| In recent years,with the continuous acceleration of China’s military modernization and informatization,the upgrading of weapons and equipment has become more frequent.As an important part of the development cycle of weapons and equipment,test and evaluation can effectively reflect the performance,effectiveness and operational applicability of weapons and equipment.The main task of the test data preprocessing is to organize the observation test data collected in the shooting range and the simulation test data generated by simulation into index data that can be used for subsequent effectiveness evaluation calculations.At present,most domestic test data preprocessing software faces some problems,such as lack of operations,poor ease of use,and limited computational efficiency.The thesis sorts out various types of test data preprocessing requirements,and combines model-driven related theories,and proposes a set of solutions that can be used for rapid construction of test data preprocessing procedures.The main contents are as follows:(1)Proposing a set of development methods for model-driven test data preprocessing processThis method takes the experimental data preprocessing development process as the research object,combined with model-driven engineering related theories,and sorts out several key technologies needed in the data preprocessing process development process based on the model-driven architecture.In this method,first,we introduced the PIM model definition method.Then,we explained the process method of PIM model to executable code conversion.Finally,we summarized the various component templates needed in the model conversion process to supplement the model conversion technology.(2)Proposing a development method for distributed data processing process based on FlinkIn order to solve the problem of insufficient single-machine computing efficiency when facing experimental big data processing scenarios,the thesis proposes a set of data preprocessing process parallelization methods for distributed architecture based on the model-driven data preprocessing process development method and combined with Flink.This method combines the Flink programming model and Jet template technology to parallelize the development of component templates used in the development of model-driven preprocessing procedures.Then a set of software environment to support the distributed computing of preprocessing process is constructed to realize the application of the preprocessing process constructed by the user in the distributed environment.(3)Proposing a set of model recommendation techniques based on spectral clusteringIn order to improve the construction efficiency of the PIM model during the development of the test data preprocessing process,the thesis proposes a set of model recommendation techniques based on spectral clustering to recommend preprocessing model construction schemes that users may need.In this method,semantic-based feature representation method is used to achieve a unified representation of similar semantic features of data sets.A clustering algorithm based on spectral clustering groups similar data sets and preprocessing models to provide a recommendation basis for recommendation algorithms.The collaborative filtering recommendation technology based on two-way clustering realizes accurate recommendation of available models to users. |
