Integration of LSTM and U-Net Models for Monitoring Electrical Absorption With a System of Sensors and Electronic Circuits

The evolution of electronic systems has significantly improved the monitoring of electrical consumption, ensuring precise measurement of energy assets. This article proposes a three-step approach to improving the energy system management of an industrial plant. The first phase involves the identification of hot surfaces in electrical panels using nondestructive instrumentation for thermographic measurements. The second phase develops an electronic system consisting of current sensors and a temperature control system, a control unit made with Raspberry Pi integrated with Arduino, and a data transmission system using a 4G LTE Internet device capable of measuring the electrical current used by a machine in the plant. The electronic system measures current absorption (CA) by the lines of the electrical switchboards. The final phase focuses on the implementation of deep learning (DL) method based on the integration of a long short-term memory (LSTM) model for the forecasting of hourly CA and a U-Net model for supporting the classification and detection of anomalies identified during thermographic assessment. The results demonstrated substantial improvements in the accuracy of power consumption classification and the potential to optimize power distribution and reduce consumption accordingly. This classification results in reduced energy consumption and improved process quality.