RBF Fire Prediction System Based on Improved PSO Optimization

The current fire status prediction system is inaccurate. Incorrect predictions can lead to fire expansion, threatening lives and property. This paper builds a fire state prediction system using a radial basis neural network multi - sensor data fusion algorithm optimized by an improved particle swarm. It takes temperature, smoke, and carbon monoxide concentrations as inputs, and outputs the probabilities of no fire, smoldering fire, and open flame. A fuzzy inference system is added to correct potential neural network biases. An improved particle swarm optimization algorithm with nonlinear dynamic adaptive inertia weights is used to address the algorithm's shortcomings. Simulation experiments show that when predicting open - flame states, the improved system's average absolute percentage error is 0.169, the root - mean - square error is 0.0021, and the average absolute error is 0.031, indicating significant performance improvement.