Application Research Of 6 Fuzzy Logic Controllers Parallel Genetic Algorithms On One PC-based Cluster Of Workstation In Multisensor Multitarget Tracking

Genetic algorithm (GA) can solve effectively the NP-hard problem of combinatorial optimization. GA has become one of the best tools for searching satisfactory solution. With the goal of improving performance, fuzzy logic controllers (FLC) are used for modeling GA control parameters throughout the run. Architecture parallelization can decrease the long running time of GA for time-critical conditions. In this dissertation, a novel multi-deme parallel fuzzy GA (MDPFGA) is developed to improve the efficiency and real-time performance of fuzzy GAs (FGAs) based data association algorithm for multisensor multitarget tracking. The framework of MDPFGA-based data association algorithms comes into being. These proposed algorithms are all designed and run on one cluster of workstation (COW) with message passing interface (MPI), and are tested by simulation experiments of passive location (either static or dynamic) in the scenario with multiple targets and multiple observation stations.First, research status and application feasibility are summarized for the MDPFGA and the COW. The domain of multisensor multarget tracking algorithm is introduced from the viewpoint of information fusion. The engineering application background of the objectives in the emulation is explained. The main idea and work of this dissertation is also introduced.In this dissertation, special self-learning algorithm is designed for 6FLC-FGA.Thus an automatic design approach is provided for the fuzzy logic system of 6FLC-FGA. Followed the forgone work, the MDPFGA with 6 fuzzy logic controllers (6FLC-MDPFGA) is proposed based on one PC cluster of workstation. The research work is focused on important parallel parameters of the algorithm, such as population size, migration rate and the frequency of individual migration. Empirical values and rules of parameter selection are determined conditionally. Moreover, qualitative analysis is made for different parameters with respect to the different effects on solving results. Furthermore, an extendable common algorithm platform is built.For multisensor multitarget tracking systems, the data association problem in target state estimation can be formulated as a generalized S-dimensional assignment problem. When it is represented as a sort of constrain combinational optimization problem, the data association problem can be studied by the corresponding 6FLC-MDPFGA. Based on the maximum likelihood approach, the static measurements association uses one 6FLC-MDPFGA to determine both the static S-dimensional assignment and the track formation for multiple targets tracking. With the idea of multiple hypotheses tracking, the dynamic measurements-tracks association uses a composite 6FLC-MDPFGA to manage tracks of multiple targets. Further, a novel generalized S-dimensional assignment algorithm synthesizes above algorithms. In addition, a multidimensional clustering assignment algorithm based on a 6FLC-MDPFGA is proposed by using clustering techniques. Simulation results show the application feasibility of 6FLC-MDPFGA in multisensor multitarget tracking.Finally, a new model of common hardware platform, based on double DSP and FPGA cluster, is proposed for hardware algorithm of 6FLC-MDPFGA. Some pendent problems are discussed. These problems concern both the 6FLC-MDPFGA and the application for multisensor multitarget tracking. Moreover, the perspective for further research work is presented.