# A Fault Warning Method for Electric Vehicle Charging Process Based on Adaptive Deep Belief Network

^{1}

^{2}

^{*}

## Abstract

**:**

## 1. Introduction

## 2. Problem Description

## 3. Electric Vehicle (EV) Charging Fault Warning Based on Adaptive Deep Belief Network (ADBN)

#### 3.1. Structure and Training Process of DBN

_{1}, the hidden layer ${\mathit{H}}_{1}$ as the visible layer of RBM

_{2}and the hidden layer ${\mathit{H}}_{2}$ form RBM

_{2}, and so on. The lines in Figure 2 represent the weights between the connected neurons, ${\mathit{W}}_{k}=\left\{{w}_{i,j}\right\}\in {\mathit{R}}^{n\times m}$ is the connection weight between the visible layer and the hidden layer of the kth RBM, and ${\mathit{A}}_{k}=\left\{{a}_{i}\right\}={\mathit{R}}^{n}$ and ${\mathit{B}}_{k}=\left\{{b}_{j}\right\}={\mathit{R}}^{m}$ are the visible layer bias and hidden layer bias of the kth RBM. Thus, only three parameters are required to determine an RBM.

_{1}receives information on the EV’s required voltage, required current, charging current and temperature, and trains each RBM in a bottom-up sequence using a layer-by-layer greedy learning algorithm to achieve the extraction of high-level features of the input data and the update of the connection weights of the training network. The output data is the predicted charging voltage. In the fine-tuning phase, BPNN takes the predicted charging voltage as the input and the actually measured charging voltage as the output, and continuously adjusts and optimizes the network parameters from top to bottom in the way of supervised learning.

#### 3.2. NAdam Algorithm

#### 3.3. Pearson Coefficient

#### 3.4. EV Charging Process Fault Warning Process

- Obtain the historical data of EV charging process, and divide it into normal charging data and fault charging data.
- Data normalization of normal charging data and fault charging data.
- Constructing the normal charging voltage model ADBN1 and the normal charging current model ADBN2 for EV by two stages of pre-training and fine-tuning using normal charging data.
- Input the fault charging data into the constructed ADBN1 and ADBN2 models to predict the charging voltage and charging current, calculate the Pearson coefficient between the predicted charging voltage and charging current and the measured charging voltage and charging current, perform fault warning when the Pearson coefficient is less than the set expectation value. Calculate the ratio between the number of fault warnings and the actual number of faults to test the warning performance of the models.
- The ADBN model that meets the requirements is applied to real-time charging fault warning for the EV.

## 4. Implementation of Fault Warning for EV Charging Process

#### 4.1. Data Selection and Pre-Processing

#### 4.2. Training of ADBN Model

_{1}and ADBN

_{2}are 15. Considering the accuracy of model prediction and training time, some network parameter settings of ADBN

_{1}and ADBN

_{2}are shown in Table 3.

_{1}and ADBN

_{2}. Therefore, in this paper, we take ADBN

_{1}as an example to tests the influence of different number of hidden layers L and different number of hidden layer units n of ADBN

_{1}on the accuracy of charging voltage warning with the same parameters in Table 3. The number of hidden layers and the number of hidden layer units of ADBN

_{2}are also determined.

_{1}networks with L of 1, 3 and 5 are constructed. Set the number of units in the hidden layer of ADBN

_{1}to 100 and test with conducted with the same historical normal data and fault data. The accuracy of ADBN with different hidden layers is plotted with the number of iterations as a curve, as shown in Figure 4.

_{1}is highest when L = 3, the lower accuracy when L = 1, and the lowest accuracy with a large change when L = 5. The reason for the large change of accuracy at L = 5 may be that the DBN model is more complex due to the increase of the number of hidden layers, which makes the optimization process of NAdam tortuous. According to the above analysis, the number of hidden layers of ADBN

_{1}is set to three, and then the influence of the number of units of different hidden layers on the accuracy is determined. Figure 5 is the change of accuracy corresponding to the number of units in different hidden layers.

