2009 IEEE RealTime

1 Real-Time Power Electric System Modeling, Assessment and Reliability Prediction Danling Cheng, Yishan Liang, Dan Zhu, Robert P. Broadwater Abstract-- Given a large and complex plant to operate, a realtime understanding of the networks and their situational reliability is important to operational decision support. This paper introduces using a unified model to help operators to minimize real-time problems, and also provide a mimic-real time play-back mechanism for system planners on how to upgrade the grid. Real-time simulation architecture is described to identify and predict where problems may occur, how serious they may be, and what is the possible root cause. Multi-views of visualization are plotted for the system area in order to obtain good situational awareness of the scenario. Index Terms-- Real-time simulation, Integrated System Model, Monte Carlo simulation, Contingency Analysis, Reliability Prediction, Visualization, Situational Awareness, Distribution System I. INTRODUCTION W ith load growth, aging systems, increased costs, and deregulation, public utility systems are experiencing increased stress. In order to continue to provide reliable service at affordable costs utilities must look to developing more intelligence in the design and operation of systems. Utility systems may contain tens of thousands of transformers and thousands of miles of cables/lines. Given such a large and complex plant to operate, a real-time understanding of the networks and their situational reliability is important to operational decision support. Conventional reliability evaluation theory reflects the longterm average reliability level of an electric system. In realtime operations, analyzing the up-to-minute reliability level and predicting what might be wrong in the near future is important. Not only could this help operators to minimize problems, but it would also provide useful information to system planners on how to upgrade the grid. Historically in utility engineering, separate environments are used for real-time operations, system planning, and reliability assessment. In other words, everybody accesses the same electric system but with different models. The different naming conventions of the same device and data structures Danling Cheng, Yishan Liang, and Robert P. Broadwater are with Department of Electrical and Computer Engineering, Virginia Tech, Blacksburg, VA 24060 USA (e-mail: dcheng@vt.edu, yshliang@vt.edu,dew@vt.edu). Dan Zhu is with Electrical Distribution Design, Blacksburg, VA 24060 USA (e-mail: dan-zhu@edd-us.com). 978-1-4244-3811-2/09/$25.00 ©2009 IEEE make it expensive and error-prone to maintain these models [1]. With such fragmented models it is difficult to keep a consistent view of the system in order to maintain situational awareness. Furthermore, modeling management needs emerge when more participants attempt to mimic real-time plant-wide simulation environments with models that provide a consistent view of the system. This paper describes an approach to electric power system real-time monitoring, system situation assessment, and reliability prediction. The simulation system is a collection of network topology, on-line measurement, historical data, event storage, analysis and prediction that is assembled to achieve the following objectives: • • • • • Provide a consistent system model to operator, system planner, and reliability engineer Insure that the model integrity is consistent with realtime SCADA measurements Identify and predict where problems may occur, how serious they may be, and what are the possible root causes. Provide visualization for enhancing the situational awareness of system operators Provide a historical scenario reproduction mechanism to the system planners II. INTEGRATED SYSTEM MODEL BASED ON REAL-TIME CONDITIONS A. Unified Model The challenges in distribution network integrity and real-time operation spotlight the need of using an Integrated System Model (ISM) to perform system-wide analysis. The discrepancies among conventional separate models of different working groups has provided only a few circumstances in which the planning and real-time studies have provided the same solution [1]. The ISM model described in this paper is a unified model for design, planning, operation, and control [2]. In the power systems modeled in this paper, one model integrates network, looped, and radial systems. Various algorithms operating on the ISM model exist in a distributed processing environment in order to achieve the desired real-time monitoring and analysis response. 