基于EMD-KELM-EKF与参数优选的用户侧微电网短期负荷预测方法

A Short-Term Load Forecasting Method for Micro-Grid Based on EMD-KELM-EKF and Parameter Optimization

短期负荷预测是微电网经济调度的重要组成部分,预测误差将直接影响运行经济性。相对于大电网环境,在用户侧微电网实施短期负荷预测的难度更高。提出了一种基于经验模态分解、扩展卡尔曼滤波及核函数极限学习机的组合短期负荷预测模型,通过经验模态分解对随机性强的微电网负荷时间序列逐级分解为多组固有模态函数分量,采用扩展卡尔曼滤波及核函数极限学习机2种存在典型差异的预测模型对不同性质的固有模态函数分量进行预测,并采用粒子群算法实现模型参数的优选。针对用户侧微电网的环境约束,提出了离线参数寻优、周期参数更新与在线预测相结合的实现模式。通过多种类型、容量的用户侧微电网算例分析,验证了模型预测精度、周期更新稳定性与计算效率。

Short-term load forecasting is an important component of microgrid economic dispatching and the forecasting error directly affects the economy of microgrid operation. Different than large-scale power grid, it is more difficult to implement short-term load forecasting of microgrid at user side. Based on empirical mode decomposition （EMD）, extended Kalman filter （EKF） and extreme learning machine with kernel （KELM）, a combined short-term load forecasting model is proposed. In the proposed model the time-series data of microgrid load with strong randomicity is decomposed into multi sets of intrinsic mode function （IMF） components by EMD; two forecasting models, in which EKF and KELM are adopted respectively and typical differences exist, are used to forecast the IMF components with different features and the particle swarm optimization （PSO） is used to implement the optimization of model parameters. To cope with the environment constraint of microgrid at user side, a forecasting mode combining off-line parameter optimization, periodical parameter updating with on-line forecasting is put forward. The forecasting accuracy of the proposed model, the stability of its periodical parameter updating and calculation efficiency is validated by case studies of various types of microgrid at user side with different capacities.