Chaos Characteristics Analysis And Intelligent Prediction Study Of Rockburst Based On Monitoring Time-series

Rockburst, a typical mine disaster, is increasingly harmful with the expansion ofmining scale and depth, and has become a major issue to be solved in the mining. It isdifficult to establish precise mathematical model of complex damage process ofRockburst. Since various measures to monitor Rockburst have been taken in almostall of coal mines, massive data can be achieved by monitoring Rockburst. Based onhistorical time-series monitoring data, this thesis analyzes chaos characteristics ofRockburst data in reconstructed phase space, and studies intelligent algorithms topredict multiple variables monitoring Rockburst and integrated classification methodto identify Rockburst risk on chaos prediction theory.First, through analyzing and pretreating monitoring data, this thesis acquirestime-series to indicate dynamics characterize of coal damage, such as Microseismiccumulative energy, maximum energy, frequency and amplitude of electromagneticradiation amplitude, and pulse count, et al. Based on single variable time-series phasespace reconstruction, the chaos characteristic of each time-series monitoringRockburst is judged by power spectral method and principal component analysis.Second, aiming at the multi-dimension of evaluating reconstruction results andthe correlation between reconstruction variables and reconstruction parameters inphase space, this thesis proposes an improved multi-objective immune algorithm(IMOIA) to simultaneously determine reconstruction variables and parameters ofmultivariate time-series reconstruction. The algorithm is employed to two benchmarkchaos system and its effectiveness has been verified. When employing to Rockburstdata, the results show that the algorithm can simultaneously solve multiple optimalcombinations of reconstruction variables and reconstruction parameters, whichprovide foundation for chaos analysis and prediction of Rockburst monitoring data onmultivariate time-series reconstruction.Third, considering that multivariate time-series reconstruction can offset datalength inadequacy and noise influence, chaos geometric invariants of Rockburstmonitoring data are solved based on multivariate reconstruction. G-P algorithm isextended to solve correlation dimension d2and nonlinear least squares method is putforward to calculate Largest Lyapunov Exponent (LLE) of multivariate time-serieswith noise. The extended G-P algorithm is employed to Lorenz chaotic system and itseffectiveness is proved. Meanwhile, it is observed that d2of all Rockburst dataobtained by above algorithm is fraction. This shows that monitoring data are chaos and their complexity can be judged by d2value. The applicability of LLE calculationmethod is verified by employing it to Rossler coupled system. And the LLE ofRockburst data also demonstrate that these data have chaos characteristic, whichprovide a power support for short-term forecasting of Rockburst.Fourth, taking the reconstruction state phasor as input variables, multiplevariables monitoring Rockburst are predicted base on GRNN to realize indirectlyforecasting of Rockburst accident. The results of GRNN model indicate that the futurevalue of multiple variables monitoring Rockburst can be predicted under a certainlength data. And the more data types, the more variables employing to reconstruct, thegreater embedding dimension is, hence the higher the prediction precision are. Thenbased on LSSVM, MLSSVM model is proposed to predict multi-variables monitoringRockburst on small data samples. MLSSVM sets different model paraments fordifferent output, and the paraments are optimized on immune algorithm by minimizeoverall fitting error of all outputs and single-output fitting error, so as to achieveoverall optimal prediction model of each output. It is applied to predict multiplevariables monitoring Rockburst of three mines (work faces), and the results show thatMLSSVM has stronger generalization ability and can achieve small prediction error inlimited training samples.Finally, the state phasor reconstructed by multi-source time-serie is taken asinput of identification model to directly predict Rockburst risk in view of that it candescribe the system characteristic very well. Because the attributes of reconstructedstate phasor have certain complementary and redundancy, subspace selectiveensemble algorithm based on feature clustering is presented. The results indicate thatthe ensemble algorithm can effectively improve classification accuracy on differentexperimental data sets, and directly forecast Rockburst risk according to historicaldata of multi-source monitoring time-series.The complex feature of Rockburst monitoring data is studied from the view ofchaotic time-series on the basis of chaos dynamics, and the intelligent models areestablished to forecast monitoring variables and Rockburst risk based on chaoticprediction theory. This thesis beneficially explores chaos theory and intelligentprediction algorithm, and provides a theoretical basis and practical significance forshort-term forecasting Rockburst on monitoring data of mining process.