_{1}is 50, 100 and 150, the warning accuracy is above 90%, and the warning accuracy increases with the increase of the number of hidden layer units. However, when n is greater than 100, the improvement of accuracy is small. Considering the speed of model training and over-fitting problem, the network structure with the number of hidden layer units of 100 is selected in this paper.

_{1}loss value and accuracy with the number of iterations in the fine-tuning stage. It can be seen from the observation curve that the loss error decreases rapidly from 0 to the 5th iteration, and gradually tends to a stable value after 10 iterations. Finally the loss error reaches 0.0087 at 50 iterations, and the average accuracy of voltage fault warning reaches 98.61% and stabilizes at this time.

_{2}is the same as ADBN

_{1}, the accuracy of charging current fault warning can reach 97.75%, which proves that ADBN can be adapted to the corresponding charging data when predicting the physical quantities in different charging processes. Therefore, the structure of ADBN

_{2}can be the same as ADBN

_{1}.

#### 4.3. Method Validation and Comparison

_{1}and ADBN

_{2}to predict the output, and judge the Pearson coefficient with the actual measured output in order to analyze whether it can accurately warn the fault. The change of Pearson coefficients is shown in Figure 7 and Figure 8.

_{1}and DBN

_{2}are DBN models for normal charging voltage and normal charging current of EV, and BPNN

_{1}and BPNN

_{2}are BPNN models for normal charging voltage and normal charging current of EV. The fine-tuning learning rate of DBN model is 0.6 and one momentum parameter is 0.25, and the rest of the network parameters and network structure are the same as ADBN. BPNN adopts a single hidden layer structure, the corresponding number of nodes is 15-9-1, the learning rate is 0.05, and the number of iterations is 1000.

