IEE International Conference on Advances in Power System Control, Operation and Management, November 1991, Hong Kong

KNOWLEDGE-BASED ALGORITHMS FOR SECURITY/ECONOMY CONTROL

OF DISTRIBUTION SYSTEMS

C. S. Chang T.S. Chung

Department of Electrical Engineering Hong Kong Polytechnic

ABSTRACT

In this paper, several knowledge-based on-line algorithms are described, which assist the system operator to perform comprehensive distribution controls in a distribution SCADA environment with real-time databases. Functions contained in the proposed architecture of the expert system are described by a block diagram in Figure 1. A typical Hong Kong distribution network with zone substations and open-ended multiple outgoing feeder groups is used as an example for testing the validity of the approach. 'The paper first reviews an expert system (11 and describes the extension for providing comprehensive controls of the zone substation. By generalizing the procedures for the reallocation of outgoing feeder loads, adequate security margins are achieved for every apparatus on the zone- substation busbars against overloads during outage of a supply transformer. With additional rules, these procedures have been extended for the removal of existing busbar/feeder overloads, restoration of customer loads after each feeder fault and after repair, busbar reconfigurations after outage of a supply transformer, and minimization of distribution losses for one feeder group at a time. Overall control of the expert system provides a menu-driven architecture which allows the user to ask "what-if' questions by calling up any of these functions for feeder load reallocation. Using a scaling factor known as TREND, the architecture allows continuous security monitoring and assessment to be made for each substation load variation. The architecture is supported by a security-constrained optimization which minimizes the overall distribution loss for each zone substation, and a procedure of fast voltage prediction which is being developed for detecting excessive voltage differences between zone substations before switching.

INTRODUCTION

In a dispatch center, system operators control the system through a distribution SCADA environment with real-time databases. Running on a personal computer, an expert system as an operational aid is described in this paper aiming to save operators from laborious and repetitive security monitoring and assessment of system states. Thus many schemes for relieving overloads, preventive or corrective control plans may be tested in an on-line environment to give gross estimations to the operators. This furnishes the system operators with clear pictures on the distribution network improving the security and efficiency during operation. Furthermore, this is particularly meaningful during load variations and emergency conditions where smooth control should be implemented rather quickly. The expert system provides graphical displays, a user-friendly environment to facilitate further development, and a knowledge acquisition subsystem to enable editing of

a Knowledge-base with little or no assistance from a knowledge engineer.

Functions contained in the proposed architecture of the expert system are described by a block diagram in Figure 1. Using the local 132/llkV zone substation ( Figure Z ) as a test example, the expert system contains five main functions that are carried out on each zone substation:

o busbar/feeder load monitoring o system security monitoring o contingency analysis o preventive/corrective action

o distribution Iosa reduction and synthesis and analysis

ni i ni miza t ion

An expert system [ I ] was developed to provide suggestionh to system operators for carrying out preventive controls on a zone substation when it is delivering firm load. Loads on all the outgoing feeders connected to each zone substation are reallocated in order to provide sufficient security margins for every apparatus on the substation busbars to avoid overloads when ;I supply transformer (Txl. TxZ, Tx.3 or Tx3) failed. Should a supply transformer actually fail, switching of the busbar interconnectors and bus-sectioning switches (busbar reconfiguration) would be carried out at relative ease, without the risks of overloading any other apparatus on the busbars nor interrupting consumer. In this paper, functions of this expert system have been redefined and generalized for all substation load conditions. I n particular, this new extension enables the expert system to perform the above mentioned functions at continuous time>, to warn system operators against potential insecurities, and to recornmend switching plans for substation/feetler controls.

To provide a comprehensive control environment. the expert system is menu-driven to enable system operators to ask "what-if' questions for investigating effects of different scenarios or unexpected changes of operating condition. In each case, various control plans may be studied, and remedial actions involving other zone substations may be investigated. After accepting the final control plan of his own choice, the system operator updates the databases and executes the expert system for the next time instant.

To be in line with the present operating policy, there are mechanisms in the proposed architecture for minimizing the distribution loss for each feeder group, and for a zone substation.

