Genetic algorithm-based non-linear auto-regressive with exogenous inputs neural network short-term and medium-term uncertainty modelling and prediction for electrical load and wind speed

Electrical load and wind power forecasting are a demanding task for modern electrical power systems because both are closely linked with the weather parameters, such as temperature, humidity, and air pressure. The conventional methods of electrical load and wind power forecasting are useful to handle dynamic and uncertainties in un-regulated energy markets. However, there is still need of relative improvement by incorporating weather parameter dependencies. Considering above, a genetic algorithm-based non-linear auto-regressive neural network (GA-NARX-NN) model for short-and medium-term electrical load forecasting is presented with relative degree of accuracy. Causality, a new modelling technique, is employed for monthly and yearly wind speed patterns predictions and long-term wind speed forecasting. Real-time historical electrical load and weather parametric data are used to critically observe the performance of the proposed models compared to various state-of-the-art forecasting schemes. Numerical simulations are conducted that validates the proposed models based on various error calculation methods, such as mean absolute percentage error, root mean-square error, and variance (sigma(2)). The quantitative comparison with five traditional techniques for electrical load and wind speed forecasting reveals that the GA-NARX-NN method is more accurate and reliable.