Prototipo de macro y micro medición inteligente de energía de bajo costo para redes eléctricas de distribución para la caracterización del consumo con el fin de hacer proyecciones de la demanda de energía

Abstract

This work focuses on the design and construction of an advanced electrical energy metering system capable of functioning as both a macro and micro energy meter. The data acquisition system (DAS) and sensing stage consist of an ESP8266 microcontroller and three PZEM-004T devices, one for each phase. The innovation of the system lies in its ability to monitor electrical variables in real time and subsequently perform demand forecasting, thereby improving energy management and efficiency in distribution networks. Initially, a prototype was developed with the capability to capture electrical variables such as voltage, current, active, reactive, and apparent power, power factor, frequency, and energy at various points in the electrical network, such as transformer outputs and consumption points in homes and businesses (macro and micro metering). The developed prototype includes an application that can be installed on desktop and mobile platforms, enabling users to visualize real-time electrical consumption behavior and other monitored variables. The system also stores this data in a repository for further analysis, facilitating demand forecasting and promoting improved consumption habits. To ensure the accuracy of the device, extensive testing was conducted, comparing the prototype's results with those obtained from a certified measuring instrument. These tests allowed the determination of the device's precision class, confirming that it meets the necessary requirements for energy measurement in the aforementioned distribution network points. Subsequently, the system's measurement data was analyzed to identify consumption patterns and trends in energy demand behavior. Various statistical and econometric methods, such as ARIMA and Holt-Winters, as well as non-statistical methods like Random Forest, XGBoost, and neural networks, were implemented and compared. These methods employed different mathematical techniques to deliver results within a confidence interval and with relatively low error rates. To validate the performance of the implemented algorithms, real power demand data was used, obtained from the Sinergox platform by XM, as well as simulated consumption data generated by the measurement prototype through the provided API and the corresponding Python libraries. The time series used for algorithm training covered a four-month period, and the predictions spanned 7 to 10 days. These predictions considered the type of day and atypical dates, such as holidays, where energy consumption experiences significant variations. The results demonstrated the system's efficiency in making accurate predictions and adapting to any time series or day of the week, even under unconventional demand conditions. The developed system is projected as a low-cost, highly efficient solution for monitoring electrical variables and demand forecasting, establishing itself as a valuable tool for efficient energy management in electrical distribution networks.