A quantum neural network-based approach to power quality disturbances detection and recognition

• Published in: Physica scripta Vol. 100; no. 7; pp. 75102 - 75122
• Main Authors: Li, Guo-Dong, He, Hai-Yan, Li, Yue, Li, Xin-Hao, Liu, Hao, Wang, Qing-Le, Cheng, Long
• Format: Journal Article
• Language: English
• Published: IOP Publishing 01.07.2025
• Subjects: power quality disturbances; power quality disturbances detection; power quality disturbances recognition; quantum machine learning; quantum neural network
• ISSN: 0031-8949; 1402-4896
• DOI: 10.1088/1402-4896/adda9e

As an emerging technology force, quantum algorithms have shown great potential and unique advantages in many fields of application. Power quality disturbances (PQDs) affect the security and stability of the power system, which may lead to equipment damage, system failures and economic losses. The accurate detection and recognition of PQDs is a guarantee for the safe operation of the system. In order to explore the application of quantum algorithms in the field of power quality, an improved quantum neural network(QNN) is proposed for PQDs detection and recognition. The QNN model consists of three parts: quantum encoding, quantum variational, and quantum measurement. Firstly, the classical data is mapped to the high-dimensional quantum feature space through the encoding layer, so that it is transformed into quantum data. Subsequently, a unique quantum circuit is constructed in the quantum variational layer to adjust the rotation angles and entanglement of data and ancilla qubits to realise quantum information transformation. The quantum information obtains the expected value through the measurement layer, which is used for model optimization and subsequent disturbance classification. The experimental results show that the detection accuracy of PQDs by this method reaches 99.75%, the recognition accuracy of single disturbance and mixed disturbance reaches 97.85% and 95.50%, respectively. The model uses fewer parameters to obtain high accuracy under the same conditions compared to classical machine learning models. In addition, comparative tests in different noise environments show the robustness of the model.