Today’s Grid Relies on Network Analysis (which relies on a lot of data) Jay Britton, Pat Brown, John Moseley, Milos Bunda Introduction Yesterday’s power grid was overbuilt to achieve reliability. Today’s grid depends on automation to operate at tighter margins and under a wider range of supply and demand patterns – with network analysis as a cornerstone of this automation. The future is clearly toward more and more critical business processes relying on network analysis to operate closer to system limits. The data management that is required to support these network analysis processes is very challenging and, as currently implemented, requires too much valuable senior engineering time to assure necessary accuracy. As a consequence, IEC TC57 WG13, along with EPRI, ENTSO-E and various other parties, have been working for some time on a strategy for improved information management for network analysis. The result is a set of revolutionary CIM-based standards aimed at improved model quality, reduced net labor requirements and a systematization of data that can support the ever-smarter grid of the future. This article summarizes the CIM strategy for integrated network model management. A power grid is an enormous, complex, interconnected machine which we all depend on to deliver power reliably, while at the same time societal values and concerns drive policies such as deregulation, open markets, energy diversity, carbon reduction, emission control and the like. Implementation of these policies requires sophisticated automation, and an increasingly critical part of this automation is based on mathematical modeling of the grid as a whole system. This holistic analysis is necessary because a power grid is made up of millions of parts that are bound together in instantaneous interdependence by the physics of electricity, and no decision can be taken about one part without some impact on the rest. The business processes that analyse the grid are complex because of the necessity of simulating the physics of electricity on a very large scale, and because ownership and operating responsibility of the grid is distributed among many participating grid entities. The eastern interconnection in North America, for example, stretches from the Atlantic coast to the Rocky Mountains and is made up of grid parts owned by hundreds of different utilities, plus participating generators and consumers of many types. It is divided into a hierarchical arrangement of planning and operating entities with a variety of different responsibilities. In carrying out these responsibilities, each entity must assemble accurate models tailored to a variety of specific purposes. Each model is huge – millions of individual datum must be accurately assembled in order for each analysis to be accurate. Much of the analysis is ‘real-time’, meaning that the models of the present system operation must be assembled and solved within tight time limits. The size of models coupled with changing power grid conditions would alone make the maintenance of accurate models a significant challenge, but in addition, the data required to produce models for any given analysis always come from multiple organizational sources. Behind each analysis, therefore, is a complex information management process that involves participation of many entities in order to maintain the model required for each particular purpose. Because of the number of different departments and corporate entities involved, standards for exchange of network information are necessary, and developing these standards has been a focus of IEC CIM work for some time. But interoperability requirements go way beyond the ability to exchange and process someone else’s data. Most current modelling processes have far too many manual steps – too many points of data entry, too many manual checks, too many partially automated data transformations. A CIM central objective from its inception has been to reduce the amount of time senior power system engineers (a scarce commodity!) need to spend doing data management that could be automated. The key to reducing labour (while simultaneously increasing network model quality) is to assure that the parts of the overall information that come from different sources will fit together without manual intervention and can be assembled into all the different kinds of models that are needed. Metaphorically, what practitioners need is a versatile and reliable building block data architecture, and this is what CIM has developed in addition to a format for exchanging the building blocks. The CIM approach to network models is to design modular data such that each logical part of the overall data can be supplied from its natural source and can be reliably plugged together with parts from other sources to make all of the different kinds of analytical models required in today’s grid. This article outlines the architecture behind the CIM network building blocks and reports on the current state of its development and deployment. Problem Description Broadly speaking, the CIM goal for network modeling is to define network analysis data in terms of building blocks – let’s call them ‘model parts’ – which satisfy the following two requirements: 1. Each model part is designed to fit into a ‘model framework’ that allows model parts to be ‘composed’ to form the required input data for all of the kinds of network analysis that are required. 