Elsevier

Energy

Volume 86, 15 June 2015, Pages 232-239
Energy

Predicting the wind power density based upon extreme learning machine

https://doi.org/10.1016/j.energy.2015.03.111Get rights and content

Highlights

  • ELM (extreme learning machine) is applied to predict wind power density.

  • Outlines of two Weibull methods are used as training and testing data.

  • The capability of ELM is appraised compared to 3 different approaches.

  • ELM can be utilized effectively in applications of wind power predictions.

Abstract

Precise predictions of wind power density play a substantial role in determining the viability of wind energy harnessing. In fact, reliable prediction is particularly useful for operators and investors to offer a secure situation with minimal economic risks. In this paper, a new model based upon ELM (extreme learning machine) is presented to estimate the wind power density. Generally, the two-parameter Weibull function has been normally used and recognized as a reliable method in wind energy estimations for most windy regions. Thus, the required data for training and testing were extracted from two accurate Weibull methods of standard deviation and power density. The validity of the ELM model is verified by comparing its predictions with SVM (Support Vector Machine), ANN (Artificial Neural Network) and GP (Genetic Programming) techniques. The wind powers predicted by all approaches are compared with those calculated using measured data. Based upon simulation results, it is demonstrated that ELM can be utilized effectively in applications of wind power predictions. In a nutshell, the survey results show that the proposed ELM model is suitable and precise to predict wind power density and has much higher performance than the other approaches examined in this study.

Introduction

Nowadays, the current trend of fossil fuels consumption as well as environmental problems have further motivated incentives towards developing renewable and clean energy sources. As an appropriate renewable energy source, wind energy is being largely harnessed in various regions around the world to enhance the sustainability and reduce some negative environmental issues raised by excessive exploitation of fossil fuels. Owing to the outstanding nature of wind energy which is free, environmental friendly and inexhaustible, countries are performing tremendous efforts to assign a high priority to wind energy harnessing [1]. In fact, adaptation and utilization of the wind energy has received considerable attention all over the world as an alternative source to meet the energy demand. Wind energy is now viewed as the fastest growing source because its development has been characterized by a remarkable rate and this trend is expected to continue. Although investment on wind energy to generate electricity is the cheapest one among all renewables, inappropriate locations for wind turbines installation eventuates in losing huge amounts of money. Therefore, for providing secure wind energy utilization as well as enhancing the efficiency of wind energy markets, wind resources evaluation particularly in terms of realizing wind speed and power predictions is an imperative task [2]. Nevertheless, since the frequency distribution of wind speed may provide different wind power densities for the same wind speed, the knowledge of wind power density would be further reliable. Wind power density resembles the level of accessible energy at the site for converting to electricity by using wind turbines.

Despite the fact that a vast number of mathematical models have been suggested to simulate the wind energy related parameters, there are still disadvantages of the models such as being very demanding in terms of computation time. As a consequence, during the last decade, a large number of scientists worldwide have applied artificial intelligence and computational intelligence techniques to estimate, simulate and optimize various important elements in the realm of wind energy.

Shamshirband et al. [3] performed an investigation on the probability density functions of wind speed and directions based upon the ANFIS (adaptive neuro-fuzzy inference system). They compared the results of ANFIS technique with four wind distribution approaches of Weibull, Frechet, Gumbel and Joint probability density functions. They concluded that ANFIS can represent the wind speed and direction distribution favorably. Karki et al. [4] developed a probabilistic methodology to determine the wind variability and measure the wind power commitment risk throughout the operation of wind energy conversion systems. Chang [5] employed the PSO (particle swarm optimization) technique to assess the wind energy in Taiwan based upon the Weibull distribution function. He calculated the Weibull parameters, the wind speed distribution as well as many important parameters for wind energy evaluation. The obtained results indicated that the PSO method is viable in applications of wind energy. Ma et al. [6] utilized wind power data for all locations distributed across Ireland to suggest some scenarios for attaining the knowledge of error and fluctuation distribution of predicted short-term wind power. Bigdeli et al. [7] predicted the wind power time series for a wind farm located in Alberta, Canada. They developed some hybrid models by combining the NN (neural network) with ICA (imperialist competitive algorithm), GA (genetic algorithm) and PSO techniques. The achieved results showed that the hybrid NN-ICA outperforms other hybrid models. Mohandes and Rehman [8] applied three approaches: the PSO, the AIM (Abductory Induction Mechanism) and the PER (Persistence) for forecasting the 12-h ahead wind speed in Saudi Arabia. They found close agreement between the predicted wind speed data and the measured ones. Petković et al. [9] applied the ANFIS to control and design the wind generator system with CVT (continuously variable transmission). The ANFIS scheme regulated the CVT ratio to achieve the highest power generation efficiency of the wind turbine with extracting maximum wind energy. Bhaskar and Singh [10] proposed an approach, consisted of two steps, to predict wind power based upon the AWNN (adaptive wavelet neural network) as well as the FFNN (feed-forward neural network).

