Wind Speed Models Considering Dependence And Their Applications In Reliability Evaluation Of Generating Systems

With energy exhausting and environment protection becoming more severe, thedevelopment and utilization of wind energy have received considerable attention inrecent years. However, the wind is an unstable and variable energy source, and alarge-scale integration of wind farms has a significant effect on power system reliabilityin a different manner of conventional energy sources. Therefore, it is important to doin-depth study on the effect of large-scale integrated wind power on the reliability ofpower systems containing wind energy.Currently, wind power is developing towards the direction of development oflarge-scale and high penetration. It is an urgent need to build a more accurate reliabilitymodel of wind farm (WF) when the quantity of WFs and their installed capacityincrease. Based on this, the thesis makes wind, which is the key influence factor of windpower output, the research target, presents the wind speed model for single WF andwind dependence model for multiple WFs, and applies them to power system reliabilityassessment containing WFs.A kernel density estimation (KDE) method, which is a nonparametric model, forestimating the probability density function of wind speeds is presentted. The method is adata-driven approach without any assumption for underlying the stochasticcharacteristic of wind, and is capable of uncovering the statistics hidden in the historicaldata. An acceptance-rejection technique is developed to simulate wind speed, and anonlinear optimization method and linear interpolation technique are used to improvethe sampling efficiency and convergence speed.The accuracy and applicability of theproposed method is verified using the wind data at six sites. The results indicate that theKDE can describe the wind speed distributions with a high accuracy and excellentrobustness. The fitting statistics of R_a~2for six sites are more than0.99. The averageabsolute errors of annual energy production (AEP), loss of load expectation (LOLE) andloss of energy expectation (LOEE) are0.47%,0.22%and0.39%, respectively.Wind speed has not only distribution property but also time persistence nature. Topreserve the temporal dependence of wind speed time series (WSTS), a continuous statemarkov chain model (CSMC) for generation of WSTS is proposed. Firstly, the methodtransforms the collected WSTS into a normal domain, and then establishes a continuoustransition kernel (CTK), which reflects the conditional probability distribution of wind speed data in adjacent time, and finally obtains the CSMC of WSTS. The methodovercomes the drawback of tradeoff between model accuracy and complexity existingin discrete state markov chain model (DSMC). Based on this method, the thesis extendsthe CTK from normal distribution to a variety of distribution types using copulafunctions, and obtains the copula-based CSMC of WSTS. The results prove that theCSMC method can offer a satisfactory fit for both probability distribution and temporaldependence, and reduce the complexity of the modeling.The extensive utilization of wind energy leads to the integration of multiple windfarms into a power system in a region. Due to the geographic locations of wind farms,wind speeds at different sites show certain dependence between each other, which has aconsiderable impact on the reliability of power systems containing wind energy. A newtechnique for establishing multivariate distribution of wind speeds using copulafunctions is presented to model wind speed dependence in the thesis. The copulamethod is used to separate the wind speeds into the marginal distributions of each singlewind speed and dependence structure, which leads to the effectiveness and simplicity indealing with the dependence in a multivariate distribution. The case study results usingtwo actual wind speed data indicate that the copula model can give a validrepresentation of multiple wind speeds.Considering the time dependence of single WSTS and spatial dependence betweenthem, two methods based on autoregressive and moving average model (ARMA) andCSMC model, respectively, are proposed to describe multivariate wind speed timeseries (M-WSTS). Firstly, ARMA model or CSMC model is used to characterize thetime dependence of a single WSTS, and Copula functions are then applied to capturethe spatial dependence by constructing the joint probability distribution. Finally, amultivariate ARMA (M-ARMA) model or multivariate CSMC (M-CSMC) model canbe obtained by combing the time dependence model and spatial dependence model.Case studies show that the M-ARMA model can offer satisfactory fit for the statisticalproperties including autocorrelation and joint probability distribution, but the model hasthe drawback of needing a large number of sample data, whereas the M-CSMC modelrequires less wind data but only preserve the temporal dependence of single WSTS.A reliability evaluation model of power system containing multiple wind farms isproposed based on Monte Carlo simulation method, which can be used to consider thewake effect of wind farm and the outage of wind turbine generators. The IEEEreliability test system (IEEE-RTS79) is used to verify the proposed model. The results show that the model can be directly used to evaluate the reliability effect on the powersystems containing of wind farms.