Six-phase mmc circulating current suppression method and system based on pir control

Abstract

The application discloses a six-phase MMC circulating current suppression method and system based on PIR control, and relates to the technical field of motor driving control. The method comprises the following steps: continuously monitoring six operating parameters of six bridge arms in a six-phase motor; when any phase open-circuit fault exists, five interference operating parameter sequences of the other five bridge arms are screened; according to a moving window, the five interference operating parameter sequences are intercepted and respectively input into corresponding five interference feature prediction intelligent agents to obtain five first predicted interference feature groups; the five predicted interference feature groups are continuously analyzed, the combined stability is calculated, and an optimization strategy is configured to perform PIR control optimization to obtain five optimal PIR control parameters; and the circulating current suppression control is performed according to the five optimal PIR control parameters, and real-time updating is performed. The application solves the technical problem that the traditional controller cannot effectively suppress the direct current offset and asymmetric harmonic components after the single-phase open-circuit fault, and improves the fault-tolerant operation ability and steady-state performance of the system.

Description

Technical Field

[0001] This invention relates to the field of motor drive control technology, specifically to a six-phase MMC circulating current suppression method and system based on PIR control. Background Technology

[0002] The demand for high-power, high-reliability drive equipment in offshore power generation systems is becoming increasingly urgent. Multiphase motors and their associated converter technologies have attracted widespread attention due to their high power density and strong fault tolerance. Dual three-phase permanent magnet synchronous motors, consisting of two sets of three-phase windings and driven by a six-phase power converter, can maintain operation under partial faults. Modular multilevel converters, characterized by high output voltage levels, low harmonic content, and strong scalability, have been widely used in high-voltage, high-power scenarios. Applying a six-phase modular multilevel converter topology to dual three-phase permanent magnet synchronous motor drives can effectively improve system voltage levels and waveform quality, adapting to the operational requirements of complex conditions such as offshore wind power.

[0003] However, existing systems suffer from circulating current suppression failure when a single-phase open-circuit fault occurs. After the faulty phase arm current returns to zero, the current of that phase is redistributed to the other five healthy phases, causing the system to evolve from symmetrical six-phase operation to asymmetrical five-phase operation, and generating complex asymmetrical circulating current components such as DC offset, fundamental frequency disturbance, and second harmonics between the bridge arms. Traditional proportional-integral or proportional-resonant control strategies are mainly designed for the second harmonic negative sequence component under symmetrical operating conditions, lacking the ability to suppress DC offset and asymmetrical harmonic components. At the same time, the current reconstruction after the fault exacerbates the uneven energy exchange between the bridge arms and capacitor voltage fluctuations, creating dynamic coupling between voltage balance and circulating current suppression; harmonic subspace currents are significantly excited, causing additional losses and torque ripple, and existing methods lack a unified collaborative suppression framework. Summary of the Invention

[0004] This invention addresses the technical problem of existing technologies lacking accurate modeling and effective suppression mechanisms for asymmetric circulating current components when a single-phase open-circuit fault occurs in a six-phase modular multilevel converter drive system. It provides a six-phase MMC circulating current suppression method and system based on PIR control.

[0005] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: In a first aspect, the present invention provides a six-phase MMC circulating current suppression method based on PIR control, comprising: The six operating parameters of the six bridge arms in the six-phase motor are continuously monitored. When any phase open circuit fault exists, the five interfering operating parameter sequences of the other five bridge arms are obtained. According to the moving window, the five interference operation parameter sequences are extracted and respectively input into the five corresponding interference feature prediction agents to obtain five first predicted interference feature groups. Further analysis of the five predicted disturbance feature groups was conducted, the combined stability was calculated, and an optimization strategy was configured to optimize PIR control, resulting in five optimal PIR control parameters. Circulation suppression control is performed based on five optimal PIR control parameters, and the parameters are updated in real time.

[0006] Secondly, the present invention provides a six-phase MMC circulating current suppression system based on PIR control, comprising: The monitoring module is used to continuously monitor six operating parameters of the six bridge arms in the six-phase motor. When any phase open circuit fault exists, it filters out the five interfering operating parameter sequences of the other five bridge arms. The feature prediction module is used to extract the five interference running parameter sequences according to the moving window and input them into the corresponding five interference feature prediction agents to obtain five first predicted interference feature groups. The parameter optimization module is used to further analyze the five predicted interference feature groups, calculate the combined stability, and configure the optimization strategy to optimize the PIR control, thereby obtaining five optimal PIR control parameters. The circulating current suppression module is used to perform circulating current suppression control according to five optimal PIR control parameters and update them in real time.

[0007] The beneficial effects of this invention are: Compared to existing technologies, this invention first establishes a real-time data acquisition mechanism for single-phase open-circuit faults by continuously monitoring the operating parameters of six bridge arms and filtering five interference operating parameter sequences when an open-circuit fault occurs. Secondly, by using a moving window to capture the interference operating parameter sequences and inputting them into an interference feature prediction agent, it accurately predicts the circulating current DC offset and second harmonic component of the healthy bridge arm after the fault, overcoming the deficiency of existing technologies in accurately modeling asymmetric circulating currents. Thirdly, by calculating and merging stability and optimizing PIR control parameters, the integral term eliminates DC offset, and the resonant term locks in characteristic frequency harmonics, solving the problem that traditional controllers cannot effectively suppress asymmetric circulating current components. Finally, circulating current suppression control is performed according to the optimized PIR control parameters and updated in real time, achieving decoupled synergy between voltage balance and circulating current suppression. This suppresses the surge in harmonic subspace current, reduces additional losses and torque ripple, and improves the system's operational stability and fault tolerance under single-phase open-circuit fault conditions. Attached Figure Description

[0008] Figure 1 A schematic flowchart of the six-phase MMC circulating current suppression method based on PIR control provided by the present invention; Figure 2 A schematic diagram of the structure of the six-phase MMC circulating current suppression system based on PIR control provided by the present invention.

