By T.S. Sidhu & M.S. Sachdev
The University of Saskatchewan is one of a few North American Universities which feature power system protection as an integral part of their undergraduate and graduate curricula. Its increasing importance for power system engineers, has ensured its inclusion. A proper use of protective relay devices can substantially reduce the impacts of faults and other disturbances on the operation of a power system.
The earliest protective devices were fuses which were -- and still are -- used in many situations to isolate faulted equipment. This development was followed by the evolution of circuit breakers equipped with series trip coil. Later, electromechanical relays were introduced which controlled the trip coils of the circuit breakers. Although research in the development of sold state electronic relays started in the 1950s, they were not immediately accepted by utilities because of high failure rates and inappropriate designs. But with the introduction of integrated circuits and their use in relays, designs were modified to rectify the deficiencies found in solid state relays. This led to the acceptance of solid state relays by the electric power utilities for use on most power systems.
The development of microprocessor-based relays has received considerable attention since the digital technology became available (1,2,3). These relays are now used by most utilities. And manufacturers currently produce relays that protect lines, transformers, generators, reactors, capacitors, and other devices.
A major portion of the present research in power system protection is devoted to the development of new microprocessor-based relays (4). Two members of the Department of Electrical Engineering at the University of Saskatchewan and several graduate students are actively conducting research in power system protection. To facilitate this activity, two courses are taught. A well-equipped laboratory (5) is used to impart hands-on-experience and means of verifying new developments. The facilities were developed during the last fifteen years and are kept up-to-date to keep pace with the new technology. Emtp software is also available in the lab and is used extensively for testing and evaluating new relay algorithms and designs.
This paper describes the use of emtp for testing and evaluating designs of microprocessor-based relays. The design, development and testing activities associated with producing new algorithms and relay designs are also briefly described. The use and importance of emtp simulations during these activities is highlighted. Three examples of how emtp is used for testing new transformer relays, designing artificial neural network-based direction discriminators and for testing a generator winding protection algorithm are included.
Design, Development, and Testing Activities
Relay design activity at the University of Saskatchewan has always included the mathematical development of relaying techniques. Once satisfied with the basics of a technique, it is programmed in a high-level language and tested using appropriate data. Data are produced by simulating faults and abnormal conditions of a power system using emtp. The relay calculates the power system voltages and currents as inputs and can interact with the power system model to initiate operations, such as opening and closing the circuit breakers. Records and graphical displays of the relay inputs, intermediate results and relay outputs are available for review and further use. Fig.1 illustrates the use of emtp for conducting off-line studies of both the relaying algorithm and relay design.
Development activity includes the implementation of relaying techniques on a DSP. Appropriate logic is then included. The DSP, housed in a PC, now performs as a relay. Each PC in a the lab is equipped with one or more DSP boards that use Texas Instruments' TMS32020, TMS32025, TMS32030, TMS32031, or Motorola's DSP56001 chips. Each PC also acts as a developer's interface with its DSP board.
DSP boards can be programmed using either the assembler level or a high level language, such as C. A graphics-oriented debug monitor is provided with each DSP board. This facilitates the debugging and testing of programs in real time. The relay programs can be tested by downloading data stored in file on to the memory of DSP boards. Input data samples, stored in the file, are taken sequentially by the relay program for processing. Test data is then obtained using emtp, which simulates the required condition or fault in a power system
A microprocessor-based relay, a software controlled device, can be implemented by using a general-purpose hardware for microprocessor-based relays (6). This hardware functions as a relay. Exactly what type of relay depends on the software modules used in its design. After implementation, the relay can be tested by using a playback simulator that reads emtp simulated data files, converts the data to analog signals and amplifies them to levels typical of a power system. This testing evaluates the entire relay including the analog and digital interface portions, whereas the testing during design and development concentrates on the relay algorithm and logic.
Thus emtp simulations form an integral part of the relay design, development and testing activity.
Application Examples
Emtp has been used for several applications related to relay design, development and testing at the University of Saskatchewan. Three examples are provided in this section.
Data testing for a transformer winding protection algorithm
Emtp was used to model a three-phase transformer and then simulate magnetizing inrush, external and internal faults in the transformer. The simulated data were then used to test the transformer protection algorithm and design (7). The model of a N-winding single-phase transformer, shown in Figure 2, was used. The important features of the model are:
(i) N-q single-phase two-winding ideal transformers are involved, provide the correct transformation ratio of the windings 2,3.....N with respect to winding 1.
(ii) Each winding has an associated leakage-impedance branch, involving resistance and inductance.
(iii) Saturation and magnetizing current effects are confined to a single non-linear reactor in the winding 1 circuit.
(iv) Core losses are confined to a constant linear resistance which is in parallel with the saturation branch.
For modeling of a three-phase transformer, three single-phase transformers were connected in the desired configuration. Faults on the terminals of the transformer and external faults were simulated by closing switches provided at the desired fault points. For simulating the magnetizing inrush conditions, the non-linear reactor in winding-1 was replaced with a hysteresis element. A partial winding fault in the transformer was simulated by replacing the short-circuited winding with a combination of two windings. For example, consider a winding that has a resistance of R2, and inductance of L2, and a transformer ratio of N2, with respect to winding 1. Also, consider that x% is short circuited. This can be replaced by a combination of two windings as shown in Figure 3.
