摘要： 提出一个基于人工免疫原理的RBF网络预测模型AIP-RBF，该模型使用新的克隆选择算法和免疫抑制策略，通过隐层可行解的抽取算法EAHLFS，能在聚类数目未知的情况下，生成RBF网络隐层。给出了改进的隐层节点重要度(SHLN)概念，用于指导RBF网络第2阶段的训练过程。与传统的基于聚类算法的预测模型比较，AIP-RBF具有更快的收敛速度和更高的预测精度，在实际盾构施工地面沉降预测中得到了验证。

关键词: RBF网络, 人工免疫原理, 回归, 预测模型

Abstract: This paper proposes an artificial immune principle based RBF network prediction model named AIP-RBF which uses the new clonal selection algorithm and immunosuppressive strategy. By using Extraction Algorithm of Hidden Layer Feasible Solution(EAHLFS), AIP-RBF is able to get the hidden layer of RBF without knowing the clustering number. The paper presents an improved Significance of Hidden Layer Node(SHLN) used in the second phase of the RBF training process. Compared with the traditional clustering algorithm based prediction model, AIP-RBF has faster convergence speed and higher prediction precision which has been verified to predict the actual shield construction ground subsidence.

Key words: RBF network, artificial immune principal, regression, prediction model

中图分类号: 

• TP18