Uncertainty Of Wind Power Forecasting And Power System Economic Dispatch

Wind intermittency creates uncertainties in wind power forecasting (WPF). Additional operating cost and risk have to be incurred in the alternative electrical power system with large scale renewable energy sources. This thesis carries out a study on wind power forecasting and its uncertainty analysis, and system economic dispatch incorporating WPF uncertainties. Main contributions includes:1. Establishing the model of wind power forecasting and its uncertainty analysis based on relevance vector machine (RVM)Influences on modelling process caused by accuracy and size of training samples are firstly analyzed. Based on this, a training sample selection mechanism is presented for small sample learning. This thesis proposes a model of wind power forecasting and its uncertainty analysis based on VVRVM. This model is capable of predicting deterministic future wind power as well as its fluctuation range under given confidence level. Case study shows that the proposed method can well simulate the power generation process of wind under variable meteorological conditions, importantly with small amount of relevance samples. In this way, accuracy of deterministic forecasting, reliability and sharpness of uncertainty analysis are improved.2. Presenting four methods to improve accuracy of wind power forecasting and its uncertainty analysis(1) A dynamic method is proposed based on wind scenarios identification is proposed, for improving model adaptabilities towards variable and diverse wind.Sensitivity of WPF error with respect to wind and power curve are analyzed. Two clustering algorithms are employed, which are K-means and Spectral clustering, to identify wind scenarios. Two index, termed as data distribution density and data distribution evenness, are defined to evaluate the clustering quality. Based on this, a cluster number self-seeking algorithm is presented and improves traditional clustering method which determining cluster number only by subjectivity. With clustered wind scenarios, WPF and its uncertainty analysis is conducted by the proposed strategy that off-line refined training and online dynamic forecasting.(2) A wind turbines grouping method considering correlation of wind flow is proposed. This mitigates problems in large wind farm (or cluster), who suffers severe uncertainties aroused from complex flow. Problems might be lack of NWP representativeness and computational efficiency in model training.Characteristics of wind speed distribution and variation are analyzed. Prevailing wind coordinate system is designed coupling with wind farm flow information. Wind turbine grouping model is built up based on self-organization mapping neural network. This helps selection of location and numbers of NWP calculation points.(3) A NWP correction method is proposed, considering uneven distribution of wind speed prediction deviation in different wind speed section. In this way, NWP accuracy is improved with any effort on difficult mesoscale system assimilation.(4) To minimize subjectivity on forecasting performance, this thesis optimizes RVM kernel parameter bases on GA and PSO.3. Modeling dispatching cost considering wind power uncertaintiesDefinition and mathematical model of WPUIC and WPUDC are proposed. These two cost incorporate future uncertainty estimates into power generation cost, also overcoming the difficulty when integral terms in cost model to be solved in a traditional dispatching manner. WPUIC is to quantify the cost to increase unit wind power into grid. WPUDC measures balancing cost because of wind power forecasting uncertainty. WPUDC features in its convex properties, and also capabilities of tracking cost changes with variable external conditions by adjusting model parameters. External conditions might include complex weather and diverse wind farms. WPUDC parameters reflects average level and PDF shape of forecasting.4. Building up system economic dispatch model bases on uncertainty of wind power forecastingThis thesis proposes economic dispatch for power system, which coordinates operating risk and cost. Decision is made according to distribution of future uncertain range. Dispatching task is to trade-off the contradiction between benefits and risks of integrating more wind generation. More wind would be dispatched when WPF uncertainty is predicted to be low, vice versa. In this way, wind resources are well dispatched. The dispatching objective function has analytical differential expression and enables equal incremental principle to be used in future power system. ED solving method is proposed based on GMA. With GMA solving model, WPUDC based ED is able to be validated with traditional integral ED. Also, accuracy and efficiency of GMA are validated with PSO and GA. Case study shows that the proposed ED method reduces operating cost of power system and wind power curtailment. This improvement benefits from integrating the future uncertainties under variable external condition into ED.