Study On The Gas Concentration Prediction System In The Upper Corner

Because of the complex geological conditions of coal mine,various disasters may happen during mining,such as gas explosion,dust,water inrush and other disasters.Excessive gas concentration seriously threatens the safety of coal mine,especially the corner of comprehensive mining face,which is easy to cause gas explosion because of the special geographical position and spatial structure.This subject focuses on the gas concentration at the upper corner,the gas concentration detection equipment and prediction algorithm model have been studied and improved,under the support of national key R&D project—Research on safety and dust-proof metering technology of new methane ventilation instrument in mines(2017YFF0205500)and the support of Anhui Institute of Optics and Fine Machinery.In terms of detection equipment,different detection methods have been studied and screened,the overall design scheme and complete the system construction are determined.Firstly,the hardware selection of portable gas concentration detection system based on TDLAS technology is completed,the hardware circuit is designed and built,including the laser module for concentration detection,the conditioning module for signal amplification,the data storage module for temporary storage and the communication module for connecting the upper computer.Secondly,the upper computer software can display and predict the interface of data,realizing the functions of data display,prediction and storage.Finally,the test result shows that the designed portable detection equipment can complete the detection of gas concentration and achieve the expected goal.In terms of gas concentration prediction,gas concentration is predicted based on various of prediction algorithms.Considering many factors affect gas concentration value,BP neural network prediction model is adopted to approximate the non-linear relationship,which is between gas concentration and the influencing factors in actual measurement,therefore a more complete forecasting model is established.The circulation procedure is used to determine the number of neurons in the hidden layer,which avoids the appearance of gradient sparsity and gradient diffusion caused by too many layers.However,the established model still has the problem that practical features are difficult to extract.In order to solve the problems above,the ARMA model is used to improve the BP network for feature extraction,making the feature extraction more accurate,so that the accuracy of prediction results will be improved.The stability and fluctuation items are obtained through ARMA analysis,which is involved in adjusting the weight and threshold values of BP network.However,the prediction model based on the combination of ARMA and BP network will still fall into the local minimum value and can not get the optimal solution.In order to solve the problems above,the RBM model is applied to the process of BP network training.The use of RBM model can get a practical feature label,avoiding some irrelevant and duplicate features,which may be caused by subjective factors in the feature selection.This improvement can not only get higher detection accuracy,but also prevent the BP network from falling into the local minimum value.Finally,a variety of commonly used prediction algorithms are analyzed,on qualitative analysis,the curve of prediction are been used to describe results.On quantitative analysis,variance,root mean square error,average absolute error and standard deviation are been used to describe prediction accuracy of various prediction algorithms.The results show that the improved BP network based on RBM works better and has obvious advantages compared with other algorithmsthes.The improved BP neural network based on RBM is applied in the portable detection system designed in this paper,and the feasibility of the algorithm is verified by experiments,and the expected goal is achieved.This paper has 64 pictures,17 tables and 78 references.