Multi-level systematic protection method and system for modular multi-level converter

Abstract

The invention discloses a multilevel systematic protection method and system for a modular multilevel converter, and relates to the technical field of power system protection, and the method comprises the steps: protection starting: judging whether to trigger the protection starting or not through monitoring the capacitor voltage peak-to-peak values of all sub-modules of a single-phase single bridge arm in real time; fault severity classification: taking the bridge arm current after protection starting as an input feature, inputting the input feature into a shared classification head model based on a LightGBM classifier, and carrying out preliminary classification on faults; fault identification: according to a preliminary classification result, taking a capacitor voltage peak-to-peak value, a bridge arm current and a bridge arm reactor voltage drop of a sub-module after protection starting as characteristic quantities, respectively inputting the characteristic quantities into different models, and identifying a specific fault type; and protection response: adopting a differential protection response strategy according to a specific fault type. The invention provides a multi-level systematic protection method and system which can quickly and accurately identify MMC multi-level faults and take corresponding protection measures.

Description

Technical Field

[0001] The present invention relates to the technical field of power system protection, and in particular to a multi-level systematic protection method and system for a modular multi-level converter. Background Art

[0002] With the continuous adjustment of energy structures and the increasing demand for power system interconnection, flexible DC transmission technology has attracted widespread attention and application due to its unique advantages. As the core equipment of flexible DC transmission systems, MMC has become the most mainstream converter topology in current flexible DC transmission systems due to its modular design, strong scalability, and low output harmonics.

[0003] MMC converters typically consist of a large number of submodules (SMs) connected in series, each consisting of two insulated-gate bipolar transistors (IGBTs) and a capacitor. This structure enables MMCs to achieve high voltage levels, good output waveform quality, and low switching losses. However, as the number of submodules increases, the probability of internal converter failures also increases. If an MMC failure is not detected promptly and appropriate protective measures are not implemented, it will lead to abnormal system operation, even equipment damage and system crashes, seriously affecting the safe and stable operation of the power system. Various faults in the MMC will seriously affect the safe and stable operation of the converter station and may even paralyze the entire transmission system, resulting in huge economic losses.

[0004] In particular, MMC failures can be divided into the following levels: Module-level failures: These include open-circuit failures of the upper and lower IGBTs of the submodule (SM). These failures have a relatively small impact and develop slowly. Bridge arm level faults: including valve short circuit faults, valve group grounding faults, bridge arm reactor interphase short circuit faults, etc. These faults have a greater impact and develop quickly; Equipment-level faults: These include AC-side ground faults, AC-side phase-to-phase faults, DC-side ground faults, and DC-side phase-to-phase faults. These faults have a wide impact range and develop rapidly.

[0005] Currently, protection methods for MMC faults mainly include traditional protection methods such as differential protection, overcurrent protection, and voltage anomaly protection. However, due to the complex topology and diverse fault types of MMC, existing protection methods have the following shortcomings: Traditional protection methods are difficult to fully cover all types of MMC faults; For module-level faults, especially IGBT open-circuit faults, the recognition sensitivity of traditional protection methods is insufficient; Existing protection strategies fail to fully consider the differences in fault severity, making it difficult to achieve differentiated protection responses; In the fault identification process, the recognition speed and accuracy of existing methods are difficult to meet actual needs, especially for bridge-level and equipment-level faults that require rapid response.

[0006] Therefore, there is an urgent need for a multi-level systematic protection method that can accurately identify various types of MMC faults, quickly respond and take corresponding protection measures to improve the safety and reliability of the flexible DC transmission system. Summary of the Invention

[0007] The present invention provides a multi-level systematic protection method and system for modular multi-level converters, which overcomes the shortcomings of the MMC protection scheme in the prior art and provides a multi-level systematic protection method and system that can quickly and accurately identify MMC multi-level faults and take corresponding protection measures.

[0008] In order to achieve the above-mentioned object, the present invention adopts a technical solution: a modular multi-level converter multi-level system protection method, comprising the following steps: Protection startup: By real-time monitoring of the peak-to-peak value of the capacitor voltage of all submodules in a single-phase single-bridge arm, it is determined whether the protection startup is triggered; Fault severity classification: the bridge arm current after protection is activated is used as the input feature and input into the shared classification head model based on LightGBM classifier to perform preliminary fault classification; Fault identification: Based on the preliminary classification results, the peak-to-peak capacitor voltage, bridge arm current, and bridge arm reactor voltage drop of all submodules after protection is activated are used as feature quantities and input into different models to identify the specific fault type. Protection response: adopt differentiated protection response strategies based on the specific fault type.

