Comparative analysis of methods for modelling the short‐term probability distribution of extreme wind turbine loads

We have tested the performance of statistical extrapolation methods in predicting the extreme response of a multi‐megawatt wind turbine generator. We have applied the peaks‐over‐threshold, block maxima and average conditional exceedance rates (ACER) methods for peaks extraction, combined with four extrapolation techniques: the Weibull, Gumbel and Pareto distributions and a double‐exponential asymptotic extreme value function based on the ACER method. For the successful implementation of a fully automated extrapolation process, we have developed a procedure for automatic identification of tail threshold levels, based on the assumption that the response tail is asymptotically Gumbel distributed. Example analyses were carried out, aimed at comparing the different methods, analysing the statistical uncertainties and identifying the factors, which are critical to the accuracy and reliability of the extrapolation. The present paper describes the modelling procedures and makes a comparison of extrapolation methods based on the results from the example calculations. Copyright © 2015 John Wiley & Sons, Ltd.     

Short‐term forecasting of surface layer wind speed using a continuous random cascade model

This paper describes a statistical method for short‐term forecasting (1–12 h ahead) of surface layer wind speed using only recent observations, relying on the notion of continuous cascades. Inspired by recent empirical findings that suggest the existence of some cascading process in the mesoscale range, we consider that wind speed can be described by a seasonal component and a fluctuating part represented by a ‘multifractal noise’ associated with a random cascade. Performances of our model are tested on hourly wind speed series gathered at various locations in Corsica (France) and the Netherlands. The obtained results show that a better modeling of the noise term based on cascade process enhances the forecast; furthermore, there is a systematic improvement in the prediction as compared with reference models. Copyright © 2011 John Wiley & Sons, Ltd.     

Implementation of a Model Output Statistics based on meteorological variable screening for short‐term wind power forecast

A combination of physical and statistical treatments to post‐process numerical weather predictions (NWP) outputs is needed for successful short‐term wind power forecasts. One of the most promising and effective approaches for statistical treatment is the Model Output Statistics (MOS) technique. In this study, a MOS based on multiple linear regression is proposed: the model screens the most relevant NWP forecast variables and selects the best predictors to fit a regression equation that minimizes the forecast errors, utilizing wind farm power output measurements as input. The performance of the method is evaluated in two wind farms, located in different topographical areas and with different NWP grid spacing. Because of the high seasonal variability of NWP forecasts, it was considered appropriate to implement monthly stratified MOS. In both wind farms, the first predictors were always wind speeds (at different heights) or friction velocity. When friction velocity is the first predictor, the proposed MOS forecasts resulted to be highly dependent on the friction velocity–wind speed correlation. Negligible improvements were encountered when including more than two predictors in the regression equation. The proposed MOS performed well in both wind farms, and its forecasts compare positively with an actual operative model in use at Risø DTU and other MOS types, showing minimum BIAS and improving NWP power forecast of around 15% in terms of root mean square error. Further improvements could be obtained by the implementation of a more refined MOS stratification, e.g. fitting specific equations in different synoptic situations. Copyright © 2012 John Wiley & Sons, Ltd.     

Short‐term spatio‐temporal prediction of wind speed and direction

This paper aims to produce a low‐complexity predictor for the hourly mean wind speed and direction from 1 to 6 h ahead at multiple sites distributed around the UK. The wind speed and direction are modelled via the magnitude and phase of a complex‐valued time series. A multichannel adaptive filter is set to predict this signal on the basis of its past values and the spatio‐temporal correlation between wind signals measured at numerous geographical locations. The filter coefficients are determined by minimizing the mean square prediction error. To account for the time‐varying nature of the wind data and the underlying system, we propose a cyclo‐stationary Wiener solution, which is shown to produce an accurate predictor. An iterative solution, which provides lower computational complexity, increased robustness towards ill‐conditioning of the data covariance matrices and the ability to track time‐variations in the underlying system, is also presented. The approaches are tested on wind speed and direction data measured at various sites across the UK. Results show that the proposed techniques are able to predict wind speed as accurately as state‐of‐the‐art wind speed forecasting benchmarks while simultaneously providing valuable directional information. Copyright © 2013 John Wiley & Sons, Ltd.     

Prediction of the extreme wind speed in the mixed climate region by using Monte Carlo simulation and measure‐correlate‐predict method

The extreme wind speed at an offshore location was predicted using Monte Carlo simulation (MCS) and measure‐correlate‐predict (MCP) method. The Gumbel distribution could successfully express the annual maximum wind speed of extratropical cyclone. On the other hand, the estimated extreme wind speed of tropical cyclones by analytical probability distribution shows larger uncertainty. In the mixed climate like Japan, the extreme wind speed estimated from the combined probability distribution obtained by MCP and MCS methods agrees well with the observed data as compared with the combined probability distribution obtained by the MCP method only. The uncertainty of extreme wind speed due to limited observation period of wind speed and pressure was also evaluated by the Gumbel theory and Monte Carlo simulation. As a result, it was found that the uncertainty of 50 year recurrence wind speed obtained by MCS method is considerably smaller than that obtained by MCP method in the mixed climate. Copyright © 2014 John Wiley & Sons, Ltd.     