2. Each model part has one natural source that could be assigned responsibility for the quality and availability of the model part. Call this party the ‘model authority’. Requirement #1 expresses the basic goal – all types of cases can be constructed straightforwardly by putting together the right selection of building blocks. Requirement #2 is basic information management wisdom. It is always desirable to have a single well-known original source for each datum. This reduces the chance of conflicting data and provides natural points of quality control. (Note: this does not mean that all data comes from one source – it only means that each datum has a single source.) Satisfying requirement #2 requires an analysis of where the input data for a typical network analysis case comes from. Over the lifetime of a given network component, it will be represented in many thousands of analytical cases. CIM started with a simple question: across all these cases, why would a given component ever be represented in a different way? In current practice, where different models are maintained by different parties, this is commonplace. The result, however, is more engineering labour, greater difficulty in validating data, and difficulty in comparing results generated by different parties. This observation leads to two more key requirements: 3. In designing model parts, the aspects of models that should be invariant over all the cases should be separated from those that change. a. These invariant aspects correspond to the characteristics that are inherent in the system ‘as constructed’ and which don’t change except by new construction events. b. The remaining variant aspects make up the hypothesis which every study has about the particular operating condition for the grid that is being studied. 4. Component identification must be consistent across all studies, forming a reliable basis for comparing the content of analytical results and tracing all data back to original sources. The largest number of network cases study the present state of the grid. Many operating studies, though, look at future conditions ranging from an hour ahead out to a year ahead, and in planning contexts, studies may project as much as ten years into the future. Studies of past conditions are also sometimes required. These differing time frames gives us another key requirement: 5. Model parts must be able to represent the changes in the network over time. Different studies cover different parts of the grid. No study ever analyses the entire grid in detail – each one has a focus and models that part in depth while it simplifies its view of the rest of the grid. To take two extremes of this as examples, a bulk power analysis might represent all high voltage lines and large generating plants, but simplify all sub-transmission and distribution to net substation ‘load’, while at the other end of the voltage scale a distribution feeder analysis might represent every customer transformer, but simplify the transmission to a single power source. Different parts of the grid are owned by or under the responsibility of different parties, and these parties are usually the logical source for data about their area of responsibility. There is no party that could logically be assigned responsibility for all of the grid. Hence: 6. The design of model parts must allow partitioning of like data according to its logical model authority. The most common divisions are regional (by ownership or operating responsibility) and electrical (as in transmission vs sub-transmission vs distribution). There are many kinds of studies being run. They are differentiated primarily by the way that the initial case is assembled. Some aspects of setup are always similar – for example, selecting the parts of the network that must be represented and selecting the time frame being represented. Others may be very specialized – such as setting up a generation pattern based on outcomes of a specific market. In all cases, though, the CIM goal is to enable automation that reduces the need for manual babysitting of the process. 7. The design of model parts must enable creation of appropriate automated procedures for each different network analysis business process. Finally, no significant revision of network modelling practice could possibly be accomplished all at once. Any practical plan for change will have to involve an evolution from current state, which leads to a final requirement: 8. The design of model parts in the future vision for network model management must also support present practice and any intermediate stages required for migration to the future vision. In sum, CIM pursues a revolutionary vision: a framework of non-overlapping plug-together model parts representing the grid and its operating characteristics through all voltages levels, across all participants, over all time frames and serving all analytical requirements. The CIM Building Block Network Data Architecture This section provides highlights of the CIM approach to network model management. (An overview is all that space allows here; for greater depth see the references cited at the end of the article.) Model Part Basics A CIM model part is a set of CIM data objects with these minimum characteristics: The model part objects are governed by a subset of the CIM information model for power systems operations. (The specific information model subset defines the model part’s ‘type’.) The model part object instances together satisfy some holistic purpose, such as describing ‘the equipment that makes up the bulk power grid of Belgium’, or ‘the schematic diagram objects for Airport substation’. The model part data comes from one source, such as ‘the Belgian TSO’ or …. This core definition of model part is loose enough so that, per requirement #8, today’s network analysis processes which are primarily based on exchange of complete power flow cases can be expressed as exchange of model parts. The CIM building block vision, however, is based on two other important and innovative aspects of CIM model parts: A model part may contain objects which define associations to objects in other model parts. These ‘dangling references’ are the essential