2. Description of the Survey
3. Analysis and Discussions
4. Summary and Final Remarks
Agent technology has been enlightened as the correct vector to promote decentralization, au- tonomous operation and active management activities in power distribution system opera- tion. As a matter of fact, an adequate agent-based modeling can produce flexible, extensible, and robust systems, which are features of utmost importance to a smooth modernization of power systems. Moreover, the agent paradigm can provide a well-established notion of intelli- gence/smartness to be progressively applied along with the modernization of power distribution systems. Hence, in this document, an extensive survey about the applications of agent tech- nology to power engineering is provided, aiming at helping engineers and academia to identify gaps in the state of the art to be explored in the future. This survey is presented and discussed highlighting the applications more directly related to the scope of the power distribution sys- tems.
The document is organized as follows. In section 2, the developed survey is described, highlighting the size/range of the survey and main sources of information. Then, in section 3, discussions and analyzes about the state of the art are developed. Final remarks are outlined in section 4.
2 Description of the Survey
Recently, a representative literature survey on agent-based systems applied to power engineer- ing was published1 by the IEEE Power Engineering Society’s Multi-Agent Systems Working Group, created in June 2005. The related publications were sought and categorized based upon their applications in protection, modeling, simulation, distributed control, monitoring and di- agnoses. The survey included publications dated from 2001 to 2005 of the Intelligent Systems Applications to Power Systems (ISAP) conference proceedings as well as publications dated from 1998 to 2007 covering some IEEE transactions and some Institution of Electrical Engi- neers (IEE) journals. Hence, a similar survey was developed considering publications dated from 2001 to 2011 of the ISAP conference proceedings plus the publications dated from 1998 to 2011 of the IEEE Transactions on Power Systems, IEEE Transactions on Power Delivery, Institution of Engineering and Technology1 (IET) Generation, Transmission & Distribution, Electric Power Systems Research (EPSR) and International Journal of Electrical Power & Energy Systems (IJEPES). Table 1 summarizes the outcomes of the survey emphasizing the functional separation between power generation & transmission engineering and power distribution engineering2. Naturally, additional conference proceedings and journals were sought for the sake of completeness.
Table 1: Bibliographic survey of multi-agent systems applied to power engineering problems
Abbildung in dieser Leseprobe nicht enthalten
Table 1 outlines in totality 100 publications, from which 23 were directly conducted aiming at power distribution system applications. In comparison with classical power engineering research fields such as reliability modeling, these figures suggest that the marriage between agent technologies and power engineering is far from being mature. This conclusion was expected given how agent-based systems have been maturing conceptually during the past few years, the interdisciplinary requirements to building practical applications under the agent paradigm, and the innovative character of applying such paradigm in power engineering solutions. On the other hand, although some overlapping work can be found, these figures also indicate that consistent research efforts have been made to exploit the agent paradigm to improve power engineering. These efforts cover from specific solutions proposed to improve the protection of small industrial systems78 until abstract frameworks devised to manage entire bulk power systems41.
3 Analysis and Discussions
Most probably, the application of agents to power system protection and primary control is the best starting point in what regards examining the evolution of academia in understanding the actual potential of agent technologies to power engineering. As highlighted in1, Wooldridge’s classical definition of an agent does not clearly distinguish agents from a number of existing softwares and hardware systems. Hence, under the quoted definition, controller devices or protective relays could be considered agents in the sense that they exhibit a certain degree of autonomy, they are situated in an environment (the power system), and they react to changes (voltage and/or current signals) in the environment. The same reasoning applies, for instance, to an excitation system of a synchronous generator, a buck-boost transformer or a capacitor bank placed in a distribution feeder. Thus, following this crude definition, the term agent can be misapplied as a wrapping concept over control devices and protective relays, thereby originating multi-agent systems from interactions employed in legacy control and protective schemes. Although this can be arguably viewed as beneficial in providing different abstractions and ways of understanding existing systems, we share the harsh belief of some authors (e.g.1 ) which state that renaming existing or new systems built over existing technologies as agents offers almost nothing in terms of concrete engineering benefit.
Not surprisingly, one of the first research contributions37 about agents to bulk power system protection applies a cooperation of agents abstracted over a wide range of equip- ments/components such as line agents, bus agents, current transformer agents, potential trans- former agents, circuit breaker agents, current transformer data collector agents, and so forth, without an explicit representation of goal-directed behaviors for all of them. In power distri- bution systems, one example is the control and protection scheme described in79 where a network split in zones, as proposed in101, is managed by monitoring agents, communication agents, breaker agents and relay agents, which similarly lack goal-directed behavior. Also, in several further works, model validation through computer simulations (if any) did not imply that the agents were provided with data encapsulation, code encapsulation, separated threads of control, or the ability to run autonomously without being evoked externally (e.g. [97, 102]). Despite these issues, those works should not be seen with criticisms since they provided contri- butions to their respective fields. In fact, possible misconceptions might be merely faced as a consequence of the academic knowledge of the epoch in which the studies were performed.
