Hybrid Discrete and Continuous Control for Power Systems
W. H. ESSELMAN
Abstract： The control of the electricity supply from generation to end-use has been
an engineering and mathematical challenge for many decades. The continuous
increase in power requirements, system inter connection sand technological options
prompted the search for discrete and continuous control systems. With the emergence
of competition and limits on infrastructure construction, new methods of system
control are needed to improve performance, to achieve minimum cost and desired
reliability and to meet environmental requirements. The purpose of this paper is to
describe a few electrical system control problems and introduce them to the Discrete
Event Control research community with the expectation that useful approaches and
valuable results will be developed and directed to the beneﬁt of electric power system
Keywords: intelligent controls, power systems, electric systems
Generation and delivery of an economic and reliable supply of electricity are the corner-stones of electric utility operation. Achieving these combined goals during a
period of increasing electricity use and competition represents a growing challenge.
Advanced control principles must be applied to meet that challenge. To this end, the
Electric Power Research Institute (EPRI) has maintained an interest in Intelligent
Control method logies, some of which have been successfully applied to non-electric
system problems, which could also address a number of the electricity supply
1.1. Intelligent Control Concepts
During the past decade, signiﬁcant strides in advanced controls and digital computer technology created an interest in the potential of Intelligent Control
technologies. The key question became: Can these advanced methodologies be used
to create Control Assistants (Figure 1.) that supplement the human operator’s
capability to select more economic operating conditions while ensuring timely
response to system faults. A resultant joint National Science Foundation
(NSF)-Electric Power Research Institute (EPRI) initiative was undertaken that
stimulated a series of projects on Intelligent Controls (National Science Foundation,
1992; in press). Research was recommended
on projects aimed at achieving a high degree of autonomy, reasoning under
uncertainty, performance in a goal seeking manner, decision making at a higher level
of abstraction, data fusion from a multitude of sensors and adaptation in a
In the context of power system control, an ―intelligent‖ control system was deﬁned as one that could:
Perform tasks as well or better than a human operator or remove the need for continuous human attention,
Provide assistance to the operator in the selection of optimum control solutions to complex economic and environmental issues,
Provide advisory information to the operator on optimum corrective actions during fault and upset conditions.
Within these broad deﬁnitions , studies were supported on neural networks, fuzzy
logic and genetic algorithms along with exploratory investigations of discrete event
systems. Some approaches to intelligent control that are being investigated by the
joint EPRI=NSF projects include:
Supervisory planning: with the plan either pre-stored, generated in real-time by automated reasoning, or learned from human examples;
Qualitative modeling: using expert systems, either rule-based or frame-based, and either completely or incompletely deterministic (for instance, using fuzzy logic);
Computational intelligence & machine learning: using neural networks,
genetic algorithms, and other evolutionary or reinforcement learning methods,
sometimes combined with mathematical optimization techniques.
Among their research goals are:
methods for learning and adaptation in heterogeneous systems,
methods for reasoning and planning,
analysis of interaction of multiple agents, both human and machine,
techniques for transformation of data into knowledge,
methods for rule generation and modiﬁcation,
automatic knowledge interpretation,
techniques for development of qualitative and quantitative models,
methods for autonomous process operation,
intelligent sensors and actuators, and
tools and techniques for ICS veriﬁcation and validation.
Some of the aspects of the control and operation of the electric utility systems
will be described with the hope of stimulating new ideas and directions for future
1.2. Discrete Events
The purpose of this paper is to review a number of utility system problems so that
Discrete Event Control experts are encouraged to conceive innovative methods to
address utility system control issues. A useful starting point is the deﬁnition of
Discrete Event Systems given by P. Ramadge and W. Wonham (1989):―A Discrete
Event System (DES) is deﬁned as a dynamic system that evolves in accordance with
abrupt occurrences, at possibly unknown, irregular intervals, of physical
events.‖―Abrupt occurrences at unknown intervals‖ characterize the operation of an
electric power system. Some such occurrences are initiated by the system operator as
they select transmission lines, alternate equipment or speciﬁc units for power
generation. Others—line faults or equipment failures—are mitigated by prompt
autonomous and operator controlled actions that select alternate transmission lines,
adjust frequency or increase operating capacity. Autonomous control of variables,
such as system frequency and generation levels, act in conjunction with operator
actions to adjust for changes. For most cases, no signiﬁcant disruption of the
electricity supply is apparent to the user.
