基于加权决策树算法的调度指令风险评估方法

Risk assessment method of dispatching orders based on weighted decision tree algorithm

针对海量电力数据人工分析困难、调度人员难以评估调度指令安全风险、传统风险评估方法适用性差的问题,提出了一种基于加权决策树算法的调度指令风险评估方法。对海量电力数据进行多维指令画像并建立调度指令专家知识库,通过该数据训练决策树模型找到合适的调度指令风险评估规则,使用该规则对新的调度指令进行风险评估,并通过引入代价敏感的学习方法解决了样本不平衡问题。实验表明,所提方法可以避免人工分析庞大的实时电力数据流,准确高效地对调度指令进行风险评估。在构建的不平衡数据集上,使训练集和测试集的准确率保持在95%以上,并将小样本的召回率由91.7%提高到了100%,避免因小样本分类错误导致的严重问题。

Aiming at the difficulty of manual analysis of massive power data,the problem for dispatching personnel to make risk assessment on dispatching orders and the poor applicability of traditional risk assessment methods,this paper proposes a risk assessment of dispatching orders based on weighted decision tree algorithm. Multi-dimensional order profile is made and the expert base of dispatching orders is established to train the decision tree model,which is to find the appropriate rule for risk assessment of dispatching orders. Then the rule can be used to make risk assessment on new dispatching orders. The introduced cost-sensitive training method solve the problem of unbalanced samples. The experiments show that the proposed method can avoid manual analysis of real-time power data flow and make accurate and efficient risk assessment on dispatching orders. The classification accuracy of train set and test set was kept above 95% and the recall rate of small samples was increased from 91.7% to 100% on the test set,which was to avoid serious problem caused by misclassification of small samples.