mechanism for defining how the model part will connect with other model parts. A model part may also have an association to a model framework that defines the role of the model part – e.g. ‘my role is to cover the Belgian territory’ or “my role is to cover…”. In this vision, the two most recognizable characteristics of a building block model part are its type, which defines the kind of information it contains, and the portion of the grid that it represents. The grand idea is that model parts may be assembled in different ways to create all the necessary analytical models. Creating a Framework of Model Parts CIM encourages (but per requirement #8 does not require) entities participating in a grid to get together and formally define their areas of modelling responsibility so that each datum has a single source and each set of data from a given source fits together with each other set. This formal definition is called a ‘model framework’. SW N-NE S-C S-SW W-SW Central South NE E-NE E-C West N-C W-C W-NW North East E-SE S-SE NW N-NW The primary agreement on framework is made at the level of type ‘EQ’ (equipment) model parts which define what grid equipment is represented in a model. Figure 1 shows a very basic framework dividing a bulk power grid among 9 TSO/ISO participants. The region of modelling responsibility assigned to each TSO is called a ‘frame’ and the framework is completed by 12 ‘boundary’ parts, one for each place where two TSO’s have ties, and therefore have to agree about how their adjacent models fit together. SE Figure 1. An interconnection model partitioned among 9 TSOs/ISOs. Both frames and boundaries in a framework define roles that are filled by EQ model parts. Consider for a moment what is required for two TSOs to marry their models properly. For each tie, they have to sit down and agree about some minimal common bit of modelling that neither party will change without consulting the other. In fact, what this does is to create three sets of data – the part that A can change by itself; the part that B can change by itself; and the part can only change by A and B acting together. In CIM, each of these three roles are captured by an EQ model part. Roughly, it goes like this: A and B together define an EQ boundary model part containing the objects that make up their agreement on mutual modeling. A and B separately then define EQ frame model parts, each of which may be individually validated as adhering to the agreement between A and B simply by verifying that their dangling references all properly terminate in the boundary model part. In this scheme, the fact that validation only requires testing against the immediate boundary model parts for the frame is very important, because it means that a set of consistent model parts can be established without ever assembling the whole framework in one place. CIM exchanges do not require that this sort of partitioning occur by TSO, or by any other specific criteria. In fact, the expectation is that partitioning will ultimately take place in layers, as is illustrated in Figure 2, in which the model parts of the top blue layer are composed from a purple layer framework, and model parts of the purple layer are composed from a further red layer framework. In most cases, given the nature of electric grids, the top layers would tend to be the higher voltages and the lower layers would be representations of lower voltage parts of the grid, but CIM only provides the partitioning mechanism – each group of collaborating model sources may tailor this layering to their own purposes. Figure 2. Layered Partitioning of a Model Framework TSOs will tend to be a logical choice at some layer of transmission partitioning, but it often makes sense to partition within a TSO based on a criteria of separating the bulk power model (which interconnection studies would be interested in) from local detail (which is only of interest in the TSO’s own studies). The local interest building block is used only in certain studies and it is convenient to separate it, in order to avoid having to extract it or equivalence it over and over again for interconnection usage. In a large interconnection, there might also be a layer above TSOs, representing reliability authorities, planning authorities, or markets that have model responsibility. Partitioning down the voltage levels may be continued into the distribution system and even into customer premises, providing a means to support completely seamless Transmission to Distribution modelling. This is an important feature for the future, where the clear trend is that transmission and distribution are less separable from one another. Seamless modelling would enable, for example, a straightforward method of processing distribution models to get an accurate real-time picture of what part of a transmission ‘load’ was wind, or solar, or under a demand constraint, etc. One final wrinkle deserves mention. Engineers currently often depend on ‘equivalents’ to create models of appropriate size for a given analysis. Equivalents may imply anything from manual simplification to mathematically derived models. Equivalents are also troublesome, as a rule. They take time to build and test, and they always pose an accuracy risk compared to models with greater fidelity. The first CIM idea is that, through creating the right framework of parts, models can be built up with minimal waste by selecting the model parts that should be in view, and most equivalents will then be unnecessary. But the second CIM idea is that when equivalents are necessary, they are created as the output of equivalent analysis in the form of CIM model parts which fit into the same overall framework. Partitioning Data by Types of Model Parts Model parts are also modularized by the kind of data they contain. (There