Publication: How accurate are predictions regarding energy consumption in households?
/
Publication: How accurate are predictions regarding energy consumption in households?
For its research project TETRA EnergAI, Howest is developing several different cases around AI and energy management. One of them is predicting energy consumption and energy production in residential homes, using Artificial Intelligence (AI). In this article, we highlight how the possibilities for measuring and predicting have evolved.
In the past, household energy consumption was not considered flexible enough to be managed or planned. The emergence of new technologies such as electric vehicles, heat pumps, PV panels, home batteries, etc. is changing that now.
Main section
Quick facts
/
Tetra project
/
Nelectra article
Impact of digital meter and capacity tariff
The introduction of the digital meter and the capacity tariff has also increased consumers' awareness of their energy consumption. Previously, it was difficult for households to know exactly how much energy they consumed and when. However, the digital meter provides detailed, real-time data, allowing consumers to gain insight into their consumption patterns and make better-informed decisions to optimize their energy use.
The capacity tariff, introduced on January 1, 2023, by the Flemish Energy Regulator VREG, has brought significant changes to consumers' energy consumption. Previously, the electricity bill was mainly calculated based on the amount of electricity consumed (kWh). With the capacity tariff, peak consumption is now also taken into account, or in other words, the amount of electricity consumed simultaneously.
The goal of the capacity tariff is to encourage consumers and businesses to spread their consumption and avoid peaks.
For owners of solar panels (PV panels), the arrival of the digital meter and the associated injection tariffs meant a significant change. While the old reversing meter allowed excess energy to be fed back into the grid without direct costs, the digital meter separately records how much electricity is drawn and how much is injected.
Increase self-consumption
Those who want to make optimal use of their PV panels can apply various strategies to increase self-consumption:
• Use of smart appliances, such as washing machines, dishwashers, and electric boilers. By only operating these when the solar panels produce the most energy, households can increase their self-consumption.
• Installing a home battery can help store excess energy for later use, reducing dependence on the grid and increasing self-sufficiency.
• By using smart thermostats and energy management systems, households can better align their energy consumption with the production of their solar panels.
• Small adjustments in daily life, such as cooking or vacuuming during sunny hours, can contribute to higher self-consumption.
The LSTM model
Due to the possession of the digital meter or an EMS (Energy Management System), many households have a timeseries dataset of their electrical consumption. When predicting household consumption, a Long Short-Term Memory (LSTM) model can be used, for example.
An LSTM model is a type of recurrent neural network (RNN) specifically designed to process sequential data and recognize patterns over longer periods. Here is an overview of how an LSTM model works with neurons:
• Cells and Gates: an LSTM consists of cells that store information over time. Each cell has three gates: the input gate, the output gate, and the forget gate. These gates control the flow of information in and out of the cell.
• Forget Gate: decides which information should be removed from the cell state. This helps the model forget irrelevant information.
• Input Gate: determines which new information should be added to the cell state. This ensures that important new information is stored.
• Output Gate: regulates which information from the cell state is used to generate the final output.
Through these mechanisms, an LSTM model can effectively learn from sequential data and recognize patterns that extend over longer periods.
Bottom section
Be careful with general predictions
But predicting the consumption pattern of households is not so simple for the following reasons:
• The electricity demand of individual households is generally very variable. There is a significant difference between different households, and the consumption pattern of households depends on various parameters, including the number of people, remote work, family composition, and the type of electrical appliances.
• The consumption behavior of households is strongly influenced by various factors such as the type of heating, the method for heating sanitary water, the ownership of an electric vehicle, and the presence of solar panels. For example, the general pattern of a household with a night-time boiler looks very different from the general pattern of a household with gas central heating.
• Households also constantly deviate from their general patterns. For instance, people may choose to cook an hour earlier or later due to changes in their daily lives. Less predictably, there can be sharper deviations from general patterns, for example, due to a change in household occupancy, people going on vacation, etc.
• The consumption of households has a pattern with daily, weekly, and seasonal cycles.
• When the consumption prediction of one household is compared to the actual consumed value, we see that the predictions of both peaks and troughs are consistently an underestimate of the actual values. The prediction is able to follow the trend of the consumption pattern, but the accuracy of the prediction is not very precise.
Due to the low accuracy of household predictions, the gains made by using Machine Learning algorithms are very limited compared to a rule-based algorithm.
As the household has more sub-measurements or LCT (Low Carbon Technologies) such as heat pumps, EV charging stations, PV panels, home batteries, etc., the predictions become more accurate and can be targeted through an Energy Management System.
You can find the publication of the article atthe website of NElectra