Smart Grid Data Communication and Analytics

With the popularity of smart electrical appliances and home energy management systems, there have been a massive amount of data generated about the electricity consumption. This data can be beneficial for the utility companies as it provides the behaviour patterns of customers, and thus useful decisions can be made to optimize the load on the grid. In this work, we propose and implement a two-way communication system between the transformer agent (TA), attached to a neighbourhood’s electric transformer, and its customer agents (CAs), attached to each house in that neighbourhood. Once data is collected at the TA, it is communicated to the utility which can control and suggest any changes in consumption behaviours. In our system, CAs form a self-healing mesh network with the TA using IP-based Wi-Fi, while TAs communicate with the utility headquarters using the LTE network. Our system is implemented in compliance with the IEEE 2030.5 standard requirements, also known as smart energy profile 2.0. We have performed several tests across the Carleton University campus. We have also tested and implemented this system in real neighbourhoods in Ottawa, including Sandcherry and Viewmount sites to prove the system’s operation and reliability.The data obtained from the communication system is stored in a database hosted by IBM Cloud services. Our aim in this work is not only to communicate the data but ii to further process it and help the utility companies design better demand side management (DSM) programs to ensure efficient transmission and distribution of energy. This solves the problem of balancing electric demand and supply at the grid and also reduces peak demands, which helps lower the electricity bills for the consumers. In this context, we analyze the household electricity consumption data to forecast energy consumption for short-term (hours/days ahead) and long-term (weeks/months ahead). To this end, we use and compare seven different machine learning models predicting the energy consumption: linear regression, polynomial regression, support vector regression (SVR) using linear kernel, SVR using Gaussian kernel(SVR-G), SVR using the polynomial kernel, feed-forward neural networks (FFNN), and recurrent neural networks (RNN) using long-short-term memory (LSTM) neurons. To measure the accuracy of these models, we compute three different error metrics: the normalized mean absolute percentage error (NMAPE), the normalized root mean square error (NRMSE), and R2also known as the coefficient of determination. We then propose a novel approach for short-term load forecasting by combining the power of multiple models and evaluate its performance on a real energy consumption dataset that is publicly available by Massachusetts Institute of Technology (MIT). Results show that our proposed model performs better than existing models for time series energy forecasting.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.