A dynamic response analysis method of an AC-DC hybrid power grid

Abstract

This invention discloses a dynamic response analysis method for AC / DC hybrid power grids. The method first searches for a fault set in the near-field region of the DC landing point of the AC / DC hybrid power grid. Then, it uses electromechanical-electromagnetic transient simulation to analyze the faults, obtaining the system dynamic response curve corresponding to each fault in the fault set. An unsupervised learning algorithm is then used to cluster the system dynamic response curves and label the categories. Finally, a random forest model is trained using the power flow dynamic information of the fault and the fault information corresponding to the fault as model inputs, and the system dynamic response curves corresponding to the faults as model outputs. By inputting real-time fault information and power flow dynamic information into the trained random forest model, the category of the system dynamic response curve corresponding to the real-time fault can be output. This analysis method can deeply explore the correlation between power flow dynamic information and fault information and the dynamic response characteristics of power grid voltage and frequency, and quickly determine the voltage and frequency response characteristics of the power grid.

Description

Technical Field

[0001] This invention belongs to the field of electrical engineering, and in particular relates to a dynamic response analysis method for AC / DC hybrid power grids. Background Technology

[0002] High-voltage direct current (LCC-HVDC) transmission based on grid-commutated converters (LCCs) plays a crucial role in China's long-distance, high-power transmission projects. However, LCC-HVDC transmission relies on the AC network for commutation voltage. When a short-circuit fault occurs in the AC system on the inverter side, the DC voltage and converter bus voltage drop rapidly, leading to increased reactive power fluctuations between the AC and DC systems. This, coupled with significant fluctuations in commutation voltage, can cause commutation failure. If the AC fault is not cleared in time, the fluctuating reactive power exchange may trigger subsequent commutation failures, resulting in a sharp decrease in DC transmission power and severe consequences such as voltage collapse in weak power grids. Furthermore, with the increasing proportion of residential and commercial loads, typical loads such as variable frequency fans and temperature control equipment have become significant contributors to system peak loads. With the integration of numerous renewable energy devices, renewable energy, as a broad load, also plays a vital role in power system voltage stability. Therefore, the increasing dynamic load, the high penetration of large-capacity HVDC transmission, the development of smart electricity consumption technologies for residential and commercial load systems, the interaction between user-side load dynamics and the system, and the characteristics of dynamic loads are increasingly posing serious challenges to power grid safety and stability. The operation of new power systems with a high proportion of renewable energy and power electronic equipment is highly random and uncertain. Influenced by factors including renewable energy, bidirectional power flow caused by demand-side response, AC / DC hybrid transmission of high power, and new electricity market behaviors, power balance may be disrupted in the short term. Static security of a power system refers to its ability to continue supplying power to loads after a fault, primarily considering whether node voltages and line power flows exceed limits after a fault. Transient stability refers to the ability of the grid bus voltage and frequency to recover to safe ranges within a short time after a large disturbance to the power system.

[0003] The safe and stable operation of power systems is typically affected by factors such as extreme weather, tree strikes, terrorist attacks, cyberattacks, equipment failures, malfunctions of protection systems, and operator errors. In extreme cases, if large disturbances are not quickly and effectively assessed and mitigated, cascading failures and large-scale blackouts may occur. Therefore, maintaining the safety and stability of the power system at all times is crucial for reliable power supply, and it is essential to accurately assess the safety and operational risks of the power system at all times to effectively address these challenges. Meanwhile, the significant increase in new energy sources and DC power electronic equipment under the new power system context leads to strong volatility, strong nonlinearity, and strong uncertainty characteristics in the power system. The increased scale of DC transmission also means that AC / DC hybrid power grids are subject to various types of large disturbances, such as AC-side faults and DC system blockages. The transient and recovery characteristics after system faults are complex and varied, posing a significant challenge to the safety and stability analysis of new power systems.

