Multivariable Gas Concentration Prediction Based On Data Fusion And AE-ELM

Coal mine resources are part of China’s precious material resources.However,mine accidents caused greatly threaten personal safety and production safety during the process of collecting coal mines.According to the disaster data,gas accidents have always accounted for a large proportion in various mine accidents,which need our attention.Therefore,finding an accurate prediction method for coal mine gas concentration is a useful tool means to avoid underground disasters.Traditional gas concentration prediction methods generally only use historical gas concentration data when forecasting,and do not consider the problem that the gas concentration value in the mine is also affected by other environmental parameters such as wind speed.For this problem,the paper focuses on the study of gas concentration in the mine,A multi-variable gas concentration prediction method with wind speed and temperature as input variables,and a model structure based on autoencoder and extreme learning machine is constructed to realize the gas concentration prediction.The research done for this paper can be summarized as the following three points:1.A method to predict the gas concentration value using the time-series data of wind speed,temperature and gas concentration is proposed.The combination of multi-sensor data fusion and phase space reconstruction method makes the predicted value closer to the actual value of gas concentration in the entire mine face.2.Propose a combined neural network structure combining autoencoder and extreme learning machine.The self-encoder used for feature extraction will effectively reduce the dimension of the original data and remove redundant information contained in the original data,which increases the accuracy of prediction and reduces the time of prediction.It is suitable for situations with larger amounts of data.At the same time,the characteristics of random learning machine’s random weight make the training time extremely short,which will further improve the real-time performance of prediction.The simulation results show that the average absolute error(MAE)predicted by the paper method is47.24% and 75.97% lower than the traditional uni-variate gas concentration prediction method and the traditional model prediction method,respectively,which improves the accuracy and efficiency of the algorithm.3.An improved method for predicting the change trend of gas concentration using fuzzy information granulation is proposed.Combining the existing uni-variate gas concentration trend prediction method with multi-sensor data fusion and phase space reconstruction techniques,a multi-input variable-based gas concentration trend prediction method is proposed to improve the accuracy of prediction.The improved method for predicting the gas concentration trend using multivariables also has a significantly lower MAE value than the method using uni-variate methods before the improvement,which achieves the purpose of improving accuracy.