Wireless Sensor Network In Fire Detection System Research And Implementation

With the enhancement theoretical study on WSNs and sensor hardware capabilitiesimproved, WSNs has been applied in actual environmental monitoring. Fire monitoringis one of the applications applying in actual environment monitoring. But firemonitoring system based on wireless sensor networks meets two problems. The firstproblem is that because the sensing ability of the low-power sensor is limited, fire alarmmay be delay or even be omission. The second problem is that because of the fire’suncertainty feature, it is difficult to determine whether fire broking out or not. Wepropose a new fire monitoring system based on wireless sensor networks which consistsof sensor data collection mechanism using time series prediction algorithm and firedetection mechanism using neural network model.We propose a new node data collection method using time series predictionalgorithm, processing the temperature and humidity data sensed by nodes in a period oftime, extracting the trend function which there is just one time variable t in this function.Then nodes send the trend function to sink. When the client receives the nodes returningdata, generates predictive function using time series trend parameters. According todifferent need, we set different time changing rate t`as the variable of predictionfunction, and predict the temperature and humidity values. The fire occurringprobability can be computed when the three parameters are taken into trained neutralnetwork as input. The output of neural network algorithm is the probability p of fire.Experiment results show that the fire monitoring system can recognize the flaming firenearly100%, and warning delay can be controlled in the30s. The slow smoldering firerecognition rate can be controlled in80%, warning delay can be controlled in1min.In order to reduce the transmission energy of nodes, we propose contour mappingin wireless sensor networks based on B-Spline curves with Tangent Constraints algorithms (TCCM). After sink broadcast a query, not all the nodes that match the queryreturn values to sink by using TCCM, just some representative nodes returninginformation to sink. The theoretical analysis shows that the expectation of therepresentative nodes returned by TCCM is only30%of the number of nodes in sensornetworks. Experiments results show that compared with the best of the currentalgorithm, the number of the representative nodes returned by TCCM is reduced by53%, and TCCM can make more the accurate contour map.