[0004] The current assessment of power system safety and stability is based on electromagnetic and electromechanical transient simulation and analysis. In recent years, the safety and stability analysis of hybrid AC / DC power systems, primarily driven by new energy sources, has received considerable attention, mainly focusing on the following aspects: modeling, control and fault protection strategies for hybrid DC transmission systems, and harmonic characteristic analysis. However, these studies primarily concentrate on the electromagnetic transient domain. In the context of increasingly electronic power systems, the extensive use of switching devices significantly limits the speed and scale of electromagnetic transient simulation. Currently, electromagnetic transient simulation is not suitable for studying the operation planning of large AC / DC power grids. A common approach is to use electromechanical transient simulation that only considers the fundamental frequency component, which still falls short in understanding the dynamic characteristics of electronically powered systems dominated by new energy sources. Therefore, hybrid electromechanical-electromagnetic simulation, employing an electromagnetic model for DC transmission and an electromechanical model for the AC side, has become the main method for analyzing the safety and stability of AC / DC hybrid power grids. However, current simulation analysis methods only perform simulation analysis under specific operating conditions and specific fault scenarios; when the number of simulation examples is large, there is a lack of effective analytical methods to quickly summarize the dynamic response characteristics of AC / DC hybrid power grids. Summary of the Invention

[0005] To address the problems existing in the prior art, the purpose of this invention is to provide a dynamic response analysis method for AC / DC hybrid power grids.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A dynamic response analysis method for an AC / DC hybrid power grid, wherein multiple DC systems in the AC / DC hybrid power grid are connected to the AC system at multiple points, includes the following steps:

[0008] Step 1) Search for fault sets in the near-field area of ​​the DC landing point of the AC / DC hybrid power grid; the fault information of each fault in the fault set includes the fault location and fault type;

[0009] Step 2) Use electromechanical-electromagnetic transient simulation method to perform fault analysis and obtain the system dynamic response curve corresponding to each fault in the fault set;

[0010] Step 3) Using the system dynamic response curve corresponding to each fault as a sample, construct a sample library, and use an unsupervised learning algorithm to cluster the samples in the sample library and label the categories to obtain the category of the system dynamic response curve corresponding to each fault.

[0011] Step 4) Train the random forest model by using the power flow dynamics information corresponding to each fault and the fault information of each fault as model inputs, and the system dynamic response curve category corresponding to each fault as model output.

[0012] Step 5) Input the fault information of the real-time fault and the power flow state information of the corresponding real-time fault into the trained random forest model, and output the system dynamic response curve category corresponding to the real-time fault.

[0013] Furthermore, the system dynamic response curve includes the system frequency response curve and the voltage response curve;

[0014] Further, in step 1), the search for fault sets near the DC landing point of the AC / DC hybrid power grid specifically includes:

[0015] Step 1.1) In an AC / DC hybrid power grid, select the AC bus connected to the DC landing point;

[0016] Step 1.2) Based on the power grid topology information, obtain the adjacency lists of power grid buses and tie lines and establish a bidirectional topology graph;

[0017] Step 1.3) For the established bidirectional topology graph, use the breadth-first search algorithm, starting from the bus connected to the DC landing point, search for all connected next-level buses and put them into the double-ended queue. Repeat this step X times to find all paths that are X non-repeating connecting lines from the starting point.

[0018] Step 1.4) Locate all busbar and tie line fault locations that are at least five tie lines away from the starting busbar. Based on the found fault locations and fault types, establish a fault set.

[0019] Furthermore, the power grid topology result information is presented in BPA format power flow file.

[0020] Furthermore, in step 1.3), X = 5.

[0021] Furthermore, the power flow dynamics information includes the system bus voltage amplitude, AC line active power, DC line active power, and system load level.

[0022] Furthermore, the electromechanical-electromagnetic transient simulation method includes modeling the AC side of the AC-DC hybrid power grid using an electromechanical transient model and the DC side using an electromagnetic transient model.

[0023] Furthermore, the fault types include three-phase permanent grounding faults and single-phase permanent grounding faults.

[0024] Furthermore, the unsupervised learning algorithm employs the K-means algorithm.

