基于模型-数据混合驱动的MMC阀损耗计算方法

Loss Calculation Method for MMC Valves Based on Hybrid Model-driven and Data-driven Approach

针对现有模块化多电平换流器(MMC)阀损耗计算方法无法兼顾准确度和计算速度,仅使用数据驱动算法又会导致机理不明、可靠性低的问题,提出了一种基于机理模型和数据混合驱动的MMC阀损耗计算方法。该方法物理过程明确,充分利用了现场数据隐含信息,能够快速准确计算不同工况、不同控制策略下的阀损耗。首先推导了MMC阀损耗的计算公式,并引入热电类比模型,获得了不同冷却水温下的阀损耗理论计算结果。然后,结合全连接神经网络,利用实际MMC换流站数据对理论结果进行了修正。最后,以某MMC换流站实际工程作为算例,利用该工程实际参数和现场数据,对所提出的计算方法进行验证,证明了该方法的有效性和准确性。

Current MMC valve loss calculation methods are either time consuming or inaccurate,and besides,only adopting data-driven calculation methods may lead to low reliability and unclear mechanism,thus a MMC valve loss calculation method based on hybrid model-driven and data-driven approach is proposed in this paper.The method has a clear physical process,and enables quick and accurate calculation of MMC valve losses under different working conditions and different control strategies.Firstly,the calculation formulas of MMC valve losses are derived and the thermoelectric analog model is introduced into the model.Secondly,results of the mechanistic model under different temperatures of cooling water are obtained analytically.Then,the fully connected neural network is innovatively combined with the obtained mechanism model,and subsequently,the results of mechanistic model are corrected by utilizing measurement data of a MMC converter station.Finally,by using the real parameters and field data of an exemplary actual MMC converter station,the validity and accuracy of the proposed hybrid model-driven and data-driven approach are verified.