Open Access

ARTICLE

Fault Diagnosis Method for Photovoltaic Grid-Connected Inverters Based on MPA-VMD-PSO BiLSTM

Jingxian Ni, Chaomeng Wang, Shiqi Sun, Yuxuan Sun, Gang Ma^*

Nanjing Normal University, Nanjing, 210023, China

* Corresponding Author: Gang Ma. Email:

Energy Engineering 2025, 122(9), 3719-3736. https://doi.org/10.32604/ee.2025.066971

Received 22 April 2025; Accepted 06 June 2025; Issue published 26 August 2025

Abstract

To improve the fault diagnosis accuracy of a PV grid-connected inverter, a PV grid-connected inverter data diagnosis method based on MPA-VMD-PSO-BiLSTM is proposed. Firstly, unlike the traditional VMD algorithm which relies on manual experience to set parameters (e.g., noise tolerance, penalty parameter, number of decompositions), this paper achieves adaptive optimization of parameters through MPA algorithm to avoid the problem of feature information loss caused by manual parameter tuning, and adopts the improved VMD algorithm for feature extraction of DC-side voltage data signals of PV-grid-connected inverters; and then, adopts the PSO algorithm for the Then, the PSO algorithm is used to optimize the optimal batch size, the number of nodes in the hidden layer and the learning rate of the BiLSTM network, which significantly improves the model’s ability to capture the long-term dependent features of the PV inverter’s timing signals, to construct the PV grid-connected inverter prediction model of PSO-BiLSTM, and predict the capacitance value of the PV grid-connected inverter. Finally, diagnostic experiments are carried out based on the expected capacitance value and the capacitance failure criterion. The results show that compared with the traditional VMD algorithm, the MPA-optimised VMD improves the signal-to-noise ratio (SNR) of the signal decomposition from 28.5 to 33.2 dB (16.5% improvement). After combining with the PSO-BiLSTM model, the mean absolute percentage error (MAPE) of the fault diagnosis is reduced to 1.31%, and the coefficient of determination (R²) is up to 0.99. It is concluded that the present method has excellent diagnostic performance of PV grid-connected inverter data signals and effectively improves the accuracy of PV grid-connected inverter diagnosis.

---

Keywords

---