知识-数据融合的直流送端系统暂态过电压幅值预测

Transient Overvoltage Amplitude Prediction for DC Sending-end System Based on Knowledge-Data Fusion Model

对于含大规模风电的高压直流外送系统,换相失败、直流闭锁等故障可能导致送端交流母线出现暂态过电压,进而引发新能源连锁脱网,严重威胁电网的安全稳定运行。为刻画直流故障对系统安全稳定运行能力的影响,提出一种知识-数据融合的直流送端系统暂态过电压幅值预测方法。首先,通过修正决策树分裂过程中的分支质量衡量指标提出改进决策树算法,挖掘暂态过电压与关键电气量的映射关系,提升过电压幅值预测模型的评估性能。其次,基于系统机理模型分析送端系统功率交换情况与交流母线电压的关联关系,推导模型驱动的换流站母线暂态过电压幅值解析表达式。最后,结合理论分析方法,建立知识-数据融合模型以快速修正过电压幅值理论计算结果,提升融合模型在训练样本不足或输入特征选择不当情况下的鲁棒性。面向存在暂态过电压问题的直流送端系统,仿真结果验证了所提暂态过电压幅值预测方法的有效性。

For the high-voltage direct current(DC) sending system containing large-scale wind power,faults such as commutation failure and DC blocking may lead to transient overvoltage on alternating current(AC) bus at the sending end,leading to the tripping risk of renewable energy units and posing a serious threat to the safe and stable operation of power grid.To characterize the impact of DC faults on the safe and stable operation ability of the system,the transient overvoltage amplitude prediction method for the DC sending-end systems based on knowledge-data fusion model is proposed.Firstly,an improved decision tree algorithm is proposed by modifying the branch quality measurement index during the decision tree splitting process.The mapping relationship between the transient overvoltage and the key electric quantities is fully explored,and the evaluation performance of the overvoltage amplitude prediction model is improved.Secondly,based on the system mechanism model,the correlation between the power exchange situation of the sending-end system and the AC bus voltage is analyzed,and the model driven analytical expression for the transient overvoltage amplitude of the converter station bus is derived.Finally,combining theoretical analysis methods,a knowledge-data fusion model is established to quickly correct the theoretical calculation results of the overvoltage amplitude and improve the robustness of the fusion model in the case of insufficient training samples or improper input feature selection.The simulation results verify the effectiveness of the proposed transient overvoltage amplitude prediction method for the DC sending-end system with transient overvoltage problem.