No abstract

No abstract

According to the situation that the statistical characteristics of noise in initial alignment of sins of UKF filter is not agree with actually, the filtering precision will severely reduce or even divergent, a combination of support vector machine method of initial alignment is proposed. In this paper, the test samples are split into four groups. Three groups are trained for the first layer and the last group is trained for the second layer of support vector machine. The first layer is a group of support vector machine in parallel computing, the second layer is an information fusion of the single support vector machine in the first layer, and combined support vector machines. In this method initial alignment of strapdown inertial navigation system is achieved. Finally through the UKF filter, SVM, CSVM simulation contrast, the results show that CSVM has an improvement than a single SVM, better real-time than UKF filter and generalization ability.

Consider about the trade-off between cost and accuracy, lots of works investigate into the inertial measurement system (IMS) and Global Position System (GPS) fusion. However, with the necessary to insure the long-term on-board working precision it results in a complication between the time-consuming and complex calculation in the system design. Therefore timing has become a critical problem in the hardware design. Nowadays more and more applications require not only the accuracy but also the rapid response. This paper addresses a new way to separate the system design into three modules and applied into a parallel structure, like Microprocessor without Interlocked Pipeline Stages (MIPS) to realize the real-time performance. Using the System-on-Chip (SOC) technology, the system platform can adapt the timing consummation and signal frequency between each module, in order to estimate the velocity and position of vehicle in real-time.

The INS/GPS integrated navigation as the research object, based on in-depth analysis of the multiple model filter method, the convex optimization hybrid filtering method based on adaptive optimization. OLS-SVM method utilizes real-time to obtain the weighted value of the hybrid filter, enables the weight value can be varied with the real-time filtering effect of changes. Therefore, it can effectively improve the system robustness, thus affecting the estimation precision of the whole system. The simulation results show that, the method of state model instability and filter unreliable has strong adaptability, can effectively restrain the divergence of Kalman filter, which improves the system's accuracy and robustness.