边坡稳定性分析中，模糊点估计法能同时考虑模糊不确定性和随机不确定性因素。针对传统模糊点估计法计算工作量大的缺点，提出一种神经网络改进模糊点估计法。利用拉丁超立方抽样法和径向基函数神经网络（RBF）建立边坡安全系数的预测模型；对黏聚力和内摩擦角等模糊随机变量取λ截集，并在各截集水平对参数进行组合；利用建立的预测模型对各参数组合的安全系数进行预测；最后由统计矩点估计法计算边坡的可靠度指标。实例分析表明：改进模糊点估计法使用方便、结果可靠，且能通过增加λ截集水平的数目来提高计算精度。对于含有2～4个模糊随机变量的边坡，采用改进模糊点估计法计算可靠度时λ截集水平的数目可近似取25。

The fuzzy point estimate method can simultaneously consider fuzzy and random uncertainty in slope stability analysis. To overcome the drawbacks of a great deal of computing existing in traditional fuzzy point estimate method, an improved fuzzy point estimate method is proposed on the basis of artificial neural network. Firstly, the Latin hypercube sampling method and radial basis function (RBF) neural network are adopted to establish a prediction model for determining safety factor of slopes. Secondly, fuzzy-random variables, i.e. cohesion and friction angle, are transformed into interval numbers by the λ cut set approach, and then combined at each cut set level. The corresponding safety factor of each variable combination is obtained with the established prediction model. Finally, reliability index of slope is calculated using the point estimate method. A practical example is analyzed, showing that the proposed method is convenient and reliable to evaluate slope stability, and can be further improved to increase the computational accuracy of slope reliability index by increasing the number of λ cut set levels. For those slopes with 2-4 fuzzy-random variables, as the proposed method is used to compute the reliabilty of a slope, the number of λ cut set levels is recommended to be 25.