Research on non-invasive load monitoring method based on transfer learning strategy
The existing load monitoring algorithms have greatly improved the power decomposition accuracy in the same data set,but the model generalization is poor and the decomposition accuracy across data sets is low.Therefore,it experimentally explores whether the load monitoring model is portable,and studies a load monitoring model based on transfer learning strategy,which fixes part of the convolutional network layer and realizes the fine-tuning transfer learning of the load monitoring model.It verifies the influence of fixed layers on prediction results through experiments,as well as the effectiveness of the model on different electrical appliances.Compared with the traditional direct transfer model,the proposed model effectively reduces the load decomposition error of each device and reduces the requirement of the model on the number of load training samples.
NETL
NSTL
万方数据