Abrupt, irregular occurrences are frequently the result of natural phenomena such
asstorms, high winds, lightning and formation of ice on the power lines. But not all
such unexpected events are attributable to unusual natural phenomena. Daily power
loads in an area can also change rapidly as a result of such diverse events as increased
air temperature causing greater air conditioning loads, televised programs attracting
large audiences or the sudden increase of an industrial load.
Failures of one or more of the many constituent parts of the system place additional demands on both the economic operation and security of the system.
Shutdown of a power plant or loss of a transmission line is manifested by the sudden
change of the utility system power ﬂows. While the electrical system is designed to
accommodate or limit the effects of such events, maintaining and rapidly restoring
system security remains a major control objective. Every small percentage of the
cases result in cascading events and power outages for many users over large areas for
2.0. Utility Systems
Operation of the generation, transmission and distribution of the electric supply system is guided by the objectives of minimizing costs, achieving high reliability and
meeting environmental requirements. To accomplish these objectives, electric power
supply system operation is characterized by a combination of discrete and continuous
controls that respond to both human and automated actions. System protection
requires subsecond discrete responses to network or associated equipment faults.
System frequency, on the other hand, requires a continuous automated response to
maintain the limits prescribed by satisfactory system operation. Decision making by
human operators is guided by procedures developed after extensive computer analysis
and their knowledge of the system based on experience and intuition. These insights
are difﬁcult to match with an automated system.
2.1. Structure and Organization of Power System
From a functional viewpoint, electrical system control operations are traditionally hierarchical. System operations are divided geographically into control areas and
divided functionally into dispatching, generation, transmission and distribution
control as illustrated in Figure 2. The broadest elements of the power system subject
to control are power generating plants, transmission networks comprised of lines
operating at voltages above 115 kV, and distribution networks comprised of 4 kV
to 34.5 kV lines that deliver electricity to the customers. At each level, speciﬁc
functions are assigned that collectively provide an integrated supervisory control of
the total system. In the modern power system, the generators are interconnected
through the transmission network over large distances. Distributed control of this type
requires substantial communication capabilities among the levels of control to ensure
meeting the goals of economic and reliable operation.
Each control area is served by a local dispatcher who selects the most economical
generating units and network conﬁguration to supply daily power requirements within
his control area. Maintenance of frequency, voltage, power interchange between
control areas and system reliability are essential functions of these dispatching centers.
Control areas are served by switching stations that control the operation of speciﬁc
circuit breakers and equipment associated with the power system. With such control
stations distributed throughout the electrical system network, appropriate transmission
lines are selected to supply the electricity in the most efﬁcient and reliable manner.
Geographically the control areas, generating units and transmission lines are an
integral part of a larger network that extends over large sections of the North
American continent. Generators, which convert fuel derived energy to electricity,
produce the power to supply the load demands of the individual users. A dispatcher’s objective of producing low cost electricity is achieved by operating the most
economical generators on an essentially continuous basis. Other higher cost
generators are then operated for several hours a day to meet peak afternoon loads.
These operations are complicated by the fact that some units require several hours to
startup from a standby condition. Power plant operation of this type requires the
proper execution of many discrete control actions, attention to interlocks that disable
or enable further actions, proper sequencing of plant systems, attention to plant
temperature gradients and heat up rate for proper startup. Examples of the diverse
items requiring operator attention include the monitoring of the many plant
restrictions such as component rates of temperature change, feed water chemistry
conditions and other operating restraints.
