Estimating the electric consumption values of a textile factory with deep learning

Electrical energy is one of the most important sources that ensure the continuity of industrial processes. Since electrical energy is a very costly resource, minimizing electricity consumption is an important issue for enterprises. It is possible to reduce costs by estimating the resource values consumed in the production processes of the enterprises. Recently, Machine Learning and Deep Learning concepts are powerful Artificial Intelligence subdomains used for future prediction in any field. Therefore, in this thesis, a Deep Learning supported electrical forecasting model is designed to prevent excessive resource consumption of textile industry machines in their standby state. The proposed method estimates dynamic threshold values of electricity consumption in textile machines using the Long-Short-Term Memory (LSTM) and Sliding Window technique. The threshold values obtained with the LSTM model were compared with other Deep Learning methods such Recurrent Neural Networks (RNN) and Gated Repetitive Units (GRU) and Automated Regressive Integrated Moving Average (ARIMA) as a traditional method, then the results has been analyzed how close they are to real-time electricity consumption data at standby. The proposed model in this thesis successfully predicts electricity consumption levels and sends an interrupt signal to the Programmable Logic Controller (PLC) unit when the consumption levels reach the threshold, thus preventing excessive resource consumption.