Construction Of A Data-driven Prediction Model For Main Performances Of Particleboard

Particleboard is one of the three main wood composites that are very important in production and life.Faced with the shortage of raw materials,energy consumption and fierce competition in the industry,these approaches such as improving production efficiency,reducing production costs,enhancing product quality and product stability can be used to increase the competitiveness of particleboard enterprises.Particleboard production process has an important impact on many properties of particleboard.Therefore,how to realize the intelligent manufacturing of particleboard through the intelligent feedback and regulation of particleboard production process,so as to further effectively improve the product quality and production efficiency of particleboard,has become the concern of particleboard manufacturers and scholars at home and abroad.In this paper,based on the production records of particleboard enterprises and the performance analysis and detection data of corresponding particleboard products,the data-driven method combined with the principle of particleboard production technologyisused to reveal the mapping relationship between particleboard production process parameters and their performance,evaluate and clean the particleboard performance data,and select the most suitable prediction method of particleboard performance.An effective model for online prediction of particleboard production process parameters on its performance is constructed,and the reliability of the prediction model is verified through pilot production test.The research results lay a good theoretical and practical foundation of the online prediction of particleboard performance in the process of production,the feedback and regulation of particleboard production process,and the benefit improvement and intelligent manufacturing upgrading of the existing traditionalparticleboard enterprises.The main research contents and results are as follows:(1)In the aspect of reliability evaluation of particleboard performance data,the performance of particleboard is analyzed by using the test data of particleboard enterprises.Through reducing the interference of performance data uncertainty and unreliable factors on the prediction results,a method for evaluating and cleaning particleboard performance parameters based on IER rule is proposed,which provides a reliable guarantee for the subsequent prediction of the impact of different production processes on the properties of particleboard.Compared with the evaluation results of BP,ELM and RBF methods,the evaluation result acquired with the IER rule method has a good fitting effect,and the mean square error is only 1.84×10-4.It is proved that the evidential reasoning evaluation method while considering the reliability interval state is effective,which can be used to solve the uncertain data in the form of interval and clean the unreliable data.(2)In the aspect of optimizing particleboard performance prediction method,based on the production process parameters and performance test data of particleboard enterprises,a data-driven particleboard performance prediction method is proposed.The data with different dimensions are standardized by normalization,and then 10 prediction methods such as Random Forest,Decision Tree,K-nearest neighbor,Adaptive boosting and Support vector regression are used to evaluate the prediction results with three different evaluation indexes:mean absolute error,mean square error and mean absolute percentage error.The most suitable method for predicting the properties of particleboard is selectedusing the process parameters of particleboard.Among them,the Random Forest method shows outstanding effect on the prediction of 8 properties of particleboard.It has excellent generalization ability for the three evaluation indexes of average absolute error,mean square error and average absolute percentage error.The minimum mean square error of the eight predicted properties of particleboard is7.90×10-3,and the maximum is only 2.38×10-2.Therefore,the Random Forest is finally determined as the most suitable prediction method for the performance of particleboard.(3)In the construction of particleboard performance prediction model,in order to better find out the key factors affecting the performance of particleboard,a particleboard performance prediction model based on principal component regression and random forest is proposed.Using the principal component regression method that can effectively solve the multi-linear problem between independent variables,multi-factor regression fitting is performed on the data set of the process parameters and the corresponding board performance in the production of particleboard enterprises,and the key influencing factors of the performance of each board are obtained.Based on the optimal random forest prediction method,its effectiveness was verified,and a particleboard performance prediction model based on principal component regression and random forest method was constructed.Through the normalization and standardization of process records of particleboard enterprises and their corresponding particleboard performance test data,the establishment of matrix,the determination of eigenvalues and eigenvectors,and the calculation of eigenvalue information contribution rate,six eigenvalues with eigenvalues greater than 1 and cumulative information contribution rate greater than 60%are obtained.On the basis of random forest method,with regard to the performance and related process parameters of particleboard,six principal components are selected for multivariate linear fitting.The regression equation between particleboard production process parameters and their performances is obtained.The analysis of key influencing factors determined by the equation shows that the constructed prediction model is in line with the principle of particleboard production process and the influence mechanism of particleboard performance.(4)In the verification of the prediction model,the pilot production test data are used to verify and evaluate the prediction reliability of the optimized particleboard production process parameters and their main properties.During the pilot production under the control of six kinds of parameters of three process parameters:"moisture content","hot-pressing temperature" and "curing agent dosage",the relative errors between the predicted values of the optimized prediction model and the measured values of the properties of the particleboard in the pilot production are less than 4.1%,and most of the relative errors are between 1.3-3.8%.It is fully confirmed that the optimized prediction model can be used effectively to predict the main properties of particleboard by the production process parameters,and has good reliability.Therefore,the prediction model can realize online nondestructive prediction,performance monitoring,production process regulation and feedback for the production of particleboard,so as to provide technical guarantee for the stability of product quality,production efficiency and benefit improvement of particleboard enterprises,and also provide a theoretical basis for the intelligent manufacturing upgrading of particleboard enterprises.