Soft Sensing Of Intrinsic Viscosity Based On Atom Search Dimension Reduction Aggregation Process

In recent years,reducing production costs has not only become an urgent demand of enterprises,the production technology of polyester fiber is also the focus of academic research,the polymerization process is a key link in the production of polyester fiber,the intrinsic viscosity of reaction parameters has an important influence on the properties of polyester fiber.Through the analysis of the intrinsic viscosity content measured in the production process,we can evaluate the quality of the final product,so as to effectively reduce some unnecessary costs.So it is necessary to predict the intrinsic viscosity.However,in the process of industrial production,the measurement of intrinsic viscosity is often restricted by technology and expensive cost,so it is difficult to measure directly,resulting in the measurement of intrinsic viscosity can not be widely used in actual production.The paper offers an amazing learning model,which is coded on the basis of the binary atom search algorithm through this background,and applies it to the prediction of intrinsic viscosity in the polymerization process of polyester fiber production.First of all,with the distance meter of the nearest k-neighbor classification algorithm(KNN),combined with binary coding,atomic search algorithm is used to select features of industrial data to obtain the optimal data set.In order to collect data,the weight and the level of foolishness will be better for the best atomic search(ASO).Then,the mean square error(RMSE)fitting function is used to determine the off-line prediction model by increasing convergence and inclination of the activation function of the final learning machine.In order to further reduce the experimental error,an off-line viscosity prediction model was established to improve the prediction accuracy of natural viscosity.Based on the idea of Adaboost in ensemble learning combined with long term short-term memory(LSTM)networks,some weak trainers are introduced to create strong trainers.The main contributions of this paper are as follows(1)We propose a binary coding based method to control whether or not to select features.KNN algorithm to evaluate the correlation of features.Atomic search algorithm overcomes the shortcomings of other optimization algorithms,such as easy to fall into local optimum,poor effect and so on.A feature selection model is established to improve the interpretability of selected features.According to the fitness function of the above algorithm,the input variables are selected.Finally,select an input variable to predict the characteristic viscosity of the polyester:esterification backwater temperature T₁,EST temperature T₂,final polycondensation reactor temperature T₃,1222-h01 lead generation supply pressure P1,oligomer pipeline pressure P₂,DEG injection pressure P₃,fresh EG(ethylene glycol)addition F1,pre polycondensation reactor Ti O₂addition F₂.(2)A soft sensing algorithm for characteristic viscosity of extreme learning machine is proposed.The algorithm uses an atomic search algorithm to optimize functions of activation convergence and gradient problems and find optimal parameters,using atomic model search algorithms to avoid the effects of immune on the difference model.The input value and the threshold value of the beam hidden layer are repeatedly optimized.And the parameters are selected randomly.At the same time,aiming at advantages and disadvantages of the Re Lu function,the HLU function is improved as a new activation function of ELM.The simulation results show that this model has higher prediction accuracy than the model before optimization,which has a certain reference for the subsequent industrial production.(3)The LSTM neural network is embedded into Adaboost framework.Adaboost combines and trains weak classifiers with different LSTM blocks.In the training model process,Adaboost adjusts based on the error rate per weak classifier.The strong classifier is generated by averaging the weighted values of many training machines to get the prediction results and output the experimental prediction error.(4)Combining the above data preprocessing method with soft sensor prediction model,a complete set of intrinsic viscosity prediction model is established,which has a certain guiding role for practical industrial engineering.