Combined economic-emission dispatch problem: Dynamic neural networks solution approach

Abstract

This paper explores the performance of dynamic neural networks (DNNs) in solving the combined economic-emission dispatch (CEED) problem. The idea behind the CEED formulation is to estimate the optimal generating unit schedule in such a manner that both the fuel cost and the pollutant emission levels are minimized for a given load demand. DNNs present an elegant approach to transform the CEED optimization problem into dynamical systems whose states converge to the desired optimal generated power configuration. As compared to other neural network approaches, e.g., multilayer perceptron or radial basis function networks, complicated problem mapping or training is not necessary in DNN. Further, the convergence characteristics of DNNs are theoretically well established. Three DNN structures, namely, gradient based primal DNN, Lagrangian based primal-dual DNN, and dual DNN, are investigated in optimizing the CEED of 6 and 11 u test power systems. Compared to other optimization techniques, the investigated DNNs feature a transparent design procedure, lower model complexity, and faster convergence.