The present day digital computing resources are overburdened by the amount of calculation necessary for power system dynamic simulation. Although the hardware has improved significantly, the expansion of the interconnected systems, and the requirement for more detailed models with frequent solutions have increased the need for simulating these systems in real time. To achieve this, more effort has been devoted to developing and improving the application of numerical methods and computational techniques such as sparsity-directed approaches and network decomposition to power system dynamic studies. This project is a modest contribution towards solving this problem. It consists of applying a very efficient sparsity technique to the power system dynamic simulator under a wide range of events. The method used was first developed by Zollenkopf (^117) Following the structure of the linear equations related to power system dynamic simulator models, the original algorithm which was conceived for scalar calculation has been modified to use sets of 2 * 2 sub-matrices for both the dynamic and algebraic equations. The realisation of real-time simulators also requires the simplification of the power system models and the adoption of a few assumptions such as neglecting short time constants. Most of the network components are simulated. The generating units include synchronous generators and their local controllers, and the simulated network is composed of transmission lines and transformers with tap-changing and phase-shifting, non-linear static loads, shunt compensators and simplified protection. The simulator is capable of handling some of the severe events which occur in power systems such as islanding, island re-synchronisation and generator start-up and shut-down. To avoid the stiffness problem and ensure the numerical stability of the system at long time steps at a reasonable accuracy, the implicit trapezoidal rule is used for discretising the dynamic equations. The algebraisation of differential equations requires an iterative process. Also the non-linear network models are generally better solved by the Newton-Raphson iterative method which has an efficient quadratic rate of convergence. This has favoured the adoption of the simultaneous technique over the classical partitioned method. In this case the algebraised differential equations and the non-linear static equations are solved as one set of algebraic equations. Another way of speeding-up centralised simulators is the adoption of distributed techniques. In this case the simulated networks are subdivided into areas which are computed by a multi-task machine (Perkin Elmer PE3230). A coordinating subprogram is necessary to synchronise and control the computation of the different areas, and perform the overall solution of the system. In addition to this decomposed algorithm the developed technique is also implemented in the parallel simulator running on the Array Processor FPS 5205 attached to a Perkin Elmer PE 3230 minicomputer, and a centralised version run on the host computer. Testing these simulators on three networks under a range of events would allow for the assessment of the algorithm and the selection of the best candidate hardware structure to be used as a dedicated machine to support the dynamic simulator. The results obtained from this dynamic simulator are very impressive. Great speed-up is realised, stable solutions under very severe events are obtained showing the robustness of the system, and accurate long-term results are obtained. Therefore, the present simulator provides a realistic test bed to the Energy Management System. It can also be used for other purposes such as operator training.
Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:314581 |
Date | January 1990 |
Creators | Bousnane, Kafiha |
Publisher | Durham University |
Source Sets | Ethos UK |
Detected Language | English |
Type | Electronic Thesis or Dissertation |
Source | http://etheses.dur.ac.uk/6623/ |
Page generated in 0.0015 seconds