多工况生产过程下的即时学习能耗预测建模方法

Just-in-time Learning Energy Consumption Predictive Modeling Method in Multi-condition Production Process

针对全局能耗预测模型只适用于部分预测样本且模型计算量大的问题,引入即时学习思想,采用局部加权偏最小二乘法结合能耗模型建立临时局部能耗预测模型;改进粒子群算法的惯性权重,考虑粒子适应度、迭代次数和种群大小对粒子群算法收敛速度和收敛精度的影响,提出一种非线性变化的自适应惯性权重策略,离线计算阶段使用改进的粒子群算法(adaptive PSO,APSO)对历史样本的带宽参数进行寻优,当预测样本到来时在线更新局部模型。考虑多工况生产场景下不同工况样本之间的能耗差异性所导致的预测误差,增加工况相似性度量过程,提出局部加权偏最小二乘算法与K-means算法相结合的APSO-JITL(just-in-time learning)-CLWPLS(cluster locally weighted partial least squares)能耗预测建模方法,在预测时选取同一工况的历史样本来设计预测样本的带宽参数。通过仿真实验验证了算法有着更高的预测精度且能更好地应对多工况生产场景。

Aiming at the problem that the global energy consumption prediction model is only suitable for part of the prediction sample and the model is computationally intensive,the idea of just-in-time learning is introduced,and the local weighted partial least squares method combined with the energy consumption model is used to establish a temporary local energy consumption prediction model.The inertia weights of the particle swarm algorithm are improved,considering the effects of particle fitness,number of iterations and population size on the convergence speed and convergence accuracy of the particle swarm algorithm,a nonlinear change adaptive inertia weight strategy is proposed,and the improved adaptive PSO(APSO)is used to optimize the bandwidth parameters of historical samples in the offline computing stage,then the local model is updated online when the predicted samples are available.Considering the prediction error caused by the different energy consumption of the samples under different working conditions in multi-working condition production scenarios,and increasing the measurement process of working condition similarity,an APSO-JITL-CLWPLS energy consumption prediction modeling method combining local weighted partial least squares algorithm and K-means algorithm is proposed,and the bandwidth parameters of the predicted samples are designed by selecting the historical samples of the same working conditions during prediction.Simulation experiments show that the algorithm has higher prediction accuracy and can better cope with the multi-working production scenarios.