Planning for the Next TESP Version¶
At this stage, TESP comprises a basic framework to conduct design and evaluation of transactive mechanisms, and it is open for use by others on Windows, Linux and Mac OS X. The next version of TESP should rapidly expand its capabilities, by building on the established framework.
These are arguably the most important, as they add key features that are directly in TESP’s scope, and likely not available elsewhere integrated into a single platform. The more examples we provide, the easier it should be for others to write their own (better) TEAgents.
VOLTTRON is a standard for building automation and management systems, and it has been used to implement build-level transactive mechanisms for electricity, air and chilled water in co-simulation with EnergyPlus . A TEAgent based on VOLTTRON could manage the building-level transactive system, and also participate in the feeder-level or substation-level electricity markets on behalf of the building loads and resources. The work involves porting the Python-based VOLTTRON program to interface with EnergyPlus via FNCS instead of EnergyPlus’s built-in Building Control Virtual Test Bed (BCVTB). Then, the VOLTTRON program will need to construct bid curves for the grid market.
PowerMatcher is a transactive mechanism implemented by the Netherlands Organisation for Applied Scientific Research (TNO) . The existing code is in Java, with a custom API and message schema. TNO would have to undertake the work of interfacing PowerMatcher to the TESP, with technical support from PNNL.
TeMix is another transactive mechanism that has been implemented by a California-based company , and selected for some pilot projects. TeMix would have to undertake the work of interfacing its product to the TESP, with technical support from PNNL.
Passive Controller (Load Shedding) – GridLAB-D includes a built-in passive controller, and switches that can isolate sections of a circuit. This function would be extracted into a separate TEAgent that implements load shedding in response to a message from PYPOWER. If the bulk system capacity margin falls below minimum, or worse, if the optimal power flow fails to converge, the bulk system operator would have to invoke load shedding. In TESP, the PYPOWER simulator would initiate load shedding a few seconds prior to the market clearing time, which initiates a new GridLAB-D power flow and reduced substation load published to PYPOWER. Load shedding is a traditional approach that will reduce the system reliability indices, whereas transactive mechanisms could maintain resource margins without impacting the reliability indices.
Passive Controller (Demand Response) – the GridLAB-D passive controller already simulates various forms of price-responsive or directly-controlled loads. These would be extracted into a separate TEAgent for control of waterheaters and other loads, complementing the transactive dual ramp controller for HVAC.
Generator Controller – GridLAB-D has a built-in generator controller that is tailored for conventional (i.e. dispatchable) generators with operating, maintenance and capital recovery costs included. This has not been completely developed, but it would be useful in TESP as a separate TEAgent so that cogeneration may be included. For example, several teams are developing 1-kW generators for co-generation with residential gas furnaces (the ARPA-E GENSETS program).
Storage Controller – GridLAB-D’s built-in battery only implements a load-following mode with state-of-charge and charge/discharge thresholds. We expect to develop a more capable battery controller during 2017 as part of a Washington State Clean Energy Fund (CEF) project in collaboration with Avista Utilities and Washington State University. This new agent would be implemented and tested in TESP.
The enhancements 1, 2 and 3 are probably the most important. A VOLTTRON agent is strategic because it enables intrabuilding-to-grid transactions. It also fills a weakness in GridLAB-D’s own commercial building models, which are adequate for small-box establishments and strip malls, but not for larger buildings like the school in Section 2.3. The PowerMatcher and TeMix agents are strategic because they would show usability of TESP by others, and facilitate cross-vendor experiments.
These are also important for usability and widespread adoption of TESP.
Fig. 27 shows the data structure beneath a new case configuration graphical user interface.
TE Challenge Message Schemas – NIST has defined several classes and message schemas for the TE Challenge project . Many of these tie directly to GridLAB-D, so they are already supported via FNCS. We will continue to review all of them to ensure that TESP remains compatible with TE Challenge to the extent possible.
Solution Monitor – at present, TESP is configured and launched via script-building utilities and console commands, which are adequate for developers. The two-day simulations described in this report finish within an hour or two, but that will increase as the time horizons and system sizes increase. We plan to provide a graphical user interface (GUI) with spreadsheet interfaces for configuring TESP, live strip charts to indicate solution progress, and more convenient methods to stop a simulation.
