Using temperature and state transitions to forecast wind speed

A major issue in forecasting wind speed is non‐linear variability. The probability distribution of wind speed series shows heavy tails, while there are frequent state transitions, in which wind speed changes by large magnitudes, over relatively short time periods. These so‐called large ramp events are one of the critical issues currently facing the wind energy community. Two forecasting algorithms are analyzed here. The first is a regression on lags, including temperature as a causal factor, with time‐varying parameters. The second augments the first using state transition terms. The main innovation in state transition models is that the cumulative density function from regressions on the states is used as a right‐hand side variable in the regressions for wind speed. These two methods are tested against a persistence forecast and several non‐linear models, using eight hourly wind speed series. On average, these two models produce the best results. The state transition model improves slightly over the regression. However, the improvement achieved by both models relative to the persistence forecast is fairly small. These results argue that there are limits to the accuracy that can be achieved in forecasting wind speed data. Copyright © 2007 John Wiley & Sons, Ltd.     

Forecasting aggregated wind power production of multiple wind farms using hybrid wavelet‐PSO‐NNs

This paper describes the problem of short‐term wind power production forecasting based on meteorological information. Aggregated wind power forecasts are produced for multiple wind farms using a hybrid intelligent algorithm that uses a data filtering technique based on wavelet transform (WT) and a soft computing model (SCM) based on neural network (NN), which is optimized by using particle swarm optimization (PSO) algorithm. To demonstrate the effectiveness of the proposed hybrid intelligent WT + NNPSO model, which takes into account the interactions of wind power, wind speed, wind direction, and temperature in the forecast process, the real data of wind farms located in the southern Alberta, Canada, are used to train and test the proposed model. The test results produced by the proposed hybrid WT + NNPSO model are compared with other SCMs as well as the benchmark persistence method. Simulation results demonstrate that the proposed technique is capable of performing effectively with the variability and intermittency of wind power generation series in order to produce accurate wind power forecasts. Copyright © 2014 John Wiley & Sons, Ltd.     

Day-ahead wind speed forecasting using f-ARIMA models

With the integration of wind energy into electricity grids, it is becoming increasingly important to obtain accurate wind speed/power forecasts. Accurate wind speed forecasts are necessary to schedule dispatchable generation and tariffs in the day-ahead electricity market. This paper examines the use of fractional-ARIMA or f-ARIMA models to model, and forecast wind speeds on the day-ahead (24 h) and two-day-ahead (48 h) horizons. The models are applied to wind speed records obtained from four potential wind generation sites in North Dakota. The forecasted wind speeds are used in conjunction with the power curve of an operational (NEG MICON, 750 kW) turbine to obtain corresponding forecasts of wind power production. The forecast errors in wind speed/power are analyzed and compared with the persistence model. Results indicate that significant improvements in forecasting accuracy are obtained with the proposed models compared to the persistence method.     

Using nacelle‐based wind speed observations to improve power curve modeling for wind power forecasting

Wind power forecasting for projection times of 0–48 h can have a particular value in facilitating the integration of wind power into power systems. Accurate observations of the wind speed received by wind turbines are important inputs for some of the most useful methods for making such forecasts. In particular, they are used to derive power curves relating wind speeds to wind power production. By using power curve modeling, this paper compares two types of wind speed observations typically available at wind farms: the wind speed and wind direction measurements at the nacelles of the wind turbines and those at one or more on‐site meteorological masts (met masts). For the three Australian wind farms studied in this project, the results favor the nacelle‐based observations despite the inherent interference from the nacelle and the blades and despite calibration corrections to the met mast observations. This trend was found to be stronger for wind farm sites with more complex terrain. In addition, a numerical weather prediction (NWP) system was used to show that, for the wind farms studied, smaller single time‐series forecast errors can be achieved with the average wind speed from the nacelle‐based observations. This suggests that the nacelle‐average observations are more representative of the wind behavior predicted by an NWP system than the met mast observations. Also, when using an NWP system to predict wind farm power production, it suggests the use of a wind farm power curve based on nacelle‐average observations instead of met mast observations. Further, it suggests that historical and real‐time nacelle‐average observations should be calculated for large wind farms and used in wind power forecasting. Copyright © 2011 John Wiley & Sons, Ltd.     

