基于特征选择及ISSA-CNN-BiGRU的短期风功率预测

Short-term Wind Power Prediction: Feature Selection and ISSA-CNN-BiGRU Approach

针对风电功率随机性大、平稳性低,以及直接输入预测模型往往难以取得较高精度等问题,提出了一种基于特征选择及改进麻雀搜索算法(ISSA)优化卷积神经网络-双向门控循环单元(CNN-BiGRU)的短期风电功率预测方法。首先,利用变分模态分解(VMD)将原始功率分解为一组包含不同信息的子分量,以降低原始功率序列的非平稳性,提升可预测性,同时通过观察中心频率方式确定模态分解数。其次,对每一分量采用随机森林(RF)特征重要度的方法进行特征选择,从风速、风向、温度、空气密度等气象特征因素中,选取对各个分量预测贡献度较高的影响因素组成输入特征向量。然后,建立各分量的CNN-BiGRU预测模型,针对神经网络算法参数难调、手动配置参数随机性大的问题,利用ISSA对模型超参数寻优,自适应搜寻最优参数组合。最后,叠加各分量的预测值,得到最终的预测结果。以中国内蒙古某风电场实际数据进行仿真实验,与多种单一及组合预测方法进行对比,结果表明,本文所提方法相比于其他方法具有更高的预测精度,其平均绝对百分比误差值达到2.644 0%;在其他4个数据集上进行的模型准确性及泛化性验证结果显示,模型平均绝对百分比误差值分别为4.385 3%、3.174 9%、1.576 1%和1.358 8%,均保持在5.000 0%以内,证明本文所提方法具有较好的预测精度及泛化能力。

Given the significant randomness and non-stationarity inherent in wind power,achieving higher prediction accuracy through direct in-put of the prediction model is often challenging.Therefore,this study proposes a short-term wind power prediction method based on feature selec-tion and an improved Sparrow Search Algorithm(ISSA)optimizing Convolutional Neural Network-Bidirectional Gated Recurrent Unit(CNN-Bi-GRU).Initially,Variational Mode Decomposition(VMD)is employed to break down the original power data into subcomponents,each contain-ing distinct information.This process reduces the original power series’non-smoothness,thereby enhancing predictability.The determination of the number of decomposition models relies on the observation of the central frequency method.Subsequently,feature selection is conducted us-ing the Random Forest(RF)feature importance method for each component.This selects influential factors from meteorological features such as wind speed,wind direction,temperature,and air density to form the input feature vector.Following this,individual CNN-BiGRU prediction mod-els are established for each component.To address the challenge of manually adjusting neural network algorithm parameters due to their inherent randomness,ISSA is employed to hyperparameterize the model,enabling the adaptive discovery of optimal parameter combinations.Finally,the prediction values of each component are aggregated to derive the final prediction results.Simulation experiments are conducted using real data from a wind farm in Inner Mongolia,China.These results are compared with various single and combined prediction methods.The findings demonstrate that the proposed method achieves higher prediction accuracy,with an average absolute percentage error value of 2.6440%.Further-more,the model's accuracy and generalization are validated on additional data,with average absolute percentage error values of 4.3853%,3.1749%,1.5761%,and 1.3588%,all within 5%.This confirms the superior prediction accuracy and generalization ability of the proposed method.