Abstract: Basing on the speciality of traffic load in the paper, a traffic load forecasting model based on microcanonical annealing—Support Vector Machines（SVM） is proposed. Appropriate parameters are very crucial to SVM forecasting ability, the optimal parameters selection is achieved by random walks of demons in the state space of Microcanonical Annealing（MA） algorithm. The verification on the model with real traffic data shows that, this algorithm will offer better results with higher probability to hit the global optimum than Simulated Annealing（SA） algorithm, objective function value is also decreased faster, and has high precision.

Key words: traffic load, microcanonical annealing, Support Vector Machine（SVM）, forecasting model

摘要： 根据话务量数据的特征，首次提出了一种基于微正则退火算法和支持向量机的预测模型。支持向量机参数的选择影响其预测的能力，微正则退火算法而是通过在状态空间中随机行走的虚拟妖来实现参数的优化选择。实际的话务量数据验证表明，其搜索成功率远高于模拟退火算法，目标函数值下降更快，能在短时间内搜索到最优解，且预测精度高。

关键词: 话务量, 微正则退火, 支持向量机, 预测模型