Design And Implementation Of State Estimation Algorithm Based On Improved Particle Filter For Hybrid System

Due to the large scale and complexity, hybrid system becomes the focus of control field. Studying the state estimation method on hybrid system, and then acquiring the discrete modes and continuous dynamics real-time, not only can be easy to identify and control systems, but also can be supportive of fault diagnosis and decision when faulty modes are included. So it has great theoretical and practical meaning that studying the state estimation for hybrid system. Based on particle filter, several questions under the background of hybrid system were discussed in this paper and the main contributions included the following tips:1. The theory of particle filter and its application into state estimation of hybrid system were introduced, and then its advantages and existing problems were analyzed.2. The low transition probability can cause that no particle is in some discrete modes. To tackle this problem, an improved genetic particle filter was proposed in this paper. The similarity of genetic algorithm and particle filter was analyzed, and then the genetic operator was introduced into particle filter in order to improve the diversity of particle set and cover as more as possible particles. MATLAB was used to simulate the proposed algorithm. The simulation results showed the genetic particle filter algorithm improved the filter accuracy and shorten the sampling time.3. For the problem of unknown transition probability, an adaptive particle filter based on observation was put forward. The transition probability was expanded as status of the system and estimated on-line by the algorithm. Observation was applied to judge all possible subsequent modes. The posterior transition probability of each mode was calculated through multiplying the likelihood function by the prior transition probability, which was also adjusted during the estimation process. Simulations were carried in MATLAB and the results showed that the proposed algorithm was capable to track the true mode change more timely and get better hybrid state and transition probability estimation compared with another existing adaptive algorithm, although it has a relatively longer running.4. On the background of train operation hybrid model, taking its linear and Gaussian characteristics into consideration, the proposed adaptive particle filter based on observations and the RBPF algorithm (Rao-Blackwellization Particle Filter) were combined to estimate train operating status online. At last, train running process was simulated in MATLAB and the results showed the improved algorithm had a more accurate estimation compared with RBPF algorithm.