is a standard set of CIM model part types defined in the IEC 61970-4xx series of documents.) These different types are defined primarily on the basis that the information each type contains typically comes from a different source or a different stage in a data development processes. Broadly, there are two major categories of types, corresponding to variant and invariant data identified in requirement #3. These are illustrated in Figure 3. SC: Short Circuit DY: Dynamics DL: Diagram Layout EQ: Equipment Invariant Model Part Types SV: State Variables TP: Topology SSH: Steady-State Hypothesis Variant Model Part Types Figure 3. Partitioning of network data by types of model parts. Invariant Types Invariant types include the data that describes the qualities of the grid that are inherent in its construction and would not typically change except by new construction activity. (The term ‘invariant’ here emphasizes the goal that this information about a given element should be the same in every study that represents that element.) The foundation on which any model assembly rests is the set of Equipment (EQ) model parts that are used in defining the framework. Model parts of type EQ declare which grid components are part of the model. In other words, if a line, transformer, switch, busbar, generator, or other element is to be represented in a case, then a corresponding data object must be supplied by an EQ model part that is part of the case. These EQ data objects also describe how the grid elements are connected together electrically and give their basic steady-state characteristics. Other invariant model data is defined in model parts that reference EQ model parts, which means they have dangling associations that reference objects defined in EQ model parts. These include: Diagram layout types (DL) define how grid parts may be organized into schematic diagrams. Short circuit types (SC) define additional data required for short circuit analysis. Dynamic types (DY) define additional data required for dynamic analysis. Geographical layout types (GL) define additional data required to link the model with geographical data (GIS). In recommended practice, invariant model parts are maintained as part of business processes which reflect the result of new construction activity or which reflect the development of future planned changes. Their maintenance is typically independent of study case preparation. Thus an engineer setting up a study case might select and review the invariant model parts needed, but would not normally make any modifications to invariant model parts. Variant Types Variant types are those that the engineer will set up differently for different kinds of studies. (For example, a capacity planning study might look at a heavy load scenario while a real-time state estimator is studying current load conditions.) The variant model part types also plug into the EQ foundation model parts: Steady-state hypothesis (SSH) model parts define the input choices for power flow, consisting mainly of: o Component status o Generation and load values o Regulation targets o Limits Topology (TP) describes the bus-branch topology that results from eliminating switches from the model. State variables (SV) describes the steady-state solution. Maintaining Model Parts through Time Network cases need to be assembled covering views of the future. A future view of invariant model types is normally the present plus the set of new construction plans that should be in view for a particular study. The CIM approach is to use plans stored in an ‘incremental’ form – that is, stored as a set of changes to an EQ model part. Then any future state of an EQ model part is the latest version of the as-built model part plus the latest versions of each appropriate plan’s incremental changes. Variant data is normally set up differently. Set up processes differ for different study assumptions and are often customized to pull data from outage schedules, market outcomes, load forecasts, etc. Such set up processes will, however, always produce standard CIM variant model parts, which will then plug directly into the invariant EQ model part bases. Operations, Procedures and Audit Trails The CIM architecture facilitates the definition of generic operators on CIM datasets. For example, since all model parts follow a similar pattern, it is straightforward to define ‘composition’ as a generic operation that merges the model parts as a step in preparing a case. Similarly, ‘apply incremental’ can be an operation that adds a plan to a model part as a step in preparing a case. Even steps like ‘run my power flow’ or ‘run my network equivalent procedure’ can be expressed as operations on CIM model parts. This aspect of CIM work is still in development, but the result is expected to be a language that may be used in future tense for defining case assembly procedures and in the past tense for expressing a precise audit trail of the operations that produced a given network analysis case. This latter is an extremely important goal because today, one of the most manual chores for engineers is the painstaking process of assuring that exactly the right input has been used in producing a given case. Using CIM to Design Effective Business Processes Grid operation today employs some very complex analytical procedures. In Europe, for example, one procedure involves daily setup and execution of 24 hourly simulations of the European grid for the following day. Running these successfully, day in and day out, requires the ability to assemble cases involving many sources of data quickly and with low probability of mistakes. The CIM architecture that enables such automation has been summarized in the previous section. CIM generally does not prescribe either the processes that should be run (which vary