地理模拟作为地理环境模拟与认知的重要方法,是挖掘地理知识、揭示地理规律的重要手段.气象模拟,作为重要组成部分参与到地理环境模拟的诸多方面,影响到地理环境的系统模拟和复杂问题求解.而气象模型专业性强、模拟知识复杂,涉及到输入数据、模拟方案设置等诸多方面,所以如何高效管理及共享气象模拟知识成为亟需解决的问题.本文应用案例推理和知识图谱,统一异构的气象模拟知识,研究顾及语义和图结构相似度的气象模拟知识推荐方法,以提高气象模拟知识的共享能力.①本文基于中文气象模拟文献资料抽取了气象模拟知识,以模拟案例为载体,在知识图谱中构建气象模拟知识库;②构建气象模拟案例相似度评估模型,实现顾及语义和图结构相似度的气象模拟知识推荐,该模型一方面利用经过气象模拟语料库训练的Bert(Bidirectional Encoder Representations from Transformers)语义模型挖掘气象模拟知识案例的语义特征,另一方面提取存储在图谱中模拟案例的结构特征,并通过AHP(Analytic Hierarchy Process)层次分析法进行权重设置,从而准确的衡量气象模拟案例的相似度.③研发了可视化的气象模拟知识共享原型系统,可以根据用户需求推荐气象模拟知识的参数方案,为涉及气象过程的地理系统模拟人员提供知识参考.基于所构建的气象模拟知识库,经推荐准确性和稳定性评估,系统的推荐准确率达到91.3％,研究成果提高了气象模拟知识共享和重用能力.
Research on Knowledge Retrieval of WRF Meteorological Simulation Based on Semantics and Graph Structure
As an important method for simulating and understanding geographical environments,model based geographic simulation is essential for extracting geographical knowledge and revealing geographical laws.The meteorological process,as a crucial component of geographic environment,plays a significant role in various aspects of geographic simulation,thereby influencing the comprehensive simulation and resolution of complex problems within the geographic environment.However,meteorological models usually require a high level of expertise,involving intricate simulation knowledge and encompassing various components such as input data,simulation scheme settings,and others.Therefore,efficiently managing and sharing meteorological simulation knowledge has become the key step to improve the scientific rigor and accuracy of simulation.In this study,case-based reasoning and knowledge graph are applied to unify heterogeneous knowledge of meteorological simulation,and a retrieval method of meteorological simulation knowledge considering semantic and graph structure similarity is proposed to enhance the sharing capability of meteorological simulation knowledge.Firstly,based on Chinese meteorological simulation literature data,this study extracts meteorological simulation knowledge by using Bidirectional Long Short-Term Memory with Conditional Random Fields model(BiLSTM-CRF)and constructs a meteorological simulation knowledge base in the knowledge graph with simulation cases as the carrier of knowledge.As an example,in this study,a total of 795 cases,expressed as nodes and edges,are constructed and stored in knowledge graph.Secondly,a similarity assessment model for meteorological simulation cases is constructed to achieve the meteorological simulation knowledge retrieval considering both semantic and graph structure aspects.In detail,the model utilizes the Bert semantic model trained by the meteorological simulation corpus to mine the semantic features of meteorological simulation knowledge cases.The structural features of cases stored in the graph are also extracted,and the weights are set using the Analytic Hierarchy Process(AHP)hierarchical analysis method,so as to accurately measure the similarity of meteorological simulation knowledge cases.Thirdly,a visual prototype system for sharing meteorological simulation knowledge is developed,which can directly recommend parameter schemes for meteorological simulation knowledge based on requirements from users and provide knowledge reference for geographic system simulation personnel involved in meteorological processes.To validate the stability of the retrieval system,we selected 300 input cases,and the retrieval results provide the top 5 similar cases for each case.In this context,the accuracy of the system reaches 91.3％,improving the efficiency of meteorological simulation knowledge sharing and reuse.
knowledge graphsemantic modelknowledge recommendationknowledge sharingcase-based rea-soningmeteorological simulationgraph similarityBert model
NETL
NSTL
万方数据
