基于生成对抗网络的风电爬坡功率预测

风电的波动性和随机性，尤其是功率爬坡事件严重威胁着电网运行的安全和稳定。功率爬坡是极端天气影响下产生的，属于小概率事件。其极低的发生概率导致历史爬坡样本数量严重不足，并制约了传统功率预测模型的预测精度。针对此类问题，提出一种基于生成对抗网络的风电爬坡功率预测方案。将历史爬坡数据和模拟特征量作为输入，通过生成器和判别器的对抗训练，生成大量与历史爬坡数据特征相似的模拟爬坡数据，实现爬坡数据集的扩充。再将扩充后的爬坡数据集输入给长短期记忆神经网络算法，进行风电爬坡功率预测。通过仿真测试，验证了该方法在历史爬坡数据匮乏情况下风电爬坡功率预测的有效性。并与传统预测方法进行了对比，证明了其预测的精确性。

WIND POWER RAMPING EVENTS PREDICTION BASED ON GENERATIVE ADVERSARIAL NETWORK

Volatility and randomness of wind power, especially power ramping events, seriously threaten the sound and stable operation of the power grid. Ramping events, which accidentally, are generated under the influence of extreme weather. It’s extremely low probability of occurrence leads to a huge shortage of historical data, restricting the prediction accuracy of traditional power prediction models. To handle such problems, a wind power ramping prediction model based on generative adversarial network is proposed. Taking historical ramping data and simulated feature quantities as input, through the adversarial training of the generator and discriminator, a large number of simulated data with similar characteristics to the historical samples are generated to realize the expansion of the ramping data set. Then the expanded data set is used as input to the long and short-term memory neural network algorithm to predict ramping power. The simulation results show that the method is effective in predicting wind power climbing power in the case of lack of historical climbing data. And compared with the traditional prediction method, the accuracy of the prediction is proved.