43 1

REVIEW OF THE ORIGINAL EXPERT SYSTEM FOR PREVENTIVE FEEDER LOAD REALLOCATION r 11

I 71

As shown in Fig. 2, a typical zone substation has two long-bars each containing 16 outlets (feeders), and two short-bars each containing 10 outlets. But the actual numbers of these outlets available for out-going feeders are typically 12 and 7 respectively. The other 11 kV outlets are used for 132/11kV transformer outputs, busbar interconnectors, and power factor correcting capacitors. Each zone SubStdtiOn is normally operated as two independent substations, each comprising a long-bar and a short-bar connected to the other half zone substation via two normally-open busbar interconnector/bus- section switches. Under the normal arrangement, either 11 or 14, 12 or 13 (Fig.2) are open, while the bus section switches B1 and F2 are closed. In the event of an outage of any of the four supply transformers, switching of the busbar interconnectors/bus-section switches are carried out to supply one long bar from a single transformer, and the other long-bar plus two short-bars from the other two transformers. The above busbar reconfiguration procedure for the cases of Txl and Tx2 individually being taken out of service is illustrated in Figs. 3a and 3b.

The substation is assumed to carry its firm load of 135MVA. By maintaining the busbar load distribution as shown in Figs. 3a and 3b i.e.: 45MVA (1/3 substation load) for each long-bar, 22.5MVA (1/6 substation load) for each short-bar, the supply to all consumers is maintained without interruption with an average of 12.5% overload on the remaining apparatus after one transformer taken out of service. However, this average 12.5% overload were only valid if the 13.5 MVA load could precisely be distributed over all four busbar sections at this nominal ratio (1/3 to 1/6) between the long- and short-bars. A bus-section switch B1 or B2 is typically rated at 1200 A (22.86 MVA). The bus-section switch could be overloaded after reconfiguration by as much as 5% for a off-nominal distribution of 3 MVA between long- and short-bars whilst still maintaining the maximum SUbStdtiOn output of 135 MVA. The other overload figures after reconfiguration could be as much as 20% for a 132/11 kV transformer, 11.5% for a bus-section switch and 5% for a busbar interconnector switch. Protective relay settings could be exceeded due to such overloads.

AlDALYSIS I

Apart from the preventive function, the expert system may also be used for correcting overloads of busbar sections and/or outgoing feeders, and for reducing the distribution loss in each feeder goup. These procedures have been generalized and redefined for adoption in the proposed architecture.

Appendix 1 describes the layout of the knowledge base used in the expert system and in the proposed revised architecture.

ERoPoSED RE VISED ARCHITECI'U RE OF THE m E R T SYSTEM

Instead of furnishing mainly preventive controls at the substation firm load, a revised architecture ( Fig. 1 ) with several functions in the menu is proposed below. This new architecture deals with security and economy, preventive and corrective controls; and it has a much higher computational efficiency.

Fig.1 Functional Blocks o f Expert System

e - _ _ _ _ - _ + x y CPAPHIC DISPIAY ZOlPL SUSTAIIMIS OvFamAD m w A L

E T ~ V E a w R n o L R 6 col- RESTOUTION

Hotas: - existing communication c o m i c a t i o n being daveloped - - - - -

(1) Feeder overload control: This function is used to remove overloads of any outgoing feeder during normal operation; after a supply transformer fault, busbar reconfiguration or feeder loss minimization. The prime strategy is to detect any overload in each feeder segment, which is then eliminated by shifting the tie-switch locations from lightly loaded feeders to heavily loaded ones. A logical criterion is used to screen out "fixed, isolated, faulty or overloaded feeders which are not tit for any load transfer. This is a standard function from the menu, operating the affected feeder group to eliminate the overloads.

(2) Busbar overload control: This function is used to remove overloads under the same circumstances and with the same strategy as those in ( l ) , except that it operates within each feeder group, as well as from one feeder group to another to eliminate the overloads.

(3) Preventive control: There are certain limitations to the use of the original expert system [l] for continuous preventive controls of zone substations. The use of the security ratio (45/22.5MVA or 1/3 to 1/6) for preventive load reallocation has been an established policy in Hong Kong when supplying the firm substation load. A generalized security constraint is described in a subsequent section for all substation loads. Using this generalized constraint, each apparatus on the substation busbars is continuously furnished with adequate security margins under all loads, with preventive load reallocations recommended for all out-going feeders.

