| With the increase of renewable energy grid-connected capacity in the power system,a large number of power electronics are installed in the power system,increasing the transmission capacity of the grid.However,it also increases the stability problems of the power system,leading to the growing problem of sub-synchronous oscillation(SSO).SSO can cause harmonic pollution,damage renewable energy equipment,and pose a threat to the safe and stable operation of the grid.It is very important to quickly and accurately locate the source of SSO and adjust the relevant parameters in time to eliminate the oscillation phenomenon.Therefore,the quaternion feature set convolutional neural network(QFS-CNN)based on the temporal and spatial feature images of transient energy flows are studied in this paper.Firstly,the propagation characteristics of SSO power are studied.The composition of SSO power is described.The propagation law of SSO was quantitatively studied and the propagation characteristics of SSO were clarified by characterizing the propagation process of oscillation with SSO power.Furthermore,the propagation characteristics of SSO in multi-oscillation sources system are further studied,and the relationship between SSO power and line impedance is clarified.It provides a theoretical basis for the next step of SSO temporal and spatial feature extraction and oscillation source location.Secondly,the relationship between the transient energy and the path is analyzed,and the damping characteristics of SSO presented by the devices in the grid are studied.A two-dimensional SSO feature matrix extraction method based on the transient energy flow and energy power is proposed.In the case that the system cannot be completely observed,the SSO feature information of the whole grid is represented by the form of feature images.The temporal and spatial feature images uses limited voltage and current measurement data to match the transient energy absorbed by the positive damping characteristic devices to the location of SSO sources.Finally,a QFS-CNN oscillation sources location model based on temporal and spatial feature images is proposed to solve the problem of insufficient labeled data in grid.QFS data augmentation technique recombines temporal and spatial feature images to generate QFS feature sets and obtain more abundant training samples.The limited measurement data was used to construct the QFS feature sets,and the SSO sources location model was trained to realize the SSO sources location.The modified IEEE-39 bus wind power generation system was built with MATLAB/Simulink simulation software to verify and evaluate the location method.The simulation results show that the proposed method has high location accuracy and strong anti-noise ability under the condition of poor observability and few sample data. |
