On comparing three artificial neural networks for wind speed forecasting

Wind speed forecasting is critical for wind energy conversion systems since it greatly influences the issues such as the scheduling of a power system, and the dynamic control of the wind turbine. In this paper, we present a comprehensive comparison study on the application of different artificial neural networks in 1-h-ahead wind speed forecasting. Three types of typical neural networks, namely, adaptive linear element, back propagation, and radial basis function, are investigated. The wind data used are the hourly mean wind speed collected at two observation sites in North Dakota. The performance is evaluated based on three metrics, namely, mean absolute error, root mean square error, and mean absolute percentage error. The results show that even for the same wind dataset, no single neural network model outperforms others universally in terms of all evaluation metrics. Moreover, the selection of the type of neural networks for best performance is also dependent upon the data sources. Among the optimal models obtained, the relative difference in terms of one particular evaluation metric can be as much as 20%. This indicates the need of generating a single robust and reliable forecast by applying a post-processing method.     

A classification mechanism for determining average wind speed and power in several regions of Turkey using artificial neural networks

In this paper, average wind speed and wind power values are estimated using artificial neural networks (ANNs) in seven regions of Turkey. To start with, a network has been set up, and trained with the data set obtained from several stations—each station gather data from five different heights—from each region, one randomly selected height value of a station has been used as test data. Wind data readings corresponding to the last 50 years of relevant regions were obtained from the Turkish State Meteorological Service (TSMS). The software has been developed under Matlab 6.0. In the input layer, longitude, latitude, altitude, and height are used, while wind speeds and related power values correspond to output layer. Then we have used the networks to make predictions for varying heights, which are not incorporated to the system at the training stage. The network has successfully predicted the required output values for the test data and the mean error levels for regions differed between 3% and 6%. We believe that using ANNs average wind speed and wind power of a region can be predicted provided with lesser amount of sampling data, that the sampling mechanism is reliable and adequate.     

The application of artificial neural networks to mapping of wind speed profile for energy application in Nigeria

Modelling and prediction of wind speed are essential prerequisites in the sitting and sizing of wind power applications. The profile of wind speed in Nigeria is modelled using artificial neural network (ANN). The ANN model consists of 3-layered, feed-forward, back-propagation network with different configurations, designed using the Neural Toolbox for MATLAB. The monthly mean daily wind speed data monitored at 10 m above ground level for a period of 20 years (1983–2003) for 28 ground stations operated by the Nigeria Meteorological Services (NIMET) were used as training (18 stations) and testing (10 stations) dataset. The geographical parameters (latitude, longitude and altitude) and the month of the year were used as input data, while the monthly mean wind speed was used as the output of the network. The optimum network architecture with minimum Mean Absolute Percentage Error (MAPE) of 8.9% and correlation coefficient (r) between the predicted and the measured wind speed values of 0.9380 was obtained. The predicted monthly wind speed ranged from 0.9–13.1 m/s with an annual mean of 4.7 m/s. The model predicted wind speed values are given in the form of monthly maps, which can be easily used for assessment of wind energy potential for different locations within Nigeria.     

Transfer learning for short-term wind speed prediction with deep neural networks

As a type of clean and renewable energy source, wind power is widely used. However, owing to the uncertainty of wind speed, it is essential to build an accurate forecasting model for large-scale wind power penetration. Numerical weather prediction (NWP) and data-driven modeling are two typical paradigms. NWP is usually unavailable or spatially insufficient. Data-driven modeling is an effective candidate. As to some newly-built wind farms, sufficient historical data is not available for training an accurate model, while some older wind farms may have long-term wind speed records. A question arises regarding whether the prediction model trained by data coming from older farms is also effective for a newly-built farm. In this paper, we propose an interesting trial of transferring the information obtained from data-rich farms to a newly-built farm. It is well known that deep learning can extract a high-level representation of raw data. We introduce deep neural networks, trained by data from data-rich farms, to extract wind speed patterns, and then finely tune the mapping with data coming from newly-built farms. In this way, the trained network transfers information from one farm to another. The experimental results show that prediction errors are significantly reduced using the proposed technique.     

