Results of a Comparative Analysis of 1D-CNN and GRU Models in the Task of Predicting Production Indicators of Garment Production Based on Equipment Operating Parameters

Abstract

In modern manufacturing systems, improving the efficiency of process control is largely dependent on the use of data mining and machine learning. One of the key management tasks for manufacturing enterprises is forecasting output volumes based on equipment operating parameters and production load characteristics. Traditional forecasting methods do not always account for complex nonlinear relationships between equipment operating parameters and final production indicators, limiting their application in dynamically changing production processes. This paper examines the problem of forecasting output volumes in sewing production units based on time series of equipment operating parameters. The study is based on production data from sewing units operating in correctional facilities in the Volga Federal District. The dataset covers the period from 2014 to 2022 and includes 1129 observations with a monthly time series increment. The input features included indicators characterizing the equipment condition and operating mode, including equipment load factors, downtime, failure rate, mean time to repair, order volume, material availability, and production area staffing levels. To solve the iterative forecasting problem, two deep learning architectures were implemented: a one-dimensional convolutional neural network (1D-CNN) and a GRU-type recurrent neural network. Forecasting accuracy was assessed using the RMSE, MAE, and MAPE metrics. The experimental results showed that both models are capable of effective forecasting production indicator dynamics, but the GRU model provides higher forecasting accuracy. The obtained results confirm the potential of using neural network methods for analyzing and forecasting production processes based on equipment operating parameters.