A methodological approach for automatic weighting of variables for the planning of power transmission lines using artificial neural networks

Resumo

Weighting variables for multicriteria analysis is crucial in power transmission line planning. This study proposes using an artificial neural network (ANN) called Neuralnet to automatically assign weights to variables. Inputs for training included social, environmental, cultural, and economic factors, such as land use, slope, and indigenous areas. The ANN was trained using 2000 and 200000 samples and used cross-entropy as the error metric. Results showed that 5 hidden neurons were sufficient to generate weights with a maximum success rate of 69% for the larger sample size. However, this success rate may be due to the low quality of samples used, as many layers were not considered in the construction of transmission lines used as positive samples. Overall, the study demonstrates the potential of using ANN to automate variable weighting in power transmission line planning, which can lead to more accurate decision-making.