Research On Scenario-Based System Behavior Modeling And Composition

Requirement engineering is fundamentally essential in software development. It is desirable to have a favorable requirement engineering in order to reduce the software cost and ensure the project fulfillment. Generally speaking, an efficient requirement elicitation is the baseline of requirement modeling. However, requirement elicitation becomes complex and difficult with the rapid development of software engineering.Typically, it is publically accepted that scenario is one of the effective tools for the requirement elicitation and verification. Nevertheless, it is apt to bring the problem of requirement overlapping due to the part description and non-consistency of the submitted scenarios. During the process of scenarios combination and evolution, there possibly exist some undesired system expectation behaviors in the requirement specification. In addition, some potential requirement information may be lost.This dissertation describes a research on scenario-based requirement Modeling. The whole solution could be regarded as a machine with feedback: the control component is like state emerging, the feedback component is like a user, the input is limited to the collection of scenarios initially submitted, the output is the merged quotient automata, the input of the feedback is the new event sequence produced by the merge algorithm, the output of the feedback is the user judgment of the new event sequence. From the perspective of formal method, the input is the set of executable string, which can be summarized as an acceptable language of the formal system. The output is the language syntax which is characterized by the input string. The whole process of requirement modeling in this study reflects the idea of user center design.In this study, an abstract semantic mode is developed based on scenario together with its evolution, which is the main concern of formal semantics. Furthermore, for the sake of the part description of scenario, a category theory based semantic representation framework is presented to describe the behavior model with event trajectory combing lifecycle and point-set event structure of scenario. In this framework, Fibration functor with transition property is utilized to interpret the formal semantics of scenario with its evolution, which is conversed into categorial diagram processing.Each component in the system is assigned a labeled transition system(LTS) behavior description using scenario samples, which are extracted from the event sequences in the scenario specification. The scenario is formulated by graphical Language message sequence charts(MSC). The maximum deterministic finite state automata, which can accept positive samples, is constructed by scenario samples. The quotient automata is obtained by state merging. The new sequences produced by state merging are served as questions for users to answer. Furthermore, the validation of state merging is verified by negative samples which provide a question-answer Tracking mechanism for users, and support incremental interaction for scenarios.Unfortunately, some unexpected system behavior may be covered by the synthesized model if the state can not be recognized correctly. Similarly, some desired system behavior may not be covered by the synthesized model if the transition can not be recognized correctly. In essence, these unfavorable results produced by LTS based model are generated by the fact that the state is determined by the events between states, not the state variables. To tackle the embarrassments above, the dissertation introduces the concept of fluent which serves as the interface of event sequence and state model. Consequently, an improved LTS algorithm is presented based on state invariants derived by fluent annotation to enhance the readability of the abstract state machine.In LTS model, it is assumed that all states are acceptable. However, this assumption imposes restrictions on modeling and analysis of the non-deterministic behavior during software development. Therefore, a merging and refining approach to the non-deterministic behavior based on event model is presented. In this approach, the system behavior is described by modal transition system(MTS), and the LTSs model is extended to MTSs model. Firstly, the refinement rules between event models are defined, and then generate the merging rules. Following that, the minimum common refining pattern of the behavior model is generated using the merging rules to remove the non-deterministic behavior of the local model.Roughly speaking, the complex system is established based on component reuse and combination in software development. The dissertation describes our research that utilizes category theory to analyze the composability of software components. Two categories are constructed: one is component realization, the other is component specification. If there exists a functor between the two categories, all software components are composable. Finally, the strongest specification functor is deduced in this study.