In the modern power system all generators are interconnected through the transmission network. Electricity generated by each generator is transmitted according
to the network impedance to supply the system loads. Delivery to individual users,
shown in Figure 2, is usually controlled at the distribution level by a radial system not interconnected to other parts of the network. Development of distribution
automation and load management controls have been key research objectives for the
past few decades. Distribution systems are designed to supply a diversity of large user
and small individual loads. 2.2. Control Hierarchy Control actions and sensor
observations are broadened at each higher level control station so as to achieve system
operation that meets the combined objectives of low cost electricity and high system
These goals are usually achieved by supervisory systems executing numerous discrete control actions in coordination with a signiﬁcant number of continuous
feedback systems. Functions that require continuous attention or more complex
decisions are usually accomplished by automated feedback systems such as automatic
generation, frequency and voltage control. An illustration of discrete event systems is
the role of protective relays in the transmission network and in fossil and nuclear
power plant and their function in the security system.
2.3. Power System Operation
The following description of the operation of the utility system will be divided into two parts: Normal State and System Security Control aimed at system protection.
Power systems are considered to be in a secure and normal state when all of the user’s
demand for electricity is being supplied and the failure of a single piece of equipment
or a single line can be tolerated. Normal operation of such a complex and
geographical extensive power system, shown in Figure 3, requires many control
systems that perform their functions accurately and reliably.
But the test of system control comes when abnormal conditions occur such as a lightning stroke causing protective relay action to remove a line from service or
unanticipated power plant problems requiring rapid shut down of a generating unit.
The electrical transients caused by such events can affect major parts of the system. Control functions and actions taken by human operators must be carefully designed and orchestrated to ensure the safety of equipment and people during such events. Steps must be taken to ensure that the power system continues to function at least in part or is restored to service in the most expeditious secure way.
Extensive computer-based analyses are conducted to support the many facets of economic dispatch during normal operation and security assessment during more
vulnerable conditions. Information must be developed to provide proper parameter settings, system constraints and response requirements of both automated and human controls. These results are designed into the operating procedures and the limiting conditions become the basis of the protective system. Operator response to
contingency conditions are guided by extensive computer analysis of network
conﬁgurations and expected system response.
Protective strategies and restoration procedures are established for each postulated fault. Computer-based decision analyses to rank all the possible faults and the advisability of possible recovery actions requires more computer time than is available for a practical real-time response. The complexity of evaluating the security of a typical electrical system network can be judged by considering the many possible cases of single and multiple line faults. Evaluation of the probability of system instability or voltage collapse adds additional dimensions to this problem.
2.4. Automatic Generation Control
To accommodate smaller load changes, which do not require units to be taken on or off line, the turbine-generators can be automatically controlled to increase or decrease power outputs. As the generator loads vary, governor-controlled turbine steam valves provide the primary controls formaintaining system frequency.
Distributed controls at individual power plants act in response to system generated signals for increased or decreased turbine speed. Adjustments to the governor set points to raise or lower an output frequency are determined by the Automatic
Generation Control (AGC). These closed-loop continuous feedback controls
maintain the system frequency, area generation level and power interchange between control areas. Supervisory controls also maintain the system voltages while power factors are controlled by capacitive reactive loads distributed throughout the network.
Moment to moment levels of area generation and load requirements are matched by the AGC systems. The excess power generated in an area, called the Area Control Error(ACE), is transferred to the adjacent Control Area. These tie line loads between
are as must be adjusted to meet pre-described levels of power transfers between areas. If security conditions in one area were being challenged, power transfers would be adjusted to assist in supporting the troubled area. Control issues related to AGC and ACE operation are addressed in a paper by R. King and R. Luck (1993).
As the number of active transactions increase, energy accounting becomes more demanding as AGC controls must be supplied with the scheduled transactions. An
economicinter change scheduler supports the operator in deﬁning the interchange
agreements requiring the transfer of power to and from neighboring areas. Hourly
comparisons are made of actual generation loads and the electricity supply
2.5. Power Generation Plant
Power plant operation entails the coordinated control of many diverse processes such as combustion, feed water ﬂow and power plant emissions. These processes,
controlled by both discrete event and continuous methods, are difﬁcult to model over their complete operating range because of their non-linearity and uncertainty.
