Semantic Constrained Intelligent Selection Method Of Best Spatio-temporal Data For Landslides Assessment

In China,landslide is one of the most serious natural disasters,which causes huge losses to people's lives and property,and seriously affects urban safety and sustainable development.Landslide disaster assessment is an important part of disaster prevention and mitigation.The scope of disaster assessment includes hundreds of statistical indicators of different categories,such as the number of casualties,the number of damaged houses and so on.Comprehensive and accurate data is vital for objective economic loss assessment,which requires multidisciplinary,multi-platform,multi-sensor and multi-source data to do fusion analysis.With the rapid development of earth observation technology,the amount and type of disaster loss assessment data have increased significantly,which brings a large number of high-valued multi-modal data for disaster assessment,but also redundant,outdated,low-relevance and even invalid data.Therefore,more requirements on data integration,process and analysis have been put forward.How to automatically and efficiently select the advantageous data sets from multi-modal disaster data for evaluation indicators extraction is one of the biggest challenges of data management.The reliable data selection approach can assist in improving the decision-making ability of disaster reduction and emergency response.Traditional disaster data retrieval mainly relies on keyword matching,which is limited by the problem of semantic ambiguity of natural language,and the matching accuracy is poor.Current semantic retrieval methods only consider simple underlying data features such as spatial and temporal range of data,but lack explicit representation and formal modeling of domain knowledge semantics.It's difficult to discover and share the multi-source heterogeneous data and cannot fit the information retrieval needs of complex tasks.Additionally,the ignorance of redundancy and complementarity leads to a problem of information overload.This paper put forward a semantic constrained advantageous information selection approach to satisfy complicated disaster assessment requirements,which can deal with the multi-source and heterogeneous data.Knowledge graph technology plays a key role in the research to promote the description,discovery and reason of highlevel knowledge links.Further,the data selection and recommendation method are build based on the knowledge graph,which can break through the bottleneck problem of fast and accurate data retrieval and model data adapter in the phase of disaster emergency.A low redundancy and high accuracy data selection result can be generated by the approach.The specific studies are as follows:(1)Data intelligent selection-oriented disaster assessment association model.The logical relationships among the four elements of task,model,parameter and data are systemically analyzed and a three-domain expression model of "task-processing-data" is constructed.The hierarchical relationships among different assessment elements are described explicitly.A unified description model of assessment elements based on ontology is constructed,and the logical structure of disaster assessment knowledge graph is given.According to the requirement of data selection,a unified expression model is formed,which summarizes the multi-dimensional and multi-level semantic relations among data and implements enhanced expression of data semantics.(2)Semantic relation construction and knowledge reasoning of knowledge graph.According to the data characteristics of space-time,theme and scale,a similarity-based correlation intensity calculation method is constructed firstly,which is the basis of relation extraction.Further,a data-driven automatic generation method of disaster assessment knowledge graph is constructed,which realizes the association among tasks,models,parameters,data and other entities;To reason the three kind of high-level semantic association which named homogeneity,complementarity and causality,and the meta-path rules are designed and the meta-path-based association reasoning method is achieved.(3)Knowledge graph-based disaster assessment advantageous information automatic selection method.Based on the feature selection theory,the task-oriented data evaluation method is studied,and the data relatedness,redundancy as well as integrity evaluation indexes are defined.Further,the smart selection method of spatiotemporal data with maximum relevance and minimum redundancy is studied to achieve results optimization.Then,the multi-modal data selection method based on homogeneous relationship in knowledge graph is studied,which can be used to query the heterogeneous data with low data relatedness.Finally,an intelligent selection strategy which coupled workflow is studied to realize task-oriented multi-element data optimal selection.(4)Based on the above research findings,a prototype system is constructed and a case study is carried out taking Maoxian landslide disaster housing damage assessment as an example.Experimental analysis is carried out to verify the effectiveness of the theoretical method proposed in this paper.The experimental results show that the method can effectively utilize the correlation among assessment elements and improve the automation and intelligence level of data selection.Compared with the relatedness-based data selection method,the maximum correlation and minimum redundancy algorithm has the advantages on accuracy,recall,redundancy and integrity.In this paper,the semantic constrained advantageous information selection method for disaster assessment is presented.Theoretically,the logical model and construction method of knowledge graph of disaster assessment are constructed,and the exploration is carried out in the field of disaster knowledge engineering.Practically,the method of data retrieval and recommendation based on knowledge graph associations is put forward,which effectively improves the breadth and the accuracy of query results.