Design of BP neural network-based FPGA system for early fire detection of energy storage power station

This paper presents an FPGA-based fire detection system using a BP neural network for early detection in energy storage stations. The system analyzes temperature, smoke, and gas data with an 8-5-1 BP network structure. Trained and tested in MATLAB, it achieved 94.56% accuracy on training data. To enhance FPGA performance, the design uses fixed-point quantization, pipelining, parallel neuron calculations, and piecewise polynomial fitting for the Sigmoid function. The FPGA used 696 LUTs, reached 93.5% accuracy, and had a recognition time of $3.24 \mu \mathrm{~s}$, demonstrating strong reliability and real-time performance for early fire detection in energy storage stations.