More specifically to the applications of agent-based technology to power distribution system protection, the literature analysis had identified that the number of contributions/publications in this field of research is still limited. As one of the exceptions, we can refer to the multiagent system applied to the overcurrent protection of a 6.6 kV industrial system proposed in78. The work focuses on using overcurrent relay agents and fuzzy sets103 to search for an optimum protection capability scheme. Although agent simulation itself is not performed in sensu stricto, the idea of providing adaptive protection through agency, the cooperation among relay agents, and concerns regarding communication protocols are already discussed in the context of improving the protection schemes of power distribution systems. These concepts are present in more recent contributions such as [81,104].
In81, a coordinated protective scheme for distribution feeders is proposed using feeder agents and overcurrent relay agents. The scheme dictates that relay agents must update fault currents and maximum load currents using exchanged information whereas network topology changes are identified. For this accomplishment, source impedance information is fed down- stream the primary substations while maximum loading information is fed upstream the loads. Object-oriented programming in C++ and agent communication language (ACL) are utilized. A small test system is used to illustrate the proposed rules and the resulting relay settings update seems promising, though more information could be exhibited about how the agents actually perform their reasoning. Also, it seems that more interactions and information would be necessary to achieve a practical usage of the scheme. For instance, the scheme does not seem directly extendable for distribution feeders with Distributed Generation (DG) units since time settings are not even mentioned.
In104, the coordination of relays in power distribution systems with DG units using a multi-agent solution is proposed. Similarly to the previous works, it consists of applying the abstraction of a relay agent, but considering those along with DG agents and equipment agents to aid protective coordination. In summary, each relay agent takes sensory information, such as DG connection status, in order to choose among a list of settings. Then, some protection coordination is achieved through message exchange. The major drawback of the methodology is that agent message conveyance is required right after fault currents are identified. Besides demanding fast communications which are unfeasible to legacy systems, the idea of establishing agent interactions while the system devices are directly subjected to fault currents is, at least arguably, controversial. As a matter of fact, in case time constraints and backup protections were not adequately set, equipment life cycle might be jeopardized. Nevertheless, it is important to emphasize that some authors have been stating that while the cost of communications may not be justifiable to achieve sole protection functions, there will be situations where such cost might be justifiable if the same infrastructure is applied to provide other functions. This point of view is clearly exhibited, for instance, in the fault isolation scheme proposed in97 where a wavelet-based fault direction identification technique is applied and relay agents interchange logical signals to locate the fault and activate circuit breakers.
Apart from devising protective schemes, agent-based systems have been applied to post-fault diagnosis as well. The first found work in the area refers to the ARCHON project [105, 106] whose main outcome is a general purpose architectural framework to facilitate cooperation among computational systems for industrial applications. As a practical deployment of the framework, the ARCHON approach was used to integrate reasoning systems within a control room including an alarm analysis agent, breaker and relay supervision agent, black-out area identifier agent, and even a system restoration agent, having as target the provision of useful information to the operators regarding contingency situations. Although some authors state that the resulting post-fault diagnosis application was not truly flexible or scalable53, the general framework itself is considerably well thought in terms of deriving design concepts to structure interactions among problem solving entities (called agents in ARCHON’s context) for industrial applications. Unfortunately, recent power engineering applications or even quotations about the ARCHON framework were not found in the power engineering literature, a fact that might be explained by the lack of dissemination of ARCHON framework’s features in power engineering journals and conference proceedings.
Almost ten years ahead, the protection engineering diagnostic agents (PEDA) proposed in [53, 63, 107] were introduced. In this research, a post-fault diagnose assistant system to pro- tection engineers was developed through the integration of several protection analysis tools. The integration was performed by wrapping analysis tools in agents named protection valida- tion and diagnoses agent, incident and event identification agent, fault record interpretation agent, fault record retrieval agent, collation agent and engineering assistant agent. Despite of the straightforward (but well justified) aspect in how the authors defined the agent types, the work stands as one of the few where some multi-agent system design concepts were applied, namely the task decomposition stage of the DESIRE108 methodology. Codding was per- formed using ZEUS109 and at least one agent descriptor with functional task and exchange resource information is provided, what we believe should be a practice to be adopted in ex- hibiting research material in this area. The work itself is discussed in the context of bulk power system applications. Nevertheless, with the ongoing increase of feeder automation, it could be adapted/extended to distribution feeder applications. Other related system that could be exploited on distribution feeder automation is the conditional monitoring multi-agent system (COMMAS) currently applied to plant monitoring110 and transformer monitoring [28, 57].
