In this paper, a novel recurrent neural network (RNN) is presented to deal with a kind of nonsmooth nonconvex optimization problem in which the objective function may be nonsmooth and nonconvex, and the constraints include linear equations and convex inequations. Under certain suitable assumptions, from an arbitrary initial state, each solution to the proposed RNN exists globally and is bounded, and it enters the feasible region within a limited time. Moreover, the solution to the RNN with an arbitrary initial state can converge to the critical point set of the optimization problem. In particular, the RNN does not need the following: 1) abounded feasible region; 2) the computation of an exact penalty parameter; or 3) the initial state being chosen from a given bounded set. Numerical experiments are provided to show the effectiveness and advantages of the RNN.
In this article, we develop a continuous-time algorithm based on a multiagent system for solving distributed, nonsmooth, and pseudoconvex optimization problems with local convex inequality constraints. The proposed algorithm is modeled by differential inclusion, which is based on the penalty method rather than the projection method. Compared with existing methods, the proposed algorithm has the following advantages. First, this algorithm can solve the distributed optimization problem, in which the global objective function is pseudoconvex and the local objective functions are subdifferentially regular in the global feasible region; Moreover, each agent can have different constraints. Second, this algorithm does not require exact penalty parameters or projection operators. Third, the subgradient gains for different agents may be nonuniform. Fourth, all agents reach a consensus in finite time. It is proven that under certain assumptions, from an arbitrary initial state, the solutions of all the agents will enter their local inequality feasible region and remain there, reach consensus in finite time, and converge to the optimal solution set of the primal distributed optimization problem. Numerical experiments show that the proposed algorithm is effective.
17 April 2012 in Neural Computing and Applications
Based on the characteristic of cell voltage fluctuations in the process of electrolytic aluminum, a new method based on neural-network-genetic-algorithm (NNGA) for the optimization of cell voltage is proposed in this paper. First, the method of kernel principal component based on analysis of electrolytic aluminum process is used to determine the operating parameters. Second, in order to predict cell voltage in real time, back propagation neural network (BPNN) is used to establish the cell voltage prediction model. Third, the model of the optimization control of cell voltage is constructed, and then, genetic algorithm is used to optimize cell voltage and obtain corresponding operating conditions. Finally, the actual production data is used to perform experimental verification. The results show that the proposed method based on NNGA is effective. The process of electrolytic aluminum can operate under the optimal production conditions, and the goal of saving energy is achieved.