[0009] In the attached diagram, the components represented by each number are as follows: Monitoring module 11, feature prediction module 12, parameter optimization module 13, and circulation suppression module 14. Detailed Implementation

[0010] Example 1, as Figure 1 As shown, this embodiment of the invention provides a six-phase MMC circulating current suppression method based on PIR control, including: S10: Continuously monitor six operating parameters of the six bridge arms in the six-phase motor. When any phase open circuit fault exists, filter out the five interfering operating parameter sequences of the other five bridge arms. First, six operating parameters of the six bridge arms in the six-phase motor are continuously monitored. The six-phase motor is a dual three-phase permanent magnet synchronous motor composed of two sets of three-phase windings, which are spatially offset by 30 electrical degrees. It is driven by a six-phase modular multilevel converter. The six bridge arms correspond to the six phase windings of the motor, and each bridge arm is composed of multiple cascaded sub-modules, which are responsible for the current transmission and voltage transformation of the corresponding phase. In high-power applications such as offshore wind power, the six-phase motor and its drive system need to operate for a long time in harsh environments with high humidity, high salt spray, and difficult maintenance, which places high demands on the system's fault tolerance.

[0011] Continuous monitoring of six operating parameters across the six bridge arms of a six-phase motor refers to the real-time acquisition of key operating data such as current and voltage for each bridge arm to determine the current operating status of the system. Specifically, when a phase experiences an open-circuit fault, the bridge arm current of that faulty phase drops to zero. The current component originally borne by that phase is redistributed to the other five healthy phases, causing an increase in the current amplitude and a phase shift in the healthy phases, generating complex asymmetric circulating current components between the bridge arms. Therefore, when any phase experiences an open-circuit fault, it is necessary to filter out the five interfering operating parameter sequences of the other five bridge arms to isolate the influence of the faulty phase from the monitoring data, focusing on the interference signals generated by current reconstruction in the healthy phases, thus providing an accurate data foundation for subsequent circulating current characteristic analysis and suppression control.

[0012] Specifically, six operating parameters of the six bridge arms in a six-phase motor are continuously monitored. When any phase open circuit fault occurs, five interfering operating parameter sequences of the other five bridge arms are obtained, including: During the operation of the six-phase MMC motor, six operating parameters of the six bridge arms are continuously monitored; When any operating parameter is interrupted, it is determined that there is a single-phase interruption fault. The operating parameters of the other five bridge arms are screened and continuously recorded to obtain five interference operating parameter sequences.

[0013] First, during the operation of the six-phase MMC motor (six-phase modular multilevel converter driving a dual three-phase permanent magnet synchronous motor), six operating parameters of the six bridge arms are continuously monitored. The six bridge arms correspond to the six phase windings of the dual three-phase permanent magnet synchronous motor. Each bridge arm consists of multiple cascaded submodules, each containing capacitors and switching devices for voltage conversion and energy transfer. Optionally, the operating parameters include at least the instantaneous bridge arm current and the average voltage of the submodule capacitors for each bridge arm. These parameters are acquired in real time using current and voltage sensors, with a sampling frequency set to above 10 kHz to ensure accurate capture of transient changes at the moment of a fault.

[0014] A single-phase open-circuit fault is determined when any operating parameter experiences an open circuit. Specifically, an open circuit in any operating parameter indicates that the instantaneous current value of the corresponding bridge arm is continuously zero or continuously below a preset threshold for a duration exceeding a preset sampling period, while the voltage modulation signal corresponding to that bridge arm remains within the normal output range. In this case, the switching device or winding connection point of that bridge arm is physically disconnected, and a single-phase open-circuit fault is determined for that phase. The preset threshold can be set to 5% of the rated current to avoid false judgments caused by noise interference. The preset sampling period is set according to the fault detection response speed requirements and the system sampling frequency, for example, five sampling periods, to avoid false triggering caused by instantaneous current fluctuations and to ensure that fault identification and fault-tolerant control switching can be completed within two 0.02 seconds after the fault occurs. After fault determination, the operating parameters corresponding to the faulty phase bridge arm are marked as invalid and no longer participate in subsequent control calculations.

[0015] Simultaneously, the operating parameters of the other five bridge arms were screened and continuously recorded, resulting in five sequences of disturbance operating parameters. After a fault occurs, the current component originally borne by the faulty phase is redistributed to the other five healthy phases, causing an increase in the current amplitude and phase shift in the healthy phase bridge arms, while the voltage fluctuation of the submodule capacitors intensifies. To accurately capture the disturbance impact of this current reconstruction process on the healthy phases, the operating parameters of the faulty phase bridge arms need to be removed from the monitoring data, retaining only the operating parameters of the five healthy phase bridge arms, and continuously recorded in time series form, ultimately obtaining five sequences of disturbance operating parameters.

[0016] Specifically, each disturbance operation parameter sequence includes a sequence of historical instantaneous current values ​​and a sequence of historical average capacitor voltage values ​​for the corresponding bridge arm. The sequence length is set according to the duration of the transient process after the fault, and can be set to five seconds. This is used to completely record the current reconstruction characteristics and energy fluctuation patterns of the healthy phase bridge arm from transient response to steady-state operation after the fault occurs.

[0017] The above methods enable rapid identification of faulty phases and accurate extraction of interference signals from healthy phases, providing a data foundation for subsequent circulation characteristic analysis.