NN, as a major AI (Artificial Intelligence) approach, has been recently introduced and applied in different engineering fields. This method is capable of solving complex nonlinear problems which are difficult to solve by classic parametric methods. NNs can be trained by several algorithms including the GD (Gradient Descent), GDA (Adaptive Learning Rate Gradient Descent Momentum), GDX (Gradient Descent with Adaptive Learning Rate Back Propagation), OSS (One-Step Secant), SCG (Scaled Conjugate Gradient), CGB (Fletcher–Reeves Conjugate Gradient), CGP (Powell–Beale Conjugate Gradient) and CGF (Polak–Ribiere Conjugate Gradient Methods Gradient Descent) [11].

Nevertheless, the major disadvantage of NNs is its learning time requirement. Huang et al. [12], [13] introduced an algorithm for single layer feed forward NN which is known as ELM (Extreme Learning Machine). The ELM algorithm is able to decrease the required time for training a NN. In fact, it has been proved that by utilizing the ELM, learning becomes very fast and it produces good generalization performance [14]. Accordingly, several researchers have been attracted towards applying ELM to solve the problems in considerable scientific areas [15], [16], [17], [18], [19], [20].

A few studies have also been conducted by applying ELM to the wind energy field. Wu et al. [21] performed an investigation to develop an ELM-based model for estimating wind speed and sensorless control of wind turbine systems. Salcedo-Sanz et al. [22] combined the CRO (coral reefs optimization) with ELM to predict short-term wind speed in a wind farm situated in USA. Wan et al. [23] using ELM proposed a model for short-term probabilistic wind power forecasting in a wind farm of Australia.

Reviewing the literature indicates that there is no specific utilization of ELM in estimating wind power density. Therefore, the aim of this research work is to develop an ELM-based model in order to predict the monthly wind power density. The merit of ELM is verified by comparing its predictions accuracy with SVM (Support Vector Machine), ANN (Artificial Neural Network) and GP (Genetic Programming) successfully employed in wind energy area estimations. The wind power density values are calculated based upon the standard deviation and power density methods of the Weibull function to extract the required data for training and testing the models. Afterwards the performance evaluation is accomplished statistically by providing comparisons between the predicted results and the calculated wind powers using real data.

The ELM is a powerful algorithm with faster learning speed compared with traditional algorithms such as the BP (back-propagation). It also has a better performance too. ELM tries to get the smallest training error and norm of weights.

The organization of the remaining part of this paper is as follows: Section 2 explains the wind data and wind power estimation. Section 3 presents the description of ELM. The comparative results and discussion are brought forward in Section 4. Finally, the conclusions are presented in Section 5.

Section snippets

Wind data

Basically, wind speed is measured at a desired site by utilizing anemometers established in a wind mast. In this study, we utilize 3-h wind speed data for the period of five years. In fact, the used wind speed data for this study have been measured at the elevation of 10 m above the ground level in 3-h interval periods. In the first step of analysis, the 3-h wind speed data were averaged to obtain daily data. As mentioned, the main objective of this study was evaluating the suitability of ELM

ELM (extreme learning machine)

The ELM as a tool of learning algorithm has been introduced for SLFN (single layer feed-forward neural network) architecture [12], [36], [37]. The ELM chooses the input weights randomly and determines the output weights of SLFN analytically. The ELM algorithm has a more favorable capability in the learning speed. This algorithm does not require too much human intervention, and can run much faster than the conventional algorithms. It is capable to determine all the network parameters

Results and discussion

In the current study, an ELM is employed to establish a model for predicting the wind power density. To demonstrate the merit of the ELM approach in estimating the wind power density, its prediction performance is assessed by making comparisons with SVM, ANN and GP. SVM, ANN and GP have been widely used by many researchers in solving various problems [41], [42], [43], [44], [45], [46], [47], [48].