2 The ISM model is designed to handle system constraints in real time. The traditional system-wide analysis uses matrices to model large systems. These equations involve all levels of information from the laws of physics, to topology connections, and system level information such as the loading level. Handling real-time computations requires much coordination to track changing system conditions. To facilitate faster computation, the ISM model avoids a global view or tight-coupling in the model by employing the generic analysis paradigm with topology iterators [3]. The graph of a power system model is stored in a container [4], and real-time data are attached to the graph on-the-fly. The system components, topologies, and analyzing algorithms are then integrated into a flexible, extensible software framework [5]. Algorithms use the topology iterators offered by the graph to perform analysis calculations. Using iterators for maintaining topology and implementing algorithms with iterators makes the ISM model naturally structured for distributed processing. The ISM model discussed in this paper is different from the Common Information Model (CIM) [6]. The ISM uses an edge-edge model for topology, where as CIM uses a nodeedge model. CIM is a data exchange modeling standard that has been primarily used with transmission system models. Some recent research illustrates a trend to apply CIM to distribution systems [7]. According to this work, additional connectivity nodes need to be included in the CIM model. The ISM model is an object oriented model based on graph theory. Topology changes in the model are handled locally because only topology iterators of components directly connected to the topology change need to be updated [4]. Because topology changes are handled locally, the graph trace (GTA) based reliability analysis solution times are not significantly affected by topology changes. The topology iterators of GTA provide a straightforward means of computationally handling dependent failure rates and cascading failures. These characteristics of iterators are important to the reliability analysis simulation considered here. In section II an example will be provided of shipping an ISM circuit model object from one processor to another. B. Real-Time Field and Operating Conditions Real-time field and operating conditions have a strong impact on system reliability. Adverse weather, heavy sustained loading, environmental conditions such as moisture level, a component status such as being repaired or not - all these contribute to the failure probability of system components. System reliability may deteriorate greatly in some special field conditions or operating modes. The component failure rate model used here includes seven parameters to determine failure rate FR and is expressed as FR = f(Component Type, Age, Insulation Type, Environment, Operating Condition, Weather, Loading ) (1) The component layer [5] in the ISM model is responsible for the “get/set functionality” of the reliability parameters. III. REAL-TIME SYSTEM ASSESSMENT AND RELIABILITY PREDICTION A. Simulator Architecture The structure and the facilities of the simulator are shown in Fig. 1. A blade computer is used for distributed computation, where different processors run an instance of different algorithms such as Monte Carlo or contingency analysis, on the same circuit model in different states. The real-time simulation environment is seeking to identify the next event that will result in loss of service, where services may be prioritized as critical or non-critical. The Data Acquisition process collects network topology, measurement, load, and weather from data-base servers and the Internet. The Model Validation process ensures the data used for system analysis is as accurate and up-to-date as possible. It detects topology changes, diagnoses whether any SCADA measurement readings are inaccurate or missing, and calibrates model loads to be consistent with the SCADA measurements. The validated circuit model is then shipped to a circuit queue to wait further processing. The Monte Carlo process obtains the forecasted weather online, and uses historical load patterns to forecast hourly loads for the next 24 hours. If contingencies already exist in the system, the circuit is directly shipped to the next queue to launch a fast assessment of the contingency. Otherwise, a sequential simulation is performed to predict any contingency that may occur by utilizing the weather forecast and load data. A contingency case circuit model created by Monte Carlo is shipped to the next circuit queue to wait processing by contingency analysis. A set of contingency analysis processes is used to perform the contingency analysis. Each contingency analysis process extracts a circuit at a time from the circuit queue and uses power flow to perform contingency analysis. Alerts concerning severe, potential contingencies are provided to system operators. Time-stamped circuit models along with power flow results are stored for future reference. The Controller in Fig. 1 is used to co-ordinate the various processes. 