## 1 Introduction

Deep learning technology | Convolutional neural network (CNN) | Recurrent neural network | Deep residual network (ResNet) | Back propagation neural network particle swarm optimization |
---|---|---|---|---|

General Features | Good for the use of multisource information Strong diagnosis algorithm | The inputs are described as time-series data and this network can learn sequences that differ over time | This network is implemented having skip connections or shortcuts to jump over some layers | This network combines both back propagation neural network technique along with Particle Swarm Optimization |

Fault detection and diagnosis Method | ID CNN | Bidirectional convolutional Long Short-term memory (LSTM) | Residual Network trained by the adaptive moment estimation deep learning algorithm | BPNN combined with heuristic algorithm of PSO |

Method description | Use both feature extraction as well as post-processing of the raw signal | The CNN is applied for the extraction of the features from the original data and LSTM is used to get the correlation and then apply a fully connected layer for prediction | This technique was able to extract features automatically from the I-V curve and from the irradiance and temperature | The technique is based on using the BP neural network and then applying PSO in the hidden layer |

Advantages | It can prevent conversions. The results show that the proposed method accuracy reached 95.07% | It can deal well with prediction problems that have long sequences and it merges benefits of both CNN and LSTM. It showed that accuracy in terms of R _{2} statistical matrix (as an example) reached 0.94 | This method showed its effectiveness by comparing with other two methods showing higher accuracy. This accuracy in real study reached to 95.778% | High accuracy in fault detection as well as benefiting from combining the advantages of BPNN and PSO techniques |

Reference | [27] | [28] | [29] |

## 2 The proposed system configuration

Electrical parameters | Value |
---|---|

Short-Circuit Current (Isc) | 6.2 A |

Open Circuit Voltage (Voc) | 50.9 V |

Maximum Power Current (Imp) | 5.84 A |

Maximum Power Voltage (Vmp) | 42.8 V |

States of faults | Fault index |
---|---|

Normal | 1 |

Temperature fault | 2 |

Partial shading fault | 3 |

Cells aging | 4 |

Temperature and partial shading combination | 5 |

Temperature and cell aging combination | 6 |

## 3 The hybrid proposed fault diagnosis method

### 3.1 Back propagation-particle swarm optimization algorithm

### 3.2 Computational complexity

## 4 Data analysis

^{°}K, 358.15

^{°}K] and the series resistance values of the cells \(\in \left[ {28{ }\Omega ,{ }70{ }\Omega } \right]\). 240 samples of data are used for training of the BPNN-PSO and the remaining 60 samples are used as test samples. The main parameters used in BPNN-PSO values are displayed in Table 4.

Parameters | Values |
---|---|

Population (M) | 50 |

Number of Iterations (T) | 200 |

Inertia weight (w) | 1.8 |

Cognitive factor (\({c}_{1}\)) | 2 |

Social factor \(({c}_{2})\) | 2 |

## 5 Results and discussion

Sample | Identification parameters | Actual fault | Prediction results | Right or wrong | |||||
---|---|---|---|---|---|---|---|---|---|

\(V_{{{\text{oc}}}}\) | \(I_{{{\text{sc}}}}\) | \(P_{{\text{m}}}\) | \(V_{{\text{m}}}\) | BP | BPPSO | BP | BPPSO | ||

1 | 49.975 | 6.1943 | 246.9784 | 42.2908 | 1 | 5 | 1 | × | √ |

2 | 50.889 | 6.1903 | 247.0131 | 42.2961 | 1 | 1 | 1 | √ | √ |

3 | 46.4707 | 6.0209 | 244.1174 | 41.8009 | 2 | 2 | 2 | √ | √ |

4 | 48.8417 | 5.3103 | 230.0822 | 39.3976 | 3 | 3 | 3 | √ | √ |

5 | 50.8974 | 6.1976 | 240.3561 | 41.1569 | 4 | 4 | 4 | √ | √ |

6 | 50.2456 | 5.7783 | 237.0650 | 40.5933 | 5 | 3 | 5 | × | √ |

7 | 49.0960 | 6.0863 | 242.1384 | 41.4621 | 5 | 3 | 4 | × | × |

8 | 49.1328 | 6.1863 | 244.9137 | 41.9372 | 6 | 6 | 6 | √ | √ |

9 | 49.2950 | 6.1732 | 249.3066 | 42.6894 | 7 | 3 | 7 | × | √ |

10 | 48.8974 | 6.0981 | 238.3409 | 40.8118 | 7 | 4 | 4 | × | × |