( :OP' ri .~ llI
I F.-\( : P() \\t,: r S, 'tl' lIl ' PO\\Tr Ptllll C ()IlI H)1. Se() ul. KOI"l'.t.
I ~ I ""~I
EXPERT SYSTEMS IN ENERGY MANAGEMENT SYSTEMS B. F. Wollenberg CO l/l m l D a l a Co rpura lioll. PI\'IIltJ lllh . .\1 i lll ll'.\Ol a. [ 'SA
Abstract. The operation of power systems is largely a manual task by human operators using computerized Energy Management Systems. The principal design goal of Energy Management Systems today is how to present information to operators so they can make correct decisions in a timely manner. The use of knowledge base expert systems is proposed as one means to achieve this design goal. Examples of applications of expert systems in Energy Management Systems are discussed and techniques for more complete use of expen systems is proposed. Keywords. Energy Management System. Expert System. SCADA. Optimal Power Flow. Fault Diagnosis. Intellig~nt Alarm Processor. Restoration Assistant
INTRODUCTION On balance, the EMS is primarily a monitoring and prediction system since far more information enters the EMS than leaves. The only control outputs from the EMS are transferred to power plants in the form of set points for generators and to substations in the form of open/close commands for circuit breakers. Many of the results of analyses made on the monitored system data find their way to control actions via voice contact between operators in the control center and operators in power plants, substations and external systems .
This paper argues for the expansion of the role. of expert systems in energy management systems (see [11-). Throughout the world the operation of power systems is becoming harder due to increased complexity of the power systems themselves and the decreased margins that operators have to operate within. As a result operators are faced with more complex application programs as well as the need to respond faster in times of system emergencies. Expert systems provide one of the means to reduce the cognitive overload suffered by operators in these situations.
When viewed this way, the EMS is primarily a data presentation system with the human operators making most of the decisions based on the information presented to them. At present, the only closed loop control system is the automatic generation control system (AGC) whose function is to maintain system frequency. interchange with neighboring utility networks and to keep generation output at the most economic levels for the prevailing load demand. Virtually all of the other control outputs are made by manual action either through the supervisory control and data acquisition (SCADA) system or through voice command to another operator. An overview of the functions found in current state of the art EMS is given in Table 1.
WHAT IS AN ENERGY MANAGEMENT SYSTEM? In addition to answering the question "What is an Energy Management System?". this section will also try to answer the question "What is the most important aspect of EMS design?". An Energy Management System (EMS) is the operation and control center for an electric utility. In this paper I will assume that separate control systems exist at the power plants and distribution systems. The EMS is used for maintaining the balance between load demand and generation. scheduling generation economically. switching the high voltage transmission substations, maintaining the operating security of the transmission system, and aiding operations personnel in diagnosing problems.
Since the EMS is primarily a data presentation system for operators who will perform manual control actions. we can say that the most important aspect of EMS design is : "How to present information to human operators so they can make correct decisions in a timely manner".
A typical EMS gathers information from up to several hundred remote terminal units which are in turn monitoring tens of thousands of digital status points and analog measurement points. In addition. direct digital data links to neighboring utility EMS computers allow the EMS to monitor thousands of points in each of the neighboring systems. Along with the digital data links to remote terminal units and external computer systems, the operators of the system are in constant voice contact with power plant and substation operators as well as operators at neighboring control centers.
KNOWLEDGE ABOlJf POWER SYSTEM OPERATIONS Human operators use knowledge about power system operations to carry out their daily tasks. The knowledge possessed by operators is usually gained by long years of experience starting as apprentices or assistants to more experienced operators . Many operators have the ability to match or exceed the capability of the most elaborate programs designed to provide predictive states on the
H. F. \\'o ll e nhtTg-
power system. include:
Predictive programs. for example.
Forecasts of total load Schedules of the optimal start and stop times for generating units Estimates of the state of the electrical network based on measured data. Predictions of the electrical parameters of the transmission system after the loss of a transformer or line . The optimal way to load the transmission network so that voltage levels will not be exceeded. etc.
