Parameter tuning for dynamic digital twins in inverter‐dominated distribution grid

Abstract

This research proposed the methods based on the neural network (NN) to build the digital twins (DT) of the inverter model. The proposed methods can be divided into two groups: firstly, an online tuner for the proportional–integral (PI) controller is formulated through backpropagation (BP) algorithm; and secondly an NN-based identifier is used to approximate the non-linear functional dynamics of the targeted control loop of the inverter. The design of PI tuner is based on the deviation between the output of the model and the reference output from measurement data. Then, according to the difference, the tuner can calculate the appropriate parameter of the current and voltage controller to track the dynamic behaviour of the reference model. The NN identifier is, however, to replicate the dynamic character of the reference model by NN which is initially trained offline with extensive test data and afterwards is applied to online tuning. In order to compare the advantages of the methods by NN with the traditional ones, the system identification and parameter estimation are also utilised to build the DT of inverter model. The performance of these methods will contribute to illustrating that NN identifier is effective in the building of DT.