[0009] Furthermore, the triggering condition for the protection startup is: When the peak-to-peak voltage of the capacitor When the value of three consecutive sampling points is greater than the preset threshold When, or when the peak-to-peak voltage of the capacitor Difference When the value of five consecutive sampling points is less than the preset threshold When protection starts.

[0010] Furthermore, the peak-to-peak value of the capacitor voltage The calculation formula is:

[0011] in, In the same bridge arm Capacitor voltage of each submodule; The peak-to-peak value of the capacitor voltage Difference The calculation formula is:

[0012] in, The current sampling point number.

[0013] Furthermore, the bridge arm current 1ms after startup will be protected As input features, they are input into the shared classification head model based on the LightGBM classifier to perform preliminary fault classification, specifically including: Type I faults: include bridge arm level faults and equipment level faults; Type II failure: includes module-level failure.

[0014] Furthermore, based on the preliminary classification results, the peak-to-peak values ​​of the capacitor voltage, bridge arm current, and bridge arm reactor voltage drop of all submodules after the protection is activated are used as feature quantities and input into different models to identify specific fault types, including: For Type I faults, the peak-to-peak values ​​of the capacitor voltages of all submodules, the bridge arm current, and the bridge arm reactor voltage drop within 2ms after the protection is activated are used as feature quantities and input into a fast classification head based on the LightGBM classifier to quickly identify specific fault types, including valve short-circuit fault, valve group grounding fault, bridge arm reactor phase-to-phase short-circuit fault, AC side single-phase grounding fault, AC side phase-to-phase fault, DC side unipolar fault, and DC side bipolar fault. For Type II faults, the peak-to-peak values ​​of the capacitor voltages of all submodules, the bridge arm current, and the bridge arm reactor voltage drops within 20ms after the protection is initiated are used as feature quantities and input into a comprehensive classification head composed of a soft voting integration model based on the XGBoost, LightGBM, and CatBoost models to quickly identify specific fault types, including SM upper tube open circuit fault, SM lower tube open circuit fault, and normal operation.

[0015] Furthermore, the differentiated protection response strategy is adopted according to the specific fault type, specifically including: For arm-level faults and equipment-level faults, the protection device blocks the converter and simultaneously trips the AC circuit breaker to cut off the external power supply. For module-level faults, the protection device sends an alarm signal and automatically starts fault recording.

[0016] The present invention also adopts a technical solution: a modular multi-level converter multi-level system protection system, comprising: Protection startup module: determines whether to trigger protection startup by real-time monitoring of the peak-to-peak value of the capacitor voltage of all sub-modules of a single-phase single-bridge arm; Fault severity classification module: The bridge arm current after protection is activated is used as the input feature and input into the shared classification head model based on the LightGBM classifier to perform preliminary fault classification; Fault identification module: Based on the preliminary classification results, the peak-to-peak capacitor voltage, bridge arm current, and bridge arm reactor voltage drop of all submodules after protection is activated are used as feature quantities and input into different models to identify the specific fault type; Protection response module: Adopt differentiated protection response strategies based on specific fault types.

[0017] The beneficial effects of the present invention are: Sensitive startup: This invention proposes a protection startup mechanism based on the peak-to-peak value of the capacitor voltage of all submodules, which can sensitively detect various faults inside the MMC and improve the sensitivity of protection startup; Comprehensive coverage: The present invention constructs a multi-level systematic protection framework for MMC, covering nine typical fault conditions at three levels: module level, bridge arm level, and equipment level, thus achieving comprehensive protection against MMC faults.

[0018] Differentiated processing: The present invention adopts differentiated fault identification strategies and protection response measures according to the severity of the fault, which can not only respond quickly to serious faults but also accurately identify minor faults, thereby optimizing protection performance.