3 Fig. 1. Real-Time Simulator Architecture B. Data and Information Flow In order to maintain good real-time situational awareness, an accurate model that represents the behavior of the system is a necessity. Fig. 2 shows how the real-time data is utilized in the simulation in order to evaluate model integrity and perform reliability calculations. SCADA measurements from the utility data bases are used to tune the model to agree with real-time measurements. The SCADA data includes network transformer and primary feeder loading measurements. Device settings and status are used to update topology. Bad measurements are discarded by comparing each measurement with its historical statistical experience and a classical outlier rejection test is employed [8]. Loads are calibrated based on measurements, such that the difference between power flow calculation results for the tuned model and the actual measurements are less than predefined limits. Unusual changes in SCADA measurements are given as alerts. Low voltages or overloads at a customer service point are reported as system violations. The timestamped, tuned circuit model is then available for reliability evaluation. The reliability assessment and prediction module utilizes the current and forecasted system operating conditions to evaluate and predict whether any customer outage could happen within the next 24 hours. A geographic visual display helps to identify outaged customers. Alerts are provided for serious outage situations such as cascading outages. The time series storage module stores SCADA measurements with their corresponding power flow calculation results. These data stores are used to update the model statistical experience and are also available to system planners for offline processing. Fig. 2. Real-time Data and Information Flow Overview C. Reliability Assessment and Prediction Determining when a contingency is likely to occur and predicting its impact on the system can help to minimize or prevent serious outages. As loading has a strong impact on system reliability, accurate forecasting of the load is very important. Previous studies [9]-[10] provide a way to perform customer class based load estimation and forecasting using load patterns. Load research statistics are applied to billing cycles and kWHr consumption to estimate kW and kVar values for a given hour, day type, and month. The estimation results are improved by including the weather impacts on loads. An hourly weather load scaling factor SFW [2] is calculated as SFW = f(Temperature, Humidity, Wind speed) (2) Table I shows contingency analysis predictions compared with field measurements for a major primary feeder failure of a downtown network. The results show that the amps predicted by the contingency analysis have an average error of approximately 5%. The Monte Carlo simulation process in Fig. 1 is used to determine the most credible contingencies that may occur within the next 24 hours based on forecasted weather and load conditions. A dynamic failure rate update algorithm is used to forecast the hourly component failure rates based on equation (1). The Monte Carlo simulation was validated against a small sample of historical failure rate field data. For the system under study, Table 1 shows a comparison between feeder contingencies predicted by a 10000 year sequential Monte Carlo simulation and ten years worth of field data. It should be noted that the available field data is a very small sample. Table II shows the results of the comparison, where the units used are frequency per year. Contingency level 1 corresponds to the failure of a single feeder; contingency level 2 corresponds to the concurrent failure of two feeders, and so forth. The simulated contingencies display the same trend as the field data. The comparison indicates that the simulation is reasonable. 4 TABLE I COMPARISON OF CONTINGENCY ANALYSIS RESULT WITH THE FIELD MEASUREMENTS Feeder ID 1 2 3 4 5 6 7 8 9 10 11 12 Average SCADA Measurement (Amps) 610 566 481 418 807 0 519 606 418 470 482 488 - Model Results (Amps) 628.57 612.11 492.90 459.53 801.47 0.00 552.24 654.36 453.24 489.31 498.88 515.20 - Diff (%) 3.04% 8.15% 2.47% 9.94% 0.69% 0.00% 6.40% 7.98% 8.43% 4.11% 3.50% 5.57% 5.02% confined to substations, loads, control and protective devices. The status of each device in the schematic view is synchronized to the original geographical view, such that any control actions taken in schematic view is reflected in the geographic view, and vice versa. Note that the schematic view shown in Fig. 4 is automatically built from the