基于改进极限学习机的风电功率预测仿真研究

Simulation of Wind Power Prediction Based on Improved ELM

为有效预测超短期风电功率及其波动范围,提出一种基于模糊信息粒化(FIG)和遗传算法优化极限学习机(GA-ELM)的组合预测模型。通过对风电系统参数进行模糊信息粒化,提取各参数在时序窗口下有效分量信息的最大值、最小值和大致平均值。将各参数有效分量整合作为训练样本,并建立基于遗传算法优化极限学习机的预测模型。采用优化后的预测模型完成对下一个时序下风电功率波动范围的预测。实验结果表明,该组合预测模型可以有效跟踪风电功率变化并预测其波动范围。

To predict the range of ultra-short-term wind power fluctuation effectively, a combined forecasting model based on fuzzy information granulation （FIG） and genetic algorithm optimization extreme learning machine （GA-ELM） is proposed. The parameters of wind power are granulated by fuzzy information, and the corresponding valid information including the maximal value, the minimum value, and the general average value in time series window is further extracted. By integrating the effective components of each parameter as training samples, the GA-ELM-based prediction model is established. The range of wind power fluctuation in next time series is forecasted through using the optimized model. The experimental results demonstrate that the combined prediction model can effectively track some variations in wind power and predict the range of wind power fluctuation.