Abstract: SVM（Support Vector Machine），whether it can get a good forecasting effect when used in practical problems with the characters of small-sample is determined by whether the key parameters of the algorithm can be set successfully，this bottleneck problem has been a hindrance to the application of this algorithm in practical problems with the characters of small-sample.Based on analyzing the performance of SVM parameters as a regression estimation method，an improved SVM algorithm with APSO（Adaptive Particle Swarm Optimization） optimizing the parameters of SVM and the relevant engineering cost forecast model on the background of power engineering is proposed.With this model，a practical power engineering cost prediction simulation analysis is carried out.And compared with the traditional SVM algorithm，it proves that the power engineering cost prediction accuracy of the improved algorithm is higher and the algorithm has high operation speed.

Key words: Particle Swarm Optimization（PSO）, Support Vector Machine（SVM）, power engineering cost

摘要： 支持向量机（SVM）算法应用于具有小样本特征的实际问题时是否能获得到良好的预测效果，取决于能否成功地设置该算法的关键参数，这一瓶颈问题一直阻碍着SVM在具有小样本特性的实际工程中的应用。在分析SVM回归估计方法参数性能的基础上，提出了以自适应粒子群算法（APSO）优化SVM关键参数的改进SVM算法，并以变电工程为背景给出了相应的工程造价预测模型。运用此模型，对某实际变电工程实例进行了造价预测仿真分析，并与传统的支持向量机算法进行比较，结果说明改进的支持向量机算法具有良好的变电工程造价预测精度，且速度较快。

关键词: 粒子群算法, 支持向量机算法, 变电工程造价