Wind farm power prediction: a data‐mining approach

In this paper, models for short‐ and long‐term prediction of wind farm power are discussed. The models are built using weather forecasting data generated at different time scales and horizons. The maximum forecast length of the short‐term prediction model is 12 h, and the maximum forecast length of the long‐term prediction model is 84 h. The wind farm power prediction models are built with five different data mining algorithms. The accuracy of the generated models is analysed. The model generated by a neural network outperforms all other models for both short‐ and long‐term prediction. Two basic prediction methods are presented: the direct prediction model, whereby the power prediction is generated directly from the weather forecasting data, and the integrated prediction model, whereby the prediction of wind speed is generated with the weather data, and then the power is generated with the predicted wind speed. The direct prediction model offers better prediction performance than the integrated prediction model. The main source of the prediction error appears to be contributed by the weather forecasting data. Copyright © 2008 John Wiley & Sons, Ltd.     

Hybrid neurofuzzy wind power forecast and wind turbine location for embedded generation

Wind energy uptake in South Africa is significantly increasing both at the micro‐ and macro‐level and the possibility of embedded generation cannot be undermined considering the state of electricity supply in the country. This study identifies a wind hotspot site in the Eastern Cape province, performs an in silico deployment of three utility‐scale wind turbines of 60 m hub height each from different manufacturers, develops machine learning models to forecast very short‐term power production of the three wind turbine generators (WTG) and investigates the feasibility of embedded generation for a potential livestock industry in the area. Windographer software was used to characterize and simulate the net output power from these turbines using the wind speed of the potential site. Two hybrid models of adaptive neurofuzzy inference system (ANFIS) comprising genetic algorithm and particle swarm optimization (PSO) each for a turbine were developed to forecast very short‐term power output. The feasibility of embedded generation for typical medium‐scale agricultural industry was investigated using a weighted Weber facility location model. The analytical hierarchical process (AHP) was used for weight determination. From our findings, the WTG‐1 was selected based on its error performance metrics (root mean square error of 0.180, mean absolute SD of 0.091 and coefficient of determination of 0.914 and CT = 702.3 seconds) in the optimal model (PSO‐ANFIS). Criteria were ranked based on their order of significance to the agricultural industry as proximity to water supply, labour availability, power supply and road network. Also, as a proof of concept, the optimal location of the industrial facility relative to other criteria was X = 19.24 m, Y = 47.11 m . This study reveals the significance of resource forecasting and feasibility of embedded generation, thus improving the quality of preliminary resource assessment and facility location among site developers.     

风电功率预测技术的应用现状及运行建议

针对我国风电开发中遇到的风电接入困难、电网调度困难等问题,对国内外解决此类问题的风电功率预测技术进行了详细的阐述。指出我国急需开发风电功率预测系统,并根据我们的风电功率预测研究经验提出了我国风电功率预测宜采用风电企业和电网共同实施的运行模式。希望对我国的风电功率预测发展起到一定的促进作用。     

Regime‐switching models and multiple causal factors in forecasting wind speed

This study evaluates two types of models for wind speed forecasting. The first is models with multiple causal factors, such as offsite readings of wind speed and meteorological variables. These can be estimated using either regressions or neural networks. The second is state transition and the closely related class of regime‐switching transition models. These are attractive in that they can be used to predict outlying fluctuations or large ramp events. The regime‐switching model uses a persistence forecast during periods of high wind speed, and regressions for low and intermediate speeds. These techniques are tested on three databases. Two main criteria are used to evaluate the outcomes, the number of high and low states than can be predicted correctly and the mean absolute percent error of the forecast. Neural nets are found to predict the state transitions somewhat better than logistic regressions, although the regressions do not do badly. Three methods all achieve about the same degree of forecast accuracy: multivariate regressions, state transition and regime‐switching models. If the states could be predicted perfectly, the regime‐switching model would improve forecast accuracy by an additional 2.5 to 3 percentage points. Analysis of the density functions of wind speed and the forecasting models finds that the regime‐switching method more closely approximates the distribution of the actual data. Copyright © 2009 John Wiley & Sons, Ltd.     

