Prosumer Net Consumption Forecasting: The Impact of Behind-the-Meter Self-Consumption and Weather Forecast
Name
Yuliia Siur
Abstract
In recent years, the adoption of renewable energy sources has significantly increased. Notably, solar photovoltaic (PV) panels are gaining widespread popularity, particularly among private households. Residential rooftop PV systems enable private households to generate and utilize their own electricity. Private households with a dual role in electricity production and consumption, also known as “prosumers”, establish direct market relationships with energy companies, facilitating the sale of surplus energy to the grid and purchasing when energy production is insufficient. The boost of prosumer-driven energy generation shifts energy companies’ electricity flow management. Traditional Consumption Metering is no longer sufficient, as it fails to capture prosumers’ interactions with the grid. Instead, energy companies adopt Net Purchasing Systems. These systems measure both the electricity consumed from the grid and the electricity exported back to the grid by the prosumers. The adoption of new metering systems gives rise to the development of novel methodologies for forecasting Net Consumption. However, the task of forecasting Net Consumption presents challenges arising from the three primary factors: a) the diverse behavioral consumption patterns exhibited by private households, reflecting complexities observed in Consumption Forecasting; b) the inherent variability of solar energy production, which is influenced by fluctuations in weather patterns and solar positioning across various time frames; c) the nature of the Net Purchasing Metering does not consider behind-the-meter values of production and consumption but rather accounts for energy injected to and withdrawn from the grid. The discrepancy between total and monitored values equals the prosumers’ self-consumption of generated energy, which remains unmonitored, thereby increasing the complexity of modeling relationships indicated in (a) and (b). In our study, we focus on advancing day-ahead Net Energy Forecasting techniques using Estonian prosumers as a case study. We introduce novel types of Additive and Integrated Models by incorporating distinct input features, aiming to mitigate uncertainty originating from weather forecast variables and unmetered self-consumption. Our approach enables the models to effectively capture complex relationships between input and target variables. Experimental results provided empirical evidence of an enhanced capacity to address uncertainty originating from weather predictions and unmetered self-consumption in our Consumption model developed for the Additive method. In contrast, other models did not exhibit any improvement. These findings establish a foundation for further research focused on understanding how the models capture the relationships between input and target variables.
Graduation Thesis language
English
Graduation Thesis type
Master - Computer Science
Supervisor(s)
Novin Shahroudi, Jean-Baptiste Scellier
Defence year
2024