The above mentioned model and techniques were used to simulate various conditions in a 30 MVA, 138/13.8 kV three-phase transformer. Figure 4 shows a sample output of the primary currents for a magnetizing inrush condition in the three-phase transformer when the windings are connected in delta-wye configuration.
Training Data for an ANN-based Direction Discriminator
An artificial neural network based fault direction discriminator (8) was designed and tested as an alternative to analytical methods. The design process for fault direction discriminator consisted of the following steps.
(i) Generation and preparation of suitable training data.
(ii)Selection of a suitable ANN structure
(iii) Training of the ANN
(iv) Evaluation of the trained network using test patterns
Selection and manipulation of suitable training patterns is an issue of prime importance for neural networks. The training patterns should contain all necessary information to generalize the problem of direction discrimination. This would also enable the network to grasp and absorb the essence of a problem. A failure to do so can lead to incorrect results, unoptimized solutions and unpredictable behavior in some situations.
A sample three phase power system shown in Figure 5 was chosen for the purpose of generating current and voltage signals for forward and reverse faults. Zs1 and Zs2 represent the equivalent source impedances. The lines are 500 kV, 150 km in length, with bundled conductors (two conductors per bundle), fully-transposed construction and an earth resistivity of 100½m. The measurement system was installed on a bus-Z looking towards bus-Y. Frequency dependence of various components was included.
Training patterns were generated by simulating three phase to ground and single phase to ground faults at various locations in the forward and the reverse directions. Extensive testing revealed that the network was able to generalize the situation from the provided patterns and correctly identify the direction of other types of shunt faults as well.
Training data was obtained using an emtp-simulated power system. The data was analyzed to determine if it contained patterns covering most of the region for forward- or reverse- fault regions. This was accomplished by using a routine which calculated the voltage and current phasors and checked the relevant position of the current phasor. Fault location and fault resistance were changed to obtain the training patterns so that most of the regions were covered. The number of training patterns for both forwared and reverse faults were kept equal to avoid "skewed" training. Special care was taken to include boundary patterns. Boundary samples were generated by creating high resistance faults and solid faults near the relay. Table 1 lists various shunt faults which were simulated for generating training patterns. These faults were simulated with different initial power flow conditions.
The ANN based discriminator was trained using the data simulated by the emtp and was then tested using fault data recorded from 240 kV and 500 kV lines of two utilities. The proposed discriminator performed well indicating that the training data was suitable.
Data for Testing Generator Winding Protection Algorithm
A new algorithm for detecting shunt and series faults in the stator windings of an ac synchronous generator has already been developed at the University of Saskatchewan (9). The transient fault data simulated by the emtp was used for evaluating the performance of the algorithm. A block diagram of the circuit for simulating the power system and generating the fault data is shown in Fig. 6. Two machine models are available within the used emtp software, or EMTDC. One is a simple model of a machine, in the form of a Thevenin voltage source and its impedance, used to simulate the infinite bus on the high voltage side of a transformer. This model has limited configurations and does not provide the full dynamic response of a complete machine model. A more complete machine model is also available within the EMTDC. This model simulates a salient pole synchronous machine complete with a multi-stage turbine, exciter, and governor. The complete model of the synchronous machine was connected to the infinite bus via a step-up transformer. In Figure 6, the basic parameters of the machine that was used are listed. The machine model uses a base voltage of 13.8 kV and a base MVA of 120MVA. An R-L load, representing the Unit Service Bus load, was connected between the unit generator and unit transformer. Various types of faults were staged at three locations on the system; at the machine terminals on the machine side of the relay to simulate faults internal to the machine, on the Unit Bus to simulate close-in faults, and on the high-voltage side of the Unit Transformer, located at the midpoint of the transmission line, to simulate external system disturbances.
The simulated data were generated using a calculation step of 23040Hz in EMTDC. This allowed the generation of spectral components up too11520Hz and provided better approximation of the continuous signals that would exist in power system during a disturbance. The EMTDC output data were then pre-processed by a digital equivalent of an anti-aliasing filter which was implemented in Microsoft FORTRAN. The filtered data was then resampled at at rate of 720Hz which simulates the sample-and-hold and A/D operation of an actual relay, and provides the data in a form ready for processing by the relay algorithm. Figure 7 shows the filtered and resampled voltage and currents for an internal a-g fault at the generator terminals.
Conclusions
This paper has briefly described the design, development, implementation and testing phases associated with producing new relay algorithms and designs. The use of emtp during these phases was also highlighted. The data obtained from the emtp simulations play an important role during all phases of algorithm and relay development. With the introduction of new relay testing methods and new relay designs that include artificial neural networks, it is anticipated that the use of emtp will gain further importance as it is utilized for generating training and test patterns. ET
T.S. Sidhu and M.S. Sachdev are with the Power System Research Group at the University of Saskatchewan.