AbstractDemand response and dynamic pricing programs
are emerging in greater numbers as policy makers look for higher efficiencies and new mechanisms to integrate distributed resources into system operations. The behavior of demand-side resources (including distributed generation, storage, and load), their interaction with the power system, and their operating characteristics under different electric market scenarios raise many questions for system operators, market operators, and regulators. Computer simulation provides a powerful tool for investigating these issues. This paper discusses the use of GridLab-D as simulation environment to model the behavior of distributed resources in detail such that resources interact in new market environments while observing the physical constraints of system operation. GridLab-D provides an open source platform for integrating demand-side models with power system models and market operation models. The flexibility of this environment will be discussed as well as some of the findings emerging from having this simulation capability.
Index Termspower simulation, power modeling, power markets, energy markets, load modeling. I. INTRODUCTION significant aspect of a smart grid is its flexibility to incorporate distributed energy resources (DER generation, storage, and demand) into system operations. The vast numbers of these resources presents technical changes to how they can be effectively and reliability coordinated while preserving the integrity and resilience of the greater system. Complex systems techniques are being researched, tested, and demonstrated [1] [2]. In particular, markets can be formulated to provide an efficient mechanism for distributed decision-making that optimizes the aggregate response to changing conditions by seeking local objectives. Experimentation through the use of simulation tools can provide important insights into such system behavior, and these tools can eventually become the forerunners of tools for real-time analysis and planning studies. GridLAB-D is an open source power distribution system
Development of GridLAB-D is funded by the US Department of Energy and managed by Pacific Northwest National Laboratory, which is operated by Battelle Memorial Institute for the US Department of Energy under contract DE- AC65-76RLO1830. All registered trademarks are hereby recognized. D. P. Chassin and S. E. Widergren are with Pacific Northwest National Laboratory in Richland, Washington 99352. They may be contacted by email at david.chassin@pnl.gov and steve.widergren@pnl.gov. simulation environment [3]. It incorporates advanced modeling techniques, with high-performance algorithms to model end-use equipment and consumer behavior, coupled with distribution automation models, and software integration tools for users of many power system analysis tools. This environment is a work in progress. The most important new capabilities slated for GridLAB-D include extended quasi-steady state time-series solutions; end-use models including appliances and equipment models, consumer models, all implemented with agent- based simulation methods; distributed energy resource models, including appliance-based load shedding technology, and distributed generator and storage models; retail market simulation tools, including contract selection, business and operations simulation tools, models of SCADA controls, and metering technologies; external links to Matlab TM , MySQL TM , Microsoft TM
Excel TM and Access TM , GoogleEarth TM , GnuPlot and other text-based tools, as well as some commercial power distribution modeling systems; and the ability to run efficiently on multicore and multiprocessor machines. GridLAB-D has been validated with both existing end-use simulation and standard distribution analysis tools. In addition, the ability to integrate with many other simulation tools offers the ability to study rate cases to determine whether the price differentials are sufficient to encourage customer adoption; study the potential and benefit of deploying distributed energy resources, such as grid- responsive appliances; or study the business cases for technologies like automated meter reading, distribution automation, retail markets, or other late-breaking technologies. In addition, the tool allows one to study the interactions between multiple technologies, such as the interaction between under-frequency load-shedding remedial action strategies with appliance-based load-relief systems. The computational environment is capable of simulating distribution system behavior over time intervals ranging from a few seconds to many decades, gathering data on the interaction between physical phenomena, business systems, markets and regional economics, and consumer behaviors. The results can include many power system statistics, such as reliability metrics (e.g., SAIDI, the system average Simulating Demand Participation in Market Operations D. P. Chassin, Senior Member, IEEE and S. E. Widergren, Senior Member, IEEE A 978-1-4244-4241-6/09/$25.00 2009 IEEE 2 interruption duration index), and business metrics such a profitability, revenue rates of return, and per customer or per line-mile cost metrics. II. THE GRIDLAB-D SYSTEM A. What is GridLAB-D? GridLAB-D is a flexible simulation environment