Research And Application Of Spatiotemporal Prediction Model In Early Warning Of Coal Compound Disaster

Coal mine thermal power disasters refer to the primary and secondary disasters in coal mines due to the thermal power attributes of coal mine disaster factors exceeding the control range.Coal mine thermal power disasters seriously affect the safety of coal mine production,so it is necessary to predict coal mine thermal power disasters that may occur.This paper combines deep learning theories and methods to study the identification of gas mixtures in coal mines,and the prediction and detection of time series of marker gases.The main work of this paper is as follows:Aiming at the difficulty of multi-dimensional time series data classification,a mixed gas recognition method based on class picture matrix and convolutional neural network is proposed.Based on the time series of gas sensors,this method maps the multi-dimensional time series into a picture-like matrix,and builds four types of mapping methods based on the mutual relationship between the multi-dimensional data.Through the combination of different mapping methods,combined with the nonlinear feature extraction ability of the convolutional neural network,the mixed gas time series data is classified and identified.The experiment was conducted on a public data set of mixed gas,and the classification accuracy of different mapping combinations was compared.At the same time,the classification effects of the five models of VGG16,VGG19,Res Net18,Res Net34,and Res Net50 were compared.The experiment showed that the Res Net50 model is better.In order to predict the concentration of the iconic gas that can reflect the stage of coal mine thermal power disasters,three-dimensional spatial features and multivariate auxiliary feature analysis are added to the time convolution network,a gas concentration forecasting model based on multivariate fusion spatiotemporal convolution network was proposed.This method combines the spatial data of other monitoring points around the target monitoring point of the goaf on the basis of the historical sequence of gas concentration to extract the spatial characteristics of the target monitoring point.At the same time,the cross-correlation between the other gas data at the target monitoring point and the target gas data is analyzed as an auxiliary factor to improve the prediction accuracy.In order to verify the effectiveness of the proposed method,the experiment uses goaf model experimental data and compares it with the prediction effect of models such as long-term and short-term memory networks.The experiment shows that the proposed method has smaller errors.Detecting the abnormal pattern data segment in the target gas time series data can help to determine the stage of disaster occurrence,a time series anomaly pattern detection method based on differential rate entropy feature and generative adversarial networks was proposed.This method first extracts the differential rate entropy features of the sequence,and then uses the generated adversarial network to learn the distribution of normal patterns,and proposes a new method for calculating the anomaly score during the detection phase,that is,considering both the weighted anomaly score of the generated sample and the anomaly score of the discriminant result.Finally,it is judged whether the single-dimensional time series data segment is an abnormal mode by calculating the abnormal score of the sample.In order to verify the effectiveness of the proposed method,the experiment uses goaf model experimental data.The experiment compares the results of different time series entropy and different models.The experiment shows that the method proposed in this paper has better detection accuracy.Effective forecasting and early warning methods can reduce the harm caused by thermal power disasters in coal mined-out areas,and are of great significance to promptly notify staff to take effective protective measures.