Research On Multi-Source Navigation Information Fusion Method For Deep-Sea AUV

Autonomous Underwater Vehicles (AUVs) have a wide range of applications in marine geophysical prospecting, oceanographic survey and utilization of deep-sea space. An efficent navigation and positioning system is very important for AUVs to work in underwater environment with good reliability in terms of safety and accuracy. To improve the accuracy and reliability of the navigation system, one of the most important methods is to make the Multi-sensor data fusion methodology better. The theme of this paper is the research of Multi-sensor data fusion methodology for deep-sea underwater vehicles. The research work is mainly about nonlinear and non-Gaussian filtering, data denoising, multiple model estimation and its application to the navigation system of AUVs. The dissertation is presented as follows:1. Gaussian sum Sigma point filtering method is researched to solve nonlinear and non-Gaussian problems. The frame of nonlinear Bayesian filtering theory is deduced and Sigma point filtering methods are analyzed. Nonlinear approximation methods are studied and a high order unscented Kalman filter is set up based on the probability density function of nonlinear systems. A new Gaussian sum Sigma point filtering method is proposed by combining the high order unscented Kalman filter and Gaussian sum filtering theory. Numerical simulation examples demonstrate that the newly-developed filter has integrated advantages with respect to estimation accuracy and computational complexity and its performance excelled the existing Gaussian sum filters.2. Data denoising mehods for sensors of underwater vehicle are studied to reduce noise interference from the environmental change, machanical vibrations and multipath effect of underwater acoustic signal. How the sensors work and where the measurement errors come from are analyzed. The error models of the sensors are set up. Data denoising methods based on wavelet analysis and Emperical Mode Decomposition (EMD) are researched. The above two methods are applied to the underwater vehicle navigation system. Simulation experiments show that these two denoising methods can help to improve the estimation accuracy of the AUV integrated navigation system. The denoising method based on EMD exhibits satisfactory performance in adaptability.3. Data fusion methods based on multiple model estimation approach are researched for the integrated navigation system with unknown noise characteristics. State estimation approaches based on multiple model estimation theory are researched and the interactive multiple model (IMM) algorithm is analyzed emphatically. An improved IMM algorithm is proposed in which the idea of expected-mode augmentation, one of the model set adaptation methods, is utilized. Simulation experiment has been carried out to evaluate the improved IMM algrithm in the SINS/DVL integrated navigation system. The results show that the improved method can improve the estimation accuracy and stability of the integrated navigation algorithm while the computational complexity increases moderately.4. To resolve the problem that the model probability update is too sensitive to the measurement noises in the IMM algorithm, adaptive model switch probability methods are researched to improve the performance of IMM algorithm. The reasons why the calculation of model probability may cause errors are analyzed. The innovation filtering interactive multiple model (IFIMM) algorithm which is one of the improved IMM methods is researched. A novel improved IMM method based on Bayesian network is proposed. The hidden causal variables in the AUV integrated navigation system are utilized to set up the Bayesian network. Simulation experiments have been carried out to evaluate the improved IMM algrithm in the SINS/DVL/TAN/MCP integrated navigation system. The results demonstrate that the proposed algorithm can solve the problem of model switching hysteresis in the IMM algorithm and improve the estimation accuracy of the AUV integrated navigation system.5. The federated multiple model estimation algorithm is researched to improve the flexibility, fault tolerance and real-time performance of the multi-sensor data fusion method for the AUV integrated navigation system. The structure and design process of the federated Kalman filtering method are analyzed. The multiple model estimation is combined with federated Kalman filter to set up the federated multiple model algorithm. The federated multiple model algorithm is applied to the integrated navigation system of underwater vehicles. The error models of the integrated navigation system in which SINS is considered as the reference are deduced. Simulation experiments have been carried out to evaluate the federated multiple model estimation method in AUV integrated navigation system, and the results show that the number of position errors of the federated IMM is smaller than that of federated Kalman. A field test is designed to evaluate the proposed federated filter based on multiple model estimation in terms of its efficacy for integrated navigation system. The results show that the proposed method has significant improvement in navigation estimation accuracy and reliablility as compared to traditional federated Kalman filters.