Short term wind speed forecasting for wind turbine applications using linear prediction method

In this paper a new method, based on linear prediction, is proposed for wind speed forecasting. The method utilizes the ‘linear prediction’ method in conjunction with ‘filtering’ of the wind speed waveform. The filtering eliminates the undesired parts of the frequency spectrum (i.e. smoothing) of the measured wind speed which is less effective in an application, for example, in a wind energy conversion system. The linear prediction method is intuitively explained with some easy to follow case studies to clarify the complex underlying mathematics. For verification purposes, the proposed method is compared with real wind speed data based on experimental results. The results show the effectiveness of the linear prediction method.     

Assessment of optimum tip speed ratio in wind turbines using artificial neural networks

Wind turbine blade design depends on several factors, such as turbine profile used, blade number, power factor, and tip speed ratio. The key to designing a wind turbine is to assess the optimal tip speed ratio (TSR). This will directly affect the power generated and, in turn, the effectiveness of the investment made. TSR is suggested to be taken between 7 and 8 and in practice generally taken as 7 for a 3-blade network-connected wind turbine. However, the optimal TSR is dependent upon the profile type used and the blade number and could fall out of the boundaries suggested. Therefore, it has to be assessed accordingly. In this study, the optimal TSR and the power factor of a wind turbine are predicted using artificial neural networks (ANN) based on the parameters involved for NACA 4415 and LS-1 profile types with 3 and 4 blades. The ANN structure built is found to be more successful than the conventional approach in estimating the TSR and power factor.     

Effect of direction on wind speed estimation in complex terrain using neural networks

A method of estimating the annual average wind speed at a selected site using neural networks is presented. The method proposed uses only a few measurements taken at the selected site in a short time period and data collected at nearby fixed stations.The neural network used in this study is a multilayer perceptron with one hidden layer of 15 neurons, trained by the Bayesian regularization algorithm. The number of inputs that must be used in the neural network was analyzed in detail, and results suggest that only wind speed and direction data for a single station are required. In sites of complex terrain, direction is a very important input that can cause a decrease of 23% in root mean square (RMS).The results obtained by simulating the annual average wind speed at the selected site based on data from nearby stations are satisfactory, with errors below 2%.     

Most influential parametrical and data needs for realistic wind speed prediction

Depleting fossil fuel reserves and increasing global weather concerns has diverted mankind to look out for clean and green reserves of energy ever since the beginning of last decade. Wind holds a major role in satisfying our energy needs, however, its use as an alternate power source accounts for various issues such as deregulation of supply, frequency instability, etc. In order to nullify such effects, power engineers need to have an idea of futuristic weather conditions, especially the wind speed trend. Numerical Weather Prediction (NWP) tools such as Yearly Auto-Regressive (YAR) models when deployed for medium-term wind speed forecasting have proved their effectiveness. In this paper Artificial Neural Network based Yearly Auto-Regressive (ANNYAR) model have been used to figure out the most influential parameter's affecting wind prediction and corresponding range of yearly data set required for Time Horizon (TH) extending from 6 to 96 h. Data from area in and around ‘VABB airfield Mumbai’ has been incorporated for modelling and analysis purpose.     

Analysis of wind power generation and prediction using ANN: A case study

Many developing nations, such as India have embarked upon wind energy programs for areas experiencing high average wind speeds throughout the year. One of the states in India that is actively pursuing wind power generation programs is Tamil Nadu. Within this state, Muppandal area is one of the identified regions where wind farm concentration is high. Wind energy engineers are interested in studies that aim at assessing the output of wind farms, for which, artificial intelligence techniques can be usefully adapted. The present paper attempts to apply this concept for assessment of the wind energy output of wind farms in Muppandal, Tamil Nadu (India). Field data are collected from seven wind farms at this site over a period of 3 years from April 2002 to March 2005 and used for the analysis and prediction of power generation from wind farms. The model has been developed with the help of neural network methodology. It involves three input variables—wind speed, relative humidity and generation hours and one output variable-energy output of wind farms. The modeling is done using MATLAB toolbox. The model accuracy is evaluated by comparing the simulated results with the actual measured values at the wind farms and is found to be in good agreement.     

A novel application of artificial neural network for wind speed estimation

Providing accurate multi-steps wind speed estimation models has increasing significance, because of the important technical and economic impacts of wind speed on power grid security and environment benefits. In this study, the combined strategies for wind speed forecasting are proposed based on an intelligent data processing system using artificial neural network (ANN). Generalized regression neural network and Elman neural network are employed to form two hybrid models. The approach employs one of ANN to model the samples achieving data denoising and assimilation and apply the other to predict wind speed using the pre-processed samples. The proposed method is demonstrated in terms of the predicting improvements of the hybrid models compared with single ANN and the typical forecasting method. To give sufficient cases for the study, four observation sites with monthly average wind speed of four given years in Western China were used to test the models. Multiple evaluation methods demonstrated that the proposed method provides a promising alternative technique in monthly average wind speed estimation.     