Regarding power distribution system restoration, we initially instantiate the multi-agent system introduced in [111,112]. This work resembles a previous research developed by the same authors in the context of bulk power system restoration49 and consists of abstracting feeder agents and load agents to perform restoration functions according to the following directives.
1. Once a service interruption is assigned, load agents have to isolate their respective loads and send a message to their service feeder agents.
2. The feeder agents receive the messages from the load agents and handle the restoration using rules extracted from operator’s experience.
The proposal has the merit of being one of the earliest in the area. However, decision-making about the restoration procedures is still centralized on the feeder agents while the role of the load agents is basically to notify the feeder agent about service interruptions. In addition, dis- cussions about software modeling and the practicality of implementing one agent per secondary transformer in actual power distribution systems are neglected. Decisions about restoration are also centralized in the solutions proposed in [86, 102], where global agents102 and restoration leader agents86 were envisioned to wrap reconfiguration and decision-tree algorithms, respec- tively. The latter solution86 is assumed to follow the BDI agent architecture but beliefs, desires and intentions are not explicitly described in the agent restoration mechanisms.
A decentralized multi-agent solution for power distribution system restoration was proposed in96 where three agent types are abstracted: the generator agent (to model primary substation sources), the load agent and the switch agent. Service restoration is then achieved through interactions among these entities without the figure of a higher level entity managing the switch actions and taking into account the available transfer capacity of switches. The work is clearly JADE-oriented and the agents are described according to their initial knowledge and behaviors. Also, the simple UML class diagram of the application is presented making easy understanding the author’s implementation. Further improvements are introduced in113 with a small update to consider rules for load shedding and a DG agent. Although engineering aspects about the actual implementation of the proposed solution are not discussed, these publications succeed on describing simple behaviors and rules to aid the restoration processes in power distribution systems. Furthermore, the provided small case study is well described, so that we recommend its implementation to power system researchers interested in achieving some proficiency on implementing agent behaviors in JADE.
In concise terms, the contributions in [86,96,102,111-113] would seem more realistic in case, for instance, the load agents explicitly represented an aggregation of customers, as well as re- quirements for monitoring and main hypothesis about infrastructure were clarified. Conversely, other contributions such as in [91-94, 114] introduce fair hypotheses also offering discussions about the practicality of their proposed solutions. As a matter of fact, the authors in91 present an interesting application of agency to condition assessment and fault management, though much more focused on engineering solutions for power distribution system automation than in agent modeling itself. The concept is composed of three main aspects as follows.
1. A software object for secondary substations which encapsulates a set of object classes representing the substation hardware and following the concept of logical nodes from the IEC 61850115.
2. Functionalities based on token (permission to act and execute local functions) message conveyance between neighboring substations. The basic procedure implies that, when a substation receives a token, it executes the required function, attaches the result, and conveys information to downstream substations. After processed at the last secondary substation, the token is sent back to the primary substation.
3. An information access model which, in summary, hierarchically defines that a permission to execute a given function is conveyed from the control center to the primary substation, and further downstream the secondary substations.
Hence, fault location and isolation are obtained by sending a token from a primary substa- tion downstream secondary substations to verify fault indicators and to open normally closed switches around the faulted section. Power is restored by closing the circuit breaker at the primary substation and passing tokens upstream towards alternative supply primary substa- tions. The proposed concept was extended to include state estimation in92 and has a reduced engineering complexity in the sense that all secondary substations in a feeder are copies of a common secondary substation type. Furthermore, it provides local management at the sec- ondary substation level as well as it reduces the information and communication saturation since, for the developed functions, the control centers “see” a primary substation area as a single entity. Regarding communication issues and validation tests, the authors verified the reduction in number of hops for fault management in comparison with a totally centralized control approach, and they tested the approach using a small but very illustrative prototype implementation.
Another agent-based approach much more focused on finding engineering solutions to power distribution system automation than in agent modeling itself is provided in114. The authors introduce an hierarchical automation architecture based on the concept of intelligent logical nodes, which is envisioned as an extension of the logical node (see IEC 61850115 ) concept ap- plied to substation automation. Although the rules for fault location and power restoration can be found in previous works, the approach itself is very interesting in the sense that it establishes a direct integration of the IEC 61850 and IEC 61499116 into the so-called intelligent logical nodes. Concisely, the integration dictates that for each logical node, an intelligent logical node is implemented as a composite function block of the IEC 61499 with a database, service inter- preter and intelligence (the part responsible for decision making and negotiation). The agents communicate via services of the IEC 61850 and computer validation tests were carried out using a MATLAB117 model interfacing with function block models through custom-design
1 In 2006, the IEE merged with the Institution of Incorporated Engineers to form the IET.
2 Differently from1, papers related with the application of ant colony optimization algorithms as an approach for power system optimization problems were not included in the survey. The same applies to the direct application of reinforcement learning algorithms or other meta-heuristics.