Interpolation And Prediction Of Coal Mine Gas Monitoring Data

Underground coal mine is the majority of China's coal mine. Due to geological structure of coal resources is very complex, and groundwater, fire, gas, underground temperature, coal dust and the roof and other multi-phase media in coal stratum are impacted by dynamic work of mining engineering, natural characteristics of physical, chemical and mechanical of solid, liquid and gas in the underground environment are changed. Instability and loss of control under disturbances of in-situ stress and other external factors, result in rock fracture, roadway damage, coal and dust outburst, flooding,gas, coal mine disasters,etc.. Gas accident is a relatively serious accident hazard. In 2014, there were 47 gas accidents in coal mines of the whole country, with the death of 266 people, respectively down from 15 and 101, down 24.2% and 27.5%, respectively. Although in recent years, a significant decline in coal mine gas accidents has taken place year by year, but there are still a large amount of gas accidents, major gas accidents have not been effectively curbed, illegal coal mine production gas accidents are prone to occur, coal mine gas prevention and control situation is still severe.Because of the diversity of rock mass destruction, physical and mechanical nonlinear properties of coal rock and complexity of gas occurrence and migration process, the causes, process and some of the details of gas accidents are not very clear. there are a lot of special accident phenomena which could not be explained on the scene. The main reason is that various types of monitoring data related to gas on the spot have not been fully excavated and used, and the study of gas prediction technology is still far from adequate, forecast accuracy is not high.Through analysis of current situation of domestic and international gas prediction, coal mine gas analysis and predication's key lies in the effectiveness and practicality of gas analysis, and there are great limitations of traditional methods in the analysis of gas emission rules and predictive model building process,in selection of influencing factors, construction molding process and in fields of practical application.Rock mechanical behavior has characteristics of the nonlinear nature,gas movement rule also has nonlinear complex dynamic properties,therefore, the prediction of gas should take full account of the nonlinear characteristics of gas data, using the current nonlinear theory to find a new prediction method of gas.Given mine gas monitoring data with chaotic, using chaotic theory to do mine gas data analysis and modeling is scientific and practical. Thesis firstly carried out extensive literature research, carried on the profound analysis, the coal mine is divided into low gas mine and high gas mine, as to high gas mine it selects Shanxi Yangmei group Xinyuan coal mine as experimental object,as to low gas mine it chooses Beijing JingMei Group Muchengjian coal mine,Henan Energy and Chemical Industry Group Chengjiao coal mine and Gengcun coal mine as the subjects, these subjects covers north China, central China's main mines, which have extensive representatives;and it did in-depth analysis of the gas data chaos characteristics. Based on this it did phase space reconstruction, model eatablishment and comparative study on the gas data interpolation algorithm,finally figured out gas chaotic time series prediction model analysis, application and validation.The main work done is summarized as follows:(1)Chaos theory research.It studies the origin of chaos theory, basic features of chaotic system and time series' chaotic characteristics determination method. On this basis, the selection of gas time series sample data was classified, according to different sampling periods, respectively according to every 5 minutes, per hour and taking the mean of the daily gas time series as the basis of sample data. Through qualitative and quantitative analysis, it uses small-data method and the GP algorithm method to seek maximum Lyapunov exponent value and correlation dimension of chaotic systems feature quantity, calculated results shows that correlation dimensions of three types of gas time series samples are non-positive integers, the maximum Lyapunov exponents are greater than zero. Thus proving that gas time series has chaotic properties and the short-term prediction of chaotic systems is feasible.(2)Gas chaotic time series phase space reconstruction.It introduces the basic theory of phase space reconstruction, and does comparative analysis using delay coordinate method of two major parameter selection method---embedding dimension and delay time in the reconstruction phase space. For gas concentration time series sample of three different sampling periods, when determining delay time interval,it uses respectively autocorrelation and mutual information method to do comparison the obtained results shows that using the mutual information method does better in line with non-linear characteristics of gas concentration time series. It also introduces Cao's method and pseudo-adjacent points(FNN) method to determine the embedding dimension.The pseudo-near-point method(FNN) is relatively more intuitive, so it uses a pseudo neighboring point method(FNN)to seek embedding dimension.