Supervisory Control of Breakers/Switches. Data Acquisition
Control Generating Fre quency. Tie Flows Economic Dispatch Optimum Generation Control Transaction Scheduling Maintain Schedule of Interchange
Network Applications State Estimator Security Analysis Optimal Power Flow
Build Real- Time Network Model Find Critical Outages Correct for out-of-limit conditions. minimize cost or losses
Scheduling Applications Unit Commitment Schedule Unit On/Off Status Load Forecasting Predict system Load in future week FueVHydro Scheduling Optimal Schedule of Thermal and Hydro plants Miscellaneous Billing Calculations Record Keeping
Calculating billing for interchange Maintain Historical Records
Table 1. EMS Functions
The problem with using only human experience in operating a power system is that each of the predictions can be made only after learning how the particular phenomena reacts over a long series of encounters. Predicting the load demand. for example. requires the human operator to gain a feel for the relationship between the weather. the time of year and the load. After many years of operating experience the operators can usually predict peak load demand with better accuracy than the best prediction programs. However. the load-weather relationship is fairly stable. This is not as true for predicting the voltage at a remote substation after the loss of a key transformer during a peak load period. Here the relationship is not as stable since the network can be switched to a large variety of configurations as dictated by the need to maintain equipment and deenergize faulty components. The design of the EMS also requires a great deal of knowledge about electric power system operations. The design of the application programs usually requires years of study and engineering experience to produce programs that work correctly. The design of the human interface to be used by the operators includes knowledge about the operation of the system and the requirements of the operators as they bring up displays and take control actions using keyboards and other manual entry devices. Therefore. we could say that the knowledge of a great many designers is constantly being used in the operation of a power system along with that of the operators (see Table 2).
Applications Operators. EngSoftware ineers. Software Designers
Application Expcrience. Formal Educa tion
EMS Hard Engineers. Hardware and Sys- ware and Software tem Software Designers
Hardware and Software Design Expcrience. Formal Education
Table 2. Knowledge in an EMS System
WHY HA YE A COMPUTER? One could ask the question at the start as to why we have all of this design and engineering knowledge in a computer system in the first place? After all. human operators with simple enunciators on the wall of a contrOl room were adequate for many years.
E:-;pl'r! S\st e lll s ill EIll'rg\
The main reason for the computer is not the speed of gathering infonnation, but the timely manner in which a computerized system can analyze the data and present it in a summary fonn to a human operator. Simply placing the results of tens of thousands of digital and analog telemetered points on a wall display does not help the operator in the analysis of a system problem. The infonnation must be checked against limits and abnonnal status position tables and even then simply displaying it overwhelms the operator's capacity to understand what is happening. Large amounts of analog data must be analyzed using large numeric intensive algorithms to make any use of it to the operators. The gathering of analog measurements from a transmission system is far more useful to an operator when it is passed through a state estimator than if it is simply presented as raw infonnation. The state estimator can detect and eliminate bad data as well as provide better estimates for all transmission network quantities. Using a model of the transmission system within the computer system allows us to perfonn many predictions that could not be performed by an operator using experience alone. Security analysis is one such prediction. In security analysis a model of the transmission system is built from the results of a state estimator and is then subjected to a series of calculations that simulate the outage of major transmission and generation components. With each calculation a check is made to see if the mode led outage causes any major overloads or voltage limit violations. Other examples of prediction calculations that must be perfonned by the computer systems are the Optimal Power Flow. Unit Commitment and Hydro and Fuel Scheduling. Table 3 gives an indication of the computer resources required for typical EMS functions. COMPtJrER DEPENDENCY An important point needs to be made regarding the fact that EMS operators are dependent on the computer systems to effectively run their power system. EMS installations are typically set up with redundant computers. communication networks. human interface devices. etc. so that the failure of one element will not bring the entire system to a halt. Further witness to the fact that computer systems are necessary to the operation of a modern power system are the recent specifications from utilities that call not only for redundant equipment at the utility's major EMS site. but also for a backup system at a remote site. Thus there is a perceived need to provide backup in the event of a catastrophic event such as a fU'e or flood that took out the entire EMS building. HOW CAN WE MAKE 1HE EMS EASIER TO USE? Since the use of knowledge is a pivotal issue in the effectiveness of an EMS, one very useful way to make the
\lallagl'Ill l' lIt S\stt'IllS
EMS easier to use is to make more effective use of lhat knowledge. Instead of relying on the experience of a single operator to solve every problem. the EMS should have the experience of many operators encoded into a knowledge base that can be effectively used by the most inexperienced operators. Further, the knowledge that gets encoded into application programs must be stored in a way that allows it to be accessed and changed easily without reprogramming the application. These objectives require a rethinking of the way software for an EMS is written and the way operators interact with lhat software.
