Abstract
This thesis explores novel methodologies for improving the particle filtering algorithm and tackles state estimation and optimization problems of large-scale dynamic systems through the use of the improved particle filters. First of all, an importance density selection scheme for the particle filtering algorithm is first proposed based on the minimum relative entropy and the theorem of Taylor series expansion. By considering both the transition prior (previous states) and the likelihood (measurements), the proposed density selection scheme improves the performance of the particle filters especially when the measurements appear in the tail of the prior or the prior differs significantly from the posterior. Secondly, a particle filtering-based optimization algorithm for the multi-objective optimization problem is developed to establish a connection between the population-based optimization methods and the particle filtering algorithm. Here, the deterministic multi-objective optimization problem is represented using a state-space model. Then, samples (i.e., candidate solutions) are drawn from a distribution function, which can be computed recursively based on the performance of the prior particle set and the newly arrived observations. As the iteration progresses, the distribution function becomes more and more concentrated on the promising region of the solution space, indicating the convergence capability of the proposed algorithm. When it comes to the practical contribution, two popular state estimation problems are studied. Specifically, a daily electricity demand forecasting problem is addressed by the way of incorporating the particle filters that embed the proposed density selection scheme into a developed state-space model. In a similar vein, a problem of low elevation target tracking over the sea surface in the presence of multipath effects is considered, and a corresponding tracking mechanism is proposed based on the state-space modeling methodologies and the improved particle filters. In addition to the state estimation problems, one of the most famous problems in the area of multi-objective optimization, which is the economic and environmental load dispatch (EELD) problem on an IEEE-30 bus system, is also included in this doctoral study. Experimental results are benchmarked against several algorithms studied in the literature. Through these practical state estimation and optimization problems, the validity and effectiveness of the proposed methodologies is successfully demonstrated. Finally, recommendations for further study are enclosed.