| In nowadays’ chiller systems,the combination of intelligent building information management and data utilization has become a research hotspot.However,in the actual old air-conditioning system,due to the difficulty of system transformation,it is impossible to carry out effective building energy management and fault diagnosis.In addition,general machine learning is applied to fault detection and diagnosis,operation optimization and energy saving research.The available experimental data cannot meet the needs of most machine learning,which limits the application of machine learning to the fault diagnosis development of actual commercial operating units.This article is divided into two main parts.One is to design an airconditioning intelligent software and terminal hardware system based on embedded devices.The actual running chiller data will be communicated with the main device through the embedded device.Additional sensor data collection,preliminary data processing and analysis will be conducted through the IOT and Internet architecture to realize remote multi-unit comprehensive management.The second is to conduct a failure experiment on a chiller of a company in Taicang,which establishes a steady-state data discrimination model,removes transient data in the unstable state of the system,screens and retains the system operation to achieve stability,and uses it as the effective input of the calculation model,and then performs the data cleaning,comprehensive data preprocessing based on information gain and Gini index feature selection and chi-square correlation analysis.Try to establish a semi-supervised learning application based on the actual unit of pseudo-tags based on the DBSCAN density clustering algorithm and Pseudo-Labelling pseudo-tags in the limited experimental data to solve the problem that the actual fault data collection is not easy.It can be widely used in the fault diagnosis system of actual operation units. |