Processing Resources Needed
Raw Input Data
Raw Input Data
Raw Input Data
Security Analysis Output of State . Estimator
Optimal Power Flow
Output of State Estimator
Study Data Base
Table 3. EMS Computer Resources Requirement
One could characterize present EMS software as groups of numerical calculations that are tied together with rigid control structures. The control structures reflect lhe knowledge that engineers and programmers have about power system operations. However. that knowledge is limited to very specific conditions and does not reflect the strategic thinking needed by operators. For example: in using many large application programs such as an Optimal Power Flow (OPF) the operator must provide the strategic thinking as to what objectives are to be met and then reflect those objectives through a large number of manual entries to the OPF data before the OPF executes. Similarly, the results of large applications are usually summarized very poorly and the operator must interpret the results from a mass of output data. The result has been an increase in the complexity of large applications with a corresponding decrease in their acceptance by the operations personnel. Instead, one should design the EMS so that the strategic objectives are entered by the operator and the knowledge of how to translate them into the application program data base is made automatically for the operator as part of the program start up. Similarly, the results of an
B. F. \\'olkllbc rg
application should be analyzed for the operatnr and presented in summary form automatically.
mE USE OF EXPERT SYSTEMS IN AN EMS The encoding of knowledge and using it to aid operators requires the incorporation of ex pen systems within the EMS to such a degree not presently conceived. Three different aspects of energy management system design will be dealt with (see Table 4):
REAL TIME SCADA APPUCAnONS The operation of a power system using a SCADA system is usually a very routine task until an emergency. When an emergency occurs. the operator is flooded with alarm messages and from them must make a very quick diagnosis of the problem and decide what actions to take. This wu referred to as the "diagnosis and decision " problem in . Developments that have been recently reponed  are an intelligent alarm processor. a fault diagnosis system and a restoration advisor. These are described briefly below:
1) Control and Sequencing Logic 2) Real Time SCADA applications
3) Large Application programs
Expert System Application
Control and Sequencing Logic
Execution Control over all application software.
Real-Time SCADA Application
Intelligent Alarm Processor Fault Diagnosis Restoration Assistant
Large Application Programs
Optimal Power Flow Unit Commitment FueVHydro Scheduling
Table 4. Usinl Expen Systems in an EMS
CONTROL AND SEQUENCING LOGIC The control and sequencing logic of most application programs will have to be changed so that an expert system can exert control over the application. Virtually any control logic in the energy management system represents the application of specific knowledge about power system operations. All such instances of program logic are candidates to be replaced by expert system logic. For example. most energy management systems require the presence of logic that runs a state estimator and security analysis on a regular clock interval with an override if switching actions have occurred within a few seconds of the start. Further. the operator can call for an execution of the state estimator and security analysis whenever desired. With an expert system controlling the state estimator and security analysis. actions can be taken based on what type of switching action has taken place. what the change in load level is since the last execution. etc.
INTELLIGENT ALARM PROCESSOR(IAP) This application is being developed to provide a way to filter and prioritize alarm messages. The lAP uses an expert system to provide a simple diagnosis of the particular power system events that are generating alarms and then provides the operator with a set of brief summary alarms sorted by priority and presented as a special display called the "most urgent display". FAULT DIAGNOSIS SYSTEM This application of an expert system is designed to use circuit breaker and relay information to attempt to find the exact location of a fault. It also tries to identify possible misoperating devices. When there is insufficient information for a clear identification. the expert system provides a ranked list of possible fault locations . RESTORAnON ASSIST ANT When part or all of the power system suffers a blackout it is up to the operator to restore the system to operation. The restoration of power must proceed as much as possible using plans that the power engineering personnel have studied and developed . However. when conditions on the system prevent following all details of a plan. the operator must devise alternate restoration steps. The restoration assistant provides the operator with an evaluation of which predefmed restoration plans are most applicable as well as advice on how to modify them if needed. These three expert systems are not yet tightly interconnected with each other. That is. they all use separate knowledge bases and separate network data bases and their output is strictly to the operator. there is no transfer of results from one expert system to another. In the future this situation must be changed and will lead to a full set of expert systems that can access the same network data base and pass their results to each other. The result will be a package of expert systems as shown in figure 1 where real time data is flIst processed by an intelligent alarm processor. then various diagnostic expert systems are run depending on the nature of the disturbance. and finally the restoration expert system is called to give the operator direction on how to restore service. Together. the expert systems will provide answers to the following questions: What happened (through the analysis of alarms)
Expt'rt S\ S! t' Ill S ill
Why did it happen (through diagnosis of problems)
Ell e q.(\
What action should be taken (through restoration analysis)
Raw Scada Input Data
Fault Diagnosis Expert System
\\all a g e lllt'll! S\"S! t' Ill S
Restoration Expert System
Other , Diagnostic' , Expert , , System '
,...... , ........., ,_ ........... ,.
