A mixed back-propagation/Marquardt-Levenberg algorithm for optimizing the distribution electrical systems operation

This paper presents a new methodology to be applied to the distribution electrical systems optimization for reducing the resistive electrical losses of the systems in real-time. This methodology combines the best features of two known algorithms (back-propagation algorithm (BP) and Marquardt-Levenberg algorithm (ML)) for improving the convergence of artificial neural networks (ANN) training. Since sometimes neither the first algorithm nor the latter one is able to easily find the global solution, the convergence of the training process can be improved in such special cases with the combination of these two algorithms. A consolidated algorithm based on heuristics was adopted for obtaining the optimal topology of the network, in relation to electrical losses, taking into account several load patterns (energy demand). Finally, the optimality of the solution (distribution system configuration regarding minimal electrical losses) provided by the proposed BP/ML method is checked by using a binary integer mathematical programming method