[0019] High accuracy: This invention uses ensemble learning technology to build a three-level classification system consisting of a shared classification head, a fast classification head, and a comprehensive classification head, achieving accurate identification of various faults with an accuracy rate of up to 95%; Efficient Identification: For bridge arm-level and equipment-level faults requiring a quick response, the identification process takes only 2ms, meeting the stringent requirement of identifying faults within 1-3ms. For module-level faults that can tolerate a longer response time, the identification process is completed within 20ms. Alarm and fault recording functions are provided, providing a basis for subsequent fault location and improving system maintenance efficiency. Intelligent adaptation: The overall protection scheme based on integrated learning in this invention constitutes a complete multi-level systematized protection architecture, which is applicable to various MMC operating conditions and fault scenarios, and improves the reliability and stability of the flexible DC transmission system.

[0020] Strong practicability: The protection scheme proposed in the present invention utilizes existing measurement equipment and data, does not require additional hardware investment, reduces implementation costs, and has strong engineering practicability. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 Flowchart of the multi-level systematic protection method for MMC of the present invention; Figure 2 This is the MMC-HVDC system diagram of the Shagotan New Energy of the present invention; Figure 3 The basic structure diagram of the MMC and its submodules in the present invention; Figure 4 This is a current and voltage characteristic waveform diagram of a module-level fault in the present invention; Figure 5 This is a current and voltage characteristic waveform diagram of a bridge arm level fault in the present invention; Figure 6 This is a current and voltage characteristic waveform diagram of an equipment-level fault in the present invention; Figure 7 This is a schematic diagram of the protection startup of the present invention; Figure 8 This is a fault severity classification diagram of the present invention; Figure 9 This is a schematic diagram of the structure of the fast classification head of the present invention; Figure 10 This is a schematic diagram of the comprehensive classification head structure of the present invention; Figure 11 This is a comparison chart of the response time of the method of the present invention under typical fault conditions; Figure 12 Schematic diagram of the accuracy of fault identification results of the method of the present invention. DETAILED DESCRIPTION

[0022] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0023] Example 1, as Figure 1 As shown, a multi-level systematic protection method for a modular multilevel converter includes the following steps: Protection startup: By real-time monitoring of the peak-to-peak value of the capacitor voltage of all submodules in a single-phase single-bridge arm, it is determined whether the protection startup is triggered; Fault severity classification: the bridge arm current after protection is activated is used as the input feature and input into the shared classification head model based on LightGBM classifier to perform preliminary fault classification; Fault identification: Based on the preliminary classification results, the peak-to-peak capacitor voltage, bridge arm current, and bridge arm reactor voltage drop of all submodules after protection is activated are used as feature quantities and input into different models to identify the specific fault type. Protection response: adopt differentiated protection response strategies based on the specific fault type.

[0024] The triggering conditions for the protection start are: When the peak-to-peak voltage of the capacitor When the value of three consecutive sampling points is greater than the preset threshold When, or when the peak-to-peak voltage of the capacitor Difference When the value of five consecutive sampling points is less than the preset threshold When protection starts.

[0025] The peak-to-peak value of the capacitor voltage The calculation formula is:

[0026] in, In the same bridge arm Capacitor voltage of each submodule; The peak-to-peak value of the capacitor voltage Difference The calculation formula is:

[0027] in, The current sampling point number.

[0028] Will protect the bridge arm current 1ms after startup As input features, they are input into the shared classification head model based on the LightGBM classifier to perform preliminary fault classification, specifically including: Type I faults: include bridge arm level faults and equipment level faults; Type II failure: includes module-level failure.

[0029] According to the preliminary classification results, the peak-to-peak values ​​of the capacitor voltage, bridge arm current, and bridge arm reactor voltage drop of all submodules after the protection is started are used as feature quantities and input into different models to identify specific fault types, including: For Type I faults, the peak-to-peak values ​​of the capacitor voltages of all submodules, the bridge arm current, and the bridge arm reactor voltage drop within 2ms after the protection is activated are used as feature quantities and input into a fast classification head based on the LightGBM classifier to quickly identify specific fault types, including valve short-circuit fault, valve group grounding fault, bridge arm reactor phase-to-phase short-circuit fault, AC side single-phase grounding fault, AC side phase-to-phase fault, DC side unipolar fault, and DC side bipolar fault. For Type II faults, the peak-to-peak values ​​of the capacitor voltages of all submodules, the bridge arm current, and the bridge arm reactor voltage drops within 20ms after the protection is initiated are used as feature quantities and input into a comprehensive classification head composed of a soft voting integration model based on the XGBoost, LightGBM, and CatBoost models to quickly identify specific fault types, including SM upper tube open circuit fault, SM lower tube open circuit fault, and normal operation.

