Study On Gas Concentration Prediction Methods Based On The PSO-SVR

Gas disaster is one of the most serious disasters in coal mine safety production.Therefore,it is of great significance to research and analyze statistical characteristic of gas data,designing and building effective algorithm model to acquiring accurate prediction and improving the ability of prevention and control gas disaster in coal mine.In this thesis,three methods of gas concentration prediction are proposed based on particle swarm optimization support vector regression(PSO-SVR)for feature extraction、spatiotemporal modeling and trend prediction respectively.The specific research contents are as follows:1.Research data collection 、 analysis and processing.Making use of a variety of commonly methods to clearning the gas concentration data.Testing for linearity and gaussianity of discrete gas time series based on higher order statistics and making a foundation for the following research.2.A new gas concentration prediction model called DBN-PSO-SVR based on deep belief network is proposed which can describe deep characteristics of samples.Using deep belief network to extract information of data,results of feature extraction are the input of traditional PSO-SVR prediction model.Simulation shows that the proposed prediction method forecasting accuracy is higher and algorithm performance is better compared with traditional algorithm.3.The gas time series are proved to have spatial and temporal characteristics according to the research of gas diffusion 、 migration rule and the statistical number features of gas concentration data.An algorithm of spatiotemporal modeling is proposed based on the above analysis.A new method of establishing spatial weight matrix based on sample data is proposed simultaneously.Training set and testing set are established in accordance with spatial weight matrix and spatiotemporal delay.Obtaining predicted results by using PSO-SVR model for training and predicting.The results of simulation show that the method has large improvement on the generalization ability and prediction accuracy compared with the traditional method which only depends on the time dimension information.4.A new trend prediction model of gas concentration based on fuzzy information granulation is proposed for getting information of change tendency and change room.Constructing triangular fuzzy particles and as the input of PSO-SVR prediction model for realizing predicting the change tendency and range of gas concentration.Research spatiotemporal prediction concurrently combined with spatiotemporal modeling methodwhich proposed in the fourth chapter.The effectiveness of the algorithm is verified by the results of simulation.