geographically view shown in Fig. 3. Fig. 3. System Geographic View TABLE II FEEDER CONTINGENCY FREQUENCY COMPARISON OF FIELD DATA AND ISM MODEL Contingency Level 1st 2nd 3rd 4th Field Data (failure/year) Simulation (failures/year) 8.1 1.0 0.1 0 12.43 0.781 0.044 0.002 Fig. 4. System Schematic View IV. REAL-TIME VISUALIZATION The real-time system assessment and reliability prediction involves tremendous data processing for the system which contains tens of thousands of components. The traditional energy management system lists the contingency results in text oriented tables [11]. It is easy for the user to lose concentration in mining important information out of enormous data. The challenge exists to deliver analysis results in a quick and intuitive manner. Here we describe an approach involving multi-view visualization, such that by using different but simple views, the operator can quickly understand the real-time situation, identifying the roots of problems, and arriving at analysis based control decisions. A. Geographical versus Schematic views A geographically based full-topology model as the one shown in Fig. 3 is built to provide an overview of the system. With the instant online alarm message described in Fig. 2, this view can help the operator quickly capturing the problematic area. In order to assist the prompt control of the large and complex system, a small but equivalent system can be built by discarding the unnecessarily complexity of the geographical model [12]-[13]. A schematic view of the system shown in Fig. 3 is illustrated in Fig. 4. It is an equivalent model that contains only the essential information needed to take effective corrective actions, which include but not necessarily B. Diagnostic and Satellite Views In order to make the contingency analysis results that used listed in tabular/text displays more intuitive, visualization has been improved in recent years on displaying the percentage of overloads/outages, such as using dynamically sized pie-charts [14]-[15]. However, for dangerous situations such as cascading failures, the number of components affected by the contingency could be excessive. It is not easy to discover the underlying problems in a quick manner by checking the pre/post contingency situation of every overloaded/outaged component one by one. In order to facilitate the system diagnosis in a quick view, the cascading failure is simulated for every predicted, dangerous situation. Limit violations are checked. Burned out cables and customers that lose power in cascading failures are identified. The severity of the contingency can be quickly evaluated by how wide the affected area is and how many services are impacted. An example of a cascading simulation result is shown in Fig. 5, where the green line denotes the origination of the cascading failure, and other colored lines denote how wide the impact is expected to be. Mental connections between the failed components and the contingency caused by the violation is easily seen and understood. Detail limit violation report is available for every component upon request. 5 [4] [5] [6] [7] *Black *Green *Red *Blue *Pink - Fig. 5. Cascading Failure of Burned Cable(s)* No Power Loss Power was Lost during 1st Level of Cascading Power was Lost during 2nd Level of Cascading Power was Lost during 3rd to 5th Level of Cascading Power was Lost during Cascading Levels higher than 5 Further planning can be performed by panning the problematic area into a satellite view. In Fig. 6, a sub-set of the system (red lines) is overlapped with the real-time satellite image. The phenomenon that causes the potential trouble can be examined in this way. For example, nearby trees may need to be trimmed. [8] [9] [10] [11] [12] [13] [14] Fig. 6. Satellite View of the Problematic Area [15] V. CONCLUSION This paper describes an approach of power system real-time assessment and reliability prediction with the usage of a unified model. The validated SCADA measurements are used to update model topology and calibrate the loads in the model. A Monte Carlo simulation is used to predict what are the most probable dangerous situations that might occur in the short term. The results of contingencies are displayed in different views to achieve better diagnosis, planning, and control. Providing multiple, synchronized views, each view supporting display of analysis results, it helps the operator to understand the up-to-minute situations. By replaying historical events in the ISM simulator, the system planner is provided with a “learn the system and improve it” planning environment. VI. ACKNOWLEDGMENT The authors gratefully acknowledge Consolidated Edison for providing data used in this study. VII. REFERENCES [1] [2] [3] S. Grijalva, "Integrating