Research On Text Data Mining Of HSE In Oil And Gas Industry And Edge Computing Technology

With the advent of the information age,domestic oil and gas companies have taken the lead in completing enterprise informatization and have accumulated a large amount of data on production safety.These data are called HSE data(data on health,safety,and environment).With the vigorous development of artificial intelligence and the vigorous advancement of corporate management systems,oil and gas companies hope to automatically analyze these HSE data through machines,mine the objective laws implicit in the data,discover potential safety hazards in the company,and formulate more effective safety regulations.We uses the HSE data of a large domestic oil and gas company to study the HSE text data mining method based on neural network.In addition,the deep learning model needs to be deployed in the production environment after the training is completed.In the actual production environment,business personnel mostly use mobile devices for data processing,such as tablets and thin laptops.The computing power of these devices is related to the battery.The capacity is very limited.Another research of this subject is to use edge computing technology to perform the reasoning task of the deep learning model,by offloading the reasoning task to the edge server or cloud server for execution to minimize the global delay and energy consumption.The main work of this subject includes the following three aspects:(1)Aiming at the problem of calculating the text similarity of accident event data,we have designed a text similarity calculation method based on word2vec+L2 regularization.First,we use continuous bag-of-words model training on the HSE data set to obtain word vectors,and then introduce L2 regularization in the process of merging the word vectors into sentence vectors.The experimental results show that the method of using HSE word vector +L2 regularization to extract text features has the best performance in the calculation of text similarity of accident data.In addition,based on the experimental results of text similarity calculations,we uses web development frameworks such as Flask,Vue,Element UI to design a browser/server architecture-based HSE security regulations recommendation application.(2)Aiming at the problem of system audit data text classification,we designed an improved Ro BERTa based text classification method,by further pre-training the Ro BERTa model on the HSE data set and freezing the first three in the fine-tuning process of the downstream classification task Layer parameters and the use of linear attenuation learning rate further enhance the classification effect of the Ro BERTa model.The experimental results show that the improved Ro BERTa model has the highest F1 value in the system audit data classification.In addition,we uses Py QT5 to design a text classification application with a client/server architecture based on the results of text classification experiments,which is convenient for business personnel to use the trained model for text classification.(3)Aiming at the problem that the computing resources and battery capacity of mobile devices cannot effectively run inference tasks,we proposes a deep learning inference model based on edge computing,by offloading different sizes of inference tasks to edge servers or cloud servers to reduce the overall delay and energy consumption.Finally,we simulates different production environments by changing the key parameters of the system model and analyzes in detail the influencing factors of the optimal unloading decision.