Research on transformer oil-saturated porous medium fire detection technology based on multivariable data fusion

Transformer oil is prone to leakage during use and storage, and it often permeates porous media (such as gravel and soil), leading to rapid flame spread. The presence of porous media increases the complexity of the combustion process, making it difficult for traditional wired data transmission and single-sensor technologies to achieve accurate fire detection and prediction. This study developed a test platform for fire spread in porous media saturated with transformer oil, collected multi-source fire characteristic data, and applied artificial neural network data fusion techniques to research fire detection for transformer oil leaks in porous media. Results show that binarization and flame area extraction can convert two-dimensional image data into one-dimensional flame pixel area data, and normalization provides effective data for data fusion. The fire detection thresholds for single parameters such as temperature, heat flux, and image were set at 54°C, 0.08 kW/m², and 0 pixels of area, respectively. The LSTM neural network data fusion algorithm improved the accuracy of fire detection results to 97%, reduced detection time compared to single-parameter detection, and eliminated false detection. This paper provides significant theoretical support and practical guidance for transformer oil fire prediction technologies in complex scenarios.