[0018] S20: According to the moving window, the five interference operation parameter sequences are extracted and respectively input into the five corresponding interference feature prediction agents to obtain five first prediction interference feature groups; Secondly, following the moving window approach, the five interference operating parameter sequences are extracted and input into the corresponding five interference feature prediction agents. The moving window refers to a data sampling method that uses a fixed time length as the data extraction unit and slides forward along the time axis with a preset step size. Optionally, the window width is set according to the consistency requirements of the frequency resolution required for circulation feature analysis and the input dimension of the neural network. For example, setting the window width to 0.2 seconds and the sliding step size to 0.02 seconds is used to extract representative local data segments from the continuous time series to capture the dynamic evolution of circulation features over time. A time length of 0.2 seconds corresponds to approximately ten fundamental frequency cycles at the motor's rated speed. Although the number of fundamental frequency cycles included in the fixed time window will change accordingly when the motor speed changes, the number of sampling points remains constant, ensuring the consistency of the data dimension input to the interference feature prediction agent.

[0019] Based on this moving window, the five interference operation parameter sequences obtained above are extracted, and the latest data segment at the current moment can be extracted from each interference operation parameter sequence as the first interference operation parameter group. The first interference operation parameter group includes the instantaneous value of the bridge arm current and the average value of the submodule capacitor voltage within a 0.2-second time window.

[0020] Then, the five extracted sets of first interference operating parameters are input into the corresponding five interference feature prediction agents. These agents are prediction models built using machine learning algorithms. Each agent receives the current set of interference operating parameters as input and calculates the corresponding predicted interference features through forward propagation, including the circulating current DC offset and the amplitude of the circulating current second harmonic component. Each interference feature prediction agent corresponds to a fixed bridge arm and is pre-trained under normal system operation.

[0021] By extracting five sequences of disturbance operating parameters and inputting them into five corresponding disturbance feature prediction agents, the predicted values ​​of the circulating current DC offset and second harmonic component of the five healthy phase bridge arms under the current operating conditions can be obtained in real time. This quantifies the complex asymmetric circulating current components after a fault into quantifiable feature parameters, resulting in five first predicted disturbance feature groups. Each first predicted disturbance feature group contains the predicted value of the circulating current DC offset and the predicted value of the circulating current second harmonic component of the corresponding bridge arm, providing a precise target basis for the subsequent optimization configuration of PIR control parameters.

[0022] Specifically, according to the moving window, the five interference operation parameter sequences are input into the corresponding five interference feature prediction agents to obtain five first predicted interference features, including: Among the six pre-configured interference feature prediction agents, select and call five interference feature prediction agents from the five normally operating bridge arms. According to the preset window length, extract five first interference operation parameter groups from five interference operation parameter sequences, input them into the corresponding five interference feature prediction intelligent bodies, and output five first predicted interference feature groups.

[0023] First, from the six pre-configured interference feature prediction agents, five interference feature prediction agents from the five normally operating bridge arms are selected and invoked. The six-phase modular multilevel converter drive system is configured with a total of six interference feature prediction agents, each corresponding to one bridge arm, and used to predict the circulating current characteristics of the corresponding bridge arm under fault conditions. When a phase is determined to have an open-circuit fault, the interference feature prediction agent corresponding to the faulty phase bridge arm is automatically identified and marked as invalid, and will not participate in subsequent invocations. Simultaneously, from the remaining five healthy phase bridge arms, five interference feature prediction agents matching the healthy phase bridge arms are selected and invoked, ensuring that the interference operating parameter sequence of each healthy phase bridge arm can be input into the corresponding trained prediction model, avoiding prediction errors caused by mismatch between the model and the bridge arm.

[0024] The configuration steps for the six interference feature prediction agents include: Based on the single-phase fault record data of the six-phase motor during its historical operation, six sets of sample interference operating parameters were collected, and the corresponding sample interference feature groups were obtained and labeled to obtain six sets of sample interference feature groups. Each sample interference feature group includes circulating current DC offset and circulating current second harmonic. Based on machine learning, a predictive agent for six perturbation features is constructed. The training data is a set of six sample interference parameters, and the supervision label is a set of six sample interference features. The six interference feature prediction agents are trained and supervised and adjusted respectively until the training converges and the configuration is completed.

[0025] Specifically, the configuration process of the six interference feature prediction agents includes three stages: sample collection, model building, and supervised training.

[0026] First, during the sample acquisition phase, based on the single-phase fault record data of the six-phase motor during historical operation, six sets of sample interference operating parameters were collected, and corresponding sample interference feature groups were obtained and labeled to obtain the six sets of sample interference feature groups. Specifically, in the historical operating data of the six-phase modular multilevel converter drive system experiencing a single-phase open-circuit fault, the instantaneous value sequence of the bridge arm current and the average value sequence of the submodule capacitor voltage within a 0.2-second time window before and after the fault occurrence were extracted for each of the six bridge arms, forming six sets of sample interference operating parameters. Each set of sample interference operating parameters contains multiple samples of the corresponding bridge arm under different fault conditions.

[0027] Meanwhile, through offline analysis and annotation, the true circulation characteristics corresponding to each sample are obtained, including the DC offset of the circulation and the amplitude of the second harmonic component of the circulation. The true circulation characteristics are used as the sample interference feature group and matched and annotated with the corresponding sample interference operation parameter group to form a set of six sample interference feature groups.

[0028] The circulating DC offset is a constant component with zero frequency in the bridge arm current. It represents the bias current generated in the healthy phase bridge arm due to power redistribution during the current reconfiguration process after a fault. This bias current causes the phase current waveform to shift upward or downward, disrupting current symmetry. It is obtained by low-pass filtering the bridge arm current sequence and then averaging the results. The cutoff frequency of the low-pass filter is set, for example, to 10 Hz to filter out the AC component and retain the DC component.

