An Electrical Fire Early Warning System Based on Multi-Core Embedded Architecture and Bayesian Optimized Random Forest

This paper presents an integrated hardwaresoftware system for laboratory-scale early warning of electrical fires. The platform employs a multi-core STM32 processor, a CAN bus network, and $4 \mathrm{G} / \mathrm{NB}-\mathrm{IoT}$ modules to collect and transmit multi-source environmental data (VOCs, smoke, temperature, and CO2). Sensor noise is suppressed via adaptive Kalman filtering, and the extracted features are input into a Bayesianoptimized Random Forest model for real-time fire-risk prediction and warning. The system integrates cloud and mobile dashboards, acoustic-optical alarms, relay control, and SMS notifications to ensure rapid dissemination of warnings and timely intervention. Experimental results under laboratory conditions demonstrate that the system not only achieves over $97.3 \%$ accuracy in fire-risk identification but also precisely detects anomalous electrical signals and issues warnings within one second. By fusing multisensor data, adaptive Kalman filtering, and Bayesian-optimized Random Forest classification, this solution effectively addresses the urgent need for early detection of electrical fires in laboratory settings, offering robust accident prevention.