that can be integrated with a variety of third-party data management and analysis tools. At its core, GridLAB-D determines the simultaneous state of millions of independent devices, each of which is described by multiple differential equations solved only locally for both state and time. The advantages of the core algorithm over traditional finite difference-based simulators are: 1) greater accuracy; 2) handles widely disparate time scales, ranging from sub-second to many years; and 3) ease of integration with new modules and third-party systems. The advantage over tradition differential-based solvers is that it is not necessary to integrate all the devices behaviors into a single set of equations that must be solved. The simulation system also includes modules to perform the following system simulation functions: Power flow and controls, including distributed generation and storage End-use appliance technologies, equipment and controls Data collection on every property of every object in the system, and boundary condition management including weather and electrical boundaries. Additional planned modules are being developed to provide additional functionality, including: Consumer behavior including daily, weekly, and seasonal demand profiles, price response, and contract choice Energy operations such as distribution automation, load-shedding programs, and emergency operations. Business operations such as retail rate, billing, and market-based incentive programs. B. How does GridLAB-D work? GridLAB-D includes a suite of tools to build and manage studies, and analyze results. Existing and planned tools include: Agent-based and information-based modeling tools that allow you to create detailed models of how new end-use technologies, distributed energy resources, distribution automation, and retail markets interact and evolve over time. Tools to create and validate rate structures, examine consumer reaction, and verify the interaction and dependence of programs with other technologies and wholesale markets. Interfaces to industry-standard power systems tools and analysis systems. Extensive data collection tools to permit a wide variety of analyses.
The simulator can examine the detailed interplay of the many parts of a distribution system. Depending on the analysis problem, users can integrate highly detailed models to mitigate the danger of simplifying assumptions or use reduced-order models as needed to speed up processing and data entry. Through simulation studies, the tool is expected to find problems with market program offerings and business strategies. C. Who uses GridLAB-D? Todays power systems simulation tools study specific aspects of power system operations or market behavior, but new tools are required to provide the analysis capabilities for the forces driving change in the electric energy industry. The combined influence of fast-changing information technology, novel and cost-effective distributed energy resources, multiple and overlapping energy markets, and new business strategies result in very high uncertainty about the success of these important innovations. Concerns expressed by utility engineers, regulators, various stakeholders, and consumers can be addressed by using simulation. Some example uses included: Rate structure analysis. Multiple differentiated energy products based on new rate structure offerings to consumers is very attractive to utilities because it creates the opportunity to reveal demand elasticity and gives utilities the ability to balance supplier market power in the wholesale markets. The challenge is designing rate structures that are both profitable and attractive to consumers. The simulator provides the ability to model consumer choice behavior in response to multiple rate offerings (including fixed rates, demand rates, time-of-day rates, and real-time rates) to determine whether a suite of rate offerings is likely to succeed. Distributed resources. The advent of new DER technologies, such as on-site distributed generation, building combined heat and power systems, and grid- responsive appliance controls creates a number of technology opportunities and challenges. The simulator can be configured to evaluate the cost/benefit trade-off between infrastructure expansion investments and distributed resources investments by including the other economic benefits of DER (e.g., increase wholesale purchasing elasticity, improved reliability metrics, ancillary services products to sell in wholesale markets). Peak load management. Many peak-shaving programs and emergency curtailment programs have failed to deliver the expected benefits. The simulation environment can be calibrated to observe consumer behavior to understand its interaction with various peak shaving strategies. The impact of consumer satisfaction on the availability of peak-shaving 3 resources can be evaluated and a more accurate forecast of the true available resources can be determined. It can also evaluate the consumer rebound effect following one or more curtailment or load-shed events in a single day. Distribution automation design. The simulator offers capabilities that support the design and analysis of distribution automation technology, to allow utilities to offer differentiating levels of reliability within