Research On Prediction Of Gas Danger Level Based On Data Fusion

The current research indicates that the gas gas's change law is chaotic and can only meet short-term and inaccurate predictions.Although the existing mathematic model predicts gas data,it has improved in accuracy,but it has the disadvantages of high seriality,low level of work,and poor scalability.Therefore,this paper proposes the PCF model.The model has parallelism to meet the running time requirements;it can directly output decision-level information to ensure the length of the window from the release of warning information to the occurrence of disasters;to meet a certain degree of scalability,the model can be easily combined with other models The work can also be used in coal mines with different gas concentration changes.This paper expounds the ideas put forward by the PCF model,gives the process of model building,and proves that the PCF model achieves parallelism,output decision-layer conclusions,and extensible requirements on the basis of maintaining the correct rate.Firstly,based on the non-linear and non-stationary characteristics of raw gas data,the filtering method based on wavelet analysis was used to reduce noise.In view of the chaotic characteristics of gas data,phase space reconstruction and neural network-based time series data prediction methods are used at the data level,and the embedding dimension and delay time are calculated using the C-C algorithm,and different types of neural networks are considered,and the embedding dimension is considered.The relationship between the prediction step size and the experimental step size determines the optimal parameter ratio.The test results show that this parameter has a better prediction result.Then the gas hazard rule is defined and the classifier output dimension is determined under this rule.Due to the high nonlinearity of mapping operations,neural network-based classifiers and support vector machine classifiers are used.Taking into account the robustness and generalization of the model,a dimension reduction module was added before the classification.In order to make the dimensionality reduction module's impact on the model's running time as small as possible,a method for finding the optimal priori conversion matrix T is proposed,it is proved that the algorithm has achieved The model's requirements.Finally,the evidence generated by the classification module is sometimes highly conflicted.By analyzing the structural characteristics of the PCF model itself,an evidence discount method based on error sources suitable for this model is proposed,and the basic model of this model is constructed using this method.Probability distribution function(BPA),experiments show that the new evidence produced by this BPA effectively reduces the conflict between the original evidence,improves the significance level of the final conclusion and the correct rate of the final conclusion.Through the experimental comparison,it is found that the model can be changed.The type and number of classifiers to enhance,streamline,and enhance the model's ability to meet the requirements of scalability,but also found that this increase is effective within a certain interval.The theoretical analysis shows that the parallelism of the PCF model is reflected in the parallel design of the classification module.