State-of-charge prediction of batteries and battery-supercapacitor hybrids using artificial neural networks

The state-of-charge (SOC) of batteries and battery-supercapacitor hybrid systems is predicted using artificial neural networks (ANNs). Our technique is able to predict the SOC of energy storage devices based on a short initial segment (less than 4% of the average lifetime) of the discharge curve. The prediction shows good performance with a correlation coefficient above 0.95. We are able to improve the prediction further by considering readily available measurements of the device and usage. The prediction is further shown to be resilient to changes in operating conditions or physical structure of the devices.     

Application of artificial neural networks for the prediction of performance and exhaust emissions in SI engine using ethanol- gasoline blends

This study deals with artificial neural network (ANN) modeling of a spark ignition engine to predict the engine brake power, output torque and exhaust emissions (CO, CO₂, NO_x and HC) of the engine. To acquire data for training and testing of the proposed ANN, a four-cylinder, four-stroke test engine was fuelled with ethanol-gasoline blended fuels with various percentages of ethanol (0, 5, 10,15 and 20%), and operated at different engine speeds and loads. An ANN model based on standard back-propagation algorithm for the engine was developed using some of the experimental data for training. The performance of the ANN was validated by comparing the prediction dataset with the experimental results. Results showed that the ANN provided the best accuracy in modeling the emission indices with correlation coefficient equal to 0.98, 0.96, 0.90 and 0.71 for CO, CO₂, HC and NO_x, and 0.99 and 0.96 for torque and brake power respectively. Generally, the artificial neural network offers the advantage of being fast, accurate and reliable in the prediction or approximation affairs, especially when numerical and mathematical methods fail.     

Application of a control algorithm for wind speed prediction and active power generation

The main objective of the work described in this paper is to offer a new method of prediction of wind speeds, whilst aware that the method develops predictions in time-scales that can vary from a few minutes to an hour. This is needed because wind energy generation is increasing its participation in energy distribution and has to compete with other energy sources that are not so variable in terms of generated active power. It is important to consider that active power demand can vary quite rapidly and different sources of electricity generation must be available. In the case of wind energy, wind speed predictions are an important tool to help producers make the best decisions when selling the energy produced. These decisions are crucial in the electricity market, because of the economic benefits for producers and consequently their profitability, depends on them. The algorithm presented in this paper is based on an artificial neural network and two types of wind data have been used to test the algorithm. In the first, data was collected from a not very windy area; in the second data was collected from a real wind farm located in Navarre (North of Spain), and the values vary from very low to high speeds. Although the algorithm was not tested with typical wind speed values measured on offshore wind farm applications, it can be concluded from the first set of results presented in this paper that the algorithm is valid for estimating average speed values. Finally, a generic algorithm for the active power generation of a wind farm is presented.     

Application of artificial neural networks for modeling of biohydrogen production

In this study, an artificial neural network (ANN) model was developed to estimate the hydrogen production profile with time in batch studies. A back propagation artificial neural network ANN configuration of 5–6–4–1 layers was developed. The ANN inputs were the initial pH, initial substrate and biomass concentrations, temperature, and time. The model training was done using 313 data points from 26 published experiments. The correlation coefficient between the experimental and estimated hydrogen production was 0.989 for training, validating, and testing the model. Results showed that the trained ANN successfully predicted the hydrogen production profile with time for new data with a correlation coefficient of 0.976.     

A novel online training neural network-based algorithm for wind speed estimation and adaptive control of PMSG wind turbine system for maximum power extraction

In this paper, an adaptive control scheme for maximum power point tracking of stand-alone PMSG wind turbine systems (WTS) is presented. A novel procedure to estimate the wind speed is derived. To achieve this, a neural network identifier (NNI) is designed in order to approximate the mechanical torque of the WTS. With this information, the wind speed is calculated based on the optimal mechanical torque point. The NNI approximates in real-time the mechanical torque signal and it does not need off-line training to get its optimal parameter values. In this way, it can really approximates any mechanical torque value with good accuracy. In order to regulate the rotor speed to the optimal speed value, a block-backstepping controller is derived. Uniform asymptotic stability of the tracking error origin is proved using Lyapunov arguments. Numerical simulations and comparisons with a standard passivity based controller are made in order to show the good performance of the proposed adaptive scheme.     

