Research On Optimization Model And Algorithm For Ground Station Data Transmission Resources Allocation

Ground station data transmission resources allocation optimization is the process of assigning data transmission equipments to each ground station based on the request of data transmission capability. For the problem that is hard to make mapping connection between ground station data transmission resources allocation and data transmission capability, the dissertation designs an optimization framework to data transmission resources allocation of ground stations based on metamodel. This framework decomposes the optimization process into three phases: modeling and computing the data transmission capability to ground station system, metamodeling of the mapping between data transmission resources allocation and data transmission capability and optimization of data transmission resources based on the metamodel. The main work and contribution include:(1) The ground station data transmission programming model based on capability and particle swarm optimization algorithmBased on the characteristic of data transmission process and resources, the ground station data transmission programming model was established following the description of constraints, with the objective of maximizing the total time of data transmission service. The model uses data transmission requirement instead of data transmission task, which can achieve the corresponding relationship between ground station data transmission capability and the optimal data transmission resources allocation schema. The improved particle swarm optimization (PSO) algorithm used for ground station data transmission programming is designed. The decimal particle coding and decoding operator were designed for particle swarm iterative optimization based on the time windows, which can transform each particle code to a data transmission executing schema. To account for the weakness of standard PSO which is prone to get into local optimization, the particle swarm dynamic adjusting strategy is designed which ensure the diversity of particles in swarm. Particle swarm individual velocity controllable strategy is designed to improve the local searching quality. To increase the rationality of result, the extending relation between data transmission resource allocation schemas is proposed, and the particle swarm initialization strategy based on the extending relation is designed. Experiment results indicate that the improved PSO can find high quality solution fast for various scale problems.(2) Designing of radial basis function neural network with multi-type activation functionsThe model of radial basis function neural network with multi-type activation functions is constructed based on the analysis of concluding and sorting to each kinds of common activated function, which includes thin plate spline fitting network (TPSFN) and gauss fitting network (GFN). The TPSFN is used to fit the mapping trend. A hybrid learning algorithm combined intelligent optimization and gradient descend method is designed for TPSFN training, and the structure adjusting strategy for TPSFN is proposed to make it have small number of hidden nodes. With the local characteristic of gauss function, the GFN is used to fit the residual error of TPSFN. To improve the fitting quality and accelerate the parameter learning process, a forward constructing algorithm based on l_k norm regularization is designed for GFN, which extends the regularized orthogonal least square algorithm, and the analytical result is deduced for the variable k in some typical values. Experiment results show that the model can combine the advantage of thin plate spline activated function and gauss activated function, and get smaller fitting errors.(3) Ground station data transmission capability evaluating metamodel based on partial order constraintsTo improve the fitting quality of radial basis function neural network with multi-type activation functions to ground station data transmission capability evaluation process, two kinds of partial order relations are designed, and the approximating model of radial basis function neural network with multi-type activation functions based on partial order constraints is constructed. The model relaxation and subgradient method are adopted to solve the problem. To increase the representativeness of training samples, a sequential experiment design method to ground station data transmission resources allocation scheme is proposed. The initialization sampling constructing algorithm based on nearly orthogonal measurement is designed. The sequential sampling subspace constructing algorithm is proposed based on the fitting error of metamodel, and the sequential sampling method combined the distance measurement and irregulation measurement is designed. The experiment results indicate that the partial order constraints can make the metamodel have more rational fitting results and the sequential experiment design make the training samples more representatively which can let metamodel have better fitting result with lesser training samples.(4) Model and approximate algorithm for ground station data transmission resources allocation optimizationTo study the characteristic of optimal ground station data transmission resource allocation schema, three kinds of optimization models are constructed, in which main factors are the total data transmission time and the constructing expense. The variable neighborhood estimation of distribution algorithms is designed to solve the problem, for the hardness to compute data transmission capability of every ground station data transmission resource allocating schema. The algorithm uses the probability distribution neighborhood structure to enlarge the searching range which can strengthen the stability to get the satisfied result. Meanwhile, the algorithm uses the individual location neighborhood structure and individual potential neighborhood structure to search the local range, which can improve the searching efficiency. The global constringency of the algorithm is proved based finite population. High optimizing stability and rapid searching speed were showed with application examples. The local region data transmission capability variety analysis method and the local region data transmission capability statistical analysis method are designed to analyze the characteristic of data transmission capability of different schemas in the neighborhood region of typical data transmission resources allocation scheme. The results of the two methods present the bound and distribution of the data transmission capability of the data transmission resource allocation schemas in studying.