Wind Park Power Prediction: Attention-Based Graph Networks and Deep Learning to Capture Wake Losses

Abstract

With the increased penetration of wind energy into the power grid, it has become increasingly important to be able to predict the expected power production for larger wind farms. Deep learning (DL) models can learn complex patterns in the data and have found wide success in predicting wake losses and expected power production. This paper proposes a modular framework for attention-based graph neural networks (GNN), where attention can be applied to any desired component of a graph block. The results show that the model significantly outperforms a multilayer perceptron (MLP) and a bidirectional LSTM (BLSTM) model, while delivering performance on-par with a vanilla GNN model. Moreover, we argue that the proposed graph attention architecture can easily adapt to different applications by offering flexibility into the desired attention operations to be used, which might depend on the specific application. Through analysis of the attention weights, it was showed that employing attention-based GNNs can provide insights into what the models learn. In particular, the attention networks seemed to realise turbine dependencies that aligned with some physical intuition about wake losses.

Data-Driven wind turbine performance assessment and quantification using SCADA data and field measurements

“…With such models the influence between neighboring turbines is described by weighting factors of edges that are learned within the GNN. The resulting GNN model can than predict the power output for each individual turbine within a wind farm (Park and Park, 2019;Bleeg, 2020;Bentsen et al, 2022).…”

Graph-based Deep Reinforcement Learning for Wind Farm Set-Point Optimisation

“…The edges of these graphs represent the relationships between turbines, either over the whole farm [6] or using grouping algorithms [22]. In recent papers, the influence of each turbine on its neighbours via these edges has been investigated through physicsinduced edge weighting [23] and attention networks [24]. Dong & Zhao [25] used a graph-style approach to split a large wind farm into sub-groups to apply more localised WFC methods, but they did not train a GNN itself.…”

“…For this to be implemented in a GNN, the positions of the wakes themselves could be approximated, or techniques such as dynamic GNN could be implemented [28]. The number of graph connections between turbines could also be increased to match the levels seen in other recent works [22,24]. Additionally, to ensure the graph layers are providing useful encodings, alternate combinations of node (turbine) attributes could be tested.…”

Graph machine learning for predicting wake interaction losses based on SCADA data

“…The advantage of data-driven models is their efficiency and the incorporation of field data of actual wind farms [6]. More recent data-driven approaches that model wake effects are graph neural networks (GNNs), which represent the wind farm as a graph [7,8].…”

“…In recent years, GNNs have been applied for wind speed forecasts and predictions [15,16], power predictions [8,17,7] and interaction loss estimations [8,7,18] in wind farms. However, the developed GNNs concerning wind farm modelling were based on simulation data.…”