Abstract:

The statedbased and pathbased software reliability evaluation models suffer from the high computational complexity and the absence of robustness for the componentbased Web system evaluation with high complex structures. A Componentbased BackPropagation neural network Reliability Model (CBPRM) with a low computational complexity and robustness is proposed. The CBPRM employs the component reliabilities as the backpropagation neural network inputs. Based on the component reliability sensitivities, the neurons are optimized dynamically and the backpropagation neural network outputs the final Web system reliability evaluation. The theory analysis and experiment results present that the computational complexity of the CBPRM is evidently lower than the contrast models and the reliability evaluating accuracy is assured for the componentbased Web system with high complex structures.

Key words: reliability model;componentbased Web system;reliability sensitivity;backpropagation neural networks