[0030] In addition, the peak-to-peak capacitor voltage, bridge arm current, and bridge arm reactor voltage drop of all submodules after protection is activated are used as feature quantities and input into different models respectively. The following offline training process is also included: First, an adversarial network is generated based on simulation and historical data to generate a large amount of sample data. Second, sample data processing and label data processing are performed. Then, an ensemble learning model based on a voting mechanism is constructed to classify fault severity. Finally, the fast classification head or comprehensive classification head model is adaptively called.

[0031] The differentiated protection response strategy is adopted according to the specific fault type, including: For arm-level faults and equipment-level faults, the protection device blocks the converter and simultaneously trips the AC circuit breaker to cut off the external power supply. For module-level faults, the protection device sends an alarm signal and automatically starts fault recording.

[0032] In one embodiment of the present invention, Shagotan New Energy sends MMC-HVDC topology as follows Figure 2 As shown in the figure, the low-voltage side includes photovoltaic, wind power generation systems and energy storage power stations. After boosting, they are connected to the AC busbar, connected to the MMC through the collection line, and then rectified by the MMC. After that, they are connected to the AC grid after the transmission line and the inversion process. The basic structure of the MMC and its submodules is shown in the figure. Figure 3 As shown, U ac is the rated phase voltage on the AC side, U dc is the rated pole-to-pole voltage on the DC side, C is the submodule capacitance, L is the bridge arm inductance, N is the number of submodules of single-phase single bridge arm, 、 and For the 1st, 2nd and Nth submodules of the upper bridge arm of phase A, 、 and The 1st, 2nd and Nth submodules of the lower bridge arm of phase A, 、 and is the bridge arm current of the three-phase upper bridge arm, 、 and is the bridge arm current of the three-phase lower bridge arm, 、 and is the valve group pressure drop of the three-phase upper bridge arm, 、 and is the valve group pressure drop of the three-phase lower bridge arm, 、 and is the three-phase AC side current, For submodules l The port voltage, For submodules lThe system uses a symmetrical unipolar connection and includes a sending-end rectifier station (MMC1) and a receiving-end inverter station (MMC2). The former uses fixed active and reactive power control, while the latter uses fixed DC voltage and fixed reactive power control. The modular multilevel converter (MMC) primarily consists of multiple submodules (SMs) connected in series to form bridge arms. Each phase consists of an upper and lower bridge arm, for a total of six bridge arms across three phases. Each bridge arm consists of N series-connected submodules and an arm reactor. Each submodule primarily consists of two IGBT switches (upper transistor T1 and lower transistor T2), two anti-parallel diodes (D1 and D2), and a submodule capacitor C.

[0033] During normal operation, the submodule operates in two states: active and bypass. When the upper transistor T1 is on and the lower transistor T2 is off, the submodule is in the active state, connecting the submodule capacitor to the circuit. When the upper transistor T1 is off and the lower transistor T2 is on, the submodule is in the bypass state, bypassing the submodule capacitor. By controlling the states of each submodule, the MMC can synthesize a stepped voltage waveform, achieving conversion between AC and DC power.

[0034] According to the location and impact range of the fault, MMC faults can be divided into three levels: module-level faults, bridge-arm-level faults, and equipment-level faults.

[0035] (1) Module-level failure Module-level faults mainly refer to faults that occur within submodules, including: Fault condition 1: The upper IGBT of the submodule is open circuit fault; Fault condition 2: Open circuit failure of the lower IGBT of the submodule.

[0036] like Figure 4 As shown in the figure, when an open-circuit fault occurs in the upper IGBT of a submodule, the submodule cannot enter the active state, resulting in the capacitor being unable to charge or discharge, and the submodule capacitor voltage will fluctuate abnormally. When an open-circuit fault occurs in the lower IGBT of a submodule, the submodule cannot enter the bypass state, causing the capacitor to remain in the circuit, and the submodule capacitor voltage will also fluctuate abnormally. Module-level faults generally do not cause an immediate system crash, but if not addressed promptly, they can lead to more serious failures.

