An intelligent sensing array for thermal runaway characteristic gas concentration prediction based on SACNN-Mamba

Abstract

To address the issue of cross-sensitivity when using gas signals for early warning of thermal runaway, an efficient algorithm based on the intelligent sensing array is proposed. The algorithmic model introduced in this paper is a fusion model that incorporates a feature self-attention module, a 1DCNN, and a Mamba module, aiming to enhance the regression prediction accuracy of the intelligent sensing array for gas mixture concentrations. The study demonstrates the effectiveness of our fusion network model in accurately predicting the concentrations of various target gases (such as H₂, CO, and C₂H₄) in gas mixtures. The coefficients of determination (R²) were 99.74%, 99.45%, and 99.48% respectively, indicating that the model fits the sensor data very well and can predict changes in the data with high accuracy. The Root Mean Square Errors (RMSE) were 2.3514, 4.1752, and 3.7164, respectively. The Mean Absolute Errors (MAE) were 1.4398, 2.3846, and 2.5088, respectively. Additionally, the Symmetric Mean Absolute Percentage Errors (SMAPE) were 2.1212, 2.6272, and 2.6515. These values indicate that the model has high predictive accuracy and demonstrates good generality and robustness. The method proposed in this work holds significant potential for application in the field of gas warning for thermal runaway in lithium batteries.