(4) Restorative control: This is the first of a two-staged customer load restoration within capacity and voltage constraints following a fault on an outgoing feeder. The expert system first isolates the faulted feeder segment by opening tie-switches at both ends, and then restores the other unserved customer loads by feeder load reallocation. Techniques developed in [4] have been modified for implementation on the local system [5]. Since load restoration involves mainly the affected feeder group, the rules are relatively simple but must ensure no violation of any capacity or voltage constraints in the rest of the substation.

(5) Reconnection after repair: This is the second stage which first recloses one of the two tie-switches ( the healthy one after repair ) at the faulted segment, and then restores the customer load at the faulted segment. So long as the voltage and capacity constraints are satisfied, preference is given to restoring the previous configuration before the feeder fault.

(6) Loss reduction: There are limited facilities for adjusting a feeder configuration to reduce losses. Several legal and feasible switching plans can be generated during inference. A sub-optimal scheme has been incorporated to take advantage of the unimodal characteristic of the feeder loss

432

with the tie-s\citcli open location. Results from a preLiouh load reallocation are fine-tuned each time by shifting the tie-switch opening location to the next feasible location for a lower loss. The procedures in [l] have been generalized to perform separate minimization of distribution losses for each feeder group. A more vigorous method for minimizing all the distribution losses connected with a zone substation has been developed and is included in the proposed architecture.

(7) Overall control: Each of the above mentioned control performs a distinct function and the user is free to choose the required control from the menu to carry out a specific operation. System operators receive real-time information of changes in the configuration, parameter and loading of each apparatus in the substation. Using in-built facilities for editing, databases containing predicted system states are updated at different times of the day. The expert system is thus well suited for variations in consumer demand: (a) to carry out controls as listed in 1 - 6; (b) to allow for some load transfer between zone substations in the event of a prolonged outage of a supply transformer. In achieving (a), locations of the normally-open tie-switches on outgoing feeders must be continuously adjusted. A measure has been developed to associate the predicted substation load with previous loads and control actions. Known as TREND in the database, prerecorded load variation factors are used for multiplication with the present value for different periods of the day. Based on these load variation factors, different apparatus of the zone substation are checked against overloads at future times to enable control measures to be formulated against such risks. Another use of these factors is for providing a basis for smoother control to prevent erratic feeder-load-transfers especially during rapid load changes.

ESTABLISHMENT OF A GENERALIZED SECURITY CDNSRAINT AGAINST LOSS OF A SUPPLY TRANSFORMER

The "1/3 to 1/6" loading ratio for long- and short-bars is over-stringent for substation loads lesser than the firm load. According to the configuration of the substation in Fig. 2, the necessary security constraint to prevent overloading of any substation apparatus after a supply transformer outage is where:

rs I T x 2

I "O

Fig.2 Zone Substation Layout

Max [Tx(l), Tx(3)] + Tx(2) + Tx(4) = < 90 MVA (1)

where Tx( I ) , Tx(2), Tx(3) antl Tx(4) are the loading o n the supply transformers Txl, Tx2, Tx.3 antl Tx1 respectively.

The "45/22.5 MVA or 1/3 to 1/6" ratio for the substation firm load ( Fig. 3 ) is merely a special case of the above. If, for a substation load of 120 MVA, each supply transformer takes an equal share of the load ( 120/3MVA ). Then,

Max [Tx(l), Tx(3)] + Tx(?) + Tx(4) = 90 MVA which is within the security constraint against the loss of one supply transformer. I f the area load density is fairly uniform, as in the city area of Hong Kong, the above would furnish approximately the least distribution loss for the zone su bs ta ti on.

