Data Mining And Feature Extraction Of Massive Wind Monitoring Information

Wind power has become the fastest-growing and most mature new clean energy. China attaches great importance to the development of wind power and the installed wind power capacity has reached the world first. However, with the increase of intermittent powers penetration, the fluctuation and randomness characteristics of wind power have dramatically increased the risk and difficulty for power system operation and dispatching. Therefore, only the comprehensive analysis of wind power characteristics from different temporal and spatial scales can provides references and key data support for improving the safety and economic level of the scheduling system. At the same time it would be benefitial for large scale wind power integration and wind energy utilization.Some methods in the view of the shortcomings and insufficient of the existing data mining and feature extraction techniques when extracted the wind power output characteristics from the massive wind power data have proposed to supplement and perfect it. This paper mainly studies four aspects of data mining and feature extraction techniques include data reduction, data clustering, data statistics and correlation analysis. Thus a relatively complete system of data mining and feature extraction techniques is established, through which we can get the wind power characteristics.The statistic and analysis research of this paper are based on the massive practical data of Jibei wind farms. By using the multiple temporal and spatial scales data mining techniques, the output data of each wind farm is organized and analyzed, the evaluation methods and evaluation indexes of the output data of wind farms are studied. The statistical laws of the wind power output characteristics are studied from three aspects, including randamness characteristic, fluctuation characteristic and correlation characteristic then the multiple temporal and spatial scales evaluation index system is built. This system takes account the levels from minute to year in the temporal dimension, from single wind farm to the whole power grid level in the spatial dimension.The randomness evaluation indexes of wind power include the distribution of wind speed, wind power and the prediction error. This paper analyzes the shortcomings of the existing prediction error distribution model then uses the nonparametric estimation method to improve it. Combine the method we proposed and the wind power point forecasting, we can obtain the fluctuation interval.The fluctuation evaluation indexes of wind power include the change rate of wind power, wind power ramp events, the extreme value distribution of wind power, the peak valley difference contribution rate and the long-term typical daily patterns of wind power, these indexes are analyzed from multiple temporal scales. It can provide a reference for many fields of power system analysing, such as frequency adjustment, peak regulation and risk assessment.The correlation analysis evaluation indexes of wind power include autocorrelation and cross-correlation. The autocorrelation coefficient and wind power state transition probability are used to analysing the output correlation properties of the single wind farm. The cross-correlation coefficient is used for analysing the output correlation properties of the wind farms located in different regions. The change rate of wind power, the standard deviation and the simultaneity factor are used for analysing the clustering effect of the wind farms.