Comparative performance of AI methods for wind power forecast in Portugal

Because wind has a high volatility and the respective energy produced cannot be stored on a large scale because of excessive costs, it is of utmost importance to be able to forecast wind power generation with the highest accuracy possible. The aim of this paper is to compare 1‐h‐ahead wind power forecasts performance using artificial intelligence‐based methods, such as artificial neural networks (ANNs), adaptive neural fuzzy inference system (ANFIS), and radial basis function network (RBFN). The latter was implemented using three different learning algorithms: stochastic gradient descent (SGD), hybrid, and orthogonal least squares (OLS). The application dataset is the injected wind power in the Portuguese power systems throughout the years 2010–2014. The network architecture optimization and the learning algorithms are presented. An initial data analysis showed data seasonality; therefore, the wind power forecasts were performed according to the seasons of the year. The results showed that ANFIS was the best performer method, and ANN and RBFN‐OLS also showed strong performances. RBFN‐Hybrid and RBFN‐SGD performed poorly. In general, all methods outperformed persistence.     

Characterizing future large,rapid changes in aggregated wind power using Numerical Weather Prediction spatial fields

A critical limiting factor to the successful deployment of a large proportion of wind power in power systems is its predictability. Power system operators play a vital role in maintaining system security, and this task is greatly aided by useful characterizations of future system operations. A wind farm power forecast generally relies on the forecast output from a Numerical Weather Prediction (NWP) model, typically at a single grid point in the model to represent the wind farm's physical location. A key limitation of this approach is the spatial misplacement of weather features often found in NWP forecasts. This paper presents a methodology to display wind forecast information from multiple grid points at hub height around the wind farm location. If the raw forecast wind speeds at hub height at multiple grid points were to be displayed directly, they would be misleading as the NWP outputs take account of the estimated local surface roughness and terrain at each grid point. Hence, the methodology includes a transformation of the wind speed at each grid point to an equivalent value that represents the surface roughness and terrain at the chosen single grid point for the wind farm site. The chosen‐grid‐point‐equivalent wind speeds for the wind farm can then be transformed to available wind farm power. The result is a visually‐based decision support tool which can help the forecast user to assess the possibilities of large, rapid changes in available wind power from wind farms. A number of methods for displaying the field for multiple wind farms are discussed. The chosen‐grid‐point‐equivalent transformation also has other potential applications in wind power forecasting such as assessing deterministic forecast uncertainty and improving downscaling results. Copyright © 2008 John Wiley & Sons, Ltd.     

Application of mesoscale ensemble forecast method for prediction of wind speed ramps

Sudden changes in wind speed, so‐called wind speed ramps, are a major concern for wind power system operators. The present study applies the mesoscale ensemble forecast method for the prediction of wind speed ramps at wind farms in Japan and evaluates the ability and utility of this method. The mesoscale ensemble forecast in this study (ENS21) consists of 21 members with a horizontal resolution of 10 km for a 5‐year period. The simulated results show that ENS21 produces better accuracy than the deterministic forecast with a horizontal resolution of 10 km (DET_L). On the other hand, the deterministic forecast with a horizontal resolution of 5 km (DET_H) also produces better accuracy than DET_L. From a practical perspective, however, the ENS21 is computationally expensive. Thus, the eight‐member mesoscale ensemble forecast (ENS8) with as same computational cost as a deterministic forecast with a horizontal resolution of 5 km (DET_H) is also evaluated. The simulated results show that ENS8 has almost same accuracy as ENS21 and DET_H in wind speed ramp forecasts. ENS8 has advantages over ENS21 and DET_H because ENS8 is computationally efficient and is able to benefit wind power operators with flexibility in the selection of probability thresholds for decision processes compared with a single. It can be concluded that the mesoscale ensemble forecast method is more useful for prediction of the wind speed ramp than the single deterministic forecast method with the same computational cost if the ensemble members are successfully selected.     

