Combined Prediction Model of Gas Concentration Based on Indicators Dynamic Optimization and Bi-LSTMs

Abstract

In order to accurately predict the gas concentration, find out the gas abnormal emission in advance, and take effective measures to reduce the gas concentration in time, this paper analyzes multivariate monitoring data and proposes a new dynamic combined prediction method of gas concentration. Spearman’s rank correlation coefficient is applied for the dynamic optimization of prediction indicators. The time series and spatial topology features of the optimized indicators are extracted and input into the combined prediction model of gas concentration based on indicators dynamic optimization and Bi-LSTMs (Bi-directional Long Short-term Memory), which can predict the gas concentration for the next 30 min. The results show that the other gas concentration, temperature, and humidity indicators are strongly correlated with the gas concentration to be predicted, and Spearman’s rank correlation coefficient is up to 0.92 at most. The average R2 of predicted value and real value is 0.965, and the average prediction efficiency R for gas abnormal or normal emission is 79.9%. Compared with the other models, the proposed dynamic optimized indicators combined model is more accurate, and the missing alarm of gas abnormal emission is significantly alleviated, which greatly improves the early alarming accuracy. It can assist the safety monitoring personnel in decision making and has certain significance to improve the safety production efficiency of coal mines.

Generating Low-Earth Orbit Satellite Attitude Maneuver Profiles Using Deep Neural Networks

“…The selection factors for the network and parameters used to model the quaternion include the number of bidirectional long short-term memory (bi-LSTM) layers, number of hidden units for each bi-LSTM, maximum epoch number, mini-batch size, initial learning rate, drop rates of the first and last drop-out layers, and frequency of the feature data [ 28 , 29 , 30 ]. As the reverse order of time-series data is also meaningful, learning using bi-LSTM was used because of its high prediction accuracy of time-series data [ 31 , 32 ]. Next, the frequency of the feature data was chosen as the selection factor for optimization.…”

Gas Graph Convolutional Transformer for Robust Generalization in Adaptive Gas Mixture Concentration Estimation

“…In some practical scenarios, researchers categorize the properties of gas mixtures in those settings as a single gas, fulfilling the need for gas safety monitoring. For instance, in coal mines, flammable gases are collectively referred to as “coal mine gas”, and their concentration is estimated. , However, this approach is not always feasible, particularly in industrial environments where the challenge intensifies. For example, in chemical manufacturing plants or petroleum refineries, the production processes often result in a complex blend of gases, varying in type and concentration.…”