,the global sensitive and influence trend of process parameters are analyzed to clarify the conflict relationship between optimization goals and the importance of process parameters.In optimization,the ability of NSGA-II to solve multi-objective optimization problems including conflict relations is used to obtain the Pareto frontier of multiple objectives for the sake of balancing their conflict relationship.Finally,the proposed method is used to obtain the Pareto frontier for the maximum oil yield and minimum energy consumption in the extraction process of Camellia oil,and provides multiple process parameters for manufacturers.(4)In order to improve the convergence efficiency of the optimization method for oil processing engineering,a clustering algorithm and ES-HEM-assisted parallel(or multiple points)sequence optimization method(DEAM-EGO)is proposed.In this method,the weight method is used to convert the expected improvement(EI)function of the standard EGO algorithm into multiple single-objective optimization problems,and a reasonable weight function is designed to obtain multiple candidate samples.Moreover,in order to improve the diversity and convergence of the solution,the above candidate samples are classified through a density-based clustering algorithm(DBSCAN),then the ES-HEM is used to evaluate the response value of the classified samples,and finally the candidate samples with the smallest response value in each category and the noise samples are used as filling samples.In addition,in order to improve the robustness of DEAM-EGO in dealing with noise data in actual oil processing engineering,Kriging regression is selected to replace the interpolation Kriging in the standard EGO in modeling.Finally,the performance of the proposed method is verified by five numerical examples and camellia oil processing experiments,and the results show that the proposed method has high solving efficiency and can provide technical guidance for improving the yield and quality of camellia oil.Finally,this work is summarized and the future research is prospected.