Main Article Content
Electricity system planning is very important for electricity providers (PLN). One of them is electricity load forecasting. Backpropagation artificial neural network is one of the best methods used in electricity load forecasting because it can give high accuracy values. In application, backpropagation neural networks often provide poor convergence speed values during the training process. Therefore, it is necessary to do various combinations of training functions to accelerate the convergence of network training. In this study, a backpropagation neural network model was developed with a combination of gradient descent training functions (traingdm, traingda, traingdx). The architecture of this network model uses 24 inputs, 1 hidden layer consisting of 16 neurons and 1 output. This model uses peak load data from Pemecutan Kelod Substation and the number of kWh sold in the South Bali area as an input variable. The results show that the best model of the neural network is using the traingdx training function. In this model, the MSE training is 1.03x10-8 and with a training convergence speed is 4 seconds and MAPE testing is 6.24% with a network accuracy is 93.75%.
This work is licensed under a Creative Commons Attribution 4.0 International License.