[0037] (2) Bridge arm level fault Bridge arm level faults mainly refer to faults occurring in the bridge arm circuit, including: Fault condition three: valve short circuit fault; Fault condition four: valve group grounding fault; Fault condition five: interphase short circuit fault of bridge arm reactor.

[0038] like Figure 5As shown, arm-level faults often cause sudden changes in arm current and abnormal fluctuations in submodule capacitor voltage. For example, a valve short-circuit fault can cause a large number of submodules to short-circuit, rapidly increasing arm current. A valve block grounding fault can lead to abnormal grounding potential, causing imbalance in arm current and voltage. An interphase short-circuit fault in an arm reactor can create a direct current path between different arms, causing severe overcurrent. Arm-level faults are often more severe than module-level faults and require a rapid response.

[0039] (3) Equipment-level failure Equipment-level faults mainly refer to faults that occur at the connection between the MMC and the external system, including: Fault condition six: single-phase grounding fault on the AC side; Fault condition seven: AC side phase-to-phase fault; Fault condition eight: DC side single-pole fault; Fault condition nine: bipolar fault on the DC side.

[0040] like Figure 6 As shown, equipment-level faults often cause severe distortion in system current and voltage. For example, a ground fault on the AC side can cause a decrease in phase voltage and an increase in phase current; a phase-to-phase fault on the DC side can cause a collapse in DC voltage and a surge in DC current. Equipment-level faults are the most severe type of fault and require a very short response time (1-3ms) to promptly isolate the fault and prevent equipment damage and system failure.

[0041] The multi-level fault systematization method proposed in this invention includes four steps: protection initiation, fault severity classification, fault identification, and protection response. Each step is described in detail below.

[0042] (1) Protection start Protection startup is the first step in the entire protection plan. Its purpose is to detect potential faults in a timely manner by monitoring abnormal changes in key electrical quantities and trigger subsequent protection processes.

[0043] In this embodiment, the peak-to-peak value of the capacitor voltage of all submodules of a single-phase single bridge arm is selected ( ) as the characteristic quantity of protection startup. Figure 7 As shown, two protection start criteria are defined: Criterion 1: When the peak-to-peak value of the capacitor voltage of all submodules When the value of three consecutive sampling points is greater than the preset threshold When , the protection is triggered to start; Criterion 2: When the peak-to-peak difference of the capacitor voltages of all submodules (i.e. two adjacent sampling points The change in the value of the threshold is less than the preset threshold at five consecutive sampling points. When the protection is triggered.

[0044] Among them, the preset threshold and preset thresholds It can be determined based on system parameters and operating experience. For example, the threshold Can be set to avoid normal system operation 30V, in order to accurately and quickly identify partial ground short circuit faults, the threshold Can be set to 0.2V.

[0045] Using three consecutive sampling points for judgment can effectively avoid false starts due to random interference or measurement errors, improving the reliability of protection activation. Furthermore, using peak-to-peak voltage and its differential as two indicators for judgment can balance the detection capabilities of both slowly developing and rapidly developing faults.

[0046] (2) Fault severity classification After the protection is initiated, the second step is to preliminarily classify the severity of the fault so that differentiated handling strategies can be adopted for different types of faults.

[0047] like Figure 8 As shown, this embodiment divides MMC failures into two categories: Type I faults: These include arm-level faults (fault conditions 3, 4, and 5) and equipment-level faults (fault conditions 6, 7, 8, and 9). These faults develop rapidly, have severe impacts, and require a quick response. Type II faults: include module-level faults (fault conditions 1 and 2), characterized by weak fault characteristics and slow fault development, which allows for a longer identification time.

[0048] Fault severity classification uses a shared classification head model based on the LightGBM classifier, with input being bridge arm current data within 1ms of protection activation. LightGBM is an efficient gradient boosting decision tree algorithm with advantages such as fast training, low memory usage, and high classification accuracy. It is particularly suitable for protection systems that require rapid response.

[0049] The shared classification head model analyzes the amplitude, rate of change, harmonic content, and other characteristics of the bridge arm current data to determine whether the fault belongs to Class I or Class II. The classification result will determine whether to use a fast or comprehensive classification head for further fault identification.