[0029] The second harmonic component of the circulating current is the AC component in the bridge arm current with a frequency twice the fundamental frequency. It represents the periodic circulating current fluctuation caused by impedance imbalance between bridge arms after a fault. This fluctuation circulates between each phase bridge arm, causing additional losses and capacitor voltage fluctuations. It is obtained by performing a Fast Fourier Transform on the bridge arm current sequence and extracting the amplitude at the second harmonic. The sampling window length is set, for example, to 0.2 seconds, and the frequency resolution is set, for example, to one-tenth of the fundamental frequency, to ensure the accuracy of the second harmonic component amplitude extraction.

[0030] Secondly, in the model building phase, six interference feature prediction agents are constructed based on machine learning. Optionally, each interference feature prediction agent adopts a backpropagation neural network structure. The number of input layer nodes is determined according to the dimension of the sample interference running parameter group, and is set to 4,000, corresponding to the total number of current and voltage data points collected within 0.2 seconds at a sampling rate of 10 kHz. The hidden layer is set to three layers, with 512, 256, and 128 nodes in each layer, respectively, and the activation function is a linear rectified function. The number of output layer nodes is set to two, corresponding to the predicted value of the circulating current DC offset and the predicted value of the circulating current second harmonic component, respectively. The network structure of the six interference feature prediction agents is the same, but the network weight parameters are trained independently, each corresponding to a different bridge arm.

[0031] During the supervised training phase, six sets of sample interference parameters are used as training data, and six sets of sample interference features are used as supervision labels. The six interference feature prediction agents are trained and supervised for adjustment until the training converges and the configuration is completed.

[0032] Specifically, for each interference feature prediction agent, the set of sample interference operation parameters for its corresponding bridge arm is used as input, the labeled sample interference feature set is used as the expected output, and the mean squared error is used as the loss function. The network weight parameters are iteratively updated through backpropagation and an adaptive moment estimation optimizer. During training, the sample data is divided into a training set and a validation set. When the loss function value of the validation set no longer decreases for twenty consecutive iterations, the training is considered to have converged, and the current network weight parameters are saved as the final configuration of the interference feature prediction agent.

[0033] For example, since there is a highly nonlinear and complex correlation between the bridge arm operating parameters and the circulation characteristics, and the neural network model has significant advantages in multi-level feature abstraction and complex pattern recognition, a neural network model can be selected to construct the interference feature prediction agent.

[0034] Specifically, the interference feature prediction agent mainly consists of an input layer, a feature extraction layer, and a feature output layer. The input layer receives a normalized set of interference operating parameters, which includes the instantaneous value sequence of the bridge arm current and the average value sequence of the submodule capacitor voltage within a 0.2-second time window. The sampling frequency is set to 10 kHz, with each window containing 2,000 sampling points. Each sampling point contains both current and voltage values, resulting in 4,000 nodes in the input layer. The feature extraction layer employs a multi-layer fully connected neural network structure. The number of hidden layer neurons is adaptively configured according to the dimensions of the input features: 512 neurons in the first hidden layer, 256 in the second, and 128 in the third. Each layer uses a linear rectified function as the activation function to introduce nonlinear transformation capabilities, and a dropout layer is embedded between network layers with a dropout rate of 0.2 to effectively suppress model overfitting and improve its generalization performance. The feature output layer uses a linear activation function to map the final abstract features into two continuous values, which are used as the predicted value of the circulating current DC offset and the predicted value of the circulating current second harmonic component, respectively.

[0035] During training, key hyperparameters included a learning rate of 0.001, a training epoch count of 200, and a batch size of 64. The learning rate was set to balance training stability and convergence speed; the number of training epochs ensured the model fully learned the nonlinear mapping between the bridge arm operating parameters and circulation characteristics; and the batch size balanced training efficiency with memory resource consumption. Specifically, a supervised learning training method was adopted. Sample interference operating parameter sets were collected from historical single-phase fault records as the input sample set. Simultaneously, offline analysis was used to obtain the corresponding real circulation characteristics, including the circulation DC offset and the amplitude of the circulation second harmonic component, forming a sample label set. The input sample set and the corresponding label sample set were divided into training, validation, and test sets in a 7:2:1 ratio.

[0036] Furthermore, the sample interference operation parameters from the training set are used as input, and the corresponding real circulation characteristics are used as supervision signals. The network weight parameters are iteratively optimized using a backpropagation algorithm combined with an adaptive moment estimation optimizer. A mean squared error loss function is used to measure the deviation between the predicted interference characteristics and the real circulation characteristics. The training process is monitored using a validation set. Training is terminated when the validation set loss function value no longer decreases for twenty consecutive iterations, resulting in a converged interference feature prediction agent. This interference feature prediction agent can effectively capture the complex nonlinear relationship between the bridge arm operation parameters and circulation characteristics, achieving accurate prediction of the circulation DC offset and second harmonic component.

[0037] Ultimately, through the above methods, the six interference feature prediction agents have mastered the nonlinear mapping relationship between the operating parameters and circulating current characteristics of the corresponding bridge arm under the single-phase open-circuit fault condition, and can output accurate predicted interference features in real time during actual operation.

[0038] Furthermore, according to a preset window length, five first interference operation parameter groups are extracted from five interference operation parameter sequences, input into the corresponding five interference feature prediction intelligent bodies, and five first predicted interference feature groups are output. The preset window length is set to 0.2 seconds based on the consistency requirements between the frequency resolution required for circulating current feature analysis and the input dimension of the neural network. This window length can fully cover the transient response stage during the current reconstruction process after a fault and the characteristic fluctuation cycle after entering steady-state operation.

