The Feedback Neural Network Optimization Method And Its Application In Job-shop Scheduling And Routing Of ATM Network

The feedback neural network optimization algorithm is a new and potential approach for combinatorial optimization problem. This dissertation studies the algorithm's theory and application in job shop scheduling problem (JSP) and virtual path (VP) routing in ATM network. The main work is described as follows:1) To the job-shop scheduling problem, all restrictions are considered, and a new computational energy function that consists of row inhibition, global inhibition, unsymmetrical inhibition and column inhibition is given. An improved Hopfield neural network (HNN) method for JSP is presented, and it keeps the steady output of neural network as a feasible solution satisfied with resource and sequence restrictions for JSP.2) In order to avoid HNN to be sunk into local minimum, the approach based on a stochastic neural network for JSP is developed. It introduces the simulated annealing algorithm into the DHNN of JSP. During the course of search, the change of the maximal finishing time for JSP is considered. The algorithm for Gantt graph is proposed to find the maximal finishing time of JSP, which avoids the overlap in the graph directly drawn in terms of the cost trees, and assures that the solution is the half active scheduling. The simulation results show that the approach can keep the steady output of neural network as a global solution for JSP, but it's convergence is slow.3) To improve the convergence of the method based on the stochastic neural network, the discrete Hopfield neural network with transient chaos method (TDNN) for JSP is developed. The simulated annealing algorithm is replaced with the chaotic optimization search, and the chaos is produced by the self-inhibition in TDNN. In the chaotic search, the maximal finishing timeof JSP is also considered. By changing the threshold, when TDNN is degenerate into DHNN, the system is close to the global solution, and will be stable at the place of the optimal or near-optimal solution. The simulation indicates that it is capable of the global search, and its convergence is also more quickly.4) The proposed HNN algorithm for JSP is applied to the communication network, and then TDNN method for VP routing in ATM network is put forward. Compared with other results given, the simulation results show that it is capable of faster computation, stronger convergence and lower computation complex.5) The realization of the feedback neural network optimization method on neurocomputer and the digital computer is studied. The HNN for JSP is realized on neurocomputer. By solving example, the parallel computation ability and the validity of solving JSP are confirmed. Accordingly, it provides a feasible way for settling the larger JSP. In addition, the feedback neural network optimization software is developed, and it is applied to solve an actual JSP from a machinery factory. The results show that it can effectively get the near-optimal solution.