[0050] (3) Fault identification Fault identification is the core step of this protection scheme. Its purpose is to further identify the specific fault type based on the preliminary classification and provide a decision basis for subsequent protection response.

[0051] Type I fault identification: If the fault is determined to be a Class I fault, the fast classification head is used for further identification. Figure 9 As shown, the fast classification head is based on the LightGBM classifier, and the input is three electrical quantity data within 2ms after the protection is started: the peak-to-peak value of the capacitor voltage of all submodules , bridge arm current and bridge arm reactor voltage drop .

[0052] The rapid classification head identifies specific fault types by analyzing the time domain and frequency domain characteristics of these three electrical quantities, including valve short-circuit fault (operating condition three), valve group grounding fault (operating condition four), bridge arm reactor phase-to-phase short-circuit fault (operating condition five), AC side single-phase grounding fault (operating condition six), AC side phase-to-phase fault (operating condition seven), DC side unipolar fault (operating condition eight), and DC side bipolar fault (operating condition nine).

[0053] The design of the fast classification head focuses on ensuring recognition speed while also balancing accuracy. Experimental verification shows that the head can complete fault identification within 2 milliseconds with an accuracy rate of 99%, meeting the requirement for rapid response to severe faults.

[0054] Type II fault identification: If the fault is determined to be a Class II fault, a comprehensive classification head is used for further identification. Figure 10 As shown, the comprehensive classification head is based on the soft voting integration of three models: XGBoost, LightGBM and CatBoost. The input is the peak-to-peak value of the capacitor voltage of all submodules within 20ms after the protection is started. , bridge arm current and bridge arm reactor voltage drop data.

[0055] The comprehensive classification head analyzes data over a longer time window to identify a submodule's top IGBT open-circuit fault (condition 1), a submodule's bottom IGBT open-circuit fault (condition 2), or normal operation. The integration of three distinct models leverages the strengths of each to improve identification accuracy and robustness.

[0056] The design of the comprehensive classification head focuses on improving recognition accuracy, especially for module-level faults with less obvious characteristics. Experimental verification shows that the classification head can complete fault identification within 20 milliseconds with an accuracy rate of 95%, providing a reliable basis for subsequent troubleshooting.

[0057] (4) Protection response Protection response is the last step of the entire protection plan. According to the results of fault identification, corresponding protection measures are taken to ensure the safe operation of the system.

[0058] like Figure 11 As shown in the figure, for Type I faults (fault conditions 3 to 9), the protective measures taken are to block the converter and trip the AC circuit breaker. Blocking the converter means stopping the PWM pulses, causing all IGBTs to cease conduction and prevent the fault current from continuing to flow through the converter; tripping the AC circuit breaker disconnects the MMC from the AC system, completely isolating the fault. These measures are designed to cut off the fault current path as quickly as possible, preventing equipment damage and system failure.

[0059] For Class II faults (fault conditions 1 and 2), the protective measure is to issue an alarm signal and automatically initiate fault recording. Since module-level faults do not immediately cause system crashes, the system can continue to operate. However, an alarm signal is issued to notify maintenance personnel, and electrical waveform data at the time of the fault is recorded to provide a basis for subsequent locating the specific faulty submodule.

[0060] The following is a specific implementation case to illustrate the implementation and effects of the present invention in detail. Consider a ±320kV / 300MW flexible DC transmission system with the following MMC converter parameters: Rated DC voltage: ±320kV Rated capacity: 300MW AC side rated voltage: 166kV Number of submodules per bridge arm: 100 Submodule capacitance: 5mF Bridge arm reactor: 60mH Sampling frequency: 10kHz (i.e. sampling period is 0.1ms) Protect startup parameters: Preset threshold 1: Avoid normal operation , about 30V; Preset threshold 2: 3 times the peak-to-peak rate of change of the capacitor voltage of all submodules during normal operation, which is approximately 500 V / ms.

[0061] Fault severity classification: Input data: bridge arm current data within 1ms after protection is started, a total of 10 sampling points; Classification model: LightGBM, with parameters set to max_depth=5, n_estimators=50, learning_rate=0.1.

[0062] Type I fault identification (quick classification head): Input data: within 2ms after protection starts 、 、 Data, a total of 20 sampling points; Classification model: LightGBM, with parameters set to max_depth=5, n_estimators=50, learning_rate=0.1.

