基于改进AdaBoost.RT和KELM的风功率预测方法研究

Research on Wind Power Forecasting Method Based on Improved AdaBoost.RT and KELM Algorithm

为了提高风功率预测精度及预测模型的泛化能力,提出基于改进Ada Boost.RT算法的风功率预测方法,可以有效提高弱学习算法的性能。首先建立核极限学习机(kernel extreme learning machine,KELM)模型,并用改进蝙蝠算法对其参数进行优化,通过引入局部搜索和莱维飞行使算法具有更好的搜索能力和跳出局部最优的能力。在此基础上进一步通过Ada Boost.RT算法生成多个KELM个体(即基学习器),在训练过程中不断调整每个基学习器的权重及训练集中每个样本的权重。最后用训练好的基学习器来对测试样本进行预测,并集成得到最终结果。从不同时间尺度应用不同月份的风电场数据进行仿真测试,同时与前馈(back propagation,BP)神经网络、支持向量机、极限学习机等预测模型对比,仿真结果表明所提方法具有较好的预测精度及泛化性能。

In order to improve forecasting accuracy and generalization ability, a wind power forecasting approach based on improved AdaBoost.RT algorithm is put forward in this article, to enhance performance of weak learning algorithm effectively. At first, a kernel extreme learning machine （KELM） model optimized with improved bat-inspired algorithm with better search ability and performance of jumping out of local optima by introducing local search and Levy flight was established. On this basis, several KELM models （called base learners） were built according to AdaBoost.RT algorithm. Base learners＇ weights and weight of every sample in training subset were modified to achieve optimal model. Finally, forecasting results were obtained by integrating prediction results of base learners. Simulation experiment for different time scales using data of different months was conducted. Compared with B＇P neural network, support vector machine, and extreme learning machine prediction models, the prediction results prove excellent accuracy and generalization performance of the proposed approach.