Development of Artificial neural networks approach for estimation of EHV/UHV transmission line enrgization peak overvoltages

Abstract

In India, the available generation and installed capacity has been increased many folds in the last two decades. Next?higher 765 kV transmission lines are planned to reinforce the existing 400 kV networks. The systems are interconnected and operated closer to their performance limits. Maintaining system security and facilitating efficient system operation have been challenging tasks. The system’s vulnerability to potential blackouts presents new challenges for prompt and effective power system restoration. Blackouts have increased in recent years in terms of severity and frequency of occurrence due to heavier system loading. The impact of prolonged blackouts on the public, on the economy, and on the power system itself makes rapid, effective restoration very important. Restoration to normal operating conditions is to be carried out as fast as possible to ensure system security and to avoid prolonged interruption of power supply. Bulk power system restoration is very difficult and complex. In the process of restoration, switching transient overvoltages occur, which may be detrimental and may lead to unsuccessful operation of line energization. Electromagnetic Transients Program (EMTP) is used for computation of both switching and temporary overvoltages. EMTP simulations are extensively carried out during planning studies for the purpose of various transient analyses. Planning studies aim at designing protective equipment to specified limits as per the utility practice. However, during day?to?day operation, such studies by the operators are not common due to the detailed data required and the large computational time involved. Parameters which influence overvoltages during energization are interdependent, and this eliminates the idea of deriving equations to get the peak values. Thus, the only choice left is using machine?learning techniques such as neural networks. This thesis presents development of an alternative ANN approach for estimation of peak overvoltages during line charging for EHV (400 kV) and UHV (765 kV) lines. In the proposed methodology, the Levenberg–Marquardt method is used to train the multilayer perceptron. The developed ANN is trained with the simulated results and tested for typical cases. The pattern is also adopted to classify the switching overvoltages generated during line energization into two classes as safe and unsafe. This method applies a learning?based nonlinear classifier, the Support Vector Machine (SVM). The incorporation of both ANN and SVM approaches will help the operator to select the mode of energizing EHV/UHV transmission lines within the safe limits. Simulated results of an equivalent 24?bus Indian grid system presented clearly show that the proposed technique can estimate the peak values of switching overvoltages with good accuracy.