Research On Test Sequence Automated Generation For High-Speed Railway Automatic Train Operation System

Driven by information technology,control technology and other high and new technologies,the Automatic Train Operation(ATO)system for high-speed railway comes into being.The ATO system for high-speed railway can reduce line operation cost and improve railway transportation capacity,which is the future development direction.Safety is the eternal theme for railways and testing is an important technical method to ensure function correctness of the ATO system for high-speed railway.Test sequences guide the entire test work and the quality of them have a great effect on test quality and efficiency.How to automatically generate high-quality test sequences and improve test efficiency have always been the key issue for train control system test work.As an independent institution,Beijing Jiao Tong University has done the test assignment of the ATO system for Beijing-Shenyang high-speed railway.On the basis of these research work,this thesis summarizes the test method of the ATO system and the standards in editing test sequences.Aiming at heavy workload and low efficiency problem in manually writing test sequence,the research is conducted from the perspective of test sequences generation and judgment by using Recurrent Neural Networks(RNN)and expert system.The results show that the method can improve the test efficiency effectively.In detail,the dissertation carries out the following research work:(1)A test sequence automatic generation algorithm based on RNN is proposed.The thesis converts the test sequence automatic generation problem into the prediction problem with time series and selects RNN variant —— Long Short-Term Memory(LSTM)as a test sequence automatical generation model.The neural network inputs are ordered events and output is a test sub-sequence.Test sequence can be obtained by string the test sub-sequences along the test path.The paper uses the Beijing-Shenyang high-speed railway test sequences as a data sets to train the neural network.Finally,we prove the method feasibility by comparing and analyzing differences between test sequences generated by the model and manually edited test sequences.(2)The thesis proposes a test sequences judgment algorithm based on expert system.The knowledge for length and coverage of test sequences is represented by the production rules and knowledge base is established.The paper selects the forward reasoning to judge whether the test sequences meets the knowledge rules.For test sequence which does not meets the standards,the expert system will points out the problem in the judgment result and assists the tester to regenerate a new test sequence.(3)Combined with the test sequences automatic generation algorithm and the test sequences judgement algorithm,the thesis uses the C# language to program test sequence auxiliary generation tool.The tool can automatically generate test sequences based on train control system engineering data and test cases.The actual results show that the tool can improve the efficiency by 17.95% compared with manual.