| China has put forward the goal of "carbon peaking and carbon neutrality",and the development of distributed generators represented by wind power and photovoltaic generators has become one of the key strategies in the field of power and energy.Distribution networks with distributed generators have the characteristics of complex topology,flexible operation mode,uncertain output and double change of power flow,and the fault presents weak characteristics and high-frequency transient characteristics,which is quite different from the traditional distribution network and greatly increases the difficulty of fault detection and type identification.Therefore,this paper conducts the following research on fault detection and type identification methods for distribution networks with distributed generators:(1)Aiming at the problem that fault samples in the distribution networks with distributed generators is difficult to obtain and is insufficient,the control strategy of distributed generators is studied,and the typical arc model and characteristics are analyzed;Based on the topology and network parameters of typical small-resistance grounding distribution network system,a PSCAD/EMTDC simulation model of 10 k V small-resistance grounding system with distributed generators is established,considering the complex high resistance and arc fault,which provides a dataset for the subsequent training and testing of deep learning models.In addition,based on the Hilbert-Huang transform method,the time-frequency characteristic analysis of the fault signal is carried out,and it is proved that the Hilbert-Huang transform can accurately detect the time,frequency and amplitude information of the abrupt and nonstationary disturbance signals,which lays a foundation for the subsequent fault type identification algorithm.(2)In view of the problem that the output of distributed generators has strong nonlinearity and the traditional fault detection method is no longer applicable,a fault detection method of distributed generators distribution network based on long short-term memory network is proposed,which directly acts on the time domain signal,and realizes the self-extraction of the characteristics of voltage and current timing data,which can realize fault line selection and section positioning including high resistance and arc fault.The detection accuracy of fault line selection reached 99.77%,the average accuracy of fault section positioning reached 99.58%,and the detection time of a single sample was 0.4ms,which meets the requirements for fault detection in distribution network.In addition,when accessing large-scale distributed generators sources,the detection accuracy is still stable at more than98% and is not affected by noise interference.When the distribution network topology changes dynamically,the fault detection rate of 98% can still be achieved,which proves the generalization and robustness of the model.(3)Aiming at the weak characteristics and high-frequency transient characteristics of distribution network faults with distributed generators sources,a fault type identification method based on convolutional neural network is proposed.Based on the time-frequency feature analysis method of Hilbert-Huang transform,a time-frequency energy matrix is constructed to extract the fault signal,and the deep feature extraction and hierarchical feature learning are extracted by convolutional neural network,which effectively avoids the interference factors between different features.The fault classification accuracy reached99.58%.After adding large-scale distributed generators,the classification accuracy can reach more than 99% and has good noise immunity,when the distribution network structure changes,the classification accuracy can still reach more than 98%,and the average classification recognition time of a single fault sample is 0.5ms,which meets the requirements of actual fault type identification.In addition,the model can be migrated to the application of fault phase selection,and the fault phase selection rate of 99% is realized,which proves the adaptability and generalization ability of the model.Compared with other algorithms,the proposed method has more advantages in fault detection and fault type identification for distribution networks with distributed generators,which can achieve fast and high accuracy fault detection and type identification even in the case of high impedance and arc faults. |
