Real-time Toxic Gas Concentration Prediction Research For Point Source Emission Based On Deep Learning

Accurate and efficient gas concentration prediction model plays an important role in emergency rescue of sudden toxic heavy gas leakage.However,it is difficult for the existing dispersion models to achieve accuracy and efficiency requirements at the same time.In order to solve this dilemma,new models used for heavy gas concentration prediction based on deep learning theory were proposed in this paper.Otherwise,the prediction results were coupled with geographic information system to provide more intuitive decision-making basis for emergency rescue operationsTaken non-urban environment as research object,the heavy gas eoncentration prediction models based on deep belief network and convolutional neural network were built respectively.All the study in this part was carried out on the basis of Prairie Grass experiment data.The key parameters in both deep learning models were determined by cross validation method and grid optimization method.The results of models are compared with the real data,and the similarity between the predicted data and original data is evaluated.In addition,the existing computational fluid dynamics method and the prediction models based on traditional machine learning were used to predict the concentration of the same leakage scene under the same conditions.The deep learning models were analyzed and compared with the existing prediction methods in terms of time and accuracy to evaluate its advantages and disadvantages.The results show that the CNN model has significant advantages.Then,the toxic heavy gas concentration prediction model for urban environment was studied.CFD model was used to simulate a large amount of chlorine gas leakage data which was used to build concentration prediction model.On the one hand,it solves the problem that the quantity of urban gas dispersion experimental data is too small to construct a model.On the other hand,it greatly reduces the time cost while ensuring that the prediction accuracy is similar to that of CFD model.It provides a new way to predict the concentration of toxic heavy gas leakage.In addition,on the basis of the above research,the genetic algorithm was introduced to realize the adaptive optimization of the structure and training parameters of CNN model.In addition,the weight and bias initialization method was optimized which further improved the prediction effect of the CNN model.The prediction results of the CNN optimization model were compared with other models.The results show that the CNN optimization model has the best performance.Finally,the predicted results of CNN prediction model were divided according to the hazard level of chlorine concentration.The results were coupled with the two-dimensional building model constructed in Arc map to realize the visual expression of chlorine gas leakage and dispersion.It provides richer and more intuitive decision-making basis for emergency rescue after toxic heavy gas leakage.