深度学习辅助水成膜泡沫灭火剂低温预警

Deep Learning Assisted Early Warning for Low Temperature of Aqueous Film Forming Fire Extinguishing Agent

摘要: 为了避免水成膜灭火剂落入过低的储存温度，需要对其在未来一日内可达到的最低温度做出预报。通过假设下一日最低储存温度与当日相同，在此基础上再输入两个罐体温度传感器和当地天气预报数据，引入一个高度简化的三层BP神经网络处理数据以预测下一日最低罐温。实验表明，以经典的二阶回归算法评判，引入神经网络显著改善了预测精度；以python3代码实现模型，在多种自主可控平台CPU上单线程运行可在10 min内得到预测结果，表明其具有较高的可移植性和执行效率。

Abstract: In order to prevent the aqueous film forming fire extinguishing agent (AFFF) from falling into low storage temperatures, it is necessary to predict the minimum temperature that they can reach in the next day. By assuming that the lowest storage temperature for the next day is the same as that of the current day, two tank temperature sensors and local weather forecast data are input, and a highly simplified three-layer BP neural network is introduced to process the data to predict the lowest tank temperature for the next day. Experiments have shown that the introduction of neural networks significantly improves prediction accuracy based on classical second-order regression algorithms; Using Python 3 code to implement the model, single-threaded running on multiple autonomous and controllable CPU platforms can obtain prediction results within 10 minutes, indicating its high portability and execution efficiency.