Modeling and forecasting the mean hourly wind speed time series using GMDH-based abductive networks

Wind speed forecasts are important for the operation and maintenance of wind farms and their profitable integration into power grids, as well as many important applications in shipping, aviation, and the environment. Modern machine learning techniques including neural networks have been used for this purpose, but it has proved hard to make significant improvements on the performance of the simple persistence model. As an alternative approach, we propose here the use of abductive networks, which offer the advantages of simplified and more automated model synthesis and transparent analytical input–output models. Various abductive models for predicting the mean hourly wind speed 1 h ahead have been developed using wind speed data at Dhahran, Saudi Arabia during the month of May over the years 1994–2005. The models were evaluated on the data for May 2006. Models described include a single generic model to forecast next-hour speed from the previous 24 hourly measurements and an hour index, which give an overall mean absolute error (MAE) of 0.85 m/s and a correlation coefficient of 0.83 between actual and predicted values. The model achieves an improvement of 8.2% reduction in MAE compared to hourly persistence. The above model was used iteratively to forecast the hourly wind speed 6 h and 24 h ahead at the end of a given day, with MAEs of 1.20 m/s and 1.42 m/s which are lower than forecasting errors based on day-to-day persistence by 14.6% and 13.7%. Relative improvements on persistence exceed those reported for several machine learning approaches reported in the literature.     

基于人工神经网络模型的风速预测

利用在达坂城风电场30m轮毂高处的1min实测风速数据,采用人工神经网络模型ANN对未来短时间风速进行预报。通过对风速反复训练与检测来选择一组合适的模型参数,并对模型进行了误差分析。研究结果表明,使用BP神经网络对未来风速进行短时间预测能够达到较好的效果,误差较ARMA模型更精确,但是对于突变信息的处理能力仍然有限。     

Short‐term forecasting of categorical changes in wind power with Markov chain models

As penetrations of renewable wind energy increase, accurate short‐term predictions of wind power become crucial to utilities that must balance the load and supply of electricity. As storage of wind energy is not yet feasible on a large scale, the utility must integrate wind energy as soon as it is generated and decide at each balancing time‐step whether a change in conventional energy output is required. With high penetrations of wind energy, utilities must also plan for operating reserves to maintain stability of the electricity system when forecasts for renewable energy are inaccurate. Thus, a simple forecast of whether the wind power will increase, decrease or not change in the next time‐step will give utility operators an easy tool for assessing whether changes need to be made to the current generation mix. In this work, Markov chain models based on the change in power output at up to three locations or lags in time are presented that not only produce such an hourly forecast but also include a measure of the uncertainty of the forecast. Forecasts are greatly improved when knowledge of whether the maximum or minimum wind power is currently being produced and the intrahour trend in wind power are incorporated. These models are trained, tested and evaluated with a uniquely long set of 2 years of 10 min measurements at four meteorological stations in the Pacific Northwest and perform better than a benchmark state‐of‐the‐art wind speed forecasting model.Copyright © 2013 John Wiley & Sons, Ltd.     

Improving wind speed forecasts from the Weather Research and Forecasting model at a wind farm in the semiarid Coquimbo region in central Chile

Accurate predictions of the wind field are key for better wind power forecasts. Wind speed forecasts from numerical weather models present differences with observations, especially in places with complex topography, such as the north of Chile. The present study has two goals: (a) to find the WRF model boundary layer (PBL) scheme that best reproduces the observations at the Totoral Wind Farm, located in the semiarid Coquimbo region in north‐central Chile, and (b) to use an artificial neural network (ANN) to postprocess wind speed forecasts from different model domains to analyze the sensitivity to horizontal resolution. The WRF model was run with three different PBL schemes (MYNN, MYNN3, and QNSE) for 2013. The WRF simulation with the QNSE scheme showed the best agreement with observations at the wind farm, and its outputs were postprocessed using two ANNs with two algorithms: backpropagation (BP) and particle swarm optimization (PSO). These two ANNs were applied to the innermost WRF domains with 3‐km (d03) and 1‐km (d04) horizontal resolutions. The root‐mean‐square errors (RMSEs) between raw WRF forecasts and observations for d03 and d04 were 2.7 and 2.4 ms^?1 , respectively. When both ANN models (BP and PSO) were applied to Domains d03 and d04, the RMSE decreased to values lower than 1.7 ms^?1 , and they showed similar performances, supporting the use of an ANN to postprocess a three‐nested WRF domain configuration to provide more accurate forecasts in advance for the region.     

