go back

Simulation-based Design and Evaluation of a Smart Energy Manager

Tobias Rodemann and Kai Kitamura, "Simulation-based Design and Evaluation of a Smart Energy Manager", EuroCAST, 2019.


Decreasing prices for Photo Voltaic (PV) systems promise an affordable solution for a low emission lifestyle. With the rise of electric mobility a sufficiently large PV system could even be used to power both private homes and Battery Electric Vehicles (BEVs). In practice, the fluctuating nature of PV systems requires a storage element to balance energy flows through periods of over-production (more power is produced than required) and no production (e.g. during the night). In this work we want to use a BEV’s internal battery instead of a stationary battery solution. An energy manager (EM) is used to control the flow of energy between house to BEV. The EM should be operating to minimize energy costs by maximizing self-consumption of PV power, to minimize charging peaks and to maximize customer satisfaction. The first objective refers to the basic energy costs of the building and for recharging the car minus any money earned by selling excessive PV power. The second objective tries to avoid the generation of high peaks in energy demand, which could destabilize the local energy grid (especially if many similar controllers are operating in the same sub-grid.) The third objective is handled by a customer satisfaction index (CSI) that compares the user’s (modelled) requirements with the result of the charging process. The CSI considers intermediate and final State-of-Charge (SOC) values of the car battery. While developing a controller that handles a single objective well is relatively straightforward, finding a good balance for all three objectives in many different scenarios is rather challenging. Quite obviously, the development and validation of such a controller can not be done with a real system due to the disturbances to the inhabitants and the limited control over external factors like weather and user behaviour. We have therefore decided to employ a simulation system - a so-called Digital Twin- that models relevant physical effects for building and car. Our simulator uses the Modelica language which is well-suited for the modeling of multi-physics systems, but is not a good tool to develop complex controllers. In a mid-term perspective, we believe that energy manager systems need to employ latest methods from the machine learning and data analysis field like prediction, learning, and optimization. These methods can easily be applied in high-level languages like Python. In order to link a controller written in Python with a detailed physical building simulation we are using the Functional Mockup Interface (FMI). Originally developed for connecting different simulation tools, FMI is used to link two different progams (Modelica physical simulation and the Python EM logics) via a connecting interface. Once the simulation model is completed, the controller can be developed without any change to the simulation model. Our studies show that handling all three objectives is a tough challenge. If one only charges with excess PV power, any rainy day would lead to low customer satisfaction (BEV is not properly charged), and charging from the grid produces higher costs. An obvious solution is to use a prediction of building energy demand and PV harvest to estimate the amount of excess PV power available until the next usage of the BEV. The remaining power could then be taken from the grid with a flat power load distribution over time. Unfortunately, predicting a buildings power demand and PV power production comes at a substantial cost. Prediction of building energy demand requires data over longer periods of time (months or even years) and possibly substantial computing power. A weather forecast or a prediction of PV system output necessitates an internet connection. All of this would drive costs up and lead to additional data protection / privacy concerns. In this work we used our digital twin development environment to investigate a number of different controller strategies in terms of their performance regarding the three objectives defined above.

Download Bibtex file Per Mail Request


By continuing to use the site, you agree to the use of cookies. more information

The cookie settings on this website are set to "allow cookies" to give you the best browsing experience possible. If you continue to use this website without changing your cookie settings or you click "Accept" below then you are consenting to this.