Skip to main content
10 September 2021
R&D Division, Hitachi America, Ltd.
In recent years, energy regulations, environmental policies, and technology advances have driven a significant rise in the penetration level of distributed energy resources (DERs) in distribution systems, including solar photovoltaic panels (PV), energy storage systems, electrical vehicles, etc. These transitions have brought various challenges to utilities, e.g., reverse power flow, over-voltage profiles, etc. Meanwhile the modern grid digitalization, along with the proliferation of IoT devices in energy grid is expected to generate a vast amount of data, which provide various new application opportunities for grid automation, especially those AI-based technologies.
Although AI-based technologies have achieved great success in many domains such as computer vision, and natural language processing, it has not been widely adopted in some critical infrastructures such as power grid. Before any new technologies can be applied in this field, particular in system automation control, they must be fully tested and verified to operate in an extremely reliable manner. In addition, the power system operation environment is highly complex with very high dimensional state and action spaces, especially with the proliferation of DERs, smart control devices in field.
Currently most innovative distribution system applications are developed in a separate virtual environment using offline or archived historical data set, without direct interaction with realistic power grid operation environment. So, it is necessary to have some advanced computational tools that can not only perform domain-specific system simulation, but also provide a flexible interacting environment to support the development and testing of different system analysis tools or AI-based control solutions. A cross-domain co-simulation framework can well address the gaps mentioned above. There already exist several commercial or open-source distribution system simulation software, e.g., GridLAB-D, OPENDSS, CYMEDIST, etc. However, generally these domain-specific power simulation tools provide limited openness and flexibility for expansion of third-party models, or new system applications.
The Energy Solution Lab team at Hitachi America R&D has developed an open and flexible co-simulation framework, featured with layered architecture, standardized application programming interfaces (APIs), and modular design, as shown in Figure 1. On the engine layer, different analytic engines can be deployed, such as distribution system simulator to provide realistic simulation environment for testing/verification, distribution system analytics functional modules, etc. The interaction between the engine layer and use-centric application layer is coordinated within a Python environment through these well-defined APIs. These applications featuring with user-centric GUIs can be as simple as time-series power flow study, or as complicated as system-wide integration capacity analysis, AI-based system control, etc.
Figure 1: System architecture of the proposed co-simulation framework
We demonstrate the flexibility of this co-simulation framework through a use case - Integration Capacity Analysis .
As a key distribution system analysis application – ICA, for which is also known as hosting capacity analysis (HCA), is to quantify the capability of distribution system to host additional DERs at specific locations within power system operation limitations, e.g. voltage violation, thermal limit, etc. Typically, the iterative simulation approach is preferred considering its high accuracy . The ICA not only requires a reliable grid modeling and simulation environment, but also a complex analysis process. As shown in Figure 2, the ICA application requires interaction between customized analytical module and power system simulator, which takes care of high-fidelity grid modelling.
Figure 2: Integration Capacity Analysis workflow
In these customized analytical modules, various machine-learning or AI-based technologies can be flexibly applied. As one simple example, in order to overcome the drawback of having a trade-off between accuracy and computational speed, multiplicative binary search is applied for node-by-node hosting capacity calculation. Figure 3 summarizes some key metrics comparison for the system-wide ICA study in IEEE-123 node system. The binary search achieves less computational time and less power flow simulations, while having a better accuracy. Additionally, the proposed framework allows the usage of high-performance clusters with multi-core systems to further improve the run-time performance.
Figure 3: Performance comparison of linear and binary search algorithms on IEEE 123 node feeder system*
*These results are from initial tests. Performance was improved even more after further integration with the simulation engine.
More details about this co-simulation framework, and case studies can be found in . The co-simulation framework successfully interweaves domain-specific grid simulation with AI-based technologies, which is effective for any realistic analysis of smart grid applications. Meanwhile, the high-fidelity simulation environment can automatically generate massive amount of model simulation data for system dynamic study, AI-based model training, etc.
In the future, more distribution system applications are expected to be developed on this platform, leveraging AI-based technologies, e.g., AI-based Volt-Var Optimization, machine-learning-based power prediction, etc.
Many thanks to Dr. Bo Yang and Dr. Panitarn Chongfuangprinya of the Energy Solutions Lab with whom this work is being conducted.