Research On Decision Level Fusion Algorithms In Wireless Sensor Networks

Data fusion for Wireless Sensor Networks (WSN) enables removing the redundant information from multiple sensors, thus to reduce the amount of data communication and improve the accuracy of collected data. Consequently, data fusion becomes a research hotspot of WSN, which has the weakness of energy limited. The decision level fusion is performed in the highest level of data processing, which has the merits of little communication, low computation complexity, no restrict for sensor categories, strong real-time capability and so on, and the decision level fusion has been widely used in the area of target monitoring and tracking, attributes identification and so on.This paper focus on the research of decision level fusion in WSN, we focus on target monitoring and classifying problem in an event-driven monitoring environment, and we studied in detail the local fusion algorithm for each sensor and the decision-level fusion algorithm.Firstly, this paper gave a brief introduction to the traditional data fusion algorithms in WSN such as the Bayesian inference, D-S evidence theory and the fuzzy sets, and the advantages and disadvantages of each method were reviewed. In view of the traditional fuzzy sets are not precise enough when dealing with the inaccuracy of initial data, we proposed the idea of using intuitionistic fuzzy sets to express the initial data. The existing fuzzy sets construction methods are mainly based on expert decision or focus on specific applications such as image processing, which are not suitable for the WSN. Therefore, this paper proposed a data distribution based intuitionistic fuzzy sets construction algorithm. The proposed algorithm will determine the hesitancy degree according to the standard deviation of training sample set of the monitored attributes, so as to transfer the original attribute value of sensor into intuitionistic fuzzy value according to the relationship between fuzzy sets and intuitionistic fuzzy sets.Secondly, an intuitionistic fuzzy information aggregation based local fusion algorithm in WSN is proposed to deal with the multi-attribute decision-making in each sensor. First, the concept of category similarity is suggested to determine the weight for each attribute; then a category similarity weighed TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) intuitionistic fuzzy decision-making algorithm was proposed for local decision-making. The proposed algorithm improved definition the positive ideal solution and negative ideal solution of traditional TOPSIS, thus to obtain more accurate multi-attribute classification results. Simulation results show that the classification accuracy of the proposed local fusion algorithm has been improved compared with traditional fuzzy fusion and Intuitionistic Fuzzy Weighted Aggregation (IFWA).Lastly, a clustering and fuzzy logic based decision-level fusion is proposed to perform decision level fusion of the whole network. Firstly,k-Means algorithm is used to form the nodes into clusters, which can significantly reduce the energy consumption of intra-cluster communication; In order to improve the accuracy of decision-level fusion, fuzzy inference is used to determine the weight of each cluster.