Development of an adaptive neural-fuzzy inference system based meta-model for estimating lightning related failures in polluted environments

Owing to the random nature of the lightning phenomenon, the statistical based methods such as Monte Carlo simulation are the best tool to perform the lightning-related studies. However, Monte Carlo simulation is complex and time consuming, especially, for larger networks where a lot of regions must be investigated, separately. Although some researches have been carried out to evaluate the lightning flashover outage rates based on artificial intelligence (AI) techniques, but, so far, any work has not been reported to perform a risk assessment of lightning surges by means of AI. This study presents a meta-model based on adaptive neural-fuzzy inference system (ANFIS) that can be utilised to perform the lightning-related risk analysis with a good accuracy. The inputs of ANFIS are tower footing resistance, critical flashover voltage and the rate of lightning occurrence; and its outputs include the failure risk of both the insulation and arrester. The trained ANFIS can be very efficient in terms of time and computational resources for the further analysis, like placement of arresters or insulation coordination studies, on the electrical network on which the training is performed. To demonstrate the effectiveness of the ANFIS meta-model, an optimisation method, based on genetic algorithm, has been implemented in order to find the optimum location of a specified set of line arresters on power network. Also, in order to consider the environmental condition, that is, pollution and altitude, a new method to determine the importance of each node of interest in protection is introduced.