Ehsan Taslimi , Renani (2018) *Power prediction using the wind turbine power curve and data-driven approaches / Ehsan Taslimi Renani.* PhD thesis, University of Malaya.

PDF (The Candidate's Agreement) Restricted to Repository staff only Download (1592Kb) | ||

| PDF (Thesis PhD) Download (5Mb) | Preview |

## Abstract

Wind energy as one of the promising energy sources, has attracted great attention because it is pollution-free and abundant. Moreover, it shows considerable potential for supplying electricity to meet the demand. The high dependence upon the wind, however, results in variation of the wind power due to the intermittent nature of the wind. The volatility of wind power over time jeopardizes the reliability of the power systems. Therefore, the prediction of the wind power is required. Wind turbine power curve representing the relationship between the wind speed and power can serve as a tool for prediction. In this thesis, a new parametric model, called modified hyperbolic tangent (MHTan), is proposed to approximate the wind turbine power curve. To obtain the unknown vector of parameters of the MHTan, three heuristic optimization algorithms are employed to minimize the sum of squared residuals. An alternative way to estimate the coefficients of MHTan is through maximum likelihood estimation (MLE) and the probability density function of wind speed. In this method, firstly, Weibull density function is utilized to model the wind speed and then several methods are applied to estimate the parameters of the wind speed distribution. To evaluate the performance of the Weibull parametersâ€™ estimator methods, two sets of data are considered, one based on simulated data with different random variable size and the other based on actual data collected from a wind farm in Iran. Secondly, a new formula representing frequency distribution of the turbine power is derived. The formula comprises of unknown vector parameters of MHTan which can be determined based on MLE. Then, the performance of the MHTan is evaluated using actual data collected as well as three simulated data representing three different typical shapes of the power curve. In order to demonstrate the efficiency of the proposed method, it is compared rigorously with several parametric and nonparametric models. In addition, the capability of the MHTan in on-line monitoring of the wind turbine is presented. In this research, a comparison is also drawn between two different wind power prediction models, indirect and direct approaches. In the former it is necessary to forecast the wind speed at first, then the corresponding power is obtained from the wind-power curve. Since in practice turbines do not work in ideal conditions, the theoretical power curve provided by manufacturers is avoided and a power curve approximated by MHTan is used instead. Several statistical methods are used to predict wind speed and the best one is selected for prediction over longer horizons. To set up direct wind power prediction, six data-driven approaches are employed and the same procedure as in indirect approach is applied to select the best method for longer horizon predictions, up to 60-min. The results confirm the superiority of the direct prediction models. Moreover, a hybrid feature selection technique is proposed to choose the necessary subset of inputs so that the important information is retained. This technique is a combination of mutual information and neural network where its effectiveness is examined with several linear and nonlinear feature selection methods.

Item Type: | Thesis (PhD) |
---|---|

Additional Information: | Thesis (PhD) - Institute of Graduate Studies, University of Malaya, 2018. |

Uncontrolled Keywords: | Wind power prediction; Wind turbine power curve; Wind speed prediction; Data-driven |

Subjects: | Q Science > Q Science (General) |

Divisions: | Institute of Graduate Studies |

Depositing User: | Mr Mohd Safri Tahir |

Date Deposited: | 25 Sep 2018 08:35 |

Last Modified: | 03 Feb 2021 03:46 |

URI: | http://studentsrepo.um.edu.my/id/eprint/8970 |

### Actions (For repository staff only : Login required)

View Item |