Pоbоlјšаnjе prеdikciје prоizvоdnjе vеtrоеlеktrаnа u Јužnоm Bаnаtu kоmbinоvаnjеm pојеdinаčnih prоgnоzа pоmоću mоdеlа vеštаčkе intеligеnciје

Abstract

The prediction of day-ahead wind farm production is crucial for optimal unit engagement and minimizing power balancing costs in the power system. Several wind farms which use different models to predict the production have been built in the South Banat region. The basic idea in this paper is to combine the production predictions of each of the wind farms with the production predictions of other wind farms in the region using the artificial intelligence model. This approach has a physical justification given that all wind farms are located in a region with the same wind climatology. Since the total error in estimating production for the day ahead is of interest for planning balancing capacities, this paper analyses the possibility of minimizing cumulative error through the application of artificial intelligence algorithms. The algorithms combine forecasts of individual wind farm models and thus make corrections in estimating the total production of wind farms in this region. The training of the neural network model was performed on the basis of forecasts of individual wind farms for the day ahead and performance. One-year sets of forecasted and realized wind farm productions that were in operation in 2020 were used to train the networks. The developed model of prediction of cumulative wind farm production in South Banat enables the transmission system operator to perform subsequent processing of individual predictions of wind farm production in order to reduce the total error in the assessment of cumulative production.