The Chaotic Characteristic Analysis And Prediction Research Of Gas Time Series

Gas disaster is one of the important factors that threaten our country coal mine safetyproduction. Finding out the laws of the mine gas emission data to realize the accurate andreliable prediction of gas can avoid the occurrence of disasters effectively. The gasemission system is affected by many factors and much random disturbance c, it is difficultto make a prediction model by analyzing the influence factors. To solve this problems, weintroduces the chaos theory to the forecast of the gas emission, restore the space time byreconstructing the gas time series on phase space which avoid the predictive modelingexperience and subjectivity. The main work in this paper is introduced as follows:First, we choose the1024working face of one coal mine to make a study and identifythe chaos characteristics of gas emission time series by using the maximum Lyapunovindex method and power spectrum method. Make a precondition of the gas emission timeseries by using the empirical mode decomposition (EMD) method, filtering the highfrequency component as noise and keeping the others as the useful information. Set up theprediction model of each useful components respectively, and multiply all the predictedresults as the final prediction results.Then, based on the modeling characteristics analysis of feed-forward and feedbackneural network, and take the gas time series’ dynamic characteristic into considering, weproposed the modeling scheme that based on echo state network. Aiming at the problemthat the stationary phase space cannot meet the prediction requirements with the increaseof prediction length, the maximum mutual information method is proposed, in which themutual information function is taken as the individual fitness function. We can choose themost beneficial spatial structure by calculating the individual fitness value under differentm andτcombination, and the effectiveness of the proposed method is verified byexperiment.At last, we created single-step prediction model, direct multi-step predictive modeland iterative multi-step predictive model of the gas emission based on chaotic analysis,empirical mode decomposition and the echo state network. We compared the ESNsingle-step prediction model with the BP model and the ELMAN model at the same condition, and verified the superiority of ESN model by comparison. We compared thedirect multi-step predictive model with the iterative multi-step prediction model andverified the superiority of the direct multi-step predictive model.