Peak Forecasting for Electricity Loads in Jordan Using a Weighted Combination of Feed Forward Back Propagation Neural Network and Holt-Winter

Predicting electrical loads is critical since it requires the preparation of work schedules for main and minor transformer maintenance operations. This study aims to estimate the maximum evening electricity loads in Jordan in order to plan preventive maintenance. Because the data contains two sorts of patterns, trend and seasonal, methods that address both of these patterns must be applied. Long-term maximum peak forecasting for the next five years from 2022 to 2027 is performed by utilizing a weighted combination of two methods: Feed Forward Back Propagation Neural Network and Holt-Winters. The Seasonal Autoregressive Integrated Moving Average method is also used and found of less accuracy than the two other models. A near-optimal weighted combination of the Neural Network Model and Holt-Winter is then obtained by using a non-linear optimization local search heuristic model. The new model results are of high accuracy that the value of R2 was increased from 70.2 % for the Neural Network Model and 82.4 % for the Holt-Winters method to 84.3 % in the weighted combination between them. Moreover, the value of Mean Absolute Error was decreased to from 134 and 188 for the Holt-Winter and Neural Network models respectively to become 117.98. Upon the results, recommendations are provided to the decision-makers.