[0025] Compared with the prior art, the present invention has the following advantages:

[0026] This invention proposes a dynamic response analysis method for AC / DC hybrid power grids. The method first searches for a fault set in the near-field region of the DC landing point of the AC / DC hybrid power grid. Then, it uses electromechanical-electromagnetic transient simulation to analyze the faults, obtaining the system dynamic response curve corresponding to each fault in the fault set. An unsupervised learning algorithm is then used to cluster the system dynamic response curves and label the categories. Finally, a random forest model is trained using the power flow dynamics information and fault information corresponding to the faults as model inputs and the system dynamic response curves corresponding to the faults as model outputs. By inputting real-time fault information and power flow dynamics information into the trained random forest model, the system dynamic response curve category corresponding to the real-time fault can be output. This analysis method, through machine learning algorithms, deeply mines the correlation between power flow dynamics information, fault information, and the dynamic response characteristics of power grid voltage and frequency, better capturing the potential distribution of data, obtaining underlying data structure information, and quickly judging the voltage and frequency response characteristics of the power grid. This can better meet the rapid requirements of power system safety and stability assessment.

[0027] The analysis method of this invention searches for a fault set in the near-field of the DC landing point of the AC / DC hybrid power grid. Each fault in the fault set corresponds to a fault location and a fault type. This can meet the analysis requirements of the dynamic response characteristics of the power grid corresponding to various fault locations and fault types. Therefore, the analysis method of this invention can be well applied to the dynamic response monitoring and analysis service of power grids with strong volatility, strong nonlinearity and strong uncertainty characteristics. Attached Figure Description

[0028] Figure 1 A flowchart of the dynamic response analysis method for AC / DC hybrid power grids;

[0029] Figure 2 A system block diagram of an AC / DC hybrid power grid;

[0030] Figure 3KDE density plots of response extremum occurrence time under different three-permanent N-1 fault conditions;

[0031] Figure 4 The graph shows the response curves of the maximum relative power angle difference of the entire system under different three-permanent N-1 fault conditions;

[0032] Figure 5 The graph shows the minimum voltage response curves of the entire system under different three-terminal N-1 faults;

[0033] Figure 6 A statistical chart showing the minimum voltage recovery time under different fault conditions;

[0034] Figure 7 The graph shows the maximum voltage response curves of the entire system under different three-terminal N-1 faults;

[0035] Figure 8 The graph shows the lowest frequency response curves of the entire system under different three-terminal N-1 fault conditions;

[0036] Figure 9 The graph shows the highest frequency response curves of the entire system under different three-terminal N-1 fault conditions;

[0037] Figure 10(a) is a statistical chart of the time it takes for the lowest frequency to recover to 49.9 Hz under different fault conditions;

[0038] Figure 10(b) is a statistical chart of the time it takes for the highest frequency to recover to 50.1 Hz under different fault conditions;

[0039] Figure 11 The statistical distribution of system stability under different N-1 fault conditions;

[0040] Figure 12(a) shows the clustering results of the highest frequency clusters;

[0041] Figure 12(b) shows the results of the lowest frequency clustering. Detailed Implementation

[0042] Example 1

[0043] A dynamic response analysis method for AC / DC hybrid power grids, where multiple DC systems are connected to the AC system at multiple points, such as... Figure 1 As shown, it includes the following steps:

[0044] Step 1) Search for fault sets in the near-zone of the DC landing point of the AC / DC hybrid power grid; the fault information of each fault in the fault set includes the fault location and fault type; the fault types include three-phase permanent grounding faults and single-phase permanent grounding faults.

[0045] Step 2) Use the electromechanical-electromagnetic transient simulation method to perform fault analysis and obtain the system dynamic response curve corresponding to each fault in the fault set; the system dynamic response curve includes the system frequency response curve and voltage response curve; the electromechanical-electromagnetic transient simulation method includes modeling the AC side of the AC-DC hybrid power grid using an electromechanical transient model and modeling the DC side using an electromagnetic transient model.

