Research On Distributed Diffusion Unscented Kalman Filter

With the increasing demand for fast and reliable data access,sensors are gradually moving towards lightweight,intelligent,and low-cost trends.The development of wireless communication and sensor technology has enabled a large number of low-cost and low-power sensors to communicate with each other and complete tasks through collaboration,forming a sensor network.Compared to other data fusion methods,distributed filtering operates without a local fusion center,and not only has the characteristics of low energy consumption,high reliability,and good robustness,but also can share more complete,accurate,and timely data between multiple platforms.The distributed estimation based on diffusion strategy has better flexibility and robustness because it does not require the network to have a circular structure and does not require constraints on each node and its neighbors to converge to the same value.When wireless sensor networks are affected by external events,sensor nodes may exhibit data anomalies.When the communication and computing capabilities of the sensor itself are limited by various factors,data collection and transmission limitations may occur.Moreover,the practical application of filters is also limited by the ubiquitous nonlinear models.Based on this,this thesis studies distributed diffusion unscented Kalman filtering for sensor network anomalies and resource constraints in nonlinear models.Firstly,this thesis introduces the wireless sensor network model and system model,and describes the process of unscented Kalman filtering,including three steps of sampling,updating,and prediction.It also describes the unscented Kalman filtering under the distributed diffusion strategy in detail.Secondly,aiming at the attack scenarios in the network,the secure distributed unscented Kalman filtering against network attacks is studied,and a clustering based diffusion unscented Kalman attack detection algorithm is proposed.In the diffusion process,a two-level loop mechanism is applied to state estimation attack detection and covariance matrix attack detection to assign greater weight to the local estimates of trusted neighbor nodes,thereby reducing the impact of network attacks on estimation performance.Simulation results show that the algorithm is suitable for all four attack models,including false data injection attacks,random attacks,reverse attacks,and replay attacks,with good estimation results when the security node information is unknown.Finally,aiming at the problem of sensor resource constraints,combining event triggering strategy and quantization strategy,this thesis proposes a resource constrained diffusion unscented Kalman filter,which is a diffusion unscented Kalman filter based on adaptive event triggering and quantization strategy.Select an event triggering scheme or quantization scheme based on the similarity between the current time and the previous time.When the similarity is high,an event triggering scheme is selected.The sensor node uses an adaptive event triggering method based on each item to determine which information items to transmit,thereby achieving dimensionality reduction of the transmitted data.When the similarity is low,a quantization scheme is selected.The sensor node uses a jitter quantization model to exchange quantized information with neighboring nodes,thereby reducing communication costs.In addition,the convergence of the algorithm is proved through performance analysis,and the communication cost is calculated.Numerical simulations have verified that this method can significantly reduce communication costs at the expense of smaller estimation accuracy.