A review on the forecasting of wind speed and generated power

In the world, wind power is rapidly becoming a generation technology of significance. Unpredictability and variability of wind power generation is one of the fundamental difficulties faced by power system operators. Good forecasting tools are urgent needed under the relevant issues associated with the integration of wind energy into the power system. This paper gives a bibliographical survey on the general background of research and developments in the fields of wind speed and wind power forecasting. Based on the assessment of wind power forecasting models, further direction for additional research and application is proposed.     

A new strategy for predicting short-term wind speed using soft computing models

Wind power prediction is a widely used tool for the large-scale integration of intermittent wind-powered generators into power systems. Given the cubic relationship between wind speed and wind power, accurate forecasting of wind speed is imperative for the estimation of future wind power generation output. This paper presents a performance analysis of short-term wind speed prediction techniques based on soft computing models (SCMs) formulated on a backpropagation neural network (BPNN), a radial basis function neural network (RBFNN), and an adaptive neuro-fuzzy inference system (ANFIS). The forecasting performance of the SCMs is augmented by a similar days (SD) method, which considers similar historical weather information corresponding to the forecasting day in order to determine similar wind speed days for processing. The test results demonstrate that all evaluated SCMs incur some level of performance improvement with the addition of SD pre-processing. As an example, the SD+ANFIS model can provide up to 48% improvement in forecasting accuracy when compared to the individual ANFIS model alone.     

Forecasting model for wind power integrating least squares support vector machine,singular spectrum analysis,deep belief network,and locality‐sensitive hashing

Accurate wind power prediction can alleviate the negative influence on power system caused by the integration of wind farms into grid. In this paper, a novel combination model is proposed with the purpose of enhancing short‐term wind power prediction precision. Singular spectrum analysis is utilized to decompose the original wind power series into the trend component and the fluctuation component. Then least squares support vector machine (LSSVM) is applied to forecast the trend component while deep belief network (DBN) is utilized to predict the fluctuation component. By this means, the performance advantages of LSSVM and DBN can be brought into full play. Moreover, the locality‐sensitive hashing search algorithm is introduced to cluster the nearest training samples to further improve forecasting accuracy. Besides, the effect of LSSVM based on different kernel functions and the number of the nearest samples is investigated. The simulation results show that the normalized root mean square errors of the proposed model based on linear kernel function from 1‐step to 3‐step forecasting are 2.13%, 5.03%, and 7.29%, respectively, which outperforms all the other comparison models. Therefore, it can be concluded that the proposed combination model provides a promising and effective alternative for short‐term wind power prediction.     

A general method to estimate wind farm power using artificial neural networks

An artificial neural network (ANN) is trained and validated using a large dataset of observations of wind speed, direction, and power generated at an offshore wind farm (Lillgrund in Sweden). In its traditional form, the ANN is used to generate a new two‐dimensional power curve, which predicts with high accuracy (bias ～?0.5% and absolute error ～2%) the power of the entire Lillgrund wind farm based on wind speed and direction. By contrast, manufacturers only provide one‐dimensional power curves (i.e., power as a function of wind speed) for a single turbine. The second innovative application of the ANN is the use of a geometric model (GM) to calculate two simple geometric properties to replace wind direction in the ANN. The resulting GM‐ANN has the powerful feature of being applicable to any wind farm, not just Lillgrund. A validation at an onshore wind farm (Nørrekær in Denmark) demonstrates the high accuracy (bias ～?0.7% and absolute error ～6%) and transfer‐learning ability of the GM‐ANN.