Neural computing: theory and practice
Neural computing: theory and practice
| Hi-index | 0.00 |
Power system security is one of the major concerns in competitive electricity markets driven by trade demands and regulations. If the system is found to be insecure, timely corrective measures need to be taken to prevent system collapse. This paper presents an approach based on a counterpropagation neural network (CPNN) to identify and rank the contingencies expected to reduce or eliminate the steady-state loadability margin of the system, making it prone to voltage collapse. It has been shown that unlike other artificial neural networks (ANN) paradigms, which start with random weights, CPNN is very sensitive to initial weights. To reduce the dimension and training time, a novel feature selection method, based on the coherency existing between load buses with respect to voltage dynamics, is employed to select significant input features for the CPNN. Once trained, the CPNN is found to rank voltage contingencies accurately for previously unknown system conditions very fast. Due to its fast training, the proposed CPNN will be particularly useful for power system planning studies, as a number of combinations can be tried within a small time frame. The effectiveness of the proposed approach has been demonstrated on IEEE 30-bus test system and a 75-bus practical Indian system.