风电场输出功率异常数据识别与重构方法研究

摘要

Electric power big data is an important resource for electric power development and comes from the procedures of electricity production and energy utilization. Wind power operating data is the major part of electric power big data. With the dramatic increase of wind power penetration, it is of great significance for wind farm operation, control and integration research by collection, processing and analysis of real historical operating data from wind farms. However, amounts of data collected from wind farms usually contain abnormal data, which have adverse impact on the study of fluctuation characteristics of wind power, wind power prediction, etc. The main source of abnormal data existed in wind farm historical operation data is analyzed and a model for eliminating abnormal data based on quartile method is presented. In the cases of missing data, methods based on patterns of similarity between neighboring wind farms outputs and multi-point cubic spline are used on the basis of historical data to reconstruct the discontinuous time series respectively. The case study indicates that the presented models are efficient for eliminating abnormal data and reconstructing missing data, which can be applied in practical engineering.%电力大数据是电力发展的重要资源，数据来源于电力生产和电能使用的各个环节。风电运行数据是电力大数据的重要组成部分，随着风电穿透功率的增大，风电数据的采集、处理、分析对风电场运行、控制与并网研究有重要意义。然而，从风电场收集到的大量数据中通常包含异常数据点，这样的异常点给风电功率波动特性、风电功率预测等方面研究带来负面影响。分析了风电场历史运行数据中存在的异常数据的主要来源，并针对该实际问题，采用基于四分位算法的数学模型对异常数据进行识别。在数据缺失的情况下，以可用历史数据为基础，采用基于临近风电场出力模式性的方法和多点三次样条插值方法重构出完整的时间序列。算例分析给出了两种重构方法的重构效果以及各自的适应性，结果表明采用所提出的方法能够有效识别、剔除异常数据并重构缺失数据，对不同风电场有较强的通用性，具有一定的工程实用价值。