considerably) or the design of processes. Its focus is on standardizing the agreements on exchanged data structures on which these processes depend. However, in setting those agreements, it has considered how to enable simple error-free composition of data parts from many sources, which is the hardest part. Despite its not being prescriptive about processes, the CIM work provides some very useful guidance toward managing network analysis processes. It is a striking result, when current practice is reviewed, that very complex processes have evolved based on little or no underlying data architecture. By contrast, the main CIM lesson is simply: establish a data architecture and build processes on it -- don’t design each network analysis procedure independently. Level 3: Network Model Business Processes Integrate sources and consumers with Network Model Management. Automate business processes based on Network Model Management. Level 2. Network Model Management Repositories for model parts. Facilities for managing model parts and assembling cases. Hub for CIM compliant exchange of data. Level 1. Data Management Responsibilities Organize modeling responsibility among cooperating entities. Define CIM framework agreements. Define case assembly procedures. Level 0: CIM Semantic Standards Shared definition of data structure and meaning. Standards for Model Part data exchange. Methodology for network model management. Figure 4. Four Layer Approach to Establishing a Network Model Data Architecture. The right approach can be viewed in four layers. CIM standards are at level 0. Everything else builds on CIM as the common language of network modelling. IEC standards provide the core, but it is important to understand that some business process goals may require extensions of the data model, and CIM supports this as well. At level 1, data management responsibility is organized among the participants in modelling processes. Participants decide which sources are responsible for what master data parts and establish framework agreements about how the data parts fit together. They also define participant roles in key business processes, such as ‘assembling future planning base cases’ or ‘updating the model of my territory at my ISO’. At level 2, the implementation work begins based on the standards of level 0 and the process agreements of level 1. The key mechanism to be organized is network model management functionality at each participant. Sharing of model parts among disparate analytical processes almost certainly requires some new data management capability which can manage and exchange model parts and build cases for a variety of purposes from model parts. Finally, at level 3, it becomes productive to invest in more sophisticated automation of the agreed upon business processes. In other words, levels 0, 1, 2 fully enable organized use of shared model parts, but within processes that are essentially managed manually. In some cases, this will be fine, but in others, automation will be desirable and is practical to achieve because of the groundwork laid by the lower levels. For more in-depth information, the reader is referred to the recently published EPRI report on this subject [1]. Of course, as was made clear in requirement #8, it is not possible to convert instantaneously from existing practice to this vision. Realistically, each utility entity will start with some basic changes and move forward step by step. Fortunately, CIM standards support evolution well. It is not necessary to complete one level (of Figure 2) before moving to the next. It is only necessary to complete the parts of lower levels that directly support a particular higher level goal. In fact, it is not advisable to try to implement a comprehensive stack all at once. It will normally be best to pick a core initial objective that is achievable, such as a network model manager function (at level 2) for a TSO as-built model. This will require some set of decisions at levels 0 and 1, and some planning around how a more complete stack is going to be evolved, but such an initial step is achievable and valuable by itself, and provides a key piece on which to build. ERCOT and CIM for Network Analysis On September 1, 2009, as part of a new node-based market implementation, the Electric Reliability Council of Texas (ERCOT) adopted the Network Model Management System (NMMS) – a first of its kind software solution for supporting power system data modeling processes. NMMS is based on CIM and supports all aspects of the model life-cycle including the data entry, validation, creation of power system models, testing and data finalization, and archival of model data changes. Nearly four and a half years of planning, designing, developing and integrating went into development of the NMMS. Its design is a response to both internal ERCOT goals for increased efficiency in its internal data management processes and preparation for the new challenges posed to the electric power industry in the area of power system data management. These latter include “smart-grid” related technologies, integration of phase-angle measurements into electric grid operations, additional market products like distributed resources, analysis methodologies for solar generation forecasting and management techniques for operating energy storage devices. With the broad scope of these challenges and the need to integrate grid operations systems and align them on a common modeling syntax for the electric power grid, ERCOT elected to pursue a design based on the IEC TC57 CIM. ERCOT’s approach leverages and extends the CIM. The result can be seen in Figure #: The NMMS repository of master network data is based on the CIM information model. Input to NMMS from member TSOs, generators and other sources is standardized on CIM. Most