An Interpretable Deep Neural Network Fault Diagnosis Method for Air Conditioning Systems

Abstract

The invention discloses an interpretable deep neural network fault diagnosis method for an air-conditioning system, and relates to the technical field of fault diagnosis of HVAC systems. CNN) extracts the feature information of the HVAC operating data, and uses the absolute gradient weighted class activation map (Grad‑Absolute‑CAM) as a visual classification diagnosis standard, and uses the absolute gradient weighted class activation map to visualize the fault feature information of the HVAC system. The fault diagnosis standard corresponding to each fault data in the one-dimensional convolutional neural network fault diagnosis model is obtained through the absolute gradient weighted class activation map, so that the fault diagnosis model becomes interpretable. The method of the present invention uses a one-dimensional convolutional neural network model for fault diagnosis, and the accuracy of the fault diagnosis result is high; at the same time, the fault characteristics are visualized through the absolute gradient weighted activation mapping, and the fault diagnosis criteria are obtained, which can enhance the interpretability of the model .

Description

technical field [0001] The invention relates to the technical field of fault diagnosis of heating, ventilating and air-conditioning systems, and more specifically, relates to an explainable deep neural network fault diagnosis method of an air-conditioning system. Background technique [0002] The HVAC system is a multi-equipment operating system, and some failures often occur during the operation process. The occurrence of these failures will not only affect the normal operation of the system, but also increase the total energy consumption of the building. With the rapid development of building intelligence, as well as the improvement of data accumulation and calculation speed, various data-driven deep learning fault diagnosis methods have been applied to HVAC systems, such as long short-term memory network (LSTM), deep neural network, etc. (DNN) and deep learning algorithms such as convolutional neural network (CNN). These deep learning methods have received increasing att...

AI Technical Summary

Problems solved by technology

[0004] In order to overcome the shortcomings of poor interpretability of deep learning black box models

Images