基于气液逸出物图像识别的锂离子电池火灾早期预警

Early Warning of Lithium-ion Battery Fire Based on Image Recognition of Gas-liquid Escape

锂离子电池在事故条件(过充、短路等)下会在安全阀处喷出气液逸出物,其中包含大量可见的白色汽化电解液以及部分无色气体,有效识别白色汽化电解液并发出预警可大大降低储能电站火灾甚至爆炸的风险。为此,提出了一种基于气液逸出物图像识别的锂离子电池火灾早期预警方法。以YOLOv3算法为基础,同时考虑到锂离子电池储能舱快速、精确识别的安全性需求,将算法中的原始Darknet53特征提取网络替换成轻量级Re XNet特征提取网络;此外,利用K-means聚类算法得到适宜的初始化锚框,加快模型的收敛速度,并结合路径聚合网络进行多尺度特征融合提升模型的检测精度,使模型对大目标和小目标均能达到良好的识别效果。实验结果显示:该方法在对实际锂离子电池储能舱汽化电解液进行识别时表现出了良好的效果,在GTX1650显卡上测试的模型预测速度可以达到每秒65帧,平均精确度均值达到83.65%,基本满足实际应用需要。研究结果对于进一步提升锂离子电池储能电站的安全性及推动电化学储能的健康发展具有一定的参考价值。

Under accident conditions such as overcharge or short circuit, lithium-ion battery will eject gas and liquid at the safety valve, which contains a large amount of visible white vaporized electrolyte and part of colorless gas. If we can effectively identify the signal and give early warning, the risk of fire and explosion in the energy storage power station can be greatly reduced. Therefore, we propose an early warning method for lithium-ion battery fire based on image recognition of gas-liquid escape. As the lithium-ion energy storage power station needs the model to be able to identify quickly and accurately, we have improved the original YOLOv3 algorithm from the following aspects. First, the feature extraction network is replaced by ReXNet from the original Darknet53. Second, the K-means clustering algorithm is used to get an appropriate initial anchor frame, which can accelerate the convergence rate of the model. Last, we use path aggregation network for multi-scale feature fusion to improve the accuracy of the model, so that the model can achieve good recognition of both large and small targets. The experimental results show that this method achieves a good effect in identifying the vaporized electrolyte of the actual lithium-ion battery energy storage tank. Besides, when the model is tested on GTX1650 graphics card, the prediction speed of the model can reach 65 frames per second and the mean average precision can reach 83.65%, which basically meet the needs of practical application. This study can provide references for improving the safety of lithium-ion battery energy storage power station and promoting the healthy development of electrochemical energy storage.