基于时序长记忆模型的风电场短期功率预测

Wind Power Short-term Prediction Based on the Long-term Memory Model the Time Series Analysis Method

随着风电的飞速发展,风电场接入电网的规模日益扩大,随之而来的是对风电功率预测准确性要求的提高,准确的风电功率预测可以更好地利用风能资源,减小风电并网对电网的不利影响。为了提高风电预测的精度,采用最大期望算法(expectation maximization algorithm,EM算法)对风电场功率历史数据进行处理,填补缺失值,替换错误数据,然后采用修正重标极差分析法即修正R/S分析方法分析数据的长记忆性,采用时间序列ARFIMA模型,然后根据预测时刻之前的功率数据,通过贝叶斯统计推断对模型参数进行估计,生成预测模型方程,进而对风电场输出功率进行预测。

With the rapid growth of wind power, more wind power plants have been connected to the power grid, followed by the requirement of increasing accuracy of wind power prediction. Accurate prediction is indispensable for the better use of wind energy resources, and reducing the negative impact on the power grid. In order to improve the accuracy of prediction of wind power, expectation maximization algorithm （EM） was apllied for processing the history data of wind power, filling missing values, and replace erroneous data. Modified rescaled range analysis was used to analyze the long memory of data. Then autoregressive integrated moving average model （ ARIMA） was used. Based on the previous prediction data of the power, a prediction model equations can be made after the model parameters are estimated by Bayesian inference. This model can be used for the prediction of wind power.