Research On The Expert Matching Selection And Optimal Allocation Model Driven By Knowledge Of Electric Power Science And Technology

There is a conflict between the high-quality and diverse review needs of Electric Power Science and Technology(EPST)research and the relatively lagging EPST knowledge management and expert selection methods.On the one hand,as technological innovation plays an important role in the transformation and upgrading of electric power systems,the construction of a new generation of power systems requires high-quality EPST innovations under the carbon peaking and carbon neutrality goal.Moreover,with the deep integration of energy and digital technology,EPST research presents a trend of interdisciplinary,which brings more diversified needs for scientific and technological consultation.On the other hand,lacking excavation,organization,and management of expert knowledge with the growth of expert databases can lead to an inaccurate description of experts’ research fields and directions,thus it is difficult to meet the consulting needs using traditional selection methods.To alleviate the above conflict,this paper focuses on the research of Expert Matching Selection(EMS)and Expert Optimal Allocation(EOA)models driven by EPST knowledge.The main research contents are as follows:(1)Building EPST text feature representation based on the Pre-trained Language Model(PLM).To overcome the challenges of knowledge acquisition brought by the specialization and interdisciplinary characteristics of EPST texts,this paper collects and processes a large number of EPST documents and obtains a PLM for EPST based on the masked language model.Using a manually-annotated EPST term classification dataset,the obtained PLM is verified by fine-tuning and representing the EPST texts.The PLM can improve the performance of downstream text analysis tasks.(2)Build an EPST-specific knowledge graph(KG)for Expert Optimal Selection(EOS).To realize the efficient organization and management of expert domain knowledge,this paper defines the entity-relation types from local and global perspectives,so as to establish an EPST-specific KG for EOS.For local types,an entity relation extraction model based on rule semantic analysis is constructed,key terms are extracted from expert achievements,and achievement-term relations are formed according to their functions;for global types,a distant supervised EPST entity relation extraction model based on contrastive learning is proposed,where the domain-specific language model and entity description encodings are used to improve the model performance.(3)Build an EMS model driven by EPST knowledge.To establish feature descriptions of an expert’s research field or direction and match the field-related candidate experts for a project to be reviewed,this paper proposes a content filteringbased EMS model,which uses term location and background information to enhance the expert feature vector representation.An advanced EMS model based on graph embedding is also proposed to integrate EPST-specific KG structure information and expert-term interaction information.Using the TransD and graph attention networks,the proposed model gets better embeddings for experts and terms and improves the accuracy of EMS.(4)Build an EOA model driven by EPST knowledge.To select experts suitable for project review from candidate experts,it is necessary to consider the influence of expert authority and business goal constraints.This paper proposes an authority evaluation model for EPST experts based on IF-AHP-DEMATEL and cloud model,which quantifies the comprehensive influence of experts from three indicators:main performance,major awards,and personal ability;and defines the problem of multi-objective EOS in the multi-project review scenario.An improved NSGA-Ⅱ algorithm is used to solve the problem,which can improve the diversity and convergence of the results.Using the subjective and objective combination of the weighted TOPSIS method,a review expert allocation plan that considers the preferences of decision-makers can be acquired.This paper focuses on the EOS of EPST experts.The main achievements are(1)Pre-training to obtain a domain-specific PLM suitable for EPST text feature representation;(2)Extracting entity relations from both local and global aspects and establishing an EPST-specific KG,in which the global entity relation contains high-level association information between experts and term entities;(3)Comprehensive use of the KG neighborhood structure,expert-term interaction and technical term background information to obtain the feature representation of experts’ research fields,realizing accurate EMS;(4)Complete the multi-objective EOA model considering matching degree,authority,and business constraints in complex multi-project review scenarios,and design an optimal decision-making algorithm to solve the EOA plan.The research results can provide support for the selection of EPST experts theoretically and practically,as well as references for EPST knowledge management,and help improve the efficiency of power science and technology project management.Moreover,the research methods and results also have the potential to be applied to other scientific research fields.