Research On Retrieval And Analysis Of MEP Information In Facility Management Based On Domain Ontology

A large number of complex data files will be accumulated during the operation and maintenance phases,among which most information is related to MEP systems,including the operation status,energy consumption information and maintenance knowledge of HVAC,plumbing and electrical equipment.This information can be roughly divided into two categories.One is the sequential monitoring information,such as the operation status and energy consumption information of MEP equipment,which is characterized by rapid accumulation over time and the main storage and management medium is electronic tables.The other one is unstructured operation information,such as the design and maintenance manual of mechanical and electrical systems,which is characterized by lack of computer readable knowledge,and distributed separatly in the Internet which means difficult to collate them comprehensively.The main storage management medium of them are document or pictures.Because the current method of operation and maintenance information storage is not conducive to efficient data retrieval and knowledge learning by computers,there are many use case in operation and maintenance phases,such as energy consumption data analysis and evaluation,locating component in emergency management,etc,are still needed to make decision analysis by engineer based on years of knowledge and experience accumulation,which always means requiring a large amount of manpower and material resources.To solve the above problems,combining with neural network and deep learning model,the ontology of monitoring information and the knowledge graph of MEP have been constructed in this study.In the monitoring information ontology,the definition and integration of spatial ontology,sensor ontology,monitoring equipment and monitoring KPI are realized.The methods of sensor information parsing,transformation and retrieval based on RDF are given.AS for the establishment MEP knowledge graph,this study explores the key technologies related to deep learning,including MEP entity reorganization by Bi-LSTM model,MEP images classification and retrieval with VGG16 model,relationship extraction of MEP entities with residual neural network,and finally using neo4 j to realize storage and visualization of MEP knowledge graph.With the logic information in knowledge graph,this paper proposes an approach to automatically generate the topological chain of MEP systems with topological analysis,which can improve the efficiency of facility management(FM)activities such as locating components and retrieving related maintenance information for prompt failure detection or emergency management.In the case study section,the monitoring information ontology is developed by using the monitoring data of the existing buildings from 2017 to 2018.This paper calculates and analyses the annual,monthly and daily KPI of energy consumption.In addition,the prototype system of automatic completion of logic relationships was applied to a realworld project for validation.The results showed that the approach was able to generate topological chains of MEP systems with an average accuracy of over 80%.Overall,the monitoring information ontology and MEP knowledge graph developed in this study provide a useful framework for related research in this field,as well as contribute to the comprehensive information management and retrieval during O&M phase.