[0046] Step 3) Using the system dynamic response curve corresponding to each fault as a sample, construct a sample library, and use an unsupervised learning algorithm to cluster the samples in the sample library and label the categories to obtain the category of the system dynamic response curve corresponding to each fault; the unsupervised learning algorithm used is the Kmeans algorithm.

[0047] Step 4) Train the random forest model using the power flow dynamics information corresponding to each fault and the fault information of each fault as model inputs, and the system dynamic response curve category corresponding to each fault as model outputs; the power flow dynamics information includes the system bus voltage amplitude, AC line active power, DC line active power and bus load active power.

[0048] Random forest is an effective algorithm in machine learning, used to solve classification and regression problems. Its core principle lies in training a model that predicts data from multiple models and then averaging the outputs to provide an average prediction. This effectively corrects the overfitting problem of individual decision tree models, improving overall prediction accuracy. Random forests typically use guided clustering algorithms. Given a training set of data and a feature vector X = (x1, x2, ..., x...), ... n ), and the response vector Y = (y1, y2, ..., y3) n The training is repeated K times (n samples are randomly selected each time). For the k-th training iteration, the following algorithm is used to train the decision tree model:

[0049] (1) Randomly select n samples and obtain data features and response vector X. k and Y k .

[0050] (2) Using X k and Y k As the training set, model f is trained. k .

[0051] When using a random forest ensemble decision tree model for prediction, for unknown sample data features The average value of the above model output is used as the final prediction output, that is:

[0052] In the formula: Features of unknown sample data; To predict the output. Random forests generally outperform single decision trees in prediction, but are often considered "black box" models, sacrificing the inherent interpretability of single decision tree models.

[0053] The model input data for faults with fault location and fault type are shown in Table 1.

[0054] Table 1

[0055]

[0056] The model output data corresponding to this fault includes the system frequency response category and voltage response category, as shown in Table 2. In Table 2, Freq_type represents the frequency response category and Voltage_type represents the voltage response category.

[0057] Table 2

[0058] [Freq_type1,Voltage_type1,…]

[0059] Step 5) Input the fault information of the real-time fault and the power flow state information of the corresponding real-time fault into the trained random forest model, and output the system dynamic response curve category corresponding to the real-time fault.

[0060] Example 2

[0061] A further optional design of this embodiment is that, in this example, the fault set search for the DC landing point near-area search of the AC / DC hybrid power grid specifically includes:

[0062] Step 1.1) In an AC / DC hybrid power grid, select the AC bus connected to the DC landing point;

[0063] Step 1.2) Based on the power grid topology information, obtain the adjacency lists of power grid buses and tie lines and establish a two-way topology graph; the power grid topology results are presented in BPA format power flow files.

[0064] Step 1.3) For the established bidirectional topology graph, use the breadth-first search algorithm, starting from the bus connected to the DC landing point, to search for all connected next-level buses and put them into the double-ended queue. Repeat this step 5 times to find all paths that are 5 non-repeating connecting lines from the starting point.

[0065] Step 1.4) Locate all busbar and tie-line fault locations that are at least five tie lines away from the starting busbar. Based on the found fault locations and preset fault types, establish a fault set. If N fault locations are found, and there are M preset fault types, and both the fault locations and fault types are compatible (meaning that any of the M preset fault types can occur at any fault location), then the fault set contains a total of N*M faults.

[0066] Example 3

[0067] This embodiment uses the dynamic response analysis method of the present invention to perform dynamic response analysis on a certain AC / DC hybrid power grid, such as... Figure 2 As shown, in this AC / DC hybrid power grid, n DC systems are connected to the AC system through n landing points.

[0068] Step 1) Search for fault sets in the near-zone of the DC landing point of the AC / DC hybrid power grid; in this example, the fault types in the fault set are three-phase permanent grounding faults and single-phase permanent grounding faults, with a total of 47 faults in the fault set.

