FlexEnergy4U: How households benefit from flexible electricity rates

Support for the energy transition through flexible electricity rates

As the expansion of renewable energies progresses, their share in the electricity mix is also increasing. Wind power and solar energy in particular play a central role. However, both energy sources are highly dependent on the weather. As a result, the energy production of wind turbines and solar cells fluctuates almost continuously.

In order to maintain a balance between the production of and demand for electrical energy, methods are used to make demand more flexible: Demand Side Flexibility (DSF) and Demand Side Management (DSM). DSF describes the flexible consumption of energy over time, such as charging an electric car overnight. Exactly when and how much is charged can be flexibly arranged if the goal is for the car to be charged by a certain time. DSM, on the other hand, refers to the optimization of DSF with the aim of optimizing a certain parameter. This includes, for example, minimizing energy costs or grid-friendly behavior.

Such electricity flexibilities are already being used successfully in industry and commerce. With the introduction of Section 41a of the German Energy Industry Act (“Energiewirtschaftsgesetz”), the legislator has obliged energy suppliers to offer flexible rates for private households as well. However, in order for private households to make optimal use of flexible rates and make a significant contribution to the energy transition, three conditions must be met:

1. The individual energy demand within a day must be known precisely. The current time resolution (one meter reading per year) is not sufficient for this.
2. Energy demands must be anticipated and predicted for the near future, for example for the duration of a washing machine program, in order to optimize energy flows.
3. A measurement method must be developed that can analyze and penetrate the energy requirements of a microgrid (closed system, household) from a central point (house connection) with sufficient accuracy without installing many measuring points in the system and thus jeopardizing practicability and economic efficiency.

FlexEnergy4U: Making smart use of flexible electricity rates in private households

The IMPT – Institute of Micro Production Technology at Leibniz University Hannover and Ladon Energy GmbH are addressing precisely this issue with their FlexEnergy4U cooperation project. The project aims to optimize the flexibility of private households in the energy market in such a way that energy costs are reduced and, at the same time, a contribution is made to the energy transition. To achieve this, the goal is divided into several sub-goals:

1. Real-time visualization of energy flow: The energy flow (both positive and negative) should be available online via a simple dashboard (see Figure 1).
2. High-resolution recording of load profiles of individual consumers: This data is used to train AI models.
3. The trained models are used to break down a household’s total consumption into individual devices.
4. A forecast for energy consumption in the near future is created based on historical consumption data.
5. Make the consumption forecast available for later use with home energy management systems (HEMS). The consumption forecast can also give grid operators insights into consumption in their subgrids in order to make regulation there more efficient.

A BCG study from November 2025 estimates the potential savings for private households at up to €450 per year. These savings could increase even further in the future.

The researchers are collecting data on household energy consumption…

The research project plans to use only one sensor per household instead of equipping each device with a sensor, as the latter would be costly and uneconomical (see Figure 2). This requires suitable measurement technology, which is being developed and built by Ladon Energy GmbH. A distinction is made between two measurement systems. The sensor (see Figure 3) is capable of measuring all three phases of a standard household connection. The current and voltage per phase are sampled at 20 kHz. The sensor can be installed in the fuse box using top-hat rail mounting. The plug, which is designed as an adapter plug, was developed to record the energy signature of a single device. Similar to a radio-controlled socket, it can be connected upstream of the consumer and measures only one phase instead of three, also with a sampling rate of 20 kHz.

A set consisting of a sensor and several plugs is used to collect measurement data for individual consumers. The data from the plugs serves as training data for the AI models. Due to the large amount of data, it is important to process and structure the data in a meaningful way in order to reduce the amount of data to be transferred. Ultimately, the AI model should run directly on the sensor to avoid high data streams and only forward relevant information on consumption and forecasts.

… and train an AI with the measurement data

The researchers use the measurement data from the plugs and sensors to build a device database, which is then used to train the AI models. The database is categorized by device type, as individual models from different manufacturers differ only slightly in their energy signature. For example, a refrigerator always shows a high peak in consumption when the compressor motor starts up. The power consumption then remains constant for a certain period of time before the cooling process ends when the target temperature is reached (see Figure 4). The device database is continuously being expanded and developed.

With the completion of the first version of the device database, the research project is currently entering the phase of training the AI models. Due to the large amounts of data, the data from the sensors and plugs is stored in a proprietary data format. For training purposes, this data is first converted into a different data format. Research is currently underway to determine which data format is most suitable. After conversion, the data is used to train AI models with the help of a computing cluster at the university. The first step is to examine neural networks that are well suited for time series. The data from the sensors serves as a test to check whether the models can separate individual device signatures from the total consumption (see Figure 5).

After training, the next challenge in the project is to run the models on the sensor and then process real-time data and make it available for energy management systems.