[0039] Specifically, for each of the five healthy phase bridge arms, the latest data segment in time is extracted from the corresponding interference operation parameter sequence according to the preset window length, forming five first interference operation parameter groups. Each first interference operation parameter group contains the bridge arm current instantaneous value sequence and the submodule capacitor voltage average value sequence within a 0.2-second time window. The five extracted first interference operation parameter groups are respectively input to the five corresponding interference feature prediction agents. After receiving the interference operation parameter group of the corresponding bridge arm, each interference feature prediction agent can output the predicted value of the circulating current DC offset and the predicted value of the circulating current second harmonic component under the current operating condition through forward propagation calculation of its internal neural network. The five interference feature prediction agents output five prediction results in parallel, which are combined to form five first predicted interference feature groups. Each first predicted interference feature group corresponds to one healthy phase bridge arm and includes two parameters: the predicted value of the circulating current DC offset and the predicted value of the circulating current second harmonic component.

[0040] In summary, the above methods enable real-time prediction of the interference characteristics of the five healthy phase bridge arms, providing a precise target basis for the subsequent optimization of PIR control parameters.

[0041] S30: Continue to analyze the five predicted disturbance feature groups, calculate the combined stability, and configure the optimization strategy to optimize PIR control, and obtain five optimal PIR control parameters; Next, the five predicted disturbance feature groups obtained are analyzed. The combined stability is calculated, and an optimization strategy is configured for PIR (Proportional-Integral-Resonant) control optimization. This step compares the predicted disturbance features at the current time with those predicted at subsequent time points, quantifying the dynamic change of the system circulation characteristics. Based on this dynamic change, the optimization step size of the PIR control parameters is adaptively adjusted, enabling refined iterative optimization of the control parameters. This avoids the problems of slow convergence or insufficient optimization accuracy caused by a fixed optimization step size, ultimately yielding five optimal PIR control parameters. Each optimal PIR control parameter corresponds to a healthy phase bridge arm and includes three parameters: proportional coefficient, integral coefficient, and resonant coefficient.

[0042] Specifically, we continue to analyze the five predicted interference feature groups and calculate the merge stability, including: Continue to extract five second interference operation parameter groups from the five interference operation parameter sequences according to the preset window length, input them into the corresponding five interference feature prediction intelligent bodies, output five predicted interference feature groups, and obtain five second predicted interference feature groups. The merging stability is calculated based on five first predicted interference feature groups and five second predicted interference feature groups.

[0043] The calculation methods for merger stability include: The first merging stability is obtained by calculating the similarity between the sum of the five first predicted interference feature groups and the sum of the five second predicted interference feature groups.

[0044] First, following a preset window length, five second interference operation parameter groups are extracted from the five interference operation parameter sequences. These are input into the corresponding five interference feature prediction intelligences, and five predicted interference feature groups are output, resulting in five second predicted interference feature groups. In the preceding steps, five first interference operation parameter groups and five first predicted interference feature groups were already extracted using a moving window. To obtain the changing trend of the circulation features over time, this step requires another extraction operation after the preceding steps, at an interval of one sliding step. The sliding step is set to 0.02 seconds. After this 0.02-second interval, the latest data segment is extracted from the five interference operation parameter sequences according to the same preset window length, forming five second interference operation parameter groups. Each second interference operation parameter group differs from its corresponding first interference operation parameter group by 0.02 seconds in time, but the window length is the same, both being 0.2 seconds.

[0045] Five sets of second interference operating parameters are input to five corresponding interference feature prediction agents. Each interference feature prediction agent, after receiving the second interference operating parameter set for its corresponding bridge arm, calculates the predicted value of the circulating current DC offset and the predicted value of the circulating current second harmonic component 0.02 seconds after the current time through forward propagation of its internal neural network, thus obtaining five sets of second predicted interference features. The five sets of second predicted interference features are sequentially connected with the five sets of first predicted interference features in time, jointly describing the dynamic changes of the circulating current features between two adjacent 0.02 seconds.

[0046] Furthermore, the combined stability is calculated based on the five first predicted interference feature groups and the five second predicted interference feature groups. Specifically, the combined stability is used to quantify the overall stability of the circulation features of the five healthy phase bridge arms within adjacent time windows. It is calculated by comparing the similarity between the sum of the five first predicted interference feature groups and the sum of the five second predicted interference feature groups.

[0047] Specifically, the predicted values ​​of the circulating current DC offset and the second harmonic component of the circulating current in each of the five first predicted interference feature groups are added to obtain the first predicted feature sum vector, which contains the sum of the first circulating current DC offset and the sum of the first circulating current second harmonic component. Similarly, the predicted values ​​of the circulating current DC offset and the second harmonic component of the circulating current in each of the five second predicted interference feature groups are added to obtain the second predicted feature sum vector. Optionally, the cosine similarity between the first and second predicted feature sum vectors is calculated, and the cosine similarity is used as the first merged stability output. The value of this merged stability ranges from zero to one. The closer the value is to one, the smaller the change in the overall circulating current characteristics of the five healthy phase bridge arms within adjacent time windows, and the more stable the system tends to be. The closer the value is to zero, the greater the change in the overall circulating current characteristics, and the more the system is still in a dynamic adjustment process.

[0048] By calculating the combined stability, a quantitative basis can be provided for the subsequent optimization of the granular configuration of proportional-integral-resonant control parameters.

[0049] Furthermore, an optimization strategy is configured to optimize PIR control, resulting in five optimal PIR control parameters, including: Calculate the first optimization granularity based on the first merging stability; Based on the PIR controller, five first PIR suppression parameters are generated for the five bridge arms that are operating normally. Based on the five first PIR suppression parameters, the five first predicted interference feature groups, and the five second predicted interference feature groups, the first suppression score is obtained through analysis and processing. According to the first optimization granularity, the five first PIR suppression parameters are iteratively optimized until convergence, and the five optimal PIR suppression parameters with the largest suppression scores are obtained.

