Research On Particle Swarm Optimization Algorithm Based On ITO Stochastic Process And Its Application

Online shopping plays an important role in people's daily life under Internet+ times. The problem of packages overstocking caused by logistics distribution, is concerned after the "crazy shopping holiday"—Double eleven. The logistics distribution can be regarded as the vehicle routing problem in essence. The vehicle routing problem is the famous traveling salesman problem when we use one car and ignore the carrying capacity of the car. Currently, researchers have a lot of achievements on vehicle routing problem. But there are still shortcomings, such as falling into domain optimum easily, poor convergence and disability. In order to solve the vehicle routing problem effectively, the main research work is summarized as follows:Firstly, based on the analysis of the real world, we proposed a new vehicle routing problem model:the dynamic real-time vehicle routing problem with users' satisfaction, referred as DRVRP. In this model we map the vehicle routing problem's time window to users' satisfaction, and consider three important factors:the traffic factors, customers' demand and satisfaction.Secondly, in order to to overcome the drawbacks, such as premature, bad accuracy, and convergence, of the particle swarm optimization algorithm, we proposed an improved particle swarm optimization algorithm, which is composed of three steps:(1) conducting benefit analysis (i.e., between ancestor particles and optimal particles) to confirm the speed level in updating particles, (2) decomposing the speed, and (3) assigning the speed to different dimensions of the particle to realize asynchronous updating. Each particle has a discard probability, and is replaced after being discarded. Furthermore, disturbance mechanism and random restart strategy have been used in the improved PSO. To address the problems of combinatorial optimization, we redefined the sphere-gap transferring algorithm, and applied greedy algorithm for the initialization of particles to improve the accuracy as well as the convergence rate of the algorithm.Thirdly, we proposed a particle network algorithm based on game theory model and Newtonian mechanics. The algorithm inputs the attribute of mass and acceleration to each particle through the analysis of the particle properties, update process and convergence model of the standard particle swarm optimization algorithm, then computes the weight, resistance and gravity of each particle, and use the common characteristics of particle network to move each particle. In addition, the particle dimensions are divided into advantage and disadvantage sections, thus reducing the update dimension. Normally, update will only process the disadvantage section to keep and extend their advantage section, such that the convergence speed can be improved. For the disturbance situation, the algorithm updates its advantage section to stay away the current network so that it can jump out of local optimization. Also we use reversed strategy to deal with the collision of particles, and select the suitable speed update model by analyzing all the particles in the last stage of the algorithm.Fourthly, we improve the ITO algorithm. In older to solve the contradiction between the exploration and exploitation when ITO solve the discrete combinatorial optimization problem, we make the particle drift and fluctuation synchronously, and continue to make a small fluctuation when the feasible solution is found. In addition, we introduce ant colony information system to design the customer point selection rule thus the algorithm can be used to solve the vehicle routing problem, use cooling schedule to control the temperature variation, use associated diffusion to update drift coefficient, use hill-climbing method to update fluctuation coefficient. Furthermore, we use the method of orthogonal experiment to analyze the algorithm's parameters setting.Fifthly, we propose a particle swarm optimization algorithm based on ITO stochastic process, which contains both the advantages of the particle swarm optimization algorithm and the ITO optimization algorithm. We use the ITO algorithm's drift operator as the learning factor c1, so the wave operator as the C2, and dynamic adaptively set parameters. We introduce the gravity and acceleration to linearly decrease the inertia weight and temperature, increase the population diversity at the beginning of the algorithm and refluctuate at the end.Finally, we test the problem model and the improved algorithm we proposed. In order to verify the feasibility and effectiveness of the improved PSO and improved ITO, we test it using standard test data. In order to verify the feasibility and effectiveness of the particle swarm optimization algorithm based on ITO stochastic process, we not only test it on the problem of the traveling salesman problem and the vehicle routing problem with soft time windows, also on the dynamic real-time vehicle routing problem with users' satisfaction that we proposed. Then, in order to further observe the effect of the algorithm in the continuous optimization problem solving, also be tested on the five kind of function optimization problem. The simulation results show that the proposed algorithm is feasible and effective, and has better accuracy, convergence and stability in solving the problems.