aedifion.dynamics
Documentation of the aedifion.dynamics framework.
Introduction¶
Driven by the increasing share of renewable electricity, the sustainability and cost-efficiency of building energy systems no longer depends solely on the total amount of energy used, but also on when that energy is consumed. Since renewable energy is mainly generated when the sun is shining or the wind is blowing, electricity is greener and cheaper at certain times of the day. To address this, we developed aedifion.dynamics as part of our optimization framework.
aedifion.dynamics enables the integration of dynamic electricity prices and on-site photovoltaic (PV) generation into our autonomous building control system. To minimize electricity costs, it adds an AI-driven optimization layer on top of .controls, making building operations not just smarter, but also more sustainable and economically efficient.
The optimization layer determines the optimal setpoints and activation schedules for electricity-driven components such as chillers, air handling units, heat pumps, and battery storage systems. It does so by considering comfort constraints and estimating the storage potential of the building.
The resulting optimized control strategy is then processed by our .controls framework and implemented via the building automation system.
Figure 1: Structure of the .dynamics optimization framework
Optimization service¶
The central goal of .dynamics is to shift the electricity demand of technical building systems to times when electricity prices are lowest. To achieve this, the system leverages the building’s thermal storage capacity, along with any available electricity and hot/cold water storage.
Figure 2 illustrates this concept using a fluctuating electricity price (middle subplot). When prices are low, electrical power consumption (top subplot) is increased to charge the storage system, raising its state of charge (SOC) as shown in the lower subplot. As a result, electricity consumption can be reduced during periods of high prices by using the stored thermal energy instead.
At the core of this approach lies a scalable optimization framework. It is based on linear programming and models the building’s energy system using a set of techno-economic equations. The framework calculates the optimal charging strategy and steady-state power values for each time step. Storage systems are modeled using power-to-energy formulations, allowing different types — such as batteries, water tanks, or the building mass itself — to be defined in terms of their capacity and their charging/discharging power limits.
This optimization runs continuously during operation, enabling the system to dynamically adapt to changing environmental conditions and energy prices.
Figure 2:Example time series of a optimal storage usage under a fluctuating electricity price
Forecasting service¶
The most critical input for a mathematical energy system optimization is the predicted energy demand. To ensure the highest possible accuracy, the future time steps of the demand time series are forecasted using AI algorithms trained on historical data. Continuous forecasting is essential to adapt to unexpected events and to ensure that the optimization receives inputs reflecting realistic, up-to-date building conditions.
In addition to building energy demand, the dynamics of other relevant components are also learned — ranging from heating and cooling behavior to predicted PV electricity generation (based on solar radiation and cloud cover forecasts), and even the stochastic patterns of electric vehicle charging. In principle, any dynamics that can be observed in the data can be learned via AI to enhance the quality of the input time series for the optimization framework.
Figure 3 illustrates the principle of time series forecasting using the example of a building’s cooling demand. A daily pattern with peaks around noon is visible. The black line represents the actual demand, while the blue line shows the forecast calculated the previous day. From the point where the first forecast begins (indicated by the red dotted line), the uncertainty of the prediction increases the further it extends into the future. This growing uncertainty is reflected by the widening grey-shaded area.
Figure 3:Example forecast of a buildings cooling power demand
¶
Please feel free to contact us. We would be more than happy to discuss your case.