## 5. Conclusions

## Author Contributions

## Funding

## Conflicts of Interest

## Nomenclature

EV | electric vehicle |

BMS | battery management system |

SOC | state of charge |

ADBN | adaptive deep belief network |

NAdam | Nesterov-accelerated adaptive moment estimation |

DBN | deep belief network |

BPNN | back propagation neural network |

VMD | variational mode decomposition |

UAV | unmanned aerial vehicle |

RBM | restricted Boltzmann machine |

Adam | adaptive moment estimation |

## References

- Carbon, C.C.; Gebauer, F. The safe-range-inventory (SRI): An assistance tool for optimizing the charging infrastructure for electric vehicles. Transp. Res. Part F Traffic Psychol. Behav.
**2017**, 47, 101–113. [Google Scholar] [CrossRef] - Hu, J.; Gao, Z.W. SOC prediction of electric vehicle power battery based on data driven. Automot. Eng.
**2021**, 43, 1–9. [Google Scholar] - Tran, M.K.; Fowler, M. A Review of Lithium-Ion Battery Fault Diagnostic Algorithms: Current Progress and Future Challenges. Algorithms
**2020**, 13, 62. [Google Scholar] [CrossRef] [Green Version] - Chao, W.; Zhu, C.B.; Ge, Y.W.; Zhao, Y.P. A Review on Fault Mechanism and Diagnosis Approach for Li-Ion Batteries. J. Nanomater.
**2015**, 2015, 1–9. [Google Scholar] - Zhang, Y.X.; Li, T.Y.; Yan, X.W.; Wang, L.; Zhang, J.; Diao, X.H.; Li, B. Electric Vehicle Charging Fault Monitoring and Warn-ing Method Based on Battery Model. World Electr. Veh. J.
**2021**, 12, 14. [Google Scholar] [CrossRef] - Tran, M.K.; Mathew, M.; Janhunen, S.; Panchal, S.; Raahemifar, K.; Fraser, R.; Fowler, M. A comprehensive equivalent circuit model for lithium-ion batteries, incorporating the effects of state of health, state of charge, and temperature on model parameters. J. Energy Storage
**2021**, 43, 103252. [Google Scholar] [CrossRef] - Tran, M.K.; DaCosta, A.; Mevawalla, A.; Panchal, S.; Fowler, M. Comparative Study of Equivalent Circuit Models Performance in Four Common Lithium-Ion Batteries: LFP, NMC, LMO, NCA. Batteries
**2021**, 7, 51. [Google Scholar] [CrossRef] - Zhang, Y.X.; Li, T.Y.; Zhao, S.S.; Yan, X.W.; Zhang, J.; Diao, X.H.; Li, B. Electric Vehicle Charging Fault Diagnosis Based on BP Neural Network. J. Phys. Conf. Ser.
**2020**, 1646, 012046. [Google Scholar] [CrossRef] - Zhang, R.; Li, W.P.; Mo, T. A review of deep learning research. Inf. Control
**2018**, 47, 385–397. [Google Scholar] - Wang, T.L.; Pan, Z.F.; Yuan, X.F.; Yang, C.H.; Gui, W.H. A novel deep learning based fault diagnosis approach for chemical process with extended deep belief network. ISA Trans.
**2020**, 96, 457–467. [Google Scholar] [CrossRef] [PubMed] - Zheng, X.X.; Chen, G.N.; Ren, H.H.; Li, D.D. Fault early warning of fan vulnerable parts based on improved VMD and deep confidence network. J. Vib. Shock
**2019**, 38, 153–160. [Google Scholar] - Li, J.Q.; Chen, Y.T.; Li, S.X. Early warning of interturn short circuit fault in excitation winding of synchronous generator based on deep confidence network. Electr. Power Autom. Equip.
**2021**, 41, 153–158. [Google Scholar] - Chen, X.M.; Wu, C.X.; Wu, Y.; Xion, N.X.; Han, R. Design and Analysis for Early Warning of Rotor UAV Based on Data-Driven DBN. Electronics
**2019**, 8, 1350. [Google Scholar] [CrossRef] [Green Version] - Huang, L.; Xiang, L.Y. Method for Meteorological Early Warning of Precipitation-Induced Landslides Based on Deep Neural Network. Neural Process. Lett.
**2018**, 48, 1243–1246. [Google Scholar] [CrossRef] - Gao, D.X.; Lin, X.H. Fault Diagnosis Method of DC Charging Points for EVs Based on Deep Belief Network. World Electr. Veh. J.
**2021**, 12, 47. [Google Scholar] [CrossRef] - Shen, C.Q.; Tang, S.H.; Jiang, X.X.; Shi, J.J.; Wang, J. Application of independent adaptive learning rate optimized deep belief network in bearing fault diagnosis. J. Mech. Eng.
**2019**, 55, 81–88. [Google Scholar] [CrossRef] [Green Version] - Zhu, J.; Chen, S. Network security situation prediction method based on NAWL-ILSTM. Comput. Sci.
**2019**, 46, 161–166. [Google Scholar] - Kingma, D.; Ba, J. Adam: A method for stochastic optimization. Comput. Sci.
**2015**, 1, 1–15. [Google Scholar] - Kong, X.Y.; Zheng, F.; Zhijun, E.; Cao, J.; Wang, X. Short-term load forecasting based on deep belief network. Autom. Electr. Power Syst.
**2018**, 42, 133–139. [Google Scholar] - Yang, Z.Y.; Liu, J.Y.; Liu, Y.B.; Wen, L.L.; Wang, Z.Q.; Ning, S.C. Substation load forecasting based on adaptive deep belief network. Proc. CSEE
**2019**, 39, 4049–4061. [Google Scholar] - Qi, Y.; Gong, Y.D.; Liang, C.Y.; Li, P.F. Research on fault analysis and reliability evaluation techniques of CNC machine tools. J. Northeast. Univ. (Nat. Sci.)
**2019**, 40, 831–834. [Google Scholar]