Ultra short‐term probability prediction of wind power based on LSTM network and condition normal distribution

Considering the inevitable prediction errors in the traditional point predictions of wind power, in this paper, a new ultra short‐term probability prediction method for wind power is proposed, in which the long short‐term memory (LSTM) network, wavelet decomposition (WT), and principal component analysis (PCA) are combined together for ultra short‐term probability prediction of wind power, a conditional normal distribution model that is developed to describe the uncertainty of prediction errors. First, WT and PCA are jointly used to smooth the original time series, then the point prediction model for subsequence data based on LSTM network is proposed. It is worth pointing out that the input matrix of the model includes many features, such as wind power and wind speed, which will be helpful for improving prediction performance. After optimizing the index of the ultra short‐term probability prediction interval (PI) of wind power by particle swarm optimization (PSO), the conditional normal distribution model of prediction errors is developed. Thus, the ultra short‐term PIs for wind power are obtained. Finally, based on the data of two wind farms in China, simulation results are provided to illustrate the usefulness of the proposed prediction model. It follows from those results that the proposed method can improve the accuracy of prediction, and the reliability of probability prediction for wind power is also improved.     

Machine learning methods for short‐term bid forecasting in the renewable energy market: A case study in Italy

In liberalized markets, there usually exists a day‐ahead session where energy is sold and acquired for the following production day. Owing to the high uncertainty of its production, renewable energy (wind in particular) can significantly influence the network imbalance of the following day. In this work, we consider the problem of predicting the sum of the bid volumes for wind energy of all the producers inside the day‐ahead energy market. This is a valuable tool to be used by an energy provider in order to determine the imbalance of a market zone and, thus, properly size its bids. In particular, we focus on the estimation of the possible relationship between the meteorological forecasts and the wind power offered on the market by the companies for a market zone. We propose a machine learning model which is used to compute a 1‐day‐ahead forecast. The input‐output mapping is obtained by support vector regression. The input feature vector is defined by a suitable feature extraction technique since the meteorological forecasts are given on a lattice of thousands of geographical points. The computational experiments are performed considering the Italian market as a case study (years 2012‐2016). The results show that the proposed feature extraction technique, selecting only some geographical zones, manages to reduce the error attained using all the features. Moreover, classical statistical methods are shown to be outperformed by machine learning models. The analysis reveals also some weaknesses of the model, which may be due to other nonmeteorological factors at play.     

Comparison of simulators for variable‐speed wind turbine transient analysis

This paper presents a comparison of three variable‐speed wind turbine simulators used for a 2 MW wind turbine short‐term transient behaviour study during a symmetrical network disturbance. The simulator with doubly fed induction generator (DFIG) analytical model, the simulator with a finite element method (FEM) DFIG model and the wind turbine simulator with an analytical model of DFIG are compared. The comparison of the simulation results shows the influence of the different modelling approaches on the short‐term transient simulation accuracy. Copyright © 2005 John Wiley & Sons, Ltd.     

Evaluation of the Weather Research and Forecasting model on forecasting low‐level jets: implications for wind energy

Nocturnal low‐level jet (LLJ) events are commonly observed over the Great Plains region of the USA, thus making this region more favorable for wind energy production. At the same time, the presence of LLJs can significantly modify vertical shear and nocturnal turbulence in the vicinities of wind turbine hub height, and therefore has detrimental effects on turbine rotors. Accurate numerical modeling and forecasting of LLJs are thus needed for precise assessment of wind resources, reliable prediction of power generation and robust design of wind turbines. However, mesoscale numerical weather prediction models face a challenge in precisely forecasting the development, magnitude and location of LLJs. This is due to the fact that LLJs are common in nocturnal stable boundary layers, and there is a general consensus in the literature that our contemporary understanding and modeling capability of this boundary‐layer regime is quite poor. In this paper, we investigate the potential of the Weather Research and Forecasting (WRF) model in forecasting LLJ events over West Texas and southern Kansas. Detailed observational data from both cases were used to assess the performance of the WRF model with different model configurations. Our results indicate that the WRF model can capture some of the essential characteristics of observed LLJs, and thus offers the prospect of improving the accuracy of wind resource estimates and short‐term wind energy forecasts. However, the core of the LLJ tended to be higher as well as slower than what was observed, leaving room for improvement in model performance. Copyright © 2008 John Wiley & Sons, Ltd.     

基于CEEMDAN与CBBO-SVM的汽轮机转子故障诊断研究

为了提高汽轮机转子故障诊断的准确率和识别效率,提出了一种基于CEEMDAN(自适应噪声完备集合经验模态分解)和CBBO(混沌生物地理学优化算法)优化SVM(支持向量机)相结合的故障诊断方法。首先利用CEEMDAN对转子振动信号进行分解,提取PE(排列熵)作为故障特征值,并构造特征向量;其次将混沌理论引入到BBO(生物地理学优化算法)中,得到CBBO,通过CBBO优化SVM得到诊断模型的最优参数。最后通过ZT-3转子试验台模拟汽轮机转子故障,利用得到的4种状态下的试验数据验证优化模型的有效性与先进性。结果表明:CBBO优化SVM模型可以准确、高效地对汽轮机转子进行故障诊断;与CPSO(混沌粒子群算法)优化SVM模型相比,该方法的故障诊断准确率和识别效率更高。