A forecasting model of dissolved gases in transformer oil is established based on p-SVRM(v-Support Vector Regression Machine) algorithm and the Bayesian framework is introduced to optimally select its parameters. An evaluation mechanism combining forecasting accuracy and model simplicity is set and the improved BIC(Bayesian Information Criterion) is taken as the final evaluation function to quantify the evalua tion mechanism. The case study shows that,compared with GM(Gray Model),v-SVRM forecasting model has higher forecasting accuracy with the same small-scale samples and better performance in the proposed model evaluation function.%基于v-支持向量回归机(v-SVRM)算法建立了变压器油中溶解气体变化预测模型,并引入贝叶斯证据框架对预测模型的参数进行了优化选取.同时,结合预测模型的预测正确率及预测模型的简洁度建立了预测模型的评价机制,并利用改进的贝叶斯信息标准(BIC)作为最终的评价函数量化了评价机制.在实例中与灰色理论预测模型进行了比较,结果表明在同为小样本训练数据的情况下,v-SVRM预测模型比灰色模型有更高的预测准确率,且在所提出的评价机制里表现更好.
展开▼