Abstract:
The purpose of this study is to find the product demand forecasting model from the artificial
neural network method in a case study of a high-voltage equipment manufacturing company. And
reduce fines for late delivery of goods by using the Matlab program. The forecasting model is
established by using time series neural network method, and the nonlinear autoregressive neural
network (NARX) function model is established by combining with external input data. Then the forecast
deviation is the average deviation percentage error (MAPE) uses the least dynamics model. By using
data that collects product sales from 2019 2022 total 48 set of data are divided into 2 data sets used
in the study. by the first dataset is learning 36 datasets and the second dataset is learning 12 datasets.
The model of this study was defined with the number of hidden layer nodes as 2, 6 and 10. The time
delay was 2 and 3. Forecasting models are combined into 6 models. The best model result of each
item from NARX forecast through the learning process Levenberg-Marquardt (trainlm), including models
1, 4, 5, 4, 3, 3, and 2 sequential The results show the least average absolute percentage error at 8.32,
5.85, 2.61, 3.39, 2.91, 2.19 and 7.76 and the results of the forecasting model were analyzed to reduce
fines from selling all 7 items Able to reduce penalties for late delivery of 4 items showing the
percentage of difference in cost reduction equal to 0.28 percent 0.24 percent 0.32 percent and 0.22 percent. It can be
concluded that the prediction of the neural network inverted neural network model It can be used to
forecast product sales and product demand as well. And reduce the cost of late delivery fines for
better results."