6) A r r u xzcomIcm*TIac < b) A P T i i ~COI(TICUI*TICM Loadflow Bafora and M t a r lumbar h-conflgurat- lon--Outa;e of Tranaforur pr*vlou8ly connected CO a long-bar Short-bar

Loadflova h f o r a and Aftrr Susbar Le-CoafiguraC- i o n -- Outage of Transformar bantctad t o A

433

ESTABLISHMENT OF ECONOMIC TARGET USING

SUBSTATION SUPPLY AREA

It is often desirable to operate the distribution system with minimum loss which is to a large extent dependent upon the locations of the open tie-switches in out-going feeders. The expert system optimizes one feeder group at a time. On the other hand, the current practice requires system switching to be carried out for one zone substation at a time. The optimization scheme has thus been developed for a separate study on each individual zone substation. Information of the optimization can be used to provide a n economic target for the expert system and system operators for choosing the final switching plan. Online loadflows are used to check against overloads during each inter-zone-substatip switching.

Once again, by taking advantage of the quadratic and unimodal characteristics of feeder loop losses, the optimization is further broken down into subproblems, each corresponding to a solution of tie-switch opening locations within a small set of feeder segments [3]. In the present layout, not every tie-switch can yet be remotely controlled, a n d this poses certain complications on the subproblems. In each siibprobleni, the fractional open location within a set of feeder elements is optimized subject to network constraints, and is then rounded off to the nearest tie-switch location [3]. If an optimum for each subproblem is found outside the given set of feeder segments, another subproblem would be attempted on another set of feeder segments. This scheme is implemented on individual zone-substations with a typical configuration as that in Fig. 2 and each subproblem is solved by the program MINOS/AUGMENTED for a general security-constrained non-linear programming problem.

Distribution losses and network constraints: A constant current model [3] is used, and the current is assumed to be uniformly distributed on each feeder element. Mathematical models for voltage drops and losses are first formulated subject to individual constraints for each feeder segment. The overall voltage drops and losses are computed by aggregation subject to overall constraints imposed on the remaining apparatus of the zone substation. Feeder losses are approximated by quadratic functions of the current in its respective feeder segment [3]. Currents in each feeder segment are of course functions of the positions of the normally-open tie-switches to be optimized. Voltage drop constraints are linear functions of the currents in each feeder segment and of the positions of the normally-open tie-switches. Each feeder current must not its exceed its respective capacity. Collectively, the feeder currents must not exceed capacities of apparatus on the substation busbars such as : the load switches and supply transformers.

Securitv Preventive Constraints: See Equation (1) of the previous section.

ODtimization Resu Its; Using the data as listed in [l], Fig. 4 provides convergence plots of the optimization algorithm for different loading and substation data. Curve 1 shows the base case for a substation load equal to 80% of the firm load. Curve 2 is plotted for loading at the firm load, while curves 3 and 4 correspond to respectively 90 96 loading of the nominal station-transformer and feeder capacities. Convergence is attained after typically 5 - 7 executions of each sub-optimization.

SECURITY-CONSTRAINED OPTIMIZATION OF ZONE-

0 6

Figure 4 Convergence Pattern of Security-constrained der LOSS Optimizaticn

IMPLEMENTATION OF THE EXPERT SYSTEM

The expert system as described above is currently running on a 640 kb IBMPC with a 80286 accelerating card and a 10 Mb hard disk. Turbo PROLOG was adopted since it provides a fast compiler and an excellent debugger. The database for a zone substation in Fig. 2 occupies about 54 kb, and the performance of the expert system is fairly acceptable although the CPU time required for each problem has not been measured. However, it is noted that the CPU time is mainly taken by tasks in which database arguments are accessed and updated, graphical displays, and numbering crunching such as: loss minimization for each feeder group which has posed certain problem to the language PROLOG. Considerations of the expert system are thus restricted to a few checks such as overloads, voltage levels, phase difference, etc and loss minimization to a limited extent. More complicated calculations involving branch flows and loop losses have been kept to a minimum in the load feeder load transfer and fault restoration considerations. In-depth numerical analysis like the security- constrained optimization have been implemented in FORTRAN on VAX-11/780. Interfacing between software modules can be made through data links, which would also keep track of what has been done, what needs to be done, what computational tools/graphical interfaces, and/or what tradeoffs should be made.