[0063] Type II fault identification (comprehensive classification): Input data: within 20ms after protection starts 、 、 Data, a total of 200 sampling points; Classification model: soft voting integration of XGBoost, LightGBM, and CatBoost; XGBoost parameters: max_depth=6, n_estimators=100, learning_rate=0.05 LightGBM parameters: max_depth=6, n_estimators=100, learning_rate=0.05 CatBoost parameters: depth=6, iterations=100, learning_rate=0.05 In order to verify the effectiveness of the present invention, simulation tests were conducted on the above nine fault conditions, and the key data and time nodes in the fault identification process were recorded. Figure 12 The verification results are as follows: Protection start-up time: For Class I faults (operating conditions three to nine), the average protection start-up time is 0.56ms; for Class II faults (operating conditions one and two), the average protection start-up time is 5.25ms.

[0064] Fault severity classification accuracy: 99.8%, average classification time: 0.32ms.

[0065] The accuracy rate of Class I fault identification is 95%, and the average identification time is 1.23ms.

[0066] The accuracy rate of Class II fault identification is 97%, and the average identification time is 12.27ms.

[0067] Protection response time: For Class I faults, the average time from protection initiation to execution of protection measures is 2.03ms; for Class II faults, the average time from protection initiation to execution of protection measures is 14.32ms.

[0068] Experimental results demonstrate that the proposed protection method can quickly and accurately identify various fault types and implement appropriate protection measures, meeting the protection requirements of MMC systems. In particular, for Class I faults, which require a rapid response, the entire process from fault occurrence to protection execution can be completed in less than 3ms, effectively preventing fault escalation and equipment damage.

[0069] In order to further improve the adaptability and reliability of the protection scheme, a parameter adaptive optimization mechanism can be introduced. Specifically, it includes: (1) Self-adaptation of protection start thresholds: Dynamically adjust the protection start threshold values ​​1 and 2 according to factors such as system operating status and load level, ensuring that faults can be discovered in a timely manner and avoiding false operations caused by improper threshold settings.

[0070] (2) Classification model parameter optimization: Through online learning or regular offline training, the parameters of the classification model are optimized so that it can adapt to system parameter changes and new fault characteristics.

[0071] (3) Time window adaptation: Dynamically adjust the time window for data collection according to the system operating status and fault characteristics, and shorten the response time as much as possible while ensuring recognition accuracy.

[0072] The protection method proposed in this invention is not only applicable to conventional MMC converter stations, but can also be expanded to the following scenarios: (1) Multi-terminal HVDC transmission system: By adjusting the protection strategy and parameter settings, it is applied to the multi-terminal flexible HVDC transmission system to achieve protection of more complex network structures.

[0073] (2) Hybrid topology MMC: Applicable to MMC systems with different topologies such as half-bridge submodules, full-bridge submodules, and hybrid submodules. Only the fault feature extraction and classification model parameters need to be adjusted.

[0074] (3) Wind power and photovoltaic grid-connected systems: MMC converter protection used for new energy grid-connected systems to meet the safety requirements for renewable energy access to the grid.

[0075] (4) Power electronic grid: With the improvement of the power electronic level of the power system, the new generation of offshore wind power and new energy Shagotan transmission system have applied new topology DC converters. This protection method can be expanded to apply to the fault protection of various power electronic DC converters, providing technical support for the safe and stable operation of future power systems.

[0076] Embodiment 2, a modular multi-level converter multi-level system protection system, comprising: Protection startup module: determines whether to trigger protection startup by real-time monitoring of the peak-to-peak value of the capacitor voltage of all sub-modules of a single-phase single-bridge arm; Fault severity classification module: The bridge arm current after protection is activated is used as the input feature and input into the shared classification head model based on the LightGBM classifier to perform preliminary fault classification; Fault identification module: Based on the preliminary classification results, the peak-to-peak capacitor voltage, bridge arm current, and bridge arm reactor voltage drop of all submodules after protection is activated are used as feature quantities and input into different models to identify the specific fault type; Protection response module: Adopt differentiated protection response strategies based on specific fault types.

