Wind speed prediction based on weighted visibility graph and attention mechanism

Wind speed prediction has garnered significant attention from the wind power industry due to its wide application in wind power generation. To enhance the accuracy of wind speed prediction, this paper utilizes the visibility graph to map the time series of wind speed and obtain an adjacency matrix with weighted values. The adjacency matrix is trained by incorporating a graph attention neural network to derive a similarity matrix between nodes. The predicted values obtained through both the adjacency prediction method and neural network fitting are weightedly summed for prediction purposes. Compared to other models, this method exhibits smaller error values and enables more accurate wind speed prediction.