Integration of distributed energy resources on distribution and transmission systems

Abstract

The power system has been undergoing a significant transformation over the last two decades. The rising share of renewable production, and the gradual decommissioning of conventional power plants, creates a need for alternative providers of power system services. This need, along with the increasing presence and control capabilities of Distributed Energy Resources (DERs), calls for the seamless integration of these resources in power system operation. This thesis deals with issues related to the integration of DERs in the power system. The presented work can be broadly divided into two main areas. The first part deals with how the requirements of the provided services, and the level of available information, affect the way a DER aggregator models and controls its portfolio. The second part deals with congestion management at a Distribution Network (DN) level, where two fundamentally different approaches are investigated and their key differences are highlighted. At a system level, Primary Frequency Control (PFC) is an important ancillary service, crucial for power system stability. Decentralized PFC provision offers robustness to failures, scalability, and reduction of communication requirements. However, existing decentralized PFC methods do not address the problems of reduced efficiency, increased switching actions, and performance guarantees under such a setup. Here, a decentralized control policy that addresses the aforementioned issues is proposed. By appropriate tuning, the desired trade-off between these three objectives can be achieved. As DERs gradually become the main providers of PFC, understanding their dynamic response becomes critical. In this thesis, limitations of existing models of aggregated Electric Vehicle (EV) dynamics are investigated. It is shown that in many cases these lead to significant approximation errors. More accurate, yet still simple, models are proposed, which can decrease these errors, providing models suited for power system frequency dynamics studies. As a potential source of flexibility in modern power systems, thermal loads are a promising option. This thesis investigates two fundamentally different thermal loads control setups. In the first, Real Time (RT) dedicated metering and knowledge of building parameters are assumed. A new aggregation method that abstracts the effect of lockout constraints is proposed, without complicating the local controllers. Analytical formulations for more restrictive energy and power limits are derived. It is shown that the use of RT air-temperature feedback may lead to unstable behaviour, whereas a simple stochastic controller without such information leads to a more robust setup with equal performance. In contrast to this approach, an experimental method to model the flexibility of residential thermal loads is proposed, based on conducting Demand Response (DR) experiments and using aggregated Behind the Meter data. This is motivated by user-privacy concerns, the lack of dedicated metering, and the lack of knowledge of the buildings’ parameters. Real DR experiments were conducted on the island of Bornholm in the context of Ecogrid 2.0, which showed the considerable uncertainty in the response of the households. The proposed method embeds these uncertainties in the flexibility model through the evaluation of the DR experiments. These showed that the thermal loads have a load reduction potential of up to 1.2 kW per household, whereas the accuracy of the model was found to be acceptable for the provision of balancing or Distribution System Operator (DSO) services, given the limitations of the setup and the involved uncertainties. Congestion management mechanisms are necessary for resolving conflicts between the DSO’s security requirements and the aggregators’ objective for cost-optimal operation. Distribution Locational Marginal Prices (DLMPs) have received attention as a way to resolve such conflicts in the day-ahead market. In this thesis, focus is given on the decentralized derivation of DLMPs, because they retain user privacy and are compatible with unbundled power markets. One issue with DLMPs in practice is their slow convergence. To address this, a way to significantly accelerate convergence is proposed, which uses topological information. Simulation results indicate that the number of iterations required for convergence can be reduced threefold, making the algorithm suitable even for intra-day application. As an alternative to DLMPs, a mid-term capacity limitation DSO service is also investigated, where the interaction between aggregators and the DSO is limited to simple auctions. A method to construct offering curves that reflect the aggregator’s opportunity cost in the day-ahead and RT markets is proposed. The method relies on creating training data based on historical scenarios, and then using time-series forecasting techniques to create offering curves. Simulation results based on real EV charging data show reasonable accuracy and precision of the offering curves, given the large price volatility. The thesis results show that DER modelling detail depends on the offered service and the availability of information. In many cases, data requirements can be reduced without loss of performance, as is the case of decentralized PFC or by neglecting air-temperature for thermal loads’ control. In cases like the dynamic models of EVs offering PFC, adapting them for this specific purpose allows for increased accuracy, which is important for relevant power system studies. Additionally, some challenges associated with the real-life implementation of DLMPs are brought forward. Even though these can be partially overcome through appropriate methods, the large complexity of the mechanism, and its difficult integration with the different electricity market layers, indicates that mid-term DSO services can be a more easily realizable solution for congestion management.