关键词: 核主成分分析, 改进粒子群算法, 最小二乘支持向量机, 充填管道, 磨损风险, 组合预测模型

Abstract:

Filling pipeline wear system is a typical high-dimensional，nonlinear，strong coupling and multi-time-varying complex system.It is difficult to accurately predict the wear situation using traditional prediction methods.In order to overcome the poor applicability and insufficient prediction accuracy of traditional prediction models，and the defect of strong randomness in parameter selection and so on，this paper proposed a new method for predicting the wear risk of filling pipeline by combining the kernel principal component analysis （KPCA），improved particle swarm optimization （IPSO） and least squares support vector machine （LSSVM）.A comprehensive selection of 12 main influencing factors was used to establish the prediction index of wear risk of the filling pipeline.The KPCA method was used for feature extraction and dimensionality reduction of the influencing factors of the filling pipeline to eliminate redundant information between the data，so as to reduce the correlation between sample data and modeling accuracy impact.Established the corresponding LSSVM prediction model based on the dimensionality-reduced data，and used the IPSO algorithm with strong global search capability to optimize the model parameters to avoid the blindness of artificial parameter selection，thereby improving the model prediction accuracy and establishing KPCA-IPSO-LSSVM combined prediction model.Taking the filling system of Huangling County mining area as an example，combining the 80 sets of sample data measured in the field，MATLAB was used to train and predict the built model，and the prediction results were compared with the prediction results of the IPSO-LSSVM model，LSSVM model，and SVM model.For comparison，multi-error indicators were used to comprehensively analyze the prediction results of the four models.The research results show that the predicted value of the constructed model is basically consistent with the actual value curve.The KPCA method can effectively reduce the redundant information between the data.On the basis of retaining the original sample information to the maximum，five principal components containing 86.97% of the original information are extracted，it simplifies the calculation structure of the model.The prediction accuracy of the adopted IPSO-LSSVM model is 6.79%，the average relative error is 1.95%，and the judgment coefficient is 99.55%.Compared with other prediction models，the prediction model based on KPCA-IPSO-LSSVM has higher prediction accuracy and stronger generalization ability，which provides a more effective prediction method for the prediction of the wear risk of the filling pipeline，and provides a guiding basis for ensuring the smooth progress of the filling operation and the safe production of the mine.

Key words: kernel principal component analysis, improved particle swarm algorithm, least squares support vector machine, filling pipeline, wear risk, combined prediction model