As described previously, an expandable and modular architecture has been adopted for the expert system. Apart from eliminating the time bottleneck which would arise due to access/update of large databases, the present architecture allows for a one-to-one match between functioning blocks of distribution operations and program modules. So, it makes future additions and modifications easy. In the present structure of the expert system, a set of suggestions are presented to the operator who is prompted to make a choice using his experience. Further extension is of course to incorporate more operators experience and practice into the expert system, but the most crucial decisions would still be made by the operator.

RESULTS

The example here describes the interplay of various functions of the expert system for performing a specific duty given. First, using the built-in editor EDIT, the user

434

assigned loads to the outgoing feeders, which exceeded the capacity limits of the zone-substation busbars; and violated security limits against outage of a supply transformer:

Tx( I), Tx(2), Tx(3), Tx(4) = 49, 18.5, 43, 24.5 MVA;

Tx( 1) > 45 MVA, and Tx(4) > 22.5 MVA;

and max[Tx(l),Tx(3)] + Tx(2)+Tx(4) = 92 MVA > 90MVA

The user received the above message through the function "DATA'. Through the function "BUS", the expert system carried out the feeder-load reallocation to keep all busbar sections within security and capacity limits. The supply transformer loads then became:

Tx( l), Tx(2), Tx(3), Tx(4) = 45.1, 22.4, 45.1, 22.4 MVA

Through the function "FEEDER", no feeder overload was detected and therefore no action was taken.

The above steps provided a feasible switching plan for the zone substation. Additional switching plans could be generated as and whei required.

Through the function "LOSS', each feeder group was then optimized and to keep all busbar sections/feeders within security and capacity limits. There were little changes made to the previous supply transformer loading.

The user had earlier on specified a faulted feeder segment with the fact:

When executed, the function "FAULT' applied and isolated the faulted segment, restored the remaining customer loads:

Tx(l), Tx(2), Tx(3), Tx(4) = 44.6, 22.4, 45.1, 21.7 MVA.

Through the function "RECONNECT', the healthy tie-switch at faulted feeder segment was reclosed, and the faulted segment restored to give:

Tx(l), Tx(2), Tx(3), Tx(4) = 44.6, 22.4, 45.1, 22.9 MVA.

Both the security or capacity limits were observed in both stages of customer load restoration.

From time to time and through the funofion "VIEW", the user examined graphical displays ( Fig. 5 ) of the load distribution and the status of various breakers/tie-switches. After execution of each function, the substation configuration was automatically updatedpfor the next one.

CONCLUSIONS

I n summary, the following features of the expert system for feeder load reallocation have been developed: (i) Revised from the previous work, an overall architecture for a modular menu-driven knowledge-base system including a specific and well-defined outline of the varicas overload relieving, loss reducing, preventive/corrective characteristics as an on-line aid for distribution automation. (ii) Generalization of security constraints which guides the expert system continuously towards furnishing suff icht security margins against loss of a supply transformer. (iii) Security-constrained optimization which evaluates the optimal supply area for each zone substation and prevents remaining zone-substation apparatus from being overloaded in the event of single supply-transformer outages. (iv)A methodology developed from (i) - (iii) which guides the expert system continuously towards furnishing economic and security zone-substation operation.

Other modules such as the procedure of fast prediction of excessive voltage differences during inter-zone-station load transfer have already been covered in [l]. These modules have been designed to close the "loop" and fulfil the objectives as stated in previous sections.

REFERENCES

[ l ] C.S. Chang and T.S. Chung, " A n Expert System for On-line Security-economy Load Allocation in Distribution", Paper 89TD41-7, IEEE Transaction on Power Delivery, Jan., 1990.

[2] C.S. Chang, "Knowledge-babed Optimi Techniques for On-line Security/Eci,nc;iny Control of Distribution Systems", CEPSI'OO, Singapore, Nov., 1990.

bud f- fa3 (a) Zone S u b s t a t i c n Display (b) Feeder Group Display

F i g m e 5 Typical Printscreens of Expert system GraFhiCal outprts

435

[3] K. Aoki et. al., "Normal state optimal load allocation in distribution systems", IEEE Trans. on Power Delivery, VoLPWRD-2, No. I, January, 1087.

