Entities Recognition And Impact Prediction Approaches For Technology System-of-systems

Technology system-of-systems(TSo S)is an extension of weapon system-of-systems(WSo S)from the technical perspective and an important theoretical basis for the overall planning of weapons for organizers.Recently,the theoretical research on TSo S mainly concentrates on the strategic planning,generation method,architecture modeling,system evaluation,technological prediction and other aspects.Existing researches have two main problems.First of all,although the expert-based methods like the Delphi have been widely applied,they are defective in long cycle,high cost and incomplete coverage.The second,despite the existing multi-view based modeling framework has designed complete products,entity meta-models and relational meta-models,it does not provide solutions on how to collect entity meta-models and predict their future trends.Therefore,this paper proposes entities recognition and impact prediction approaches for technology system-of-systems.From the aspects theoretical requirements and engineering practice,the problem of recognition and prediction can be decomposed into three sub-problems that are,how to carry out prediction,what to predict and how to predict.A data-driven research framework is designed for the first sub-problem,a recognition method based on deep learning is proposed for the second sub-problem,and network dynamics and deep learning methods are presented for the third sub-problem respectively.The main work and innovations of this paper include:(1)A data-driven entities recognition and impact prediction framework is designed.In this paper,we designed an integrated multi-source technological intelligence data forecasting framework,including the domain analysis,technical entity recognition and selection,technology(emerging technology)impact prediction.Meanwhile,complex models of professional system architects and data analyst are concealed to stakeholders to share the output of each step and realize the rapid iteration of recognition and prediction.We have solved the problem on how to carry out quantitative prediction in capter two.(2)A deep learning methodology is proposed for technical entities recognition and key entities selection.Compared with the classical prediction problems with clear definition on the prediction objects,the strategic and quantitative technological trends prediction mainly solves the problem of what to predict.This paper proposes a technological entity recognition method based on deep learning,aiming at mining technological topics and directions under the background of the Internet and big data.(3)A network dynamics methodology is put forward to predict long-term impact trends of armament technologies.For medium and long term forecast demand,we have discussed the essence of the development of science and technology rule,including technology growth,priority link,fitness,time,decay and other technology factors,built the network dynamic model with four parameters.As a result,it works to predict the future technology impact trends.(4)A prediction method is present for impact trend prediction of emerging technologies based on heterogeneous network and deep learning.Aiming at the problem of inaccurate prediction of dynamic model within a short citation window,this paper further study the prediction methods of emerging technologies based on heterogeneous network and deep learning.Heterogeneous network is an extension of citation network at the data dimension,providing richer semantic information.By combining the information advantages of heterogeneous network and the prediction advantages of deep learning algorithm,we have accomplished higher prediction accuracy with shorter prediction time window.