of this information comes in the form of incremental updates of network models. Output of network models from NMMS to ERCOT and member business systems is standardized on CIM. This establishes consistency of representation throughout these business systems which reduces potential for conflicting results and enables straightforward sharing of information between these systems. Figure #. The ERCOT NMMS based on CIM. The usage of CIM (as a master data repository and data exchange standard) provides NMMS a flexible design that allows for the rapid operational integration of new devices as necessary to support grid operations. The CIM allows for not only the extension of its schema (data structure definitions), but also provides a set of inter-operational data exchange standards that the data adaptors/consumers adhere to (these allow for the easy migration of data into target systems). This reduces organizational adoption time, and produces a level of modeling consistency and accuracy that the prevalent industry practices are incapable of, and in general leads to greater operational awareness of grid conditions and the therefore greater reliability of the overall power grid system. ENTSO-E and CIM for Network Analysis Historically, system operations for all TSOs in Europe relied primarily on bilateral contracts between adjacent TSOs. Based on these agreements, TSOs cooperated in a variety of procedures for system security, capacity calculation and outage planning. Since 2007, however, some Regional Security Cooperation Initiatives have been launched as a response to a large system disturbance in 2006. Due to the increased importance of coordinated business processes for capacity calculations, outage planning and grid security calculation, the need for common grid models (CGM) and for other merged grid scenarios has been defined [2]. Understanding this, the ENTSO-E System Operations Committee in April 2012 assigned Regional Group Central Europe (RGCE) Subgroup Network Models and Forecast Tools (SG NM&FT) to “suggest a procedure and responsibilities for delivering and updating the Common Grid Models with their importance for operations and capacity calculation”, and a Task Force was established for the preparation of a technical specification. In 2013, ENTSO-E approved Common Grid Model Exchange Standard (CGMES), which specifies European network analysis procedures layered on the IEC TC57 CIM international standards for network modelling, after carefully assessing the available European and international standards for their fitness to support the identified business requirements. Figure #. Context of the Common Grid Model The design which ENTSO-E will use is illustrated in Figure #. On the left hand side individual TSOs (41 within ENTSO-E) prepare their individual CIM-based grid models (IGM) for the required time horizons, in which they incorporate future projects, planned outages, load forecast, generation schedules (based on market models or on actual market results), seasonal limits and applicable set points. After passing a centralized quality gate, the net positions and flows on HVDC interconnectors are matched to target values, either resulting from market clearing (day ahead and intraday) or from an interchange matching algorithm that was developed as part of the project. The next step is assembling the submitted model parts and creating a power flow solution for the entire pan-European grid for each required point in time. The information that is exchanged for this CGM is expressed as CIM SV and TP model parts, which refer to the CIM EQ models already available on the European Operational Planning Data Environment (a system ENTSO-E is developing to realise the exchanges needed for above mentioned business processes). The solved system states can be used to initialize a power flow tool to the state of any cross section of the European power systems, enabling regional studies, examples of which are given on the right hand side of the picture. The modular design of the processes enable the European TSOs to satisfy business needs from network planning (long term processes analyzing competing projects) to operational planning minimizing the overhead in data exchange. Without CIM based exchange profiles this would not be feasible for a constellation of 41 TSOs covering 5 different synchronous areas. Conclusion Development of CIM standards for network analysis is continuing in WG13, but it has reached the point where the cost/benefit case for adoption is compelling. It can reduce cost. It can speed up complex processes. It can reduce the potential for errors. It can free valuable engineering time to do analysis rather than data management. Comprehensive multi-lateral adoption of CIM, such as was accomplished at ERCOT and by ENTSO-E in Europe, is where the largest payoff occurs. Of course, this requires collaboration of many parties and therefore is the most difficult position to achieve. EPRI, however, has also shown in its work how a utility may benefit from a unilateral adoption of CIM, concentrating on unifying the various case building activities within the utility’s operations and planning responsibilities. The latter, of course, is a stepping stone toward multi-lateral adoption, which becomes easier as individual participants in the grid adopt CIM unilaterally. For Further Reading [1] Using the Common Information Model for Network Analysis Data Management: A CIM Primer Series Guide.”, EPRI product ID 3002002587 [2] Operational Network Codes and Market Guidelines as a result of the 3rd Energy package of the European Union: http://networkcodes.entsoe.eu/ [3] … [4] … [5] … [6] … Biographies Jay Britton is with XXX, USA Pat Brown is with EPRI, USA John Moseley is with ERCOT, USA Milos Bunda is with TenneT TSO B.V., The Netherlands