Prediction and Analysis of Time Series with the aid of Granular Parallel Support Vector Machine

The analysis and prediction of mine pressure in a fully mechanized mining face are of great significance to roof management and the safety of mine production under complex geological conditions. In order to possess the dynamic change of ore pressure data, this study proposes a method of analyzing ore pressure time series using an approximate support vector machine, namely Parallel Support Vector Machine (PSVM), and the regression analysis and prediction of ore pressure data are conducted by utilizing this method. Experiments are carried out based on the data collected in Caojiatan coal of YuBei mining company from May 2020 to June 2020. First, a parabolic fuzzy granule is introduced to the approximate principle of support vector machine (SVM) to adopt the method of cross-validation of relevant parameters optimization. Then, the optimized parameters are applied to train the time series. Finally, a regression predicting fuzzy granule with three parameters is determined to predict the dynamic changes of the pressure data. By comparing the original data with the predicted granular data, especially for the nonlinear data which is difficult to predict, the proposed method shows a better performance in mine pressure prediction and analysis.