A multi-agent system platform was developed that connects Active Buildings and Ecovats, via a Dynamic Coalition Manager, to Grid and Market Operators. This platform was used to implement and demonstrate a Flex Trading concept, where active buildings, pro-actively or on request, create and communicate their own flexibility and consumption forecast to the DCM, which does a bottom-up aggregation (per congestion point). For the interactions between the DCM, the BRP and the DSO, an extension to USEF was developed, where the combined use of both the forecast and flexibility information is used to propose a (per congestion point) optimal flexibility dispatch, rather than just asking for any flexibility that solves the problem. This multi-agent platform has been demonstrated and validated through two pilots. In Sweden for a cluster of buildings, and in the Netherlands for the Ecovat (a large seasonal thermal storage vessel that can feed District Heating Networks).
At the building level, multiple (grey-box and black-box) measurement-driven modeling approaches were developed that can be applied in retro-fit situations without requiring the intervention of a human modelling expert. Specific solutions were developed and demonstrated to overcome the problem of bad and missing data by indirectly estimating them using non-intrusive measurement from data that are available. For the multi-apartment pilot buildings we were able to forecast the overall temperature evolution – except some fast dynamics - in an accurate manner (e.g. error < 3 Degrees Celsius). Using these models, a flexibility forecast is done based on given comfort boundaries, and optimal planning is done within these comfort boundaries.
To determine the optimal amount and placement of sensors, a data-driven methodology was developed that uses statistical tests and advanced kernel independence tests to identify the most relevant sensors in a more robust and informative manner than regular correlation coefficient tests, and without requiring any building specific information.
A Grid Flex Heatpump concept was developed, that aims at enabling profile following services (e.g. balancing) with heatpumps by improving the granularity and determinism of their response to a control signal. Two variants have been proposed. One using an indirect control paradigm (outdoor sensor override) and one using a direct control paradigm (direct compressor speed control). For the direct control paradigm, very high accuracies have been achieved (up to being 98% of the time within a 3% error) but such an interface is not commonly available. The indirect control paradigm has a lesser accuracy (up to being 76% of the time within a 3% error) but it can be applied to almost all heatpumps. It requires though the creation of a heatpump signature model to determine the sensor override control signal for achieving the requested consumption. It was noted that the achievable accuracy is very much heatpump brand and model dependent (i.e. limited by the heatpump internal controller).
A date-driven grid-model free methodology was developed to enable buildings to autonomously decide on local flex activations for the real-time mitigation of local grid problems. This methodology uses a grid sensitivity map that is created from measurements without any knowledge of grid topology or cable characteristics to determine fair droop settings for each of the buildings. These fair droop settings, that are derived form the grid sensitivity map, result in a more equal spread of pains (demand to offer flex) and gains (opportunity to offer flex) than traditional droop settings.
A neighbourhood Impact Analysis Tool was developed that for a specific local grid and scenarios related to future growth of vRES and electrification of heating and transport (= flex) simulates the impact of flex activations on curtailment mitigation, self-consumption and self-sufficiency.