**Figure 2.**Structure and training diagram of deep belief network (DBN) model for normal charging voltage.

Charging Process Physical Quantities | Unit | Precision | Period/ms |
---|---|---|---|

Rated capacity of power battery | Ah | 0.1 | 250 |

Rated voltage of power battery | V | 0.1 | 250 |

Maximum allowable individual voltage | V | 0.01 | 500 |

Maximum allowable charging current | A | 0.1 | 500 |

Power battery nominal energy | kW·h | 0.1 | 500 |

Maximum allowable charging voltage | V | 0.1 | 500 |

Maximum allowable temperature | °C | 1 | 500 |

Power battery initial state of charge (SOC) | % | 0.1 | 500 |

Power battery initial voltage | V | 0.1 | 500 |

Power battery required voltage | V | 0.1 | 50 |

Power battery required current | A | 0.1 | 50 |

Charging voltage measurement value | V | 0.1 | 250 |

Charging current measurement value | A | 0.1 | 250 |

Maximum individual voltage of power battery | V | 0.01 | 250 |

Power battery current SOC | % | 1 | 250 |

Maximum individual temperature of power battery | °C | 1 | 250 |

Number | Fault Type | Fault Identification Method | Fault Description |
---|---|---|---|

1 | Charging voltage fault | The Pearson coefficient between model prediction and actual measurement is less than 0.8 | Bias fault—charging voltage is higher or lower than the required voltage |

2 | Charging current fault | The Pearson coefficient between model prediction and actual measurement is less than 0.8 | Bias fault—charging current is higher or lower than the required current |

3 | Temperature fault | The Pearson coefficient between model prediction and actual measurement is less than 0.8 | Bias fault—the measured value of temperature is widely different from the predicted value |

4 | SOC fault | The Pearson coefficient between model prediction and actual measurement is less than 0.8 | Bias fault—the measured value of SOC is widely different from the predicted value |

Description | Symbol | Value |
---|---|---|

Maximum number of pre-training iterations | - | 50 |

Fine-tune the maximum number of iterations | - | 50 |

Number of pre-trained batch samples | - | 100 |

Number of fine-tune batch samples | - | 100 |

Learning rate of pre-training | - | 0.4 |

Learning rate of fine-tuning | - | 0.001 |

Exponential decay rate of first-order moments | ${\beta}_{1}$ | 0.9 |

Exponential decay rate of second order moments | ${\beta}_{2}$ | 0.99 |

Correction parameters | $\epsilon $ | 10^{−8} |

Methods | Model Alert Accuracy (%) | Model Convergence Time (s) |
---|---|---|

ADBN_{1} | 98.61 | 424.16 |

DBN_{1} | 94.86 | 481.37 |

BPNN_{1} | 89.69 | 13.47 |

ADBN_{2} | 97.75 | 416.66 |

DBN_{2} | 94.62 | 487.94 |

BPNN_{2} | 90.65 | 13.97 |

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |

© 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Gao, D.; Wang, Y.; Zheng, X.; Yang, Q.
A Fault Warning Method for Electric Vehicle Charging Process Based on Adaptive Deep Belief Network. *World Electr. Veh. J.* **2021**, *12*, 265.
https://doi.org/10.3390/wevj12040265

**AMA Style**

Gao D, Wang Y, Zheng X, Yang Q.
A Fault Warning Method for Electric Vehicle Charging Process Based on Adaptive Deep Belief Network. *World Electric Vehicle Journal*. 2021; 12(4):265.
https://doi.org/10.3390/wevj12040265

**Chicago/Turabian Style**

Gao, Dexin, Yi Wang, Xiaoyu Zheng, and Qing Yang.
2021. "A Fault Warning Method for Electric Vehicle Charging Process Based on Adaptive Deep Belief Network" *World Electric Vehicle Journal* 12, no. 4: 265.
https://doi.org/10.3390/wevj12040265