[0050] First, based on the first merged stability, the first optimization granularity is calculated. The optimization granularity refers to the step size of each adjustment of the PIR control parameters during iterative optimization. When the first merged stability value is high, it indicates that the overall circulation characteristics of the five healthy phase bridge arms change little within adjacent time windows, and the system tends to be in a steady state. In this case, a smaller optimization granularity should be used for fine adjustment to avoid the control parameters deviating from the optimal value due to excessive adjustment. When the first merged stability value is low, it indicates that the circulation characteristics change significantly, and the system is still in a dynamic adjustment process. In this case, a larger optimization granularity should be used for rapid adjustment to converge to near the optimal parameters as quickly as possible.

[0051] Specifically, based on the first merging stability, the first optimization granularity is calculated, including: Obtain the preset optimization granularity; Based on the first merging stability, a granularity adjustment coefficient is configured, and the preset optimized granularity is adjusted and calculated to obtain the first optimized granularity. The magnitude of the merging stability and the magnitude of the granularity adjustment coefficient are negatively correlated.

[0052] First, obtain the preset optimization granularity. The preset optimization granularity is a baseline adjustment step size pre-set based on the PIR controller's parameter dimensions and engineering experience, configured during system initialization. Since the PIR controller contains three parameters with different dimensions: proportional coefficient, integral coefficient, and resonant coefficient, the preset optimization granularity is set separately for each parameter. Optionally, the preset optimization granularity for the proportional coefficient is set to 0.01, the integral coefficient to 0.1, and the resonant coefficient to 0.1. The selection of these values ​​is based on the following: the proportional coefficient typically varies between zero and ten, and a step size of 0.01 ensures a relative adjustment accuracy of one-thousandth; the integral and resonant coefficients typically vary between zero and one hundred, and a step size of 0.1 ensures a relative adjustment accuracy of one-thousandth. The preset optimization granularity provides a benchmark reference for subsequent granularity adjustments.

[0053] Secondly, based on the first merging stability, a granularity adjustment coefficient is configured to adjust the preset optimization granularity, thus obtaining the first optimized granularity. The granularity adjustment coefficient is a scaling factor negatively correlated with the magnitude of the first merging stability, and its function is to dynamically adjust the optimization step size according to the dynamic changes in the circulation characteristics.

[0054] Optionally, the granularity adjustment coefficient can be configured as follows: the value of the first merging stability is denoted as the merging stability value, which ranges from zero to one. When the merging stability value approaches one, it indicates that the circulation characteristics are stabilizing; when the merging stability value approaches zero, it indicates that the circulation characteristics are changing drastically. The granularity adjustment coefficient is determined according to the rule of subtracting the merging stability value from one. This calculation method realizes a negative correlation between the magnitude of the merging stability and the magnitude of the granularity adjustment coefficient. When the merging stability value is high, the granularity adjustment coefficient is small, the optimization step size is small, and fine adjustment is achieved; when the merging stability value is low, the granularity adjustment coefficient is large, the optimization step size is large, and the convergence speed is accelerated. At the same time, to ensure that the granularity adjustment coefficient is always within a reasonable range, a lower limit constraint is set on the granularity adjustment coefficient. If one minus the merging stability value is less than 0.1, the granularity adjustment coefficient is set to 0.1, avoiding excessively small optimization step sizes that lead to slow convergence speed.

[0055] Then, the preset optimization granularity is multiplied by the granularity adjustment coefficient to obtain the first optimization granularity. In this way, the first optimization granularity can be dynamically adjusted according to the real-time value of the first merging stability. When the system tends to a steady state, a small step size is used for fine optimization, and when the system is in a dynamic state, a large step size is used for fast approximation, thus achieving a balance between optimization efficiency and optimization accuracy.

[0056] Furthermore, based on the PIR controller, five first PIR suppression parameters are generated corresponding to the five normally operating bridge arms. Before the system enters the optimization process, the initial values ​​of the proportional coefficient, integral coefficient, and resonance coefficient of each bridge arm are calculated according to the standard design method of the PIR controller and the rated operating parameters of the five healthy phase bridge arms, forming the five first PIR suppression parameters. Specifically, the initial values ​​are calculated based on the following: the proportional coefficient is determined according to the expected bandwidth of the bridge arm current loop, the integral coefficient is determined according to the expected steady-state error elimination capability, and the resonance coefficient is determined according to the expected gain at the second harmonic characteristic frequency of the circulating current.

[0057] Secondly, based on the five first PIR suppression parameters, five first predicted interference feature groups, and five second predicted interference feature groups, the first suppression score is obtained through analysis and processing. The suppression score is used to quantify the circulation suppression effect of a set of PIR control parameters under the current circulation characteristic conditions.

[0058] Specifically, based on the five first PIR suppression parameters, five first predicted interference feature groups, and five second predicted interference feature groups, a first suppression score is obtained through analysis and processing, including: The PIR suppression prediction agent is invoked, wherein the PIR suppression prediction agent is constructed based on machine learning, using a set of sample interference feature groups and a set of sample PIR suppression parameters as training data, and using a set of labeled sample suppression scores as supervision labels for training. The five first PIR suppression parameters are combined with the five first predicted interference feature groups and the five second predicted interference feature groups, respectively, and input into the PIR suppression prediction agent. The mean of the output results is calculated to obtain the first suppression score.

[0059] First, the PIR suppression prediction agent is invoked. This agent is a machine learning-based predictive model used to evaluate the circulation suppression effect of a set of PIR control parameters under specific circulation characteristics. The PIR suppression prediction agent is pre-trained before the system is put into operation. During training, a set of sample interference features and a set of sample PIR suppression parameters are used as input data, and a set of labeled sample suppression scores is used as supervision labels.

[0060] The sample interference feature set includes multiple sets of circulating current DC offset and circulating current second harmonic component amplitudes, covering the feature distribution range under different fault conditions; the sample PIR suppression parameter set includes multiple sets of proportional coefficients, integral coefficients, and resonance coefficients, covering different combinations of values ​​within the feasible domain of the parameters; the sample suppression score set is obtained through offline simulation or experimental testing, which quantifies the circulating current suppression effect of each set of input parameter combinations in actual operation, with a higher suppression score indicating a better circulating current suppression effect.

