随着风力发电占世界发电的比重不断上升，风能的间歇性、波动性和不稳定性会影响整个电力系统的平稳运行。因此，风电功率的精准预测对于确保用户的电能质量、电网的安全运行以及电网调度具有重大意义。针对风电出力波动性强、非线性变化而导致风电功率预测准确度低的问题，提出了一种采用变分模态分解（VMD）和自适应增强长短时记忆网络（Adaboost-LSTM）的短期风电功率预测模型。

首先，对风电场高维关键气象特征进行有效提取，可以降低模型的训练难度，采用核主成分分析（KPCA）对风电场气象特征进行提取。得到的最优特征集不仅蕴含风电功率的有效信息，也能防止冗余信息的出现，有利于后续风电功率预测模型的学习与训练，同时也降低了模型的复杂度。利用VMD对历史风电功率数据进行平稳化处理，不仅降低了模型的学习难度，且有利于后续模型的训练。特征数据经过VMD分解后得到的各本征模太函数（IMF）的稳定性和噪声鲁棒性好。

其次，针对VMD和LSTM超参数难确定的问题，提出一种改进的乌燕鸥算法（ISTOA）来对预测模型的关键参数进行寻优。具体的改进措施如下：将Tent混沌映射、精英反向学习策略以及差分进化算法融入到乌燕鸥算法中，并采用两种典型的测试函数对ISTOA进行测试，结果表明ISTOA收敛速度更快、全局搜索能力显著提升。

最后，将ISTOA-LSTM作为弱预测器，通过Adaboost算法对多个弱预测器进行加权融合，集成为一个强预测器，输出动态预测结果。通过风电场的历史运行数据来验证Adaboost-ISTOA-LSTM模型在风电功率短期预测方面的可行性，结果表明：该模型预测误差相较于灰狼算法、粒子群算法、差分进化算法的各个误差有明显下降，具有较高的精度和较好的泛化能力。

As the share of wind generation in the world's electricity generation continues to rise, the intermittent, volatile and unstable nature of wind energy can affect the smooth operation of the entire power system. Accurate prediction of wind power is of great significance to ensure the power quality of customers, the safe operation of the power grid and the dispatching of power networks. In order to solve the problem of low accuracy of wind power prediction due to high volatility and nonlinear variation of wind power output, a short term wind power prediction model using variable mode decomposition (VMD) and adaptive boosting of long and short term memory networks (Adaboost-LSTM) is proposed.

First of all, the effective extraction of the high-dimensional key meteorological features of the wind farm is beneficial to reduce the training difficulty of the model. The kernel principal component analysis (KPCA) method is used to extract the meteorological features of the wind farm. The optimal feature set not only contains the effective information of wind power, but also can prevent the emergence of redundant information, is also conducive to the subsequent wind power prediction model learning and training, but also reduces the complexity of the model. The use of VMD to stabilize the historical wind power data not only reduces the difficulty of learning the model, but also facilitates the training of the follow-up model. The eigenmode function (IMF) obtained from the VMD decomposition of the characteristic data has good stability and good noise robustness.

Secondly, an improved sooty tern optimization algorithm (ISTOA) is proposed to optimize the key parameters of the prediction model for VMD and LSTM hyperparameters. Specific improvements are as follows: Tent chaos mapping, elite reverse learning strategy and differential evolution algorithm are integrated into the sooty tern optimization algorithm, and ISTOA is tested with two typical test functions. The results show that ISTOA converges faster and improves global search capability significantly.

Finally, using ISTOA-LSTM as a weak predictor, multiple weak predictors were weighted by Adaboost algorithm, and the set became a strong predictor with dynamic output. The feasibility of the Adaboost-ISTOA-LSTM model in the short-term prediction of wind power is verified by the historical operating data of wind farm. The results show that the model has higher accuracy and generalizability than the gray wolf algorithm, particle swarm algorithm and differential evolution algorithm.