Research On Road Network Intelligent Selection Based On Knowledge Learning And Reasoning

Cartographic generalization is one of the critical technologies to solve multi-scale expression of geographic information.It has experienced algorithm generalization,collaborative generalization,and process control generalization,developing towards intelligent generalization.Moreover,mature technology in artificial intelligence has opened up new ways for intelligent cartographic generalization.As a hot research area of cartographic generalization,road network selection plays an essential role in basic geographic information update,and multi-scale expression of spatial information.In order to obtain the selection results that are more in line with actual situation,cartographic experts gradually adopt more intelligent algorithms and models to improve selection results further.With the advent of the big data era,the scientific paradigm has shifted to a data-intensive paradigm.The current road network selection methods still have the following problems to be solved urgently.It is challenging to reflect the actual experience of selection process correctly,and the existed deep learning model cannot take into account both attribute characteristics and topological structure.Besides,it is difficult to realize the formal expression and reasoning of domain knowledge,and the existed methods lack intelligent selection for multi-source road network data.Given the present situation,this paper explores an intelligent road network selection method based on knowledge learning and reasoning,improving automatic cartographic generalization intelligence.The research achievements and innovations mainly include the following aspects.(ⅰ)A road network selection method is proposed based on analytic network process.Road network selection is a multi-attribute decision-making problem,and the factors influence each other.However,the integration of road network features usually cannot reflect practical cartographic experience.To solve the above problems,chapter 3 proposes a road network selection method based on analytic network process.Firstly,the feature index system and network structure of roads are constructed based on references and selection principles.Then,the feature item judgment matrix,weighted matrix,and limit hypermatrix are calculated upon analytic network process,determining stroke importance in evaluation process.Finally,density partition and connectivity preservation strategies are designed to control the road network selection process.Different from formerly decision-making model,this method compares the weight of feature items based on analytic network process,which can not only consider the objective characteristics but also obtain more reasonable selection results by learning experience from expert knowledge.(ⅱ)A road network selection method is proposed based on graph neural network.Artificial intelligence provides a strong guarantee for the intelligent upgrade of cartographic generalization.However,normal deep learning models cannot consider both attribute characteristics of road network and topological structure characteristics of non-euclidean space.To solve the above problems,chapter 4 proposes a road network selection method based on graph neural network.Firstly,cases for road network selection are designed,which are used to construct a sample database.Secondly,neighborhood stroke nodes are sampled,and their characteristics are aggregated to target stroke based on graph neural network.Then,a deep learning classification model for road network selection is designed,whose parameters are optimized by a gradient descent algorithm.Finally,selection probability is calculated according to the designed model,thus realizing automatic road network selection.Different from formerly deep learning model for road network selection,this method uses graph neural network to learn semantics,geometrics,and topological structure of road network at the same time.It can learn case knowledge from experts directly,making road network selection more intelligent.(ⅲ)A road network selection method is proposed based on ontology reasoning.Knowledge is the key to realize intelligent cartographic generalization,and road network selection cases contain rich knowledge.However,case-based reasoning is prone to noise and conflict,and some decision-making processes require manual intervention.To solve the above problems,chapter 5proposes a road network selection method based on ontology reasoning.Firstly,a road network selection ontology is constructed from cases.Then,the ontology query extension is used to identify and eliminate original cases’ noise and conflict.Finally,the reasoning manners of semantic mapping and graph query are used to map semantic and geometric characteristics onto ontology,realizing automatic road network selection.Different from formerly case reasoning method,this method expresses road network selection knowledge based on a conceptual ontology model and uses knowledge reasoning to reduce decision-making difficulty.It can further improve selection accuracy and reduce manual interaction.(ⅳ)A multi-source road network selection method based on rule reasoning is proposed.The data sources of road network are increasing in big data era,and the semantic difference in multisource data has become a new challenge for cartographic generalization.To solve the above problems,chapter 6 proposes an intelligent selection method for multi-source road networks based on rule reasoning.Firstly,feature items are calculated from stroke,their concepts and relationships are extracted,and road network selection ontology is constructed from multi-source data.Then,the conceptual similarities are calculated in semantic feature items and numerical feature items,eliminating the semantic difference from multi-source data and providing knowledge sharing.Finally,the road network selection rules are defined upon ontology and semantic web rule language to realize knowledge reasoning.Unlike the single data source selection method,this method shares domain knowledge,eliminates the semantic difference based on the ontology,and uses rules to express generalization knowledge,thus realizing intelligent selection for multi-source road network.(ⅴ)An intelligent road network selection experimental system is developed.Taking knowledge learning and knowledge reasoning as the main line,functional composition and business process are designed in experimental system.The experimental system for road network intelligent selection is designed and developed based on Qt C++ language.In detail,four core functional modules of basic operation,data processing,knowledge learning,and knowledge reasoning are designed and developed based on factory and command modes.Moreover,importance calculation model and deep learning model are integrated,and the databases of road network selection cases,samples and ontology are constructed,realizing ontology reasoning and rule reasoning.It can integrate and verify the core chapters for intelligent road network selection method.