Detecting Deficiency Data For Intelligent Model In Software Systems By Model Verification

Intelligent software systems can learn new requirements and control systems to adjust their behavior for adapting to environmental changes.The intelligent software system usually consists of two parts: traditional components and intelligent components.The traditional components reflect the determined requirements of the system,while the intelligent components are responsible for making optimal behavior decisions in dynamic environment.At the same time,the addition of intelligent components also brings many uncertain factors to the software system.It is of great significance to ensure the quality of the model.At present,in the work of ensuring the quality of intelligent software systems,researchers mainly consider factors such as uncertain demand and uncertainty of structural,but lack the attention to the uncertainty of data quality that constitutes intelligent components.This thesis focuses on the impact of defective data on the quality of intelligent software systems,the data are derived from the training set of the neural network that constitutes the intelligent components.Our work is divided into two aspects.One is to carry out model verification on the intelligent software system: Firstly,the Petri net model is established according to the determined requirements,the neural network is trained according to the provided sample data,and integrate these two parts to construct the adaptive Petri net model to describe the intelligent software system;then the fuzzy rule is extracted from the neural network and transformed it into Petri net model,we can obtain a hybrid Petri net model.Finally,the hybrid Petri net model is described by a hybrid automaton,and the system properties are verified by the KeYmaera prover.The second part is to locate the defective data through the results of the model verification: to achieve this purpose,the entire model inspection process is reversely analyzed.Firstly,some information related to the violation attribute can be obtained from the report generated by the model verification tool,and then the specific location of the information generated by the hybrid automaton or the information closely related thereto is determined;according to the relationship among the hybrid automaton,the reachability graph and the Petri net,we can locate on the Petri net related positions and transitions,and then locate the corresponding fuzzy rules;for this,a new method is proposed to locate data by fuzzy rules in our work,and we analyze its effectiveness.Through the series of processes,the model is tracked from key information points until the defect part is found.At last a simple manufacturing system is used to illustrate the proposed method.The main contributions of this paper are as follows:(1)A model verification technology for intelligent software systems which focuses on the uncertainty of data quality is proposed.The data refers to the training sample data of neural networks in intelligent components;(2)A method for locating defective data is proposed,which reversely traces the entire system through the results of the model verification to find the defective portion in the sample data used to train the neural network;(3)The model verification technique and the method of locating the defective data explain to some extent that the quality of the sample data will affect the quality of the entire intelligent software system,and provide another way to guarantee the quality of the system.