This is a GitHub repository for the project "A Novel Estimation Method for Dual Active Bridge Converter Based on Intelligent Algorithm".
Parameter estimation with various nonidealities and uncertainties in a practical circuit is a key but challenging issue in the applications of dual active bridge (DAB) converters for health diagnosis and preventive maintenance. However, most of the existing parameter extraction and identification methods for DAB converters still require a large number of experimental data and expensive measurement costs, while the established estimation model usually lacks generalization. In addition, most of the DAB converter models in previous studies adopt ideal models and ignore circuit nonidealities, which will seriously affect the accuracy and reliability of the models.
Intelligent methods can effectively mine the rule of data without considering the complex mapping relationship between DAB models and estimated parameters. Consequently, these methods are expected to become the mainstream approach for parameter estimation. This paper proposes a light, flexible, and yet reliable multiparametric estimation algorithm for isolated DAB converters. This approach combines the back-propagation neural network (BPNN) and the steady-state averaged-value model (AVM) to simultaneously predict the circuit and control parameters of a DAB converter. As a salient role of AI, BPNN possesses tremendous advantages due to its good performance in nonlinear regression, prediction, classification, and data structure exploration. To further optimize BPNN, a genetic algorithm (GA) is employed. The GA-BP-based model was trained and testified by the data from the mathematical model of DAB converter steady-state AVM, so it did not increase any computational. Furthermore, this parameter estimation method reduces the need to build experimental platforms and synchronize measurements. It should be emphasized that DAB modeling (AVM) takes into account circuit nonidealities, including dead time in the bridges, magnetizing inductance, equivalent series resistances (ESR), thereby enhancing the accuracy of the model.
Illustrative results demonstrate that the combined AVM+BPNN+GA model outperforms other existing proposals and is capable of achieving accurate results for both the estimation and prediction of parasitic and control parameters.
- MATLAB == 2021b
- PLECS == 4.5.6
- Simulink
For the DAB used as a DC transformer, the input and output voltages are nearly constant, so single-phase-shift (SPS) control is the most widely used one on account of its simplicity and fast dynamic response. Therefore, SPS control is used in this article. The topology of the DAB converter is illustrated in Fig. 1.
Fig. 1 Topology of the DAB converter and SPS control system.
The formulation of conventional modeling (both time and frequency domain) is based on the following assumptions—the circuit does not possess nonidealities and parasitics, such as magnetizing inductance, equivalent series resistances (ESR), and dead times. Compared with traditional modeling of the DAB converter, this paper proposes a reduced-order averaged-value model (AVM) that considers parasitic parameters and the effects of dead-time periods.
- (1) Effects of dead-time periods
[1] presents a comprehensive theoretical analysis of dead-time effect in the DAB converter. In [1], there are three sub-states of the dead-time effect (boost state, buck state and matching state), and each state has multiple modes. Taking the buck state of the DAB converter as an example, there are 6 modes in total (Fig. 2). We selected the two most typical modes as examples for modeling, and considered parasitism during the modeling process.
Fig. 2 Operation waveforms of the IBDC with dead-time effect in buck state. (a) Mode 1. (b) Mode 2. (c) Mode 3. (d) Mode 4. (e) Mode 5. (f) Mode 6.
- (2) Effects of Magnetizing Inductance and Core Losses
While considering the impact of dead-time, we further consider parasitics, e.g., magnetizing inductance and core losses. Fig. 3 depicts a general equivalent detailed model of DAB
converter considered in this paper.
Fig. 3 Detailed steady-state equivalent circuit of the DAB.
-
(3) Implementation of the proposed AVM based on MATLAB
Fig. 3 Simulation example of partial equations of the mathematical model of the proposed AVM
-
(4) Procedure to run this code
-
Click on code and download zip
-
Click on (The proposed AVM)
-
Click on (MATLAB-model-simulink)
-
MATLAB implementation of AVM considering high-frequency transformer parameters and parasitics
1 RUN TF.m; Parameter_preparation.m 2 RUN RAVM.slx-
Automatic PI tuner
Fig. 4 PI tuner
Fig. 5 The process of automatically tuning PID parameters
-
PLECS implementation of AVM considering circuit nonidealities;
Fig. 6 Detailed steady-state equivalent circuit of the DAB.
-
Nowadays, artificial intelligence (AI) powers many aspects of industrial applications. It has been successfully applied for power electronic systems for specific targets, such as designing time reduction, weighting factors design, performance prediction, and so on. In terms of data prediction, BP neural network has the advantages of high speed, good effect, and high precision. It has been widely used for structure optimization and result prediction.
Although BP neural networks have strong nonlinear mapping capacity and limited fault tolerance, the BP neural network algorithm has obvious limitations in practical engineering applications. Because the global error of the BP neural network is a multidimensional nonlinear function with an Stype function as the argument, there are multiple local minima on the error surface. Thus, the BP algorithm can easily fall into the local optimum.
Genetic algorithm (GA) is a highly parallel adaptive detection algorithm that mimics the theory of biological evolution in nature [27]. It is optimized to distinguish information about individual genetic changes, retaining genetic traits that are well adapted to the environment, and eliminating those that are poorly adapted. Therefore, the current problem is solved by using GA to optimize the weights and thresholds of the BP network. The combined network has an extremely strong ability to solve practical problems.
GA optimized BPNNs have been developed based on biological intelligence techniques. Owing to their advantages of learning, generalization, fast computation, and easy implementation, GA-BP neural networks have been widely used for monitoring, control, parameter estimation.
A non-invasive method based on intelligent algorithms for estimating circuit and control parameters of a DAB converter has been proposed in this paper. The proposed parameter estimation method is shown in Fig. 6.
Fig. 7 Overall diagram of the proposed parameter estimation method.
RUN main.mlx RUN main.mlx 
Fig. 8 Training process of GA-BPNN
In this article, the following models are also utilized for comparison.
We appreciate the following github repositories/IEEE DataPort(TM) a lot for their valuable code base or datasets:
If you have any questions, feel free to contact Qianyi Shang through Email ([email protected]) or Github issues. Pull requests are highly welcomed!