Hazardous liquid pipeline (HLP) leaks not only result in energy waste and environmental pollution, but also pose a threat to people's lives and property. The estimation of leakage parameters is an essential part of risk assessment and environment pollution assessment. However, current common leak detection methods are mainly based on physical models with assumptions and are susceptible to noise. Limited historical leakage data render it impossible to develop a leak model in advance. To address this problem, this study establishes a pipeline digital twin model that simulates a pipeline leak to generate leakage data. A conditional variational auto-encoder (CVAE) framework is proposed to estimate the leakage parameters based on data detected by upstream and downstream meters once the HLP leak occurs. CVAE can treat the high-dimensional detected data as labels to overcome the dimensionality problem. Based on the CVAE framework, an online real-time leakage parameter estimation tool for HLP is formed. To qualify the performance of the approach, a sensitivity analysis for the structure of the CVAE framework is evaluated. Finally, four examples demonstrate the effectiveness, stability, and applicability of the proposed method.

危险液体管道（HLP）的泄漏不仅导致能源浪费和环境污染，而且对人们的生命和财产构成威胁。泄漏参数的估计是风险评估和环境污染评估的重要组成部分。但是，当前常见的泄漏检测方法主要基于带有假设的物理模型，并且容易受到噪声的影响。由于历史泄漏数据有限，因此无法提前开发泄漏模型。为了解决这个问题，本研究建立了管道数字孪生模型，该模型可模拟管道泄漏以生成泄漏数据。提出了一种条件可变自动编码器（CVAE）框架，以在HLP泄漏发生时根据上游和下游仪表检测到的数据估算泄漏参数。CVAE可以将高维检测到的数据作为标签来解决维数问题。基于CVAE框架，形成了在线的HLP实时泄漏参数估计工具。为了验证该方法的性能，评估了CVAE框架结构的敏感性分析。最后，四个例子证明了该方法的有效性，稳定性和适用性。