[0069] Step 2) Employing an electromechanical-electromagnetic transient simulation method, the AC side of the AC-DC hybrid power grid is modeled using an electromechanical transient model, while the DC side is modeled using an electromagnetic transient model. Fault analysis is then performed to obtain the system dynamic response curves corresponding to each fault in the fault set. The system dynamic response curves include the system frequency response curve and the voltage response curve. To further study the dynamic response of the system under simulated faults, this embodiment also obtains... Figures 3 to 11 The system response diagram is shown below. Among them,

[0070] Figure 3 To respond to the KDE density plot of extreme value occurrence time, the curves represent the occurrence times and frequencies of the maximum power angle difference, the highest voltage, and the lowest frequency in the system's dynamic response after a fault in all simulation examples. According to Figure 3 It can be seen that the maximum power angle difference occurs at around 1.5s; the highest voltage and lowest frequency mainly occur at around 2.8s after the maximum power angle difference; and the lowest voltage occurs at 0.41s.

[0071] Figure 4 The envelope curves of the maximum power angle difference under different three-phase permanent ground fault N-1 conditions are shown. Figure 4 Different curves correspond to different faults. According to... Figure 4 It can be seen that after the fault occurred, the maximum power angle difference began to oscillate, with an oscillation amplitude between 5 and 30 degrees. The maximum value appeared around 1.5 seconds, and the minimum value appeared around 0.8 seconds or 2.5 seconds. The oscillation basically recovered when the simulation ended at 6 seconds.

[0072] Figure 5 The envelope curves of the lowest voltage of the entire system under different three-terminal N-1 fault conditions are shown. Figure 5 Different curves correspond to different faults. According to... Figure 5 It can be seen that a three-phase short circuit occurred in the system at 0.1 seconds, and the minimum bus voltage of the power grid dropped to 0 after the short circuit was disconnected and reclosed. After the short circuit was disconnected and reclosed, the minimum voltage can be quickly restored to the normal range, with a recovery time between [0.15s, 0.5s].

[0073] Figure 6 The system displays the statistical distribution of minimum voltage recovery time under different fault conditions, which can be used to determine the voltage recovery status and voltage stability of the power grid.

[0074] Figure 7 The system's highest voltage response curves under different three-terminal N-1 faults are shown. Figure 7 Different curves correspond to different faults. According to... Figure 7 It can be seen that after a short circuit occurs, the highest voltage drops by 0.02-0.11 pu instantaneously. After the line is broken, the voltage begins to rise and oscillate, and after about 3 seconds it returns to the initial voltage level and remains stable.

[0075] Figure 8 The envelope curves of the lowest frequency under different groups of three-phase permanent ground fault N-1 conditions are shown. Figure 8 Different curves correspond to different faults. After a fault occurs, the system's lowest frequency drops rapidly and reaches its lowest point in about 0.7 seconds. At the end of the simulation, in some fault cases, the lowest frequency has recovered to above 49.9 Hz and remained stable; in some open circuit faults, the lowest frequency is still at a low level and oscillates at 6 seconds.

[0076] Figure 9 The highest frequency envelope curves for all three-phase permanent ground fault N-1 scenarios are shown. Figure 9 Different curves correspond to different faults. After a fault occurs, the system's highest frequency rises rapidly and oscillates significantly. The overall frequency increase ranges from 0.1Hz to 0.88Hz. 0.5s after the fault occurs, the highest frequency begins to decrease with damped oscillations. At the end of the simulation, in most fault cases, the highest frequency has recovered to below 50.1Hz and remained stable; some open-circuit faults have a greater impact on the system, with the highest frequency still exhibiting slight oscillations at 6s.

[0077] Figures 10(a) and 10(b) show the statistics of the time required for the lowest frequency to recover to 49.9 Hz and the statistics of the time required for the highest frequency to oscillate back to 50.1 Hz for the first time, respectively. This information can effectively reveal the frequency recovery and stability of the power grid.

[0078] Figure 11 The statistical distribution of system stability under different N-1 fault conditions; Figure 11The main areas of concentration for the maximum power angle difference, minimum voltage, maximum voltage, and minimum frequency are shown in gray boxes. The extreme values ​​of the maximum power angle difference, minimum voltage, maximum voltage, and minimum frequency are marked by horizontal lines above and below the gray boxes, respectively. The sample distribution of the maximum power angle difference is indicated by scatter plots.

