A KPIs-Based Reliability Measuring Method for Service System

Distributed systems may experience various abnormalities due to the influence of sub-node status and complex network environment. Failure to respond in advance will lead to system failure and heavy losses. Therefore, it is very important to obtain real-time reliability values of the service system. This paper proposes a KPIs-based service system reliability measurement method. First, a KPIs feature selection method based on improved GMM is proposed to filter the KPIs data collected by the server and extract some KPIs features that play a key role in reliability measurement. Then, a reliability measurement model based on LightGBM is constructed, and the Focal Loss function is also introduced to further solve the model performance problem caused by data imbalance. Finally, a hyperparameter tuning method based on the climbing evolutionary strategy algorithm and the improved Bootstrap verification method is proposed to train the constructed reliability measurement model. Using the trained Focal-LightGBM model can quickly measure the real-time reliability of the service system. Experiments on public data sets show that compared with other reliability measurement methods, the method in this paper has better results on accuracy, MSE, R2-score, measurement time and other indicators.