Reserch On Forest Fire Multi-sensory Monitoring Based On ELM

China has a large area of forest,and the monitoring of forest fires has always been an urgent issue in China's environmental problems.It is unpredictable if there is any damage caused by such a large forest area.Therefore,it is very necessary to monitor the forest fire in real time to prevent the fire from expanding.First,the key technologies in forest fire monitoring research are analyzed,such as Internet of things technology,wireless sensor network technology,multi-sensor information fusion and LoRa technology.It also introduces several common algorithms in forest fire prediction.Then,the hardware design of the whole forest fire monitoring system includes sensor design,wireless communication module design,main controller design and 4G module design.A comprehensive analysis of the energy consumption of fire systems throughout the forest area was conducted by using solar lithium batteries.The results show that the entire system can be maintained for a long period of time.A forest fire wireless sensor network based on low-power wide-area technology was established.Due to the large forest environment,a multi-relay wireless network transmission method and protocol was designed.The gateway node interacts with the relay node information in a polling manner,and the relay node interacts with the sensing node information through polling.This ensures the orderly nature of data transmission and reception throughout the network and the stable operation of the network.The main controller sends the data from the gateway node to the 4G network to establish the forest fire monitoring platform and the forest fire monitoring database.The fire environment data of the forest area can be detected and retrieved in real time.Finally,the forest fire parameter identification model was established to make decision-level information fusion for the fire parameters in the database.Traditional forest fire prediction learning algorithms have a lot of drawbacks.For example,BP neural network has the disadvantages of complicated algorithm,slow training speed,and poor accuracy.In addition,the algorithm makes it difficult to achieve global optimum for the entire system and requires multiple corrections such as weights and thresholds.ELM is a typical single-hidden layer feed-forward neural network with fast learning speed.But it will produce different input weights and offset thresholds during the training process.Therefore,a PSO fire learning model for extreme learning forest is proposed in this thesis.A good forest fire prediction effect is obtained by using this model to make decision-level information fusion of forest fire environmental data and optimize the input weights and thresholds.The optimized ELM has a thousand times faster network training speed and higher prediction accuracy compared with the traditional method.