In the first step of analysis, the extracted wind power data from the standard deviation and power

Conclusions

The accurate wind power and speed estimations are of great significance in enhancing the financial efficiency and acceptability of the wind energy harnessing. In this paper, an efficient learning model based upon the ELM (extreme learning machine) was developed to estimate the monthly wind power density. The predictions of the ELM model were compared with those of SVM, ANN and GP approaches to determine the suitability of the established ELM model for wind power density predictions. As input

Acknowledgment

This project was supported by the High Impact Research Grant (UM.C/625/1/HIR/MOHE/FCSIT/15) from the University of Malaya and the Ministry of Science, Technology and Innovation (MOSTI), Malaysia.

References (48)

  • D.D. Wang et al.

    Fast prediction of protein–protein interaction sites based on extreme learning Machines

    Neurocomputing

    (2014)
  • R. Nian et al.

    Extreme learning machine towards dynamic model hypothesis in fish ethology research

    Neurocomputing

    (2014)
  • P.K. Wong et al.

    Modeling and optimization of biodiesel engine performance using kernel-based extreme learning machine and cuckoo search

    Renew Energy

    (2015)
  • S. Wu et al.

    Extreme learning machine based wind speed estimation and sensorless control for wind turbine power generation system

    Neurocomputing

    (2013)
  • S. Salcedo-Sanz et al.

    Feature selection in wind speed prediction systems based on a hybrid coral reefs optimization–extreme learning machine approach

    Energy Convers Manage

    (2014)
  • A. Mostafaeipour et al.

    Wind energy feasibility study for city of Shahrbabak in Iran

    Renew Sustain Energy Rev

    (2011)
  • K. Mohammadi et al.

    Assessment of solar and wind energy potentials for three free economic and industrial zones of Iran

    Energy

    (2014)
  • O. Arslan

    Technoeconomic analysis of electricity generation from wind energy in Kutahya, Turkey

    Energy

    (2010)
  • A. Malik et al.

    Economics of wind turbine as an energy fuel saver e a case study for remote application in Oman

    Energy

    (2009)
  • F.J. Liu et al.

    Validity analysis of maximum entropy distribution based on different moment constraints for wind energy assessment

    Energy

    (2011)
  • T.P. Chang et al.

    Comparative analysis on power curve models of wind turbine generator in estimating capacity factor

    Energy

    (2014)
  • A. Ouammi et al.

    Monthly and seasonal assessment of wind energy characteristics at four monitored locations in Liguria region (Italy)

    Renew Sustain Energy Rev

    (2010)
  • K. Mohammadi et al.

    Economic feasibility of developing wind turbines in Aligoodarz, Iran

    Energy Convers Manage

    (2013)
  • A. Ucar et al.

    Assessment of wind power potential for turbine installation in coastal areas of Turkey

    Renew Sustain Energy Rev

    (2010)
  • Cited by (79)

    • Randomization-based machine learning in renewable energy prediction problems: Critical literature review, new results and perspectives

      2022, Applied Soft Computing
      Citation Excerpt :

      The method is tested with synthetic and real data. Later, in [97] an ELM is proposed to estimate wind power density in wind farms. The ELM model is tested by comparing its predictions with SVR, MLP and GP.

    • Multi-layer cooperative combined forecasting system for short-term wind speed forecasting

      2021, Sustainable Energy Technologies and Assessments
      Citation Excerpt :

      The experiments were all implemented on MATLAB running on macOS Mojave. Experiment I aims to show the MLCFS's competitiveness compared to that of some benchmark single models by comparing it with the Naïve forecasting, and some other popular methods, such as ARIMA, extreme learning machine (ELM), and fuzzy neural network (FNN), which are currently used extensively [27,42]. The three parameters of ARIMA, namely the number of time lags of the autoregressive model, differencing number, and order of the moving averaging model, were decided based on autocorrelation and partial autocorrelation functions to ensure that the best ARIMA was taken for a comparison.

    View all citing articles on Scopus
    View full text