[0079] Step 3) Using the system dynamic response curve corresponding to each fault as a sample, construct a sample library, and use an unsupervised learning algorithm to cluster the samples in the sample library and label the categories to obtain the category of the system dynamic response curve corresponding to each fault; the unsupervised learning algorithm used is the K-means algorithm. Taking the system frequency response as an example, Figure 12(a) shows the highest frequency clustering result, and Figure 12(b) shows the lowest frequency clustering result.

[0080] Step 4) Train the random forest model using the power flow dynamics information corresponding to each fault and the fault information of each fault as model inputs, and the system dynamic response curve category corresponding to each fault as model outputs; the power flow dynamics information includes the system bus voltage amplitude, AC line active power, DC line active power and bus load active power.

[0081] Step 5) Input the fault information of the real-time fault and the power flow state information of the corresponding real-time fault into the trained random forest model, and output the system dynamic response curve category corresponding to the real-time fault.

Claims

1. A dynamic response analysis method for an AC / DC hybrid power grid, wherein multiple DC systems in the AC / DC hybrid power grid are connected to an AC system at multiple points, characterized in that: Includes the following steps: Step 1) Search for fault sets in the near-field area of ​​the DC landing point of the AC / DC hybrid power grid; the fault information of each fault in the fault set includes the fault location and fault type; Step 2) Use electromechanical-electromagnetic transient simulation method to perform fault analysis and obtain the system dynamic response curve corresponding to each fault in the fault set; Step 3) Using the system dynamic response curve corresponding to each fault as a sample, construct a sample library, and use an unsupervised learning algorithm to cluster the samples in the sample library and label the categories to obtain the category of the system dynamic response curve corresponding to each fault. Step 4) Train the random forest model by using the power flow dynamics information corresponding to each fault and the fault information of each fault as model inputs, and the system dynamic response curve category corresponding to each fault as model output. Step 5) Input the fault information of the real-time fault and the power flow state information of the corresponding real-time fault into the trained random forest model, and output the system dynamic response curve category corresponding to the real-time fault. The system dynamic response curves include the system frequency response curve and the voltage response curve.

2. The method of claim 1, wherein the method further comprises: In step 1), the search for fault sets near the DC landing point of the AC / DC hybrid power grid specifically includes: Step 1.1) In an AC / DC hybrid power grid, select the AC bus connected to the DC landing point; Step 1.2) Based on the power grid topology information, obtain the adjacency lists of power grid buses and tie lines and establish a bidirectional topology graph; Step 1.3) For the established bidirectional topology graph, use the breadth-first search algorithm, starting from the bus connected to the DC landing point, search for all connected next-level buses and put them into the double-ended queue. Repeat this step X times to find all paths that are X non-repeating connecting lines from the starting point. Step 1.4) Locate all busbar and tie line fault locations that are at least five tie lines away from the starting busbar. Based on the found fault locations and fault types, establish a fault set.

3. The method of claim 2, wherein the method further comprises: The power grid topology results information is presented in BPA format power flow file.

4. The dynamic response analysis method for AC / DC hybrid power grids according to claim 3, characterized in that: In step 1.3), X=5.

5. The dynamic response analysis method for AC / DC hybrid power grids according to claim 1, characterized in that: The power flow dynamics information includes the system bus voltage amplitude, AC line active power, DC line active power, and system load level.

6. The dynamic response analysis method for AC / DC hybrid power grids according to claim 1, characterized in that: The electromechanical-electromagnetic transient simulation method includes modeling the AC side of the AC-DC hybrid power grid using an electromechanical transient model and the DC side using an electromagnetic transient model.

7. The dynamic response analysis method for AC / DC hybrid power grids according to claim 1, characterized in that: The fault types include three-phase permanent grounding faults and single-phase permanent grounding faults.

8. The dynamic response analysis method for AC / DC hybrid power grids according to claim 1, characterized in that: The unsupervised learning algorithm used is the K-means algorithm.

Images