[0061] The PIR suppression prediction agent employs a backpropagation neural network structure. The number of nodes in the input layer corresponds to the dimension of the input parameters, including a set of PIR control parameters, a set of interference features at the current time step, and a set of interference features at the next time step, for a total of seven input nodes. The output layer has only one node, outputting the predicted suppression score. Through supervised training, the PIR suppression prediction agent learns the nonlinear mapping relationship from PIR control parameters and circulation features to suppression effect, enabling real-time evaluation of the suppression effect of any set of PIR control parameters under the current circulation feature conditions during actual operation.

[0062] Five first PIR suppression parameters are combined with five first predicted interference feature groups and five second predicted interference feature groups, respectively, and input into the PIR suppression prediction agent. The mean of the output results is calculated to obtain the first suppression score. Specifically, for each of the five healthy phase arms, the first PIR suppression parameters corresponding to that arm, the predicted values ​​of the circulating current DC offset and the second harmonic component of the circulating current in the first predicted interference feature group, and the predicted values ​​of the circulating current DC offset and the second harmonic component of the circulating current in the second predicted interference feature group are combined to form a seven-dimensional input vector, which is then input into the PIR suppression prediction agent. The PIR suppression prediction agent performs forward propagation calculation on this seven-dimensional input vector and can output the suppression score of that arm under the current PIR suppression parameters and current circulating current feature conditions.

[0063] Following the above method, the same operation was performed on each of the five healthy phase bridge arms, resulting in five suppression score output values. The five suppression score output values ​​were added together and divided by five to obtain the arithmetic mean of the five suppression scores. This arithmetic mean was used as the first suppression score. The first suppression score comprehensively reflects the overall circulation suppression effect of the five healthy phase bridge arms under the current PIR suppression parameter configuration, providing a quantitative evaluation basis for subsequent parameter iteration optimization.

[0064] Then, according to the first optimization granularity, the five first PIR suppression parameters are iteratively optimized until convergence, obtaining the five optimal PIR suppression parameters with the highest suppression scores. Optionally, the iterative optimization process uses a particle swarm optimization algorithm, which searches for the optimal solution in the parameter space by simulating the cooperative and competitive behavior among individuals in a swarm. Before optimization begins, for each of the five healthy phase bridge arms, an independent particle swarm is constructed, each containing thirty particles. The position vector of each particle corresponds to a set of PIR suppression parameters, including three dimensions: proportional coefficient, integral coefficient, and resonance coefficient. The velocity vector of each particle corresponds to the parameter adjustment direction and step size in the three dimensions.

[0065] During the initialization phase, the first PIR suppression parameter of each bridge arm is used as the initial global optimal position of the particle swarm of that bridge arm. The initial position of each particle is randomly generated in the vicinity of the first PIR suppression parameter of the bridge arm according to a normal distribution. The standard deviation of the normal distribution is set as the first optimization granularity. The initial velocity of each particle is randomly generated in the range from the negative first optimization granularity to the positive first optimization granularity.

[0066] After entering the iterative loop, the following operations are performed on the particle swarm for each bridge arm in each iteration: For each particle in the particle swarm, the particle's current position vector is used as a set of PIR suppression parameters. Combined with the first and second predicted interference feature sets corresponding to the bridge arm, these parameters are input into the PIR suppression prediction agent to calculate the suppression score corresponding to the particle. This suppression score is then used as the particle's fitness value. For each particle, the fitness value obtained in the current iteration is compared with its historical best fitness value. If the current fitness value is higher, the particle's current position is updated to its individual best position. For the particle swarm of each bridge arm, the position with the highest fitness value among all the individual best positions of the particles is updated to the global best position of that bridge arm. Based on the velocity update formula and position update formula of the particle swarm optimization algorithm, and combined with the first optimization granularity as the inertia weight, the velocity and position of each particle are updated. When updating the velocity, the inertia weight is set to twice the first optimization granularity, and both the individual learning factor and the swarm learning factor are set to two.

[0067] After each iteration, the sum of the suppression scores corresponding to the globally optimal positions of the five bridge arms in the current iteration is recorded as the overall suppression score for this iteration, and the overall suppression score of this iteration is compared with the overall suppression score of the previous iteration. When the improvement in the overall suppression score in five consecutive iterations is lower than a preset convergence threshold, the optimization is considered converged. The preset convergence threshold is set to 0.5%. After convergence, the PIR suppression parameter corresponding to the globally optimal position of each bridge arm during the iteration is selected as the optimal PIR suppression parameter output for that bridge arm, resulting in five optimal PIR suppression parameters.

[0068] In summary, the above methods achieve adaptive optimization of PIR control parameters, enabling the control parameters to be dynamically adjusted according to real-time changes in circulation characteristics, thus ensuring optimal circulation suppression.

[0069] S40: Performs circulating current suppression control according to five optimal PIR control parameters and updates them in real time.

[0070] Finally, circulating current suppression control is implemented using five optimal PIR control parameters, which are updated in real time. After iterative optimization, the five healthy phase arms each obtain their corresponding optimal PIR control parameters, each containing three specific values: proportional coefficient, integral coefficient, and resonant coefficient. These five optimal PIR control parameters are then sent to the corresponding arm's PIR controller, replacing the original control parameters. Based on the received proportional coefficient, integral coefficient, and resonant coefficient, and combined with the real-time feedback value of the arm current and the current command value, the PIR controller calculates the voltage compensation required for circulating current suppression. This voltage compensation is then superimposed on the arm voltage modulation signal to achieve coordinated suppression of the circulating current DC offset component and the second harmonic component.

