The Search Of Wireless Sensor Networks Based On BP Neural Network Data Collection Protocol

As the rapid development of computer networks and communication technology, wireless communication technology is attracting more and more attention. A variety of emerging wireless communications applications, including the design and the applications of WSNs has gradually become a hot topic. Wireless sensor network is a type of wireless ad hoc networks. It has the character of self-organization, self-adaptive and high fault-tolerance. In many scenarios, such as environmental monitoring, target tracking, industrial and agricultural and so on, have very broad application prospects. In wireless sensor network, how to use data fusion methods to reduce the amount of data transmission and thus make an efficient use of energy of sensor nodes to extend the life time of wireless sensor networks is a key purpose of research.Due to the fact that BP neural network can be used to approximate a curve unlimitedly, we use it for data fusion in WSNs. The sensors send the weight and threshold instead of the raw data monitored to the sink. Furthermore, the weight and threshold in the last fitting are used as the input of the new fitting to the reduce the number of neural network training steps. Simulation studies show that the proposed scheme can be effectively reduce data transmissions, so as to achieve energy efficiency in WSNs.In view of the current wireless sensor networks, most wireless sensor nodes have weak computing abilities, and their capacity of calculating is very weak. Also the BP neural network has its inherent weaknesses that a large number of calculations are inevitable. The operation of BP neural network computing at the base station side is also discussed in this thesis. The base station uses the limited data which the sensor node sends to predict the data which the sensor node does not send, or the data of other sensor nodes. Mathematical analysis and simulation studies show that the protocol is feasible.Today many data fusion schemes have been proposed, but most of them utilize various routing techniques to select energy-saving paths for data packets and data fusion techniques to reduce spatial-temporal correlation of the sensed data to reduce the energy cost of sensor nodes. The data fusion method based on BP neural network and utilizing the base station to predict data proposed in this thesis can reduce the temporal redundancy of sensory data at the source. They can also use part of the collection of data to predict other parts. This will basically reduce the data transmission between sensor nodes. This thesis has a certain reference for the related research.