Loading [a11y]/accessibility-menu.js
A Method for Compression and Deployment of Load Monitoring Model | IEEE Conference Publication | IEEE Xplore

A Method for Compression and Deployment of Load Monitoring Model


Abstract:

In recent years, with the development of load monitoring, the actual deployment of load monitoring has become a key issue to maintain the model performance while minimizi...Show More

Abstract:

In recent years, with the development of load monitoring, the actual deployment of load monitoring has become a key issue to maintain the model performance while minimizing the parameters and computations. Although existing computer vision and other fields in a wide range of attempts to modellightweighting, model compression, knowledge distillation and other methods seek to deploy models on edge devices, but the field of load monitoring has not yet been an in-depth attempt. Therefore, this paper proposes a practical load monitoring model compression deployment method. It first constructs colored V-I trajectory maps based on voltage and current signals, uses the VGG16 model commonly used in computer vision for image classification, and finally performs compressed deployment based on model compression methods. The effectiveness of the proposed method is verified by utilizing the WHITED dataset to classify five common appliances. The proposed method lays the foundation for load monitoring model deployment applications.
Date of Conference: 17-19 May 2024
Date Added to IEEE Xplore: 24 July 2024
ISBN Information:
Conference Location: Hangzhou, China

1. Introduction

Load monitoring is an important part of energy management [1]. Since its introduction, the research focus of load monitoring has changed from performance improvement based on artificial intelligence methods to practical deployment based on optimization methods [2]. However, practical deployment of load monitoring is not as simple and efficient as theoretical research, and practical deployment needs to consider practical issues. First, the number of parameters and computation volume of the load monitoring model become important bottlenecks, and the excessive number of parameters and complex computation volume will cause a great burden to the deployed system, which puts forward higher requirements to the system. Second, the performance of the load monitoring model should be good enough, and the deployed load monitoring model will run for a long time and be updated very infrequently, which should ensure that the deployed model has the optimal performance. Finally, the deployed load monitoring model should be Finally, the deployed load monitoring model should balance the number of model parameters, the amount of computation and the model accuracy, and the less parameters and computation cause the model accuracy to drop sharply is to be avoided. Therefore, the actual deployment of load monitoring becomes a key issue to maintain the model performance and minimize the parameters and calculations.

References

References is not available for this document.