When a line-commutated converter–high-voltage direct current (LCC-HVDC) transmission system with large-scale integration of renewable energy encounters HVDC-blocking events, the sending-end power system is prone to transient overvoltage (TOV) risks. Renewable energy units that are connected via power electronic devices are susceptible to large-scale cascading disconnections due to electrical endurance and insulation limitations when subjected to an excessively high TOV, which poses a serious threat to the safe and stable operation of the system. Therefore, the prediction of TOV at renewable energy stations (RES) under DC blocking (DCB) scenarios is crucial for developing strategies for the high-voltage ride-through of renewable energy sources and ensuring system stability. In this paper, an approximate analytical expression for the TOV at RES under DCB fault conditions is firstly derived, based on a simplified equivalent circuit of the sending-end system that includes multiple DC transmission lines and RES, which can take into consideration the multiple renewable station short-circuit ratio (MRSCR). Building on this, a knowledge-embedded enhanced deep neural network (KEDNN) approach is proposed for predicting the RES’s TOV for complex power systems. By incorporating theoretical calculation values of the TOV into the input features, the task of the deep neural network (DNN) shifts from mining relationships within large datasets to revealing the correlation patterns between theoretical calculations and real values, thereby improving the robustness of the prediction model in cases of insufficient training data and irrational feature construction. Finally, the proposed method is tested on a real-world regional power system in China, and the results validate the effectiveness of the proposed method. The approximate analytical expression for the TOV at RES and the KEDNN-based TOV prediction method proposed in this paper can provide valuable references for scholars and engineers working in the field of power system operation and control, particularly in the areas of overvoltage theoretical calculation and mitigation.
The development of power systems puts forward higher requirements for transient stability evaluations of power systems. The accuracy and timeliness of transient stability assessment are of great significance to the safe and stable operation of power systems. Traditional mechanistic judgment methods and criteria have strong interpretability, but they also have great limitations. They are still difficult to apply to complex power systems and are in urgent need of improvement. Artificial intelligence methods have high accuracy in stability judgment, but they have problems such as poor interpretability, and their stability judgment results are often difficult to explain. Based on the transient stability judgment mechanism of the response-driven transient energy function, this paper proposes a transient energy function stability judgment method based on a two-machine equivalent model and enhanced by a convolutional neural network. Firstly, the ST-kmeans method is used to cluster the generator sets, and the S-transformation is performed on the power angle changes of the generator sets to extract features. Then, the principal component analysis method is used to reduce the dimension of the feature data. Based on the k-means clustering method, the IEEE-39 node system generator synchronization units are grouped according to the power angle change trend of each generator after the fault. On the basis of the above methods, a two-machine equivalent model of the IEEE-39 node system is established, and the transient energy function of the two-machine system is derived. Based on the convolutional neural network, the critical energy is enhanced, and the fixed critical energy threshold is replaced by the corrected critical energy. The example results show that the transient stability prediction framework proposed in this paper can improve the scope of the application of mechanism discrimination and enhance the interpretability of the results of the intelligent method.
1 January 2025 in IET Generation, Transmission & Distribution
Large‐scale power systems typically require long‐distance transmission of electrical energy, and high‐voltage direct current (HVDC) technology is a commonly used high‐capacity means of connecting power sources to load centres. When a blocking fault occurs in an HVDC transmission system based on line commutated converters (LCC), the sending‐end system is prone to transient overvoltage (TOV) risks. This is especially severe in systems with large‐scale renewable energy integration, where excessive TOV can lead to widespread disconnection of renewable energy units, seriously threatening the safe and stable operation of the power system. Therefore, predicting the TOV magnitude in renewable energy stations (RES) under DC blocking (DCB) scenarios is of great importance for maintaining system stability and facilitating emergency control decisions. This paper first derives an analytical expression for the TOV magnitude at critical nodes in the system caused by DCB faults. Subsequently, an analytical formula is developed to characterize the relationship between the multiple renewable energy stations short circuit ratio (MRSCR) and the transient voltage rise (TVR) at the point of common coupling (PCC) of RES. Based on this, a physics‐informed neural network‐based transient overvoltage magnitude prediction (PINN‐TOMP) method for RES under DCB scenarios is proposed. The method introduces a regularization term for MRSCR into the loss function to ensure that the PINN model adheres to the physical laws and constraints governing the power system, thereby enhancing the prediction accuracy. Finally, the proposed method was tested on a real regional power system in China, and the results validated its effectiveness.