the same system by managing power closer to the point of use. III. EXISTING AND NEAR-TERM MODELING The following modules are incorporated in GridLAB-D or will soon be available. A. Power Flow Modeling The power flow component is separated into a distribution module and a transmission module. While the distribution systems are the primary focus of the tool, the transmission module is included so that the interactions between the two systems can be examined. When the integration of these two modules is complete it will be possible to examine how smart loads on the distribution system affect the transmission system. Traditionally the ability to examine interactions on this level has been limited by computational power; the use of multiple processors is how this is addressed in GridLAB-D [4]. B. End-Use Load Modeling Commercial and residential end-uses are implemented using the Equivalent Thermal Parameters (ETP) model [5]. An ETP model characterizes the heat transfers and dynamic thermal properties of a building using electronic component equivalents. These are differential models solved for both time as a function of state and state as a function of time. Currently supported residential end-uses are heat-pumps, resistance heating, electric hot water heaters, washer and dryers, cooking (range and microwave), electronic plugs and lights. Commercial loads are simulated using an aggregate multi-zone ETP model that will be enhanced with more detailed end-use behavior in coming versions. End-uses which are demand responsive require additional control code to implement the load curtailment strategies. The test case used to demonstrate this capability is based on the Olympic Peninsula Testbed Demonstration project [6], which was conducted by the US Department of Energy, Bonneville Power Administration, and a number of industry and academic collaborators. This project included a number of commercial and residential loads, which purchased electricity using either traditional fixed price, time-of-use with critical peak price (TOU/CPP), or real-time price contracts (RTP). The commercial and residential customers who used RTP required a bidding strategy that would allow them to choose the desired objective (economy versus comfort) and the setpoints for the different times of day, and days of week. This control strategy was implemented within the controller code and the simulation model must include the same control code: controllers observe the present conditions every five minutes and compute a bid price and quantity for the next five minutes. The utilitys distribution market is then cleared at the end of each five-minute interval and the clearing price is posted to all customers. Each device that bids at or above the clearing price is allowed to run, and all the devices that bid below are not allowed to run. The price response for air- conditioners was implemented by adjusting the setpoint temperature and the water heaters simply increase the probability of curtailment as the price went up. The simulators end-use models for the air-conditioners and water heater RTP use similar market bid/response control code as implemented in the Olympic Peninsula project. Controllers must be implemented to provide the additional capability that is not provided by the generic end-use model, as described above. A similar approach was used for both commercial building HVAC systems as well as the municipal water pumping loads. The details of these implementations are provided in the previously cited report. C. Market Systems The simulator implements a generic double auction for all market clearing processes. In addition, an owner object is used to keep journals for arbitrary categories of financial transactions. Devices submit price/quantity bids to buy or sell energy on the distribution system. Loads bid to purchase energy over the coming five-minute interval at a particular price. Loads that have non-synchronous generation bid the expected load at the price to dispatch the generating unit. All other loads use a 24-hour price history to determine the expectation price by evaluating the difference between the objective and current conditions and adjusting the bid up or down based on how well the objective is being met. Generator used a cost function to adjust their bid prices for the generators available capacity. The bulk supply is obtained at the wholesale price for the available capacity on the feeder under the present conditions. The complete supply and demand curves are assembled at five-minute intervals and the market is cleared, with the clearing price being published. The clearing quantity establishes the expected feeder load and any distribution generation dispatch request. Transaction logs are used to track energy costs for customers, distribution generators, and the utility. These logs are used to gather monthly, yearly, and life-cycle cost and revenues, as appropriate. IV. INTEGRATING OTHER SIMULATION MODULES At its core, GridLAB-D can determine the simultaneous state of millions of independent devices, each of which is described by multiple differential equations solved only locally for both state and time. The advantages of the 4 approach used over traditional finite difference-based simulators are: 1) greater accuracy; 2) handles