[0077] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and it should be understood that the scope of protection of the present invention is not limited to such specific descriptions and embodiments. Those skilled in the art can make various other specific variations and combinations based on the technical teachings disclosed in the present invention without departing from the essence of the present invention, and such variations and combinations are still within the scope of protection of the invention.

Claims

1. A multi-level systematized protection method for a modular multi-level converter, characterized in that: The following steps are involved: Protection startup: By real-time monitoring of the peak-to-peak value of the capacitor voltage of all submodules in a single-phase single-bridge arm, it is determined whether the protection startup is triggered; Fault severity classification: the bridge arm current after protection is activated is used as the input feature and input into the shared classification head model based on LightGBM classifier to perform preliminary fault classification; Fault identification: Based on the preliminary classification results, the peak-to-peak capacitor voltage, bridge arm current, and bridge arm reactor voltage drop of all submodules after protection is activated are used as feature quantities and input into different models to identify the specific fault type. Protection response: adopt differentiated protection response strategies based on the specific fault type.

2. The multi-level systematic protection method for modular multilevel converters according to claim 1, characterized in that: The triggering conditions for the protection start are: When the peak-to-peak voltage of the capacitor When the value of three consecutive sampling points is greater than the preset threshold When, or when the peak-to-peak voltage of the capacitor Difference When the value of five consecutive sampling points is less than the preset threshold When protection starts.

3. The multi-level systematic protection method for modular multilevel converters according to claim 2, characterized in that: The peak-to-peak value of the capacitor voltage The calculation formula is: in, In the same bridge arm Capacitor voltage of each submodule; The peak-to-peak value of the capacitor voltage Difference The calculation formula is: in, The current sampling point number.

4. The multi-level systematic protection method for modular multilevel converters according to claim 1, characterized in that: Will protect the bridge arm current 1ms after startup As input features, they are input into the shared classification head model based on the LightGBM classifier to perform preliminary fault classification, specifically including: Type I faults: include bridge arm level faults and equipment level faults; Type II failure: includes module-level failure.

5. The multi-level systematic protection method for modular multilevel converters according to claim 4, characterized in that: According to the preliminary classification results, the peak-to-peak values ​​of the capacitor voltage, bridge arm current, and bridge arm reactor voltage drop of all submodules after the protection is started are used as feature quantities and input into different models to identify specific fault types, including: For Type I faults, the peak-to-peak values ​​of the capacitor voltages of all submodules, the bridge arm current, and the bridge arm reactor voltage drop within 2ms after the protection is activated are used as feature quantities and input into a fast classification head based on the LightGBM classifier to quickly identify specific fault types, including valve short-circuit fault, valve group grounding fault, bridge arm reactor phase-to-phase short-circuit fault, AC side single-phase grounding fault, AC side phase-to-phase fault, DC side unipolar fault, and DC side bipolar fault. For Type II faults, the peak-to-peak values ​​of the capacitor voltages of all submodules, the bridge arm current, and the bridge arm reactor voltage drops within 20ms after the protection is initiated are used as feature quantities and input into a comprehensive classification head composed of a soft voting integration model based on the XGBoost, LightGBM, and CatBoost models to quickly identify specific fault types, including SM upper tube open circuit fault, SM lower tube open circuit fault, and normal operation.

6. The multi-level systematic protection method for modular multilevel converters according to claim 5, characterized in that: The differentiated protection response strategy is adopted according to the specific fault type, including: For arm-level faults and equipment-level faults, the protection device blocks the converter and simultaneously trips the AC circuit breaker to cut off the external power supply. For module-level faults, the protection device sends an alarm signal and automatically starts fault recording.

7. A modular multi-level converter multi-level system protection system, characterized in that: include: Protection startup module: determines whether to trigger protection startup by real-time monitoring of the peak-to-peak value of the capacitor voltage of all sub-modules of a single-phase single-bridge arm; Fault severity classification module: The bridge arm current after protection is activated is used as the input feature and input into the shared classification head model based on the LightGBM classifier to perform preliminary fault classification; Fault identification module: Based on the preliminary classification results, the peak-to-peak capacitor voltage, bridge arm current, and bridge arm reactor voltage drop of all submodules after protection is activated are used as feature quantities and input into different models to identify the specific fault type; Protection response module: Adopt differentiated protection response strategies based on specific fault types.

Images