An IoT-based flood alert system utilizing intelligent caching and lightweight ai on a microcontroller unit

In flood-prone areas, the connection infrastructure is often poor, hindering real-time data transmission in IoT early warning systems. This study proposes a smart store-forward and data prioritization mechanism, implemented on the ESP32 microcontroller combined with SIM7600CE to ensure stable information transmission via the MQTT protocol. The system uses a lightweight artificial intelligence model (TinyML), compressed by quantization techniques, to help sensor nodes classify data into three warning levels: Normal, Warning and Danger. Based on the priority level, the device decides whether to store the data locally or forward it to the server, contributing to reducing network congestion and saving energy. Actual testing shows that the system achieves a 100% data transmission rate, reduces memory usage by 25%, and significantly improves warning performance compared to the traditional method. This solution is suitable for hard-to-reach areas, supporting the improvement of disaster response capacity.