widely disparate time scales, ranging from sub-seconds to many years; and 3) ease of integration with new modules and third- party systems. The advantage over traditional differential- based solvers is that it is not necessary to integrate all the device's behaviors into a single set of equations that must be solved. The main processing loop updates each object by passing to it a timestamp indicating the desired synchronization time. These synchronization calls attempt to advance each object's internal clock to the time indicated by the global clock, and if successful they returns the time of the next expected change in each objects states. An object state change is one which requires the equilibrium equations or conservation laws of the system to be re-evaluated. When an object's state changes, all the other objects in the simulator are given an opportunity to consider the implications of the change. They can update other objects properties and alter the time and state of their next change. The core continues calling objects, advancing the global clock when necessary, and continuing in this way until all objects indicate that no further state changes are expected. This is the equilibrium condition and the simulation consequently ends. A. Simulation Integration Requirements Because the simulator is an agent-based system, the simplest way to integrate other modules is to treat them as agents (types of which are called classes, and instances of which are called objects) within the simulation environment. An object must meet the following requirements to operate successfully. 1. Objects must have both a time and state function for every time-interval, i.e., it must be possible to solve the time until the next state change as well as the state at a particular future moment in time. For example, a water heater agent must always be able to compute how long until the heater coil turns on, as well as determine whether the heater coil is still on in 30 seconds. 2. Objects must be able to publish some or all of the properties that other objects depend on, as well as allow other objects to change certain other properties as required. For example, a transformer must be able to observe the voltage of both buses it is connected to. 3. Objects must be able to specific their dependencies on other objects. For example, a transformer must be able to determine its current injection from the two buses it is connected to. 4. Objects must make certain critical behaviors available to others. For example, the market agent must allow loads and generator to submit bids. It is possible to make large complex agents, such as an agent to solve powerflow using a Newton-Raphson method instead of the current forward-back sweep and Gauss-Seidel methods, although this has not yet been done. Furthermore, there is no necessity that GridLAB-D be the tool that determines how the global clock is advanced, meaning that external tools, or indeed a real-time clock can govern the simulation. The ability to handle highly heterogeneous models makes coordination of the data representation very critical. Naming conventions are currently employed, and the current suite of modules employs a relatively simple data model. Many of the interchange issues, such as unit conversion and data type conversion are handled automatically during import/export operations. However, the problem of maintaining consistent and compatible naming conventions, as well as standardizing the representation of classes of object remains a difficult one. B. Targeted Tools for Integration Two significant integration efforts are planned for the immediate future. These address two aspect of electric system operating that are not addressed by the basic suite of modules implemented early on: distribution system dynamics and wholesale markets/bulk power operations. The current plans are to begin integrating Electric Power Research Institute (EPRI) Distribution System Simulation (DSS) [7], which has been recently ported to an open-source implementation paving the way for wide-spread use of the tool, including integration with GridLAB-D. The second is a proposal by Iowa State University to integrate the AMES Wholesale Power Market Testbed [8]. In addition, an effort is being initiated to implement a Multispeak [9] compatibility module that would enable data exchange between the simulator and a number of commercial distribution system applications. Although this module would not enable direct integration at the level of the concurrent solution that GridLAB-D can provide, it does enable the exchange of models between other application packages and the simulator. V. TESTING AND VALIDATION Testing of the power system components of the tool is done in 2 steps; unit tests, and integrated tests. The unit tests are designed to ensure that the objects that represent individual components, such as a line section, are properly implementing their functions. This is especially important since the distribution system is modeled in accordance with [10], which provides a complete representation of the system. The second step in testing and validation utilizes the IEEE Distribution Test Feeders [11]. In particular the 4 node, 13 node, and 37 node test feeders were used. By using the IEEE Distribution Test Feeders we