Cuckoo Search Algorithm Optimization of Holt-Winter Method for Distribution Transformer Load Forecasting

Abstract

Reactive maintenance of distribution transformers leads to lost electricity sales and decreased customer satisfaction. Due to regulatory OPEX reduction, the Philippine distribution utility uses a limited workforce for maintenance and installation. The study used actual transformer data to forecast overloading. This paper utilized Holt-Winter forecasting method and compared the Cuckoo search algorithm (CSA) with the Genetic algorithm (GA) in optimizing MSE to obtain the optimum smoothing coefficients, $\alpha$ (level), $\beta$ (trend), and $\gamma$ (season) of the Holt-Winters (HW) forecasting method. The statistical results showed that in terms of speed of optimizing the HW model and forecast accuracy, using the CSA and GA was not statistically different from one another. An MSE of 4.78922 and 4.92180 were obtained using the CSA and GA, respectively, to optimize the HW additive type. While the MSEs of 7.90807 and 7.88312 were obtained using the CSA and GA, respectively, to optimize the HW multiplicative type. Statistical tests showed that the difference between their forecast accuracy is not statistically significant.