Real-Time Operations and Planning using SameFormat Power System Models," in Power Engineering Society General Meeting, 2007. IEEE, 2007, pp. 1-6. R. P. Broadwater, "DEW Theory - Generic Analysis with Topology Iterators for Design, Operation, and Real-Time Control of Large Scale, Reconfigurable Systems," in Course Note for Object Oriented Integrated System Analysis,Virginia Tech, 2007. R. P. Broadwater, J. C. Thompson, and T. E. McDermott, "Pointers and linked lists in electric power distribution circuit analysis," in Power Industry Computer Application Conference, 1991. Conference Proceedings, 1991, pp. 16-21. L. R. Feinauer, K. J. Russell, and R. P. Broadwater, "Graph Trace Analysis and Generic Algorithms for Interdependent Reconfigurable System Design and Control," Naval Engineers Journal, vol. 120, 2008. L. Fangxing and R. P. Broadwater, "Software framework concepts for power distribution system analysis," Power Systems, IEEE Transactions on, vol. 19, pp. 948-956, 2004. "http://www.dmtf.org/standards/cim/," p. Common Information Model. D. S. Popovic, E. Varga, and Z. Perlic, "Extension of the Common Information Model With a Catalog of Topologies," Power Systems, IEEE Transactions on, vol. 22, pp. 770-777, 2007. P. J. Rousseeuw and A. M. Leroy, Robust Regression and Outlier Detection: Wiley 1987. R. P. Broadwater, A. Sargent, A. Yarali, H. E. Shaalan, and J. Nazarko, "Estimating substation peaks from load research data," Power Delivery, IEEE Transactions on, vol. 12, pp. 451-456, 1997. A. Sargent, R. P. Broadwater, J. C. Thompson, and J. Nazarko, "Estimation of diversity and kWHR-to-peak-kW factors from load research data," Power Systems, IEEE Transactions on, vol. 9, pp. 14501456, 1994. S. Yan and T. J. Overbye, "Visualizations for power system contingency analysis data," Power Systems, IEEE Transactions on, vol. 19, pp. 18591866, 2004. T. J. Overbye, E. M. Rantanen, and S. Judd, "Electric power control center visualization using Geographic Data Views," in Bulk Power System Dynamics and Control - VII. Revitalizing Operational Reliability, 2007 iREP Symposium, 2007, pp. 1-8. S. Grijalva, "Visualization and study mode architectures for real-time power system control," in Industrial Electronics and Control Applications, 2005. ICIECA 2005. International Conference on, 2005, p. 8 pp. T. J. Overbye, "Visualization enhancements for power system situational assessment," in Power and Energy Society General Meeting Conversion and Delivery of Electrical Energy in the 21st Century, 2008 IEEE, 2008, pp. 1-4. M. J. Laufenberg, "Visualization approaches integrating real-time market data," in Power Systems Conference and Exposition, 2004. IEEE PES, 2004, pp. 1550-1555 vol.3. 6 Biographies Danling Cheng received the B.S.E.E. from Huazhong University of Science and Technology, Wuhan, China in 1996, and her M.S.E.E. from Virginia Polytechnic Institute and State University (Virginia Tech), Blacksburg, VA, in 2002. Currently, she is a Ph.D. candidate in the Department of Electrical and Computer Engineering at Virginia Tech. She was an instructor in Wuhan University of Technology, Wuhan, China, from 1996 to 2000. Her research interest includes power system reliability improvement, risk evaluation, real time system diagnosis and analysis. Yishan Liang received her B.S. degree in electrical engineering from Hebei University of Technology (Tianjin, China) in 1999. She worked as an Assistant Electrical Design Engineer in Tianjin Chemical Engineering Designing Institute (Tianjin, China) from 1999 to 2001. Ms. Liang received her M.S. degree from the Department of Electrical and Computer Engineering at Virginia Tech and is presently a Ph.D. student there. Her research interests include computer applications in power systems, transformer motoring, diagnosis and analysis. Dan Zhu received the bachelor degree in communication engineering from South China Normal University, China in 2000. She received her master’s degree in 2003 and doctoral degree in 2007 in electrical engineering from Virginia Tech, Blacksburg, VA. She was with the Electric Power Bureau, Jiangmen, China as an assistant engineer. She currently works in Electrical Distribution Design, Inc. Blacksburg, VA. Her research interest includes power system reliability improvement. Robert P. Broadwater received the bachelor’s degree in 1971, the master’s in 1974 and the doctoral degrees in 1977 in electrical engineering from Virginia Polytechnic Institute and State University (Virginia Tech), Blacksburg, VA. He is currently a Professor of Electrical Engineering at Virginia Tech. He develops software for analysis, design, operation, and real time control of physical systems. His research interests are object-oriented analysis and design and computer-aided engineering.