Load estimation in electricity distribution substations using deep learning: The case of Azerbaijan

Electrical energy is generated in diverse power plants and can be sent to residential or industrial regions through distribution lines, but it is not stored in a regulated manner. Owing to the advancing technology, burgeoning heavy industry, and growing population, our need on electrical energy is escalating, just as our reliance on all forms of energy is increasing. As a result, countries worldwide, including our own, are making efforts to boost the production of electrical energy. As energy production rises, certain essential production resources are diminishing at an equivalent rate. Smart grids enable real-time monitoring of energy distribution and transmission through the use of smart meters, which in turn allows for immediate assessment of the supply and demand balance. Given the rising demand for electricity, it is crucial to implement methods that promote energy conservation and minimize losses throughout the distribution and transmission of energy. Energy demand forecasting is a method commonly employed to enhance energy efficiency. Short-term energy forecasting is becoming increasingly important among many methodologies. This paper presents the development of a consumption forecasting model based on deep learning. The model utilizes consumption data from one substation and six substations in the Shirvan region of Azerbaijan. The data is collected every 15 minutes using smart meters, along with weather data. Therefore, the deep learning based model achieves improved performance. Individual models were created for each substation and a reduction center, resulting in a success rate exceeding 93%.