Wind power is fluctuant and random. It is easy to be influenced by system disturbance. Wind power with high penetration is interconnected into grid will change the boundary condition for traditional power system operation and control, and it will bring huge challenge for stable power system operation. Accurate power forecasting and reasonable power control are two key ways to solve above problems. While, accurate power forecasting is an important basis of reasonable power control.
To improve forecasting precision and power control performance, our research team proposes the following roadmap:
(1) For the influences of forecasting error on wind power control, probability distribution characteristics of forecasting error for different wind farms and he features of wind power variation for different wind generators are analysed and considered for wind power control. Also, risk is introduced to decrease the impacts of wind power control on stable system operation. It can make wind power control more accurate and efficient.
(2) The idea of big data should be used. By data mining, dynamic features of wind power variation are extracted. Then the causality between different features and wind power variation is analysed to establish different power forecasting models, which are suitable for different regular patterns of wind power variation. It can improve the applicability of power forecasting.
(3) Because time-space function of wind speed is time variant, the maximum forecasting error is difficulty to be controlled. Especially if there are some singularities or singular surfaces, the forecasting result will be unusable. Spatial-correlation wind power forecasting can be used for the supplement to traditional sequential wind power forecasting. The time-space coordination should be considered for wind power forecasting. Global information and trend features are employed, and it can make up the shortage of traditional method only considering local information and section features. Then the robustness of wind power forecasting will be promoted.
Following jobs are done based on above thoughts:
(1) A forecasting power and risk based wind power control method is proposed. Firstly, confidence interval of forecasting wind power is evaluated by statistic analysis on probability distribution of wind power forecasting error. Based on this, the probability of positive and negative error of each wind farm can be calculated. Because forecasting error may cause the cost for reserve, load shedding, unit outage and wind power curtailment, an objective function of wind power control to minimum the risk cost caused by error. Then by optimizing, power control command can be distributed reasonably.
(2) An architecture of wind power control system based on data fusion is designed. The principle for the architecture is information integration, platform integration and control integration. Firstly, unified information model is established to combine the information of all equipments in wind farm and substation. Secondly, software structure is designed to combine the functions including wind farm supervisor control and data acquisition (SCADA), wind power forecasting, active power control and reactive power control. Then wind power control method based on forecastingpower is adjusted for application in wind farm. Finally, the wind power control system based on data fusion is developed and has been applied at a wind farm in Jiuquan wind power base in Gansu Province.
(3) The limits of existing sptial-correlation based wind power forecasting methods are analysed from the point of the view of the research thought and models. Then the feasibility and applicable conditions of wind speed forecasting using the spatial-correlation features are discussed. And a new idea and framework of sptial-correlation based wind power forecasting is proposed. It establishes different forecasting models according to different spatial-correlation features off-line by analysing historical data and background information. And appropriate model will be selected on-line by feature matching.
(4) To improve the effectiveness and robustness of shape classification for wind speed time series, unit window feature and trend feature are defined. Then a two-layer symbolic idea considering section and trend features is proposed to simply describe the feature of wind fluctuation.
(5) For sptial-correlation based wind speed forecasting, a new wind speed forecasting method is proposed, which establishes models for different shapes off-line and selects appropriate model by feature matching on-line. Firstly, according to dynamic features of wind and external conditions, wind speed time series are classified into different patterns. Secondly, different forecasting models are established and corresponding parameters are optimized for different patterns off-line. Then features of wind speed variation in recent time are extracted. By feature matching, appropriate model will be used for on-line application, and the forecasting result at reference location in recent time window is used to forecasting wind speed at objective location.