Study On Short-Term Forecasting And Dispatching Of Wind Power

Wind energy is a clean, inexhaustible source of energy. Compared with other distributed energy resources, wind power is with the most mature development techniques, and its installed capacity has been rapid increasing. In the year of 2010, the new and total installed capacities of wind power are 18.9GW and 44.7GW respectively in China, which made China be the countury with the largest wind power capacity in the world. With wind power embedded, power system would face a series of challenges mainly due to the intermittency nature of wind power. In general, a certain percentage of wind power penetration could cause voltage and frequency fluctuations, frequency deviation, harmonic problem, etc.Researches on how to improve the utilization level of wind power and lower the impact to the grid are important topics. Lower the prediction error to an acceptable level is the most effective way to improve the utilization level wind power. In addition, use other power souces to absorb wind power fluctuations and build the reasonable dispatching models also could improve wind power development.This dissertation discusses short-term wind speed forecasting in wind farm, and reaseraches on its performance improvement. Combined with the analysis of wind turbine's power curve and wake effects, we build a path from wind speed prediction to wind power prediction. And we also do researches on probability distribution of wind power forecasting error, and power system dynamic economic dispatch embedded with wind power. Major work and contributions of this dissertation include:1)Two global statistic prediction models for short-term wind speed are proposed, they are improved GMDH networks and support vector regression (SVR) combined with chaos theory, respectively. Compared to the traditional GMDH networks, a feedback loop is employed and the neurons are fuzzified, and the new networks are applied to short-term wind speed prediction. Use chaos theory to analysis wind speed time series to determine the best input dimensions of predictin models, and then use the SVR models to predict wind speed. These two models both improve short-term wind speed prediction accuracy.2) A local prediction method based on optimal neighborhood in phase space for short-term wind speed prediction in wind farm is proposed. The local predictor builds prediction models with the similar points in historical data to the prediction points. Combined with the phase space reconstruction technique, this approach gets the optimum neighborhood through considering of proportion of false neighbors, which could guarantee the high similarity between neighbors and prediction state points, and when the SVR model has a good capability of nonlinear fitness. The results have demonstrated the accuracy of the proposed very short-term wind speed prediction method in comparison with that offered by other prediction methods.3) Based on the local prediction method of optimal neighborhood in phase space, a multivariate local lredictor with correlation selected is proposed. It sifts multivariate time series by correlation principle to reconstruct multivariate phase space, and searches the neighborhood of the prediction state points to build the support vector regression models. Using the data of real-world collected from a wind farm, the results show that the proposed method could improve searching efficiency of local predictor which could find much more similar neighbor points. And combine with SVR model which could provide good capability of nonlinear fitness, it can improve the accuracy of short-term wind speed prediction effectively.4) Based on the analysis of the wind turbine power output curve and the wake effects which influences to wind farm power output, we study the short-term wind power prediction method based on power curve and the direct wind power prediction method, respectively. And we also discuss the wind direction prediction method and its error function. At last, the two prediction methods are compared both in single turbine prediction and wind farm prediction.5) In consideration of Beta probability distribution of wind power forecasting error based on the study of wind power forecasting, this dissertation studied probability distribution of it and up & down spinning reserve caused by it, and proposed a dynamic economic dispatch model of power system integrated wind power. Chance Constrained Programming was used to deal with the chance variable, and improved particle swarm optimization was employed to solve the model. The results on IEEE30 system integrated wind farms demonstrate effectiveness of the proposed model.