Wind speed forecasting with artificial neural networks (ANNs) plays important role in safely utilizing and integrating the wind power. With the rapid updated wind speed data, however, the only way to guarantee forecasting accuracy for these ANN models is re-training from scratch with an updated training dataset. Obviously, it is an inefficient work due to the resumption of constructing the new training dataset and re-training the model. To enhance training efficiency, reduce re-training cost and improve forecasting accuracy, a self-adaptive kernel extreme learning machine (KELM) is proposed in this paper. With an advanced and efficient learning process, the self-adaptive KELM could simultaneously obsolete old data and learn from new data by reserving overlapped information between the updated and old training datasets. To evaluate the efficiency and accuracy of the self-adaptive KELM, the wind speed data from three different stations are employed as a numerical experiment. The Mean Absolute Error (MAE), Mean Square Error (MSE) and Mean Absolute Percentage Error (MAPE) showed that the self-adaptive KELM with a simple structure could obtain more accurate forecasting results at a faster calculation speed than comparison models, where the proposed model decreased the MAPE values with 7.4776%, 3.5329% and 2.0900% in 1-step, 3-step and 5-step forecasting, respectively

人工神经网络（ANN）进行风速预测在安全利用和整合风能方面发挥着重要作用。但是，有了快速更新的风速数据，保证这些ANN模型的预测准确性的唯一方法是使用更新的训练数据集从头开始重新训练。显然，由于恢复了构建新的训练数据集并重新训练模型的工作，因此效率低下。为了提高训练效率，减少再训练成本，提高预测精度，提出了一种自适应核极限学习机（KELM）。通过先进且高效的学习过程，自适应KELM可以同时保留旧数据并通过保留更新的训练数据集与旧的训练数据集之间的重叠信息来从新数据中学习。为了评估自适应KELM的效率和准确性，将来自三个不同站点的风速数据用作数值实验。平均绝对误差（MAE），均方误差（MSE）和平均绝对百分比误差（MAPE）表明，结构简单的自适应KELM可以以比比较模型更快的计算速度获得更准确的预测结果，其中比较模型提出的模型在1步，3步和5步预测中分别降低了MAPE值7.4776％，3.5329％和2.0900％