Design And Implementation Of Parallel Path Planning Algorithm Based On Hopfield Neural Network

Path planning algorithm is not only a classic problem in graph theory and algorithm theory, but also in real life there are so many applications, such as ITS, logistics, communication, LBS, unmanned aerial vehicles, etc. In view of the limitations of the traditional shortest path planning algorithm on solving path planning problem, this paper proposed a path planning algorithm based on Hopfield neural network, and the algorithm is accelerated by using OpenCL(a parralle computing framework).First of all,the path planning problem with multi-objective is analyzed, and establish the mathematical model by using the unified target method to convert the problem of multiple targets for a general comprehensive goals. And the characteristics of the traffic network model is introduced, as well as the related constraint conditions and influencing factors.Secondly, this paper introduces the basic concepts and properties of Hopfield neural network, the stability of the energy function of CHNN and its feasibility for solving the convex optimization problem were analyzed. And the network iteration step length update strategy was improved.For CHNN limitations on solving the convex optimization problem, this paper proposed a algorithm named SA-CHNN, The new global search algorithms result from the combination of the genetic algorithms and simulated annealing algorithms. The experimental results show that the SA-CHNN can effectively jump out of local minima, and converge to the global optimal value.Thirdly, the parallelism of Hopfield neural network is analyzed, and its iteration process is translated into matrix operation. Use the OpenCL parallel computing framework to test different size of matrix operations, and optimize the memory access. Test results show that parallel computing model can significantly improve the speed of matrix operations compared with the traditional serial calculation. The parallel computing is applied to solving a multi-objective path planning problem, proves the algorithm in solving a multi-objective path planning problem can obtain the correct result.Finally, combined with parallel computing, path planning algorithm and GIS technology to develop a path planning simulation software.