Research On Digital Twin Model Of UHVDC System For Physical Space Complete Mapping

Loss measurement of UHVDC transmission system is an important basis for State Grid Corporation of China to achieve energy saving and consumption reduction,and its digital twin model is also helpful to achieve the development goal of "Carbon peak and carbon neutralization".However,the digitalization level of the existing UHVDC system is difficult to support the research on loss reduction.Therefore,this paper proposes a digital twin model construction algorithm for the system oriented to the complete mapping of physical space.The main contents are as follows.Firstly,the UHVDC transmission system and its loss measurement are studied.Based on the topological structure and equivalent loss circuit of UHVDC transmission system,the composition and sub-loss characteristics of loss measurement are studied.Finally,combined with the digital space prior data set of UHVDC loss measurement,the strong coupling between data is analyzed,which lays a foundation for the subsequent digital twin model based on deep learning model.Secondly,a loss metering digital twin model based on deep learning model is proposed.In view of the strong coupling and long sequence of prior data in digital space,the LSTM model is optimized,and the improved Res-LSTM digital twin model of loss metering is further proposed.The experimental results show that the Res-LSTM model reduces the MSE errors by 37.57% and 11.8% respectively compared with the LSTM model in the total loss calculation mode and the sub-loss sequence calculation mode,both of which are better than other deep learning models.Then,a modeling and application algorithm for physical space complete mapping digital twin model is proposed.Aiming at the defects of conventional deep learning models and modeling methods,a new physical space-oriented modeling and application architecture of fully mapped digital twin DTformer was proposed.On this basis,the detailed network structure of DTformer model,the complete mapping modeling and soft strategy application of DTformer model are proposed to improve the accuracy of digital twin model,the ability to deal with data loss and inversion ability.Experimental results show that all error indexes of DTformer model are far superior to conventional deep learning algorithms.Taking MSE error as an example,the maximum error is reduced by 92.55%,and the minimum error is reduced by 63.79%.Finally,an improved lightweight algorithm for digital twin model is proposed.Combined with the two theoretical bases of lightweight modules and architecture knowledge distillation,the lightweight modules with cross-channel feature interaction are selected,and the small architecture model of architecture knowledge distillation is constructed,and the knowledge is distilled into the small architecture model constructed by lightweight modules.The experimental results show that the improved lightweight algorithm can achieve the fastest operation speed,which is 7.97 times faster than DTformer model and 1.28 times faster than distilled DTformer mini model.The loss of accuracy is within an acceptable range.The algorithm in this paper realizes the modeling and loss measurement application of the digital twin model oriented to physical space complete mapping,which lays a good foundation for the subsequent research and application of loss reduction optimization.