| المستخلص: |
This paper aims to develop a fast load flow computation technique without sacrificing accuracy for various on-line applications of large power systems. Both planning and operation of any power system requires the conduct of many load flow analyses corresponding to various operating modes with different system loading conditions and network configurations. Load flow analysis is performed for the determination of steady state operating status of power systems in terms of bus voltage magnitudes and angles, real and reactive powers and the transmission line losses. The load flow analysis involves the solution of non-linear algebraic equations and hence the conventional load flow algorithms are iterative in nature. The state-of-the-art approach for load flow analysis is based on Newton-Raphson algorithm (NRLF) or its derivatives such as fast decoupled load flow. As these methods are capable of providing the steady state solution within the specified accuracy, these techniques are effectively utilized as a planning tool by various utilities throughout the world. However, these are seen to be ineffective for on-line computations of practical large power systems because of the significant computational over-head due to the inherent iterative nature of such algorithms. Even though the non-iterative DC load flow approach, derived out of NRLF is computationally faster than the conventional techniques, solution accuracy is significantly less than that of its iterative counterparts. Hence, this paper proposes to develop a fast and accurate approach for the on-line load flow analysis. It is proposed to apply artificial neural network (ANN) technique as these are seen to be non-algorithmic in nature. The multi-layer feed-forward ANN for the load flow solution used in this study has one hidden layer with 100 neurons in addition to the input and output layers. The real and reactive power demands are given as the inputs to the ANN. The output consists of the bus voltage magnitudes and angles at the load buses. The proposed ANN is trained using the conventional NRLF load flow solution of a practical power grid at various load levels. The investigations revel that the ANN as a potential tool for the on-line load flow solution of practical power systems.
|