Frequently, control of this type have been addressed by experience-based solutions that depend upon discrete switching between regions simulated by linear models.
Problems of identifying such nonlinear plant models are being addressed by research on discrete supervisory control systems by P.Antsaklis and a team of
University of Notre Dame researchers (Antsaklis, Lemmon, and Stiver, 1996). Their
proposed modeling technique, called the Hybrid Interior Point method, addresses the control of systems for which no single mathematical model would be valid over the
complete operating range (Szymanski, Lemmon, and Bett, 1998). Methods of
developing multiple sets of local models and the basis of discrete switching among the models were investigated.
The parameters of these localmodels had to be determined and discrete methods of switching among the models determined. The researchers addressed the questions
of determining the multiple model representation by decomposing the problem into
linear and quadratic subproblems . Alternating minimization techniques were used to ensure that global, as opposed to local, optima were found. The interior point (IP) technique was used to solve the linear subproblem and the Newton–Raphson (NR)
technique to solve the quadratic subproblem. While other methods have been
proposed to develop multiple models including neural networks, fuzzy logic and
genetic algorithms, the research team believes that the proposed method has an
advantage for larger complex problems. Simulation studies based on the proposed
methodology showed that a power plant, as signed to follow load, would perform satisfactorily as fuel ﬂow rate and power increased from 35% to 80% of full power. Input data obtained from 86,399 data points of rated fuel ﬂow and heat release, shown
in Figure 6, were used to train the furnace output to input relationship. This methodology resulted in the development of three different local models with each covering a different range of operating conditions. Power output resulting from the use of these three models is shown.
An example of the combined use of discrete and automated fuzzy logic
supervisory controls for turbine power operation and load change is illustrated in a proof-of-principle study completed by General Electric. A fuzzy logic control also aids the operator during turbine warm up and preparation for turbine start up. Simulator studies indicate that operation during power ramps compares favorably to the model predictive control that has been applied on a number of plants.
Discrete Event Control is a developing method that should be considered in
combination with other methods of supervisory control of the electric power system. Electric utility systems control have traditionally been based on a combination of discrete and autonomous controls. Normal control is accomplished by discrete actions of many operators functioning in distributed facilities such as control centers and power plants throughout the North American continent. The security of the electrical system is a function of the Discrete Events resulting from faulted conditions, equipment failures and unusual natural phenomena.
Recognition of the capabilities of Discrete Event Control analytical methods to
address power systems problems is in its infancy. The use of new analytical methods should consider the possible merits of the combined use of discrete event systems with neural networks, genetic algorithm or fuzzy logic. Much effort is required to bring the Discrete Event methods to the stage of general understanding and application in power systems. Security assessments could beneﬁt by the availability of a method to rapidly decompose complex decision making processes into search paths that seek desired solutions. Assigned priorities could indicate the likelihood of each path leading to a secured state. With such a method, operators could concentrate on the most likely paths and quickly determine the actions required to achieve a secure state. Continued assessment of the alternative methods of addressing this problem could lead to new insights and methods of identifying high priority search paths.
Current control methods have had the beneﬁt of years of operating experience and
have developed a level of condense among the operators. Discrete Event Controls
have features that naturally blend with utility networks that consist of many
generators, transmission lines and equipment functioning together in a uniﬁed manner.
Upsets resulting from contingency conditions can be difficult to detect, analyze and
correct. Improvements in the form of methods to assist the operators to analyze a
faulted condition and identify a corrective action should be a fertile area of research.
Rigorous analysis of all possible options for seeking optimum economic or security solutions tends to be computer intensive and requires signiﬁcantly longer
time than available in a utility operating mode. Improvements in current methods will
undoubtedly be made possible by faster computer speeds, but the real challenge is to
ﬁnd methods of analysis that assess the speciﬁc problem and conceive suitable
solutions with current computer technology.