confirmed that both the solution method and implementation are functioning correctly in the simulator. Integrated validation is much more challenging to perform. For one thing, there are no other equivalent simulation environments that can be used for reference to compare results against. In addition, the complexity of the models and the huge volume of output make direct comparison of results very difficult and probably inconclusive. This requires that higher level comparisons be employed, such as statistical techniques and comparisons of integrated results, such as 5 energy use, load shape statistics, price and load duration curves, etc. These challenges have not be overcome to date, and we foresee a need for significant work to identify a more formal approach to validating simulation systems of the complexity of GridLAB-D VI. CONCLUSION We have shown how an agent-based simulation environment can be used to examine utility demand response programs at a level of detail that has historically been difficult to do. Various incentive programs and rate structures can be tested using detailed models of the distribution system, solved concurrently with detailed models of residential and commercial loads, distribution generation, distributed energy resources, and distribution automation. We have explained how these tools are built and integrated with GridLAB-Ds core solver, and how other tools might be integrated in the future. Planned activities include integration of the EPRI DSS distribution dynamics simulation and Iowa AMES wholesale market testbed system, as well as a general purpose data exchange module using Multispeak. VII. REFERENCES [1] G. Conzelmann, G. Boyd, V. Koritarov, T. Veselka, Multi-Agent Power Market Simulation Using EMCAS, IEEE PES GM, 2005. [2] L. Kiesling and D. P. Chassin, Beneficial Complexity: A field Experiment in Technology, Institutions, and Institutional Change in the Electric Power Industry, Workshop in Political Theory and Policy Analysis, Indiana University, Fall 2008 Colloquia: http://www.indiana.edu/~workshop/colloquia/papers/kiesling_paper.pdf [3] D. P. Chassin, K. Schneider, and C. Gerkensmeyer, GridLAB-D: An Open-Source Power Systems Modeling and Simulation environment, IEEE PES T&D Conf. and Expo., April 2008. [4] D. P. Chassin, P. R. Armstrong, D. G. Chavarria-Miranda, R. T. Guttromson, Gauss-Seidel Accelerated: Implementing Flow Solvers on Field Programmable Gate Arrays, IEEE PES GM, 2006. [5] R. G. Pratt and Z. T. Taylor, Development and Testing of an Equivalent Thermal Parameter Model of Commercial Buildings From Time-Series End-Use Data, Richland, WA: Pacific Northwest Laboratory, Apr. 1994. [6] D. P. Chassin, D. J. Hammerstrom, and J. G. DeSteese, The Pacific Northwest Demand Response Market Demonstration, IEEE PES GM, 2008. [7] T. E. McDermott, An Open Source Distribution System Simulator, IEEE PES GM, 2006. [8] S. Widergren, J. Sun, and L. Testfatsion, Market Design Test Environments, IEEE PES GM, 2006. [9] G. A. McNaugton and M. E. Gordon, Development of a Real-Time framework for Enterprise Integration NRECAs MultiSpeak2 Specification, IEEE Rural Electric Power Conference, 2004. [10] W. H. Kersting, 2nd ed., Distribution System Modeling and Analysis. Boca Raton: CRC Press, 2007. [11] IEEE Radial Test Feeders- Available through the IEEE Distribution System Analysis Subcommittee: http://ewh.ieee.org/soc/pes/dsacom/testfeeders.html VIII. BIOGRAPHIES David P. Chassin (M2003, SM05) is a staff scientist with the Energy Science and Technology Division at Pacific Northwest National Laboratory where he has worked since 1992. He was Vice-President of Development for Image Systems Technology from 1987 to 1992, where he pioneered a hybrid raster/vector computer aided design (CAD) technology called CAD Overlay TM . He has experience in the development of building energy simulation and diagnostic systems, leading the development of Softdesk Energy and DOEs Whole Building Diagnostician. His recent research focuses on emerging theories of complexity as they relate to high-performance simulation and modeling in building controls and power systems and is currently responsible for the design and development of DOEs GridLAB-D simulator. Steve Widergren (M1978, SM1992) contributes to new solutions for reliable operation of electric power systems. Common throughout his career is the application of information technology to power engineering problems including, simulation, control, and system integration. He is a principal engineer at Pacific Northwest National Laboratory and Administrator for the GridWise Architecture Council a group formed to enable interoperability of automated systems related to the electric system. Prior to joining the Laboratory, he engineered and managed energy management systems products for electric power operations and supported power system computer applications in both supplier and utilities sectors. Application areas include information modeling, SCADA systems, and power system reliability assessment tools. He received his BS and MS degrees in electrical engineering from the University of California, Berkeley. He is actively involved in the IEEE Power Engineering Society and participates in standards efforts that bridge power engineering with information technology.