基于长短期记忆网络分位数回归的短期风电功率概率密度预测

SHORT-TERM WIND POWER PROBABILITY DENSITY PREDICTION BASED ON LONG SHORT TERM MEMORY NETWORK QUANTILE REGRESSION

针对风电功率确定性的点预测无法对预测结果进行风险评估以及现有静态预测模型难以描述风电功率长期相关性的现象,提出一种基于长短期记忆网络分位数回归(LSTMQR)的短期风电功率概率密度预测模型。该方法首先使用LSTMQR得到不同分位点下未来风电功率的预测结果;其次采用高斯核函数,将LSTMQR与核密度估计(KDE)相结合,进行短期风电功率概率预测,可得到未来风电功率预测点的概率密度函数,通过风电场的历史数据对所提模型以及基准模型进行对比验证,使用5种评价指标表明所提模型的预测性能更优。

Wind power forecasting is of great significance for the safe and stable operation of power systems and the optimal allocation of energy.Aiming at the fact that the wind power deterministic prediction cannot give the risk assessment of the forecast results and the existing static prediction model is difficult to describe the long-term correlation of wind power.A short-term wind power probability density prediction based on long short-term memory network quantile regression(LSTMQR)is proposed.The method first uses the long short-term memory network quantile regression to obtain the prediction results of future wind power under different quantile points.Secondly,combining LSTMQR with kernel density estimation(KDE)for short-term wind power probability prediction,the probability density function of future wind power prediction points can be obtained.The model is validated by a number of evaluation indicators based on the measured data of a domestic wind farm.The results show that the proposed model is better than the benchmark model.