REMOTE SENSING FOR FIRE PREDICTION USING ARTIFICIAL NEURAL NETWORK

• Main Author: Wahyu Wicaksono, Aria
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/73085
• Institution: Institut Teknologi Bandung

In 2019, according to Pantau Gambut Indonesia, 465,961 hectares of peatland were burned out of a total of 13.43 million hectares of peatland in Indonesia. Handling peatland fires is become one of the priority programs of the government of the Republic of Indonesia. The role of science and technology is indispensable, one of which is by making fire detection devices. The difficulty of accessing PLN electricity sources and accessing the internet are some of the challenges in designing fire detection devices. Therefore, this study aims to design solar-powered fire detection node sensors, to design a communication system between each fire detection node sensors and between fire detection node sensors and gateway, to design monitoring system for IoT-based fire detection node sensors, and to design machine learning models for fire prediction. From the experiments carried out, the obtained results are a network consisting of STM32 - L052C8T6, LoRA SX1276 microcontroller, DHT11 temperature and humidity sensor, FC-28 soil moisture sensor, water level sensor, wind direction sensor, and wind speed sensor can operate with a maximum range of up to 764.55m with an RSSI value of -134.25dBm and an SNR value of -17.25dB. The best range with a good and stable signal is 712.55m with an RSSI value of -123.25dBm and an SNR value of -9.35dB. The node sensors data are displayed in realtime on Grafana dashboard on the Raspberry Pi. An artificial neural network (ANN) model with a sensitivity value of 0.8974 and ????2 score of 0,8750 can correctly predict 6679 data from 6734 observational data