Abstract: Surface contamination is one of the main causes of flashover on outdoor insulators, since deposits accumulated on the insulator surface reduce its resistance and can lead to failure. Monitoring techniques based on parameters such as the Equivalent Salt Deposit Density (ESDD) are widely used to estimate this contamination level and warn operators before a flashover occurs, yet reliable forecasting of ESDD remains a challenging task due to its nonlinear and noisy behavior. Forecasting in power systems must be accurate, reliable, and fast when applied on a short-term or online basis, particularly in mission-critical operating environments, since it plays a significant role in decision-making and in overcoming economic and operational problems. Traditional methods have proven effective under linear or stationary assumptions, but recent challenges are nonlinear, high-dimensional, and noisier, requiring more complex approaches. To address this complexity, a hybrid method combining the Multi-Model Partitioning Filter (MMPF) with a Genetic Algorithm for Resource Allocation (GARA) is presented. The method refines the initial probabilities (weights) provided by the MMPF through an iterative, fitness-driven search for optimal weight values. The proposed approach is compared against a previously presented method that combines MMPF with Support Vector Machines (SVM), using real ESDD and electricity load measurements.

Keywords: Equivalent Salt Deposit Density (ESDD), Multi-Model Partitioning Filter (MMPF), Genetic Algorithm for Resource Allocation (GARA), Adaptive Forecasting, Electricity Load Prediction, Field Data Analysis, Mean Absolute Percentage Error (MAPE), Mission-Critical Systems

Pages: 317-325