Cascaded H-bridge Type Power Electronic Transformer Electromagnetic Interference Study Based on Parallel Simulation

Power electronic converter (Power Electronic Transformer, PET) is widely used in power systems, so the research on PET electromagnetic disturbance (Electro Magnetic Interference, EMI) cannot be ignored. PET consists of many insulated gate bipolar transistors (Insulate-Gate Bipolar Transistor, IGBT). A large number of converter levels will cause problems such as small simulation steps and long simulation times. To solve the above problems, firstly, the IGBT and its anti-parallel diodes are modeled by an 8-node model to reduce the complexity of the system; secondly, after dividing the model into fine-grained models, CUDAC coding is used to realize the parallel operation of the system on the graphics processing unit (Graphics Processing Units, GPU). Finally, the feasibility of the proposed method is compared with the PSCAD simulation results.


Introduction
In vigorously developing clean energy and environmental protection, it is an inevitable trend for clean energy DC grid connection, electric vehicle charging and large-scale energy storage system to enter the power grid.PET is one of the core devices used in power systems, which are mainly composed of a large number of semiconductor devices and high-frequency transformers, and have the advantages of multiple AC-DC port access, voltage level conversion, and power quality control [1][2][3] .Most scholars on PETs have focused on the modeling simulation of their topology and the optimization application of control strategies [4][5][6][7][8] .EMI also has an impact on the performance of PET.
Suppressing EMI in a system can improve the overall system performance and increase converter reliability.However, the number of modules, complex topology and nodes in PET systems can reduce the system's efficiency due to the high computational burden of conducting the study.Adding Line Impedance Stabilization Network (LISN) to the system for EMI detection will only increase the number of computational nodes in the system, making the computation more complex.Many scholars have improved the simulation efficiency in terms of modeling, and in the literature [9], the EMI analysis system of the converter is modeled.However, the effect of parasitic effects in EMI is ignored and lacks accuracy.The literature [10] analyzes the EMI model of the system and techniques to suppress the noise.The proposed method is not applicable to AC-DC-DC power converters.In addition to modeling to reduce the complexity of system operation, the use of GPUs can also improve operational efficiency.
In this paper, a PET system with many cascades, a large number of switching devices and high-frequency isolation transformers is computed in parallel on a GPU, and the EMI analysis of the  The overall structure diagram of the PET studied in this paper is shown in Figure 1.The rectifier circuit is connected to the three-phase voltage source.The isolated converter is connected to the load.The part in the blue box is the single-phase triplex PET structure, the orange box is the AC/DC rectifier, and the red box is the DC/DC isolated converter.In this system, The dual active bridge (Dual Active Bridge, DAB) converter containing a high-frequency isolation type transformer implements the voltage change in the system, which includes mainly 12 CM noise sources [11] .

Modular PET modeling approach
Decoupling the system's circuit and using parallel accelerated simulation on GPU can solve the problems of the small simulation step size and long simulation time.In this paper, the H-bridge and DAB are equated as a model with 8 nodes [12] , and its discrete equivalent circuit is shown in Figure 2. The detailed modeling of the IGBT has been described in the literature [13].In this paper, the same class of capacitors is represented as follows: Where  、、β、δ and  diff are parameters in the physical equivalent circuit model,  JVCT is the PN junction voltage.The expression for the diffusion capacitance on the emitter path of the BJT and the continuity diode is: Where  is the saturation current;  = / is the temperature voltage.After discretizing and equating the components, the MOSFET conductance, transconductance, BJT conductance and the Norton equivalent current source of both are obtained as: eq mos mos mosVds ds mosVgs gs eq bjt b BJT eb Where g is the conductance; g is the transconductance;  is the drain-source voltage;  is the gate-to-source voltage;  is the base-emitter voltage.
According to KCL and KVL, the nodal equations in the discrete equivalent model are:

Parallel Simulation Decomposition
Based on the decoupling given in the previous section, this section implements PET system-level simulation by modularly decomposing the entire PET based on the principle of CUDAC-coded parallel accelerated computation.In this paper, the cascaded H-bridge sub-module is separated from the corresponding bridge arm to construct a fine-grained circuit to improve numerical stability.In each step, the cascaded H-bridge sub-module accepts the current from the previous step of the bridge arm as input to obtain the output of the module and the DAB main circuit.The simulation is advanced sequentially to the end.The separation circuit is shown in Figure 3.

Simulation Analysis
The model topology of the system built in this paper is input series output parallel.The simulation time step is set to 10ns, and the system parameters are shown in Table 1.

Parallel acceleration simulation comparison analysis
Figure 4 shows the load-side output voltage comparison, with a maximum relative error of 0.7%.Figure 5(a) shows the cascaded H-bridge input-side voltage comparison with a maximum relative error of 0.04%.Figure 5(b) shows the cascaded H-bridge output-side voltage comparison, with a maximum relative error of 0.2%.The comparison of the simulation plots for the steady-state case shows that the decoupling method in this paper can be implemented in the simulation of the system by parallel calculation.LISN is located on the right side Figure 6.Diagram of common mode comparison when LISN is located on different sides Figure 6 f for the LISN is on both sides of the different input power PET system CMEMI comparison chart.It can be seen that the trend of harassment is the same when the LISN located on the right side is significantly lower than the LISN situated on the left side of the harassment value.When the frequency exceeds 1MHz will appear the maximum value of the harassment will be.
Figure 7 compares the harassment results between the module simulation and the parallel simulation when the LISN is located on different sides of the system power supply, respectively.From the result comparison graph, it can be seen that the parallel simulation and module simulation results are approximately the same.The overall harassment change trend is the same, and it can be seen that the parallel simulation results fluctuate less and the values are more accurate.The comparative analysis of the results verifies the feasibility of parallel accelerated simulation for detecting conducted electromagnetic disturbance in PET systems.

conclusion
In this paper, we first give the physical decoupling model of the key parts of the system.When calculating the system's EMI, each part's internal characteristics need to be analyzed.So the 8-node physical equivalent model is used to decouple them.The decoupled model is divided using fine-grained partitioning.The model is programmed based on C language to enable parallel simulation of the system, increasing the system's operation speed and reducing simulation time.
Comparing the system in the steady state with the module simulation, the simulation result graph shows that the maximum relative error between the parallel simulation and the module simulation is not more than 1%, and the high speed and efficiency of the system simulation can be achieved.The feasibility of the method proposed in this paper is verified.

7 Figure 1 .
Figure 1. Figure of the overall structure of PET

Figure 3 .
Figure 3. Diagram of the sub-module separated from the bridge arm

Figure 7 .
Figure 7. Diagram of Disturbance Comparison of Different Input Power Supplies Located on Both Sides of LISN