Slope reliability assessment is an efficient methodology for landslide risk mitigation. However, it is also a challenging task owing to various uncertainties and randomness of soil properties in geotechnical engineering. As the slope is a complicated system with high nonlinearity, the stability analysis with high precision and reliability fails to be obtained from the traditional response surface method. To counteract this limitation, this paper proposes a technique based on machine learning for slope reliability evaluation, namely a multi-objective grey wolf optimization-multi-kernel-based extreme learning machine model based on strength reduction method. The probability of slope failure is estimated by using the developed model when connected with Monte Carlo Simulation. Model application to probabilistic evaluation of slope was conducted by two typical case studies. The results show that as compared with Monte Carlo Simulation and other traditional techniques, the developed machine learning method, which combines the advantages of strength reduction method, the multi-objection grey wolf optimization and multi-kernel-based extreme learning machine, demonstrates better computational performance and applicability, and achieves more accurate and reliable failure probability. Further, the proposed approach can also be applied to predict probability of slope failure with consideration of different coefficient of variation and correlation of soil properties. Hence, improved failure probability can be obtained from the developed method, which could offer crucial information for decisions with regard to early landslide warning.

边坡可靠性评估是减轻滑坡风险的有效方法。但是，由于岩土工程中各种不确定性和土壤性质的随机性，这也是一项艰巨的任务。由于斜率是一个具有高非线性度的复杂系统，因此无法通过传统的响应面方法获得具有高精度和可靠性的稳定性分析。为了克服这一局限性，本文提出了一种基于机器学习的边坡可靠性评估技术，即基于强度折减法的多目标灰狼优化-基于多核的极限学习机模型。当与蒙特卡洛模拟（Monte Carlo Simulation）连接时，可以通过使用开发的模型来估计边坡破坏的可能性。通过两个典型案例研究了模型在边坡概率评估中的应用。结果表明，与蒙特卡罗模拟等传统技术相比，结合强度降低方法，多目标灰狼优化和基于多核的极限学习机的优点，开发的机器学习方法具有更好的计算能力。性能和适用性，并获得更准确和可靠的故障概率。此外，考虑到不同的变化系数和土壤特性的相关性，提出的方法还可以用于预测边坡破坏的可能性。因此，可以从开发的方法中获得更高的故障概率，