Early Indoor Fire Warning Algorithm Based on INRBO-XGBoost and Fuzzy Inference

Given the current limitations of computer vision technology in detecting early smoldering fires and the challenges posed by traditional single chemical sensors in distinguishing alarms amidst disturbances, we propose a multi-sensor fusion fire warning algorithm. This algorithm integrates an improved Newton-Raphson Based Optimizer (INRBO) with an Extreme Gradient Boosting (XGBoost) algorithm and fuzzy inference theory. Initially, the mix opposition-based learning (MOBL) method is employed to enhance the NRBO algorithm, increasing population diversity and avoiding local optima. The improved NRBO optimizer is then used for hyperparameter tuning of the XGBoost algorithm, accelerating model convergence and improving prediction accuracy. Finally, fuzzy inference theory is applied to integrate the predicted fire probabilities, yielding optimal control decisions. Compared to the PSO-XGBoost, XGBoost, and GA-BP models, the root mean square error (RMSE) of the INRBO-XGBoost model is reduced by 24.43%, 39.10%, and 32.95%, respectively. The warning accuracy for various types and regions of fires, including plastic smoldering fires, electronic equipment smoldering fires, and cotton fabric smoldering fires, exceeds 99.11%.