Wind Farm Power Control Strategy Considering Power Generation Reliability

With the development of wind power technology, installed capacity of wind power units increase constantly. Output of each unit and its health state varies a lot due to different geographical environment of these units at the same time point. Health state of units within a wind farm not only affects their output power individually, but also has an impact on the reliability of the whole wind farm output. To executive active power controlling instructions issued by dispatching department better, wind farm should allocate power to each unit reasonably in every control cycle. This paper aims at improving the reliability of wind farm operation, intensively studied prediction method of wind speed and wind power output, as well as wind farm optimal operation. By predicting wind speed and wind power output, we can get the output of individual unit under full operating status, considering system dispatching requirement and unit generation reliability, to achieve an optimal dispatch of the units in a wind farm. In this way the output power reliability of the whole wind farm is enhanced, leading to a better wind farm operation reliability. The main content of this paper involves 4 following parts:① Build up a wind speed forecasting model. This paper establishes a training matrix whose data correlate tightly between different groups using curve fitting based on SCADA original data. It forecasts wind speed using BP neural network and improved Adaboost_BP neural network respectively. The result proves that the accuracy of wind speed prediction has a great relationship of training model for different samples and time scales. The accuracy of wind speed prediction can be improved by searching optimal training sample number and using single step, multi steps training model to improve forecasting model under different conditions. The comparison result of accuracy of each prediction model shows that the accuracy of Adaboost_BP neural network is higher though it takes more time when model parameters such as time scale, training sample number and testing samples are under the same circumstances.② Build up a short-term wind power forecasting model. Use wind speed and corresponding wind power as training input to calculate the interdependency between them based on BP neural network. Also, this paper analyzes how forecasting wind speed and actual wind speed will influence result when they are respectively used as model input. Research result shows that it’s better to use forecasting wind speed as input because it can reduce error brought by training data and noise produced through sampling and data transmission.③ Build up a unit generation reliability model. Through a fitting function and abnormal status revision regarding wind speed and temperature for state parameters, a wind turbine unit state parameter system including operating condition information is established. Parameter abnormal index PAI is defined to describe state parameters, which result in the abnormal condition of state parameters and weight of this parameter. Generation reliability of a unit can be calculated out via a combination of state parameter abnormal condition and its weight④ Build up an optimal active power output operation model within a wind farm based on wind tribune grouping. A multi-objective optimal operation method of wind farm generation power using genetic algorithm, wind power output predicting data of each unit and generating reliability is brought up, on the premise of power grid dispatching information. Finally guaranteed wind farm operation security and stability through an optimal dispatch.