融合滑动窗口和MLP-AdaBoost的电力负荷预测

Electric Arc Furnace Load Forecasting Based on Sliding Window and MLP-AdaBoost

由于炼钢过程中电弧炉与LF炼钢冲击负荷的随机性,采用人工无法进行用电负荷的控制。为了能够及时调控用电负荷,尽可能节约能源和成本,采用滑动窗口的方法来划分数据,建立多层感知机(MLP)和自适应增强算法(AdaBoost)炼钢负荷预测模型,即MLP-AdaBoost方法对炼钢过程中的电力负荷进行预测。实验通过滑动窗口建立数据与之前时刻的联系,使用MLP建立弱模型,通过AdaBoost对样本权重进行再分配,通过迭代得到最终预测模型。结果表明,基于滑动窗口MLP-AdaBoost预测模型在负荷变化大、突变性强的炼钢过程中有显著效果。所提方法可为电弧炉炼钢下一步负荷量调控提供参考依据,为预测控制奠定基础,从而节约能源和成本。

Due to the randomness of the impact load of electric arc furnace and LF steelmaking in the process of steelmaking,it is impossible to control the electric load manually.In order to timely regulate the electricity load and save energy and cost as much as possible,the sliding window method is used to divide the data,and the multi-layer perceptron(MLP)and adaptive enhance⁃ment algorithm steel making load prediction model is established,that is,MLP-AdaBoost method is used to predict the electricity load in the process of steel making.In the experiment,the sliding window was used to establish the relationship between the data and the previous time,the MLP is used to establish the weak model,and the AdaBoost is used to iterate on the sample weight to ob⁃tain the final prediction model.The results show that the MLP-AdaBoost prediction model based on the sliding window has signifi⁃cant effect in the process of steel making with strong uncertainty of load variation.The proposed method can provide a reference for the next load control of eAF steelmaking,and lay a foundation for predictive control,so as to save energy and cost.