Performance prediction of a direct expansion solar assisted heat pump using artificial neural networks

This paper presents the suitability of artificial neural network (ANN) to predict the performance of a direct expansion solar assisted heat pump (DXSAHP). The experiments were performed under the meteorological conditions of Calicut city (latitude of 11.15 °N, longitude of 75.49 °E) in India. The performance parameters such as power consumption, heating capacity, energy performance ratio and compressor discharge temperature of a DXSAHP obtained from the experimentation at different solar intensities and ambient temperatures are used as training data for the network. The back propagation learning algorithm with three different variants (such as, Lavenberg–Marguardt (LM), scaled conjugate gradient (SCG) and Pola-Ribiere conjugate gradient (CGP)) and logistic sigmoid transfer function were used in the network. The results showed that LM with 10 neurons in the hidden layer is the most suitable algorithm with maximum correlation coefficients (R²) of 0.999, minimum root mean square (RMS) value and low coefficient of variance (COV). The reported results conformed that the use of ANN for performance prediction of DXSAHP is acceptable.     

Support vector machines for wind speed prediction

This paper introduces support vector machines (SVM), the latest neural network algorithm, to wind speed prediction and compares their performance with the multilayer perceptron (MLP) neural networks. Mean daily wind speed data from Madina city, Saudi Arabia, is used for building and testing both models. Results indicate that SVM compare favorably with the MLP model based on the root mean square errors between the actual and the predicted data. These results are confirmed for a system with order 1 to system with order 11.     

Estimation of wind speed profile using adaptive neuro-fuzzy inference system (ANFIS)

Wind energy has become a major competitor of traditional fossil fuel energy, particularly with the successful operation of multi-megawatt sized wind turbines. However, wind with reasonable speed is not adequately sustainable everywhere to build an economical wind farm. The potential site has to be thoroughly investigated at least with respect to wind speed profile and air density. Wind speed increases with height, thus an increase of the height of turbine rotor leads to more generated power. Therefore, it is imperative to have a precise knowledge of wind speed profiles in order to assess the potential for a wind farm site. This paper proposes a clustering algorithm based neuro-fuzzy method to find wind speed profile up to height of 100 m based on knowledge of wind speed at heights 10, 20, 30, 40 m. The model estimated wind speed at 40 m based on measured data at 10, 20, and 30 m has 3% mean absolute percent error when compared with measured wind speed at height 40 m. This close agreement between estimated and measured wind speed at 40 m indicates the viability of the proposed method. The comparison with the 1/7th law and experimental wind shear method further proofs the suitability of the proposed method for generating wind speed profile based on knowledge of wind speed at lower heights.     

Multiple architecture system for wind speed prediction

A new approach based on multiple architecture system (MAS) for the prediction of wind speed is proposed. The motivation behind the proposed approach is to combine the complementary predictive powers of multiple models in order to improve the performance of the prediction process. The proposed MAS can be implemented by associating the predictions obtained from the different regression algorithms (MLR, MLP, RBF and SVM) making up the ensemble by three fusion strategies (simple, weighted and non-linear). The efficiency of the proposed approach has been assessed on a real data set recorded from seven locations in Algeria during a period of 10 years. The experimental results point out that the proposed MAS approach is capable of improving the precision of the wind speed prediction compared to the traditional prediction methods.     

Local models-based regression trees for very short-term wind speed prediction

This paper evaluates the performance of different types of Regression Trees (RTs) in a real problem of very short-term wind speed prediction from measuring data in wind farms. RT is a solidly established methodology that, contrary to other soft-computing approaches, has been under-explored in problems of wind speed prediction in wind farms. In this paper we comparatively evaluate eight different types of RTs algorithms, and we show that they are able obtain excellent results in real problems of very short-term wind speed prediction, improving existing classical and soft-computing approaches such as multi-linear regression approaches, different types of neural networks and support vector regression algorithms in this problem. We also show that RTs have a very small computation time, that allows the retraining of the algorithms whenever new wind speed data are collected from the measuring towers.