(3)Research on cubic spline interpolation algorithm based on particle swarm in the application of gas missing data. Firstly, the advantages and disadvantages of several common interpolation methods in present are analyzed. Secondly, research on cubic spline interpolation method and particle swarm algorithm. Cubic spline interpolation curve is smooth, interpolation accuracy is high, but calculating the coefficient of cubic spline interpolation function is complicated, computational efficiency is low. Particle swarm algorithm is simple, few parameters, easy to implement, and in a relatively short period of time it can produce high-quality solution. To improve particle swarm algorithm, inertia weight of particle swarm algorithm is proposed to use a linear decreasing inertia weight to achieve balance of global search and local search. Finally, cubic spline interpolation algorithm based on particle swarm is proposed for gas missing data, which use auto-correlation analysis to determine the number of nodes of cubic spline function, and use particle swarm algorithm to calculate the coefficient of cubic spline interpolation function. Cubic spline interpolation algorithm based on particle swarm with characteristics of simple operation, easy to implement, high computationally efficiency of particle swarm algorithm, and smooth interpolating curve, high interpolation accuracy of cubic spline interpolation algorithm, is a better method to treat missing data. Through analysis of the application examples of gas missing data interpolation of different work faces in the same place and different work faces in the different places, the algorithm is valid for gas missing data interpolation. To compare with several common interpolation methods in present, cubic spline interpolation algorithm based on particle swarm is more accurate and effective.(4)Gas chaotic time series prediction model analysis, application and verification. Taking one-dimensional gas time series of the mine face under normal operating conditions in safety monitoring system's monitored data as samples, it reconstructs phase space of different sampling precision, different amounts of data samples.Combined with RBF neural network prediction method and adaptive prediction method to build predictive models, the prediction processes are implemented by MATLAB programming.Prediction results were analyzed and summarized in three-dimensional way,we draw corresponding conclusions as follows:?Through the establishment of RBF prediction model and adaptive prediction model and its application, they both achieve a high prediction accuracy. From the experimental results,Volterra adaptive prediction is superior to RBF neural network in predicting the results.In the case of a sufficient amount of data, the smaller the sampling period, the higher the accuracy of the prediction.?RBF neural network model is fitted to a certain trajectory of the chaotic system, other than training samples ability,it extrapolated no good analysis. So we should choose a large amount of data samples, but will face relatively long convergence time. Therefore, it proposes s the sample covers a wide change range as much as possible, taking the 2000-3000 samples per hour can achieve higher prediction accuracy.?Adaptive prediction model is to use currently known data and predict changes in the current to modify real-time model parameter weights, so as to achieve higher accuracy.Taking 1000-2000 as the number of samples can achieve better prediction accuracy,the sample does not need to take an infinite amount of data. Because the Volterra adaptive model can take comprehensive use of nonlinear and linear characteristics factors,which are in line with the nonlinear characteristics of chaotic time series, and has good noise immunity. Therefore, the use of adaptive Volterra model can effectively predict low-dimensional chaotic time series.?It respectively selects from two different coal mine to do gas time series chaotic properties analysis,and prediction results are verified. Experimental results show that the two prediction models established using chaotic theory phase space reconstruction have applicability to different geological conditions of different coal mine.Different types of sample predication results are consistent with the previous conclusions,further validating for nonlinear gas chaotic time series chaotic short-term predication's effectiveness. In addition, through the analysis of gas outburst time series' chaotic characteristics in abnormal state of time,we found that about 3 days before the gas outburst, gas characteristics of chaotic time series is more intense. which can prove that the variation of gas time series of chaotic characteristics, can be used as coal mine gas accident symptom judgment basis.. Implementation of gas time series short-term prediction and abnormal symptom judgment before gas accident occurs.