Other , Diagnostic ' , Expert , , System '
3) A Solution Guider Agent The case serup agent will aid the operator in selecting the type of OPF run to be made (usually by selection of the objective function). selection of appropriate control actions given the state of the system and the objective of the srudy. and fmally. the selection of the constraints for this type of srudy. Note that the case serup agent requires two types of knowledge: power system operations knowledge and OPF algorithm knowledge. Therefore the knowledge base for this expert system will probably come from more than one expert in this field. Expertise will be drawn from both system operations personnel and from OPF programming engineers. The solution analysis agent will provide the operator with advice on how well the original objectives were met and any possible corrections to the case serup to obtain a better solution. It will provide an analysis of the control actions and the resulting constrained values and will advise if proper control actions can indeed be made or constraints be brought within limit The solution guider agent will try to reassign priorities to control variables during the OPF solution so that a solution is attained when the OPF runs into difficulty. This expert system will replace many of the hard coded control variable priority selection schemes and other heuristic techniques to make the OPF a robust program.
Figure 1. Real-Time SCADA Expert Systems
Solution Guider LARGE APPliCATION PROGRAMS Experience has shown that the use of large application programs in a power system operations center often lead to frustration on the part of the operators. This is because the operator is not an expert in the use of the program or in the theory behind its implementation and therefore cannot make use of it properly. In addition. cases often arise where the operator must manually adjust the output of a program because of the inability of the program to model some of the more important constraints or control restrictions that must be observed in the actual operation. An example of this phenomena is the use of an Optimal Power Flow (OPF) in an operations center. Problems are often encountered when the operator fails to properly set up the input data for the OPF. or fails to properly interpret the output. or the OPF algorithm fails to converge to a satisfactory solution.
OPF Solution ........ Results Algorithm
Solution Analysis Agent
To make proper use of the OPF an expert system has been proposed which has three parts (see figure 2): I) A Case Serup Agent Figure 2. Optimal Power Flow Expert Systems 2) A Solution Analysis Agent
B. F. Wolle nhe rg
Power Engineering Society Winter Meeting, New York.
HOW FAR SHOUlD EXPERT SYSTEMS-GO? It is interesting to ponder how far this process can be carried. Will the entire operation of the power system be turned over to the computer? Without doubt most of us would instantly say no. However, some reflection may force us to see this in a different way.
The author's company, together with Northern States Power Company (NSP) of Minneapolis, MN have been jointly developing an Intelligent Alarm Processor. NSP will use it to analyze low voltage buses as well as line section switching. In both situations, if the logic of the lAP detects certain conditions, the expert system within the lAP will send control action requests directly to the SCADA system to take appropriate switching actions. Thus, the expert system will become a closed loop system for some well deflJled cases and will take action without the operator being in the loop. Obviously, this is a far cry from turning over the entire operation of the system to an expert system. but it has shown that such a step makes sense in this case. The evolutionary growth of expert systems within energy management systems will constantly move EMS technology in this direction.
Proceedings of the Symposium on EXPERT SYSTEMS APPUCATIONTO POWER SYSTEMS, Stockholm-Helsinki, August 22-26, 1988
Wollenberg,B.F. and Sakagauchi.T. "Artificial Intelligence in Power System Operations". ~ Proceedings. Vol 75 No 12. December 1987. pp 1678-1685
An International Survey of the Present Status and the Perspective of Expert Systems on the Power System Analysis and Techniques CIGRE SC38, WG 38 02, Task Force No 7.
Liu, C. and Darnborg, M. Development of Expert Systems as On-Line Operating Ajds Final Report, Electric Power Research Institute, RP 1999-9.
Shulte, R.P. et aI, "Artificial Intelligence Solutions to Power System Operating Problems" IEEE Transactions on Power Apparatus and Systems. Vol PWRS-2, No 4, November 1987, pp 920-926.
Talukdar, S.N. and Cardozo, E., "Artificial Intelligence Techniques for Power System Operations", Report EL-4323, Electric Power Research Institute, Jan 1986
Zhang,Z.Z., Hope, G.S. and Malik, O.P., ~ Systems in Electric Systems - A Bibliographical ~, Paper 89 WM 212-2-PWRS, 1989 IEEE
Kirschen,D.S. et al. "Artificial Intelligence Applications in an Energy Management System Environment", Proceedings of the Expert Systems Application to Power Systems Conference. Paper 17-1. Stockholm-Helsinki. August 22-26. 1988.