Cascading Collapse Effect Of Complex Network And Predictable Research Of Software Project

The application research of complex networks has been expanded in more and more fields in recent years,from the initial large-scale power network,transportation network and aviation network,to the recent Internet community network,etc.,which have attracted the attention of researchers from all walks of life.In complex networks,the large-scale cascading failures effect caused by small faults is a common phenomenon in real life,and usually results in catastrophic consequences.Therefore,in order to meet people's reliability needs for various complex networks,the study of cascading failures effects on complex networks has become a hot branch of complex networks in recent years.Modeling cascading failures effects is the cornerstone and key to analyzing cascade failures.Tracing back to the history of software development,since the first generation of software development in the 1960 s,software developers or project investors have been worried about the high risk of the final formation of software projects,according to the relevant Standish Group.According to the survey,only software projects in more than 8,000 fields were sampled and tested in the United States,and the average failure rate of total software projects was as high as 72%.Today,the success rate of massive software projects all over the world is still not ideal.In the early stage of R&D of most software projects or projects with high originality,there are only few reference cases in the development process,and most of the risks of the project can only be subjectively evaluated by the expert team through previous experience.This can have considerable accuracy in the case of low software complexity in the past,but with the development of technology,software complexity has increased exponentially,and the increase in software complexity directly leads to the complexity of software failure With the increase,the accuracy and reference of this kind of empirical supervisor evaluation are getting lower and lower.Therefore,objective and quantitative risk prediction methods have become the focus of everyone.From the perspective of system science,this article uses complex networks and other methods and technologies to study the cascading failures effect of complex software as a new research perspective,based on the current research of multi-layer perceptron technology and deep learning technology.Two different risk prediction methods in the past are proposed.Firstly,taking the Github community as an example,by collecting data from massive community projects,the risk transmission and cascading failures response of the open source project during the development process were analyzed.By focusing on analyzing the most common risk transfer modes between the two open source projects,technical association and cooperative association,combined with the collected data,it is concluded that the failure of a single project will produce a cascading failures response of a certain scale.Regarding the data of a large number of successful and failed projects in the Github open source community,through the design of reasonable features,the successful and failed project data is trained based on support vector machines,and the data cleaning and optimization methods make the trained models better.Predicting the risk of project failure provides an effective basis for the long-term development and risk assessment of the open source community.Secondly,the rapid development of deep learning has been proven to have better performance and generalization capabilities in the fields of image processing,target detection and recognition,and natural language processing.Based on the collected data and design features,this paper builds a deep convolutional neural network model and related loss function,and trains the model for project sustainability and risk prediction.Experimental results show that,compared with traditional machine learning methods,deep learning has greater advantages for nonlinear mapping modeling of features,such as the stronger generalized ability of the trained model,the more accurate project sustainability and risk prediction.