Predicting Wind Turbine Power and Load Outputs by Multi-task Convolutional LSTM Model

Woo, Seongcheol; Park, Jun-Young; Park, Jinkyoo

doi:10.1109/pesgm.2018.8586206

Cited by 23 publications

(13 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…To achieve this, the temporal portion of the LSTM's state is a 2D matrix whose entries correspond to the topology of the spatial input. This architecture has been applied successfully to wind power forecasting [18].…”

Section: Geo-temporal Predictionmentioning

confidence: 99%

Localized Convolutional Neural Networks for Geospatial Wind Forecasting

2020

View full text Add to dashboard Cite

Convolutional Neural Networks (CNN) possess many positive qualities when it comes to spatial raster data. Translation invariance enables CNNs to detect features regardless of their position in the scene. However, in some domains, like geospatial, not all locations are exactly equal. In this work, we propose localized convolutional neural networks that enable convolutional architectures to learn local features in addition to the global ones. We investigate their instantiations in the form of learnable inputs, local weights, and a more general form. They can be added to any convolutional layers, easily end-to-end trained, introduce minimal additional complexity, and let CNNs retain most of their benefits to the extent that they are needed. In this work we address spatio-temporal prediction: test the effectiveness of our methods on a synthetic benchmark dataset and tackle three real-world wind prediction datasets. For one of them, we propose a method to spatially order the unordered data. We compare the recent state-of-the-art spatio-temporal prediction models on the same data. Models that use convolutional layers can be and are extended with our localizations. In all these cases our extensions improve the results, and thus often the state-of-the-art. We share all the code at a public repository.

show abstract

Section: Geo-temporal Predictionmentioning

confidence: 99%

Localized Convolutional Neural Networks for Geospatial Wind Forecasting

2020

View full text Add to dashboard Cite

show abstract

“…Another one is the learning of characteristics in a common representation that is sufficiently expressive for all the tasks; in particular, for neuronal models, this can be enforced through a weight matrix which reflects task relationships and can facilitate subsequent learning [23]. Recently, MTL machine learning models have started to be used for several aspects of renewable energy prediction, such as ramp events [24], wind turbine output [25], solar panel outputs [26], or ultra-short PV production [27]. Similarly, Support Vector Regression has been widely used in renewable energy prediction [10,28].…”

Section: Literature Reviewmentioning

confidence: 99%

Multitask Support Vector Regression for Solar and Wind Energy Prediction

2020

View full text Add to dashboard Cite

Given the impact of renewable sources in the overall energy production, accurate predictions are becoming essential, with machine learning becoming a very important tool in this context. In many situations, the prediction problem can be divided into several tasks, more or less related between them but each with its own particularities. Multitask learning (MTL) aims to exploit this structure, training several models at the same time to improve on the results achievable either by a common model or by task-specific models. In this paper, we show how an MTL approach based on support vector regression can be applied to the prediction of photovoltaic and wind energy, problems where tasks can be defined according to different criteria. As shown experimentally with three different datasets, the MTL approach clearly outperforms the results of the common and specific models for photovoltaic energy, and are at the very least quite competitive for wind energy.

show abstract

“…A comparison between neural networks and GPs reveals no significant difference in terms of precision, but showcases the inherent ability of the GPs to produce probabilistic bounds. Woo et al [46] propose a Multi-Tasks Convolutional Long Short-Term Memory Network approach to simultaneously predict the energy output and structural load from the target wind turbine, while modeling the spatio-temporal structure of the input wind flow. The work is verified on simulations from a stand-alone NREL-5MW onshore reference wind turbine.…”

Section: Introductionmentioning

confidence: 99%

“…The predictions are delivered in a short-term horizon, i.e., few seconds ahead. In both [34,46] wake effects are not considered. Wake is tackled in [32], where a trained Variational Autoencoder (VAE) is exploited to map the high dimensional correlated stochastic variables over the wind-farm, such as power production and wind speed, to a parametric probability distribution of much lower dimensionality, with the ultimate goal of condition monitoring.…”

Section: Introductionmentioning

confidence: 99%

Virtual fatigue diagnostics of wake-affected wind turbine via Gaussian Process Regression

Avendaño-Valencia¹,

Abdallah²,

Chatzi³

2020

Preprint

View full text Add to dashboard Cite

We propose a data-driven model to predict the short-term fatigue Damage Equivalent Loads (DEL) on a wake-affected wind turbine based on wind field inflow sensors and/or loads sensors deployed on an adjacent up-wind wind turbine. Gaussian Process Regression (GPR) with Bayesian hyperparameters calibration is proposed to obtain a surrogate from input random variables to output DELs in the blades and towers of the up-wind and wake-affected wind turbines. A sensitivity analysis based on the hyperparameters of the GPR and Kullback-Leibler divergence is conducted to assess the effect of different input on the obtained DELs. We provide qualitative recommendations for a minimal set of necessary and sufficient input random variables to minimize the error in the DEL predictions on the wake-affected wind turbine. Extensive simulations are performed comprising different random variables, including wind speed, turbulence intensity, shear exponent and inflow horizontal skewness. Furthermore, we include random variables related to the blades lift and drag coefficients with direct impact on the rotor aerodynamic induction, which governs the evolution and transport of the meandering wake. In addition, different spacing between the wind turbines and Wöhler exponents for calculation of DELs are considered. The maximum prediction normalized mean squared error, obtained in the tower base DELs in the fore-aft direction of the wake affected wind turbine, is less than 4%. In the case of the blade root DELs, the overall prediction error is less than 1%. The proposed scheme promotes utilization of sparse structural monitoring (loads) measurements for improving diagnostics on wake-affected turbines.

show abstract

Predicting Wind Turbine Power and Load Outputs by Multi-task Convolutional LSTM Model

Cited by 23 publications

References 29 publications

Localized Convolutional Neural Networks for Geospatial Wind Forecasting

Localized Convolutional Neural Networks for Geospatial Wind Forecasting

Multitask Support Vector Regression for Solar and Wind Energy Prediction

Virtual fatigue diagnostics of wake-affected wind turbine via Gaussian Process Regression

Contact Info

Product

Resources

About