Kitchen Fire Prediction and Alarm System Based on Multi-Source Information Fusion and Neural Networks

Kitchen safety, a critical component of public security, faces numerous threats, with home fires posing a particularly severe risk. Many current methods are afflicted by issues such as false alarms and missed detections. Additionally, there are still notable gaps in the monitoring of kitchen safety. To address these challenges, this paper proposes a novel fire prediction approach that integrates multi-source information fusion with advanced deep learning techniques. Specifically, we leverage micro-controller technology, multi-sensor data collection, and network communication, combined with Long Short-Term Memory (LSTM) and Radial Basis Function Back Propagation (RBF-BP) neural networks. In our approach, various sensors collect feature information during a fire event, which is then preprocessed using a rate detection algorithm. The processed data is analyzed using LSTM and RBF-BP networks to adaptively learn from multiple fire-related signals and output probabilities for three fire scenarios: active fire, no fire, and smoldering fire. A fuzzy logic control system is employed to determine the occurrence of a fire, and the microcontroller sends an alarm notification to the user's mobile phone. Simulation results demonstrate that the proposed method significantly enhances prediction accuracy and adaptability compared to traditional fire detection methods and single neural network approaches.