Valuation GUI – the post-processing scripts for valuation also run from the command line, which is adequate for developers. We plan to provide a GUI that presents results in formatted tables and lists, plots variables that are selected from lists, etc. Both the solution monitor and post-processing GUIs will be implemented in Python using the Tkinter package that comes with it. This makes the GUIs portable across operating systems, and allows for user customization, just as with the Python-based TEAgents.
IEEE 1516  is a comprehensive family of standards for co-simulation, sometimes referred to as High-Level Architecture (HLA). As part of Grid Modernization Lab Consortium (GMLC) project 1.4.15, “Development of Integrated Transmission, Distribution and Communication (TDC) Models”, FNCS and other National Lab co-simulation frameworks are evolving toward greater compliance with IEEE 1516. We plan to adopt a reduced-profile, lightweight version of FNCS or some other framework in TESP, so that it will be fully compliant with IEEE 1516. This fosters interoperability among simulators and agents developed by others. However, compared to some other HLA frameworks that we have evaluated, FNCS is much more efficient, handling thousands of federated processes. For TESP, we’ll need to maintain that level of performance in the new standards-compliant framework.
Intermediate Time Aggregations – for a single feeder as described in Section 2.3, a two-day simulation produces about 1 GB in JSON metrics before compression. (CSV files would be even larger). To mitigate the growth of these files, we plan to implement aggregation in time for yearly and multi-year simulations, in which metrics are aggregated by hour of the day, season, weekday vs. weekend or holiday, and by year of the simulation. No accuracy would be lost in cumulative metrics, and it would still be possible to identify metrics for individual stakeholders.
The enhancements listed in sections 3.1 and 3.2 are of known complexity, and could be implemented within the next year, subject to resource availability (including external parties TNO and TeMix). We expect to do some prioritization at a TESP pre-release workshop on April 27, and implement the selected enhancements over a series of two six-month release cycles.
Some important longer-term enhancements are described in the next four subsections. Work on them will begin, but most likely not be completed over the next year. We are also considering a faster building simulator than EnergyPlus, and federating ns-3 to simulate communication networks. For now, both of those appear to be less important than the enhancements listed in sections 3.1 and 3.2.
The growth model described in sections 1.3 and 2.4 follows a pre-defined script, with some random variability. This is adequate for short horizons, up to a few years. Over longer terms, we’ll need an intelligent growth model that mimics the analytics and heuristics used by various stakeholders to make investment decisions. For example, the TESP user may wish to evaluate impacts from a policy initiative that will have a ten-year lifetime. That policy initiative may influence investments that have a twenty-year lifetime. It’s not possible to realistically script that kind of growth model ahead of time. Instead, we need growth model agents that will make investment decisions appropriate to the system as it evolves.
Agent Learning Behaviors¶
Participants in any market will naturally try to optimize their outcomes, or “game the system” depending on the observer’s perspective. In designing brand-new market mechanisms for transactive energy, it’s critically important to account for this human behavior, otherwise undesired and unanticipated outcomes will occur. It’s up to the policymakers to design market rules so that, with enforcement of the rules, undesired outcomes don’t occur. Currently, our agents take algorithmic and sometimes probabilistic approaches to transactions, but they aren’t smart enough to “game the system” as a human would. We have teamed with Iowa State University to investigate these agent learning behaviors beginning this year.
TESP currently uses random input variables, but the simulations are deterministic and in full detail (e.g. every house, every HVAC thermostat, every waterheater, etc.) It would be more efficient, and perhaps more realistic, to have stochastic simulations on reduced-order models as an option. This opens the door to more use of sensitivity analysis and automatic optimization routines than is currently practical. We have teamed with University of Pittsburgh to investigate the subject beginning this year, building on previous work in circuit model order reduction and probabilistic modeling.
Testing and Validation¶
Testing and validation will be a continuous process throughout the life of TESP. Some opportunities will arise through past and future pilot projects in transactive energy. Other test cases will have to be created. We expect to team with Dartmouth College in formalizing this process, and also to work with Case Western University in modeling their transactive campus project with NASA.