基于混合幂律过程的电子商务软件系统可靠性模型

Reliability Model for E-Commerce Software Systems Based on Hybrid Power Law Process

摘要: 随着电子商务的快速发展，以京东、淘宝、拼多多等为代表的电商平台已成为现代商业活动的重要载体。这些电子商务软件系统通常具有微服务化、高并发、分布式架构等技术特征，其复杂性和稳定性面临着巨大挑战。一次系统故障可能导致订单中断、支付失败等严重后果，给平台和商家带来巨大的经济损失，因此，对电子商务软件系统进行精确的可靠性评估与预测，已成为保障其业务连续性和稳定运行的关键。本文提出了一种幂律过程分别与GO (Goel-Okumoto)模型(Power Law Process - Goel-Okumoto, PLP_GO)和DSS (Delay S-Shaped)模型(Power Law Process - Delay S-Shaped, PLP_DSS)相结合的软件可靠性模型，该模型采用极大似然估计方法进行参数估计，并与传统的GO模型、DSS模型和幂律模型进行对比分析。以蒙特卡洛方法生成的电子商务软件系统仿真故障数据为案例，结果表明，本文所构建的PLP_GO模型拟合效果及预测效果均表现最优，能够有效描述电商系统在用户访问量激增、第三方服务集成等复杂场景下的可靠性变化规律，因此该模型适用于电子商务软件系统的可靠性评估。

Abstract: With the rapid development of e-commerce, platforms such as JD.com, Taobao, and Pinduoduo have become crucial carriers of modern commercial activities. These e-commerce software systems typically feature microservices architecture, high concurrency, and distributed infrastructure, presenting significant challenges in terms of complexity and stability. System failures may lead to severe consequences including order interruption and payment failures, causing substantial economic losses to both platforms and merchants. Therefore, accurate reliability assessment and prediction for e-commerce software systems have become critical for ensuring business continuity and stable operation. This paper proposes a hybrid software reliability model that integrates a Power Law Process (PLP) with the Goel-Okumoto (GO) model (referred to as PLP_GO) and the Delay S-Shaped (DSS) model (referred to as PLP_DSS). The parameters of the proposed models are estimated using the maximum likelihood estimation method. A comparative analysis is conducted against traditional GO, DSS, and Power Law models. Using Monte Carlo simulation-generated failure data of an e-commerce software system as a case study, the results demonstrate that the proposed PLP_GO model achieves the best performance in both fitting accuracy and predictive capability. It effectively captures the reliability evolution of e-commerce systems under complex scenarios such as sudden surges in user traffic and integration with third-party services. Therefore, the PLP_GO model is well-suited for reliability assessment of e-commerce software systems.