[4] C.C. Liu and K. Tomsovic, " A expert system assisting decision-making of reactive power/voltage control", Power Industry Computer Application Conference (PICA), San Francisco, May, 1085.

[SI T.S. Chung and M.K. Lau," An Expert System for Distribution System Post- fault Restoration and Loss Reduction", CEPS1'00, Singapore, Nov., 1000.

APPENDIX I LAYOUT OF KNOWLEDGE IN EXPERT SYSTEM Knowledge iii the expert system is divided into two groups (a) facts and (b) production rules. Facts are stored in a dynaniic database which is updated continually. Production rules represent the 'expert' knowledge and are incorporated into the inference mechanism of the KB system. The structure of these two groups will be discussed below.

Database Structure

Facts in the database record mainly two types of knowledge: on-line data and system constants. The on-line data will be updated automatically when the system changes its operating condition. Facts concerning system constants will be changed only if the system configuration changes, such as addition of cables and loads. For clarification, these two types of facts are separately represented s o that the database structure is more flexible and easier to maintain. To aim for an efficient search, facts are represented in "frames". These facts include:

element - represents the identity, type rating o r loading, charging-status, f:rult-status, feeder-group identity and load trends (zero for lines, transformers and buses) of an element in the power distribution system; root - represents the main bus identity,list of feeders connecting to the main bus, loading and predicted loading of a main bus in the system: feeder - represents the feeder identity, list of elements in a queue connecting to the main bus terminal of the corresponding feeder, loading, feeder-status, main bus connected, feeder-group identity and predicted loading of a feeder in the system; shift - represents the standby feeder identity, non-standby feeder identity, list of elements shifted, shifted loading, predicted shifted loading and feeder-group identity of a feeder-group in the system; connection - represents the topological connection between two elements, their identities, status of circuit breakers/switches and identity of the circuit breaker/switch.

Facts in the database, which only represent system constants are:

feeder-group - represents the feeder-group identity, lists of feeder within the feeder-group and lists of main buses to which the feeders are connecting; impedance - represents the line resistance and the corresponding element (line) identity in the system.

There is a new and third type of facts which is used to store intermediate states/values during the execution of the KB system, including:

root 1 - represents present iuading on main bus based on present system condition;

436

root 2 - represents loading on main bus after carrying out the recommended feeder load transfer. Loading invoked are based on present System condition;

root 3 - represents future predicted loading on main bus after carrying out the recommended feeder load transfer. Loading invoked are based on system condition in next time period.

Facts in the database are mainly represented in form of tuple fact representation, where each fact corresponds to a database tuple, e.g.:

element (bus (l), node, is-charged, not-faulted,..)

The number of elements in the distribution model is large. Searching the database to distinguish elements in a group of the same domain would take a long time, if each element were searched individually. Furthermore, there are numerous searching activities of this nature during inference. I n order to shorten the searching time for efficiency, members of the same domain will be represented as lists which bear characteristics of the domain. Therefore, to search out all the elements in a feeder, instead of searching around the database to find elements one by one with topological connection (i.e. "connection (A, B, on, sw( l))"), searching will be carried out by finding only the element list in the fact under the feeder identity. The choice of this fact representation structure took advantage of the list manipulation feature of Prolog which represents lists as compound objects. Lists can be easily manipulated by the user-defined or standard predicates in appending two lists into one, reversing the order of elements in a list, or chopping one list into two.

Rule Base Organization

TASK 1 : View main buses loading

TASK 2 : Feeder load transfer for present off- nominal busbar loading

Subtasks: Check present off-nominal loadhg Gather donating and receiving buses Gather, evaluate and choose feeder groups Determine load transfer Give recommendations

-TASK 3 : loading

Subtasks: Check predicted off-nominal loading

Feeder load transfer for off-nominal busbar

Gather donating and receiving buses Gather, evaluate and choose feeder-groups Determine load transfer Give recommendations

TASK 4 : Check feeder loaded above safety margin at present

TASK 5 : Feeder load transfer for predicted overloading on feeders

Subtasks: Find feeders to be overloaded Evaluate back-shift path Choose feasible path Give recommendat ions

TASK 6 : Fault restoration

Subtasks: Locate fault