Research On Data Fusion Algorithm Based On Neural Network In Wireless Sensor Networks

Wireless sensor networks(WSN) are composed of numbers of low-cost,low-power, self-organizing tiny sensor nodes via wireless or wired communication,that can perceive, monitor, collect and process the environmental information in overlay area, and transfer the result to visitors.WSN has many features, including flexible deployment, good scalability, high reliability and low-cost, which make it always arranged in harsh areas, such as disaster areas, forests, underground pipelines, so the environmental factors, human factors, and the networks robustness(nodes are too many) make the collected information redundant and wrong, that will consume lots of resources and energy, but the power, bandwidth, data storage and processing capability in WSN is limited, so it is hot to research how to reduce the redundant and false data to low the energy consumption and prolong the life of the WSN.The data fusion technology is an emerging discipline with excellent function in data processing, which can largely reduce the redundant and false data. Especially, the data fusion based on the Neural Networks, that can simulate the human brain, and has the ability of self-organizing and self-learning, its weights storage information, its parallel structure and nonlinear characteristics and so on make it become hot.Therefore, it is very important to study the technology of data fusion based on the neural network in wireless sensor networks.This paper aims on the data fusion of WSN based on the BP neural networks(Back Propagation Neural Networks, BPNN). Firstly, weights in BPNN change insensitively, that will lead to shocking and prolong the training time of neural networks, Considering the influence of multiple change rates of weight on the next weight, we improve the computing method of the change degree of weight in BP neural network. Then, the initial weight of BPNN is random, that will lead to the training time of the neural networks become long and the optimal weight is difficult to achieve and so on, so this paper takes the strategy of combining the BPNN and the DS evidence theory together, proposes the DS-BP neural network on the base of the first improved method to compute precise initial weight to short training cycles, reduce theoutput error, enhance the data accuracy. Finally, this paper applies the improved BPNN to the WSN on the base of BPNDA, in order to increase the data transmission rate, balance the energy consumption of the whole networks and prolong the life of the WSN.This paper uses the MATLAB to simulate all the improved methods, according to the result to analysis of the data fusion accuracy, the output error, the training time,the life of the networks, the data transmission rate and other indicators, results show that DS-BP neural network improves the accuracy of the data fusion, shorts the convergence time of the networks, and reduces the oscillating phenomenon, then applies the DS-BP neural network into the WSN, the result shows that it can improve the data transmission rate, reduce the network energy consumption and prolong the lifetime of the WSN.