[0071] During system operation, the operating parameters of the six bridge arms are monitored in real time and the interference operating parameter sequence is continuously updated. New interference operating parameter sets are periodically extracted according to a moving window and input into the interference feature prediction agent. Based on the latest predicted interference feature set, the combined stability is recalculated and the PIR control parameters are iteratively optimized. Once the five new optimal PIR control parameters are output, the updated parameters are immediately sent to the PIR controllers of each bridge arm, achieving real-time updates of the control parameters. Through this cyclic mechanism, the PIR control parameters can continuously adjust to follow the dynamic changes in circulating current characteristics, ensuring that the system maintains optimal circulating current suppression throughout the entire operation process following a single-phase open-circuit fault.

[0072] In summary, the embodiments of this application have at least the following technical effects: This invention first achieves accurate perception and data acquisition of the asymmetric operating state after a fault by continuously monitoring the operating parameters of six bridge arms and filtering the interference operating parameter sequences of the remaining five bridge arms when a single-phase open-circuit fault occurs. This provides a data foundation for the accurate extraction of subsequent circulating current characteristics. Secondly, this invention utilizes a moving window to capture the interference operating parameter sequence and inputs it into a pre-trained interference feature prediction agent. This enables real-time prediction of the interference feature groups of each healthy phase bridge arm, including the circulating current DC offset and second harmonic components, overcoming the shortcomings of traditional methods in accurately analyzing the circulating current components after a fault. Thirdly, this invention adaptively optimizes the PIR control parameters by calculating and merging stability and configuring optimization strategies. This allows the proportional-integral-resonant controller to automatically match the optimal suppression parameters based on the dynamic changes in the circulating current characteristics after a fault, achieving coordinated suppression of the DC offset component and asymmetric harmonic components. This solves the problem of insufficient suppression capability of traditional proportional-integral or proportional-resonant controllers under asymmetric operating conditions. Finally, this invention performs circulating current suppression control according to the optimized PIR control parameters and updates them in real time, which reduces the amplitude of the bridge arm circulating current and the fluctuation of the capacitor voltage caused by a single-phase open circuit fault, suppresses the surge of harmonic subspace current, realizes the decoupling and coordination of voltage balance control and circulating current suppression control, and improves the operational stability and fault tolerance of the six-phase modular multilevel converter drive system under fault conditions.

[0073] Example 2, as Figure 2 As shown, based on the same inventive concept as the six-phase MMC circulating current suppression method based on PIR control provided in Embodiment 1, this embodiment of the invention also provides a six-phase MMC circulating current suppression system based on PIR control, comprising: The monitoring module 11 is used to continuously monitor six operating parameters of the six bridge arms in the six-phase motor. When any phase open circuit fault exists, it filters out the five interfering operating parameter sequences of the other five bridge arms. Feature prediction module 12 is used to extract five interference running parameter sequences according to the moving window and input them into the corresponding five interference feature prediction agents to obtain five first predicted interference feature groups. The parameter optimization module 13 is used to further analyze the five predicted interference feature groups, calculate the combined stability, and configure the optimization strategy to optimize the PIR control, thereby obtaining five optimal PIR control parameters. The circulating current suppression module 14 is used to perform circulating current suppression control according to five optimal PIR control parameters and update them in real time.

[0074] The monitoring module 11 is specifically used for: Specifically, six operating parameters of the six bridge arms in a six-phase motor are continuously monitored. When any phase open circuit fault occurs, five interfering operating parameter sequences of the other five bridge arms are obtained, including: During the operation of the six-phase MMC motor, six operating parameters of the six bridge arms are continuously monitored; When any operating parameter is interrupted, it is determined that there is a single-phase interruption fault. The operating parameters of the other five bridge arms are screened and continuously recorded to obtain five interference operating parameter sequences.

[0075] Specifically, the feature prediction module 12 is used for: Specifically, according to the moving window, the five interference operation parameter sequences are input into the corresponding five interference feature prediction agents to obtain five first predicted interference features, including: Among the six pre-configured interference feature prediction agents, select and call five interference feature prediction agents from the five normally operating bridge arms. According to the preset window length, extract five first interference operation parameter groups from five interference operation parameter sequences, input them into the corresponding five interference feature prediction intelligent bodies, and output five first predicted interference feature groups.

[0076] The configuration steps for the six interference feature prediction agents include: Based on the single-phase fault record data of the six-phase motor during its historical operation, six sets of sample interference operating parameters were collected, and the corresponding sample interference feature groups were obtained and labeled to obtain six sets of sample interference feature groups. Each sample interference feature group includes circulating current DC offset and circulating current second harmonic. Based on machine learning, a predictive agent for six perturbation features is constructed. The training data is a set of six sample interference parameters, and the supervision label is a set of six sample interference features. The six interference feature prediction agents are trained and supervised and adjusted respectively until the training converges and the configuration is completed.

[0077] Specifically, the parameter optimization module 13 is used for: Specifically, we continue to analyze the five predicted interference feature groups and calculate the merge stability, including: Continue to extract five second interference operation parameter groups from the five interference operation parameter sequences according to the preset window length, input them into the corresponding five interference feature prediction intelligent bodies, output five predicted interference feature groups, and obtain five second predicted interference feature groups. The merging stability is calculated based on five first predicted interference feature groups and five second predicted interference feature groups.

[0078] The calculation methods for merger stability include: The first merging stability is obtained by calculating the similarity between the sum of the five first predicted interference feature groups and the sum of the five second predicted interference feature groups.

[0079] Furthermore, an optimization strategy is configured to optimize PIR control, resulting in five optimal PIR control parameters, including: Calculate the first optimization granularity based on the first merging stability; Based on the PIR controller, five first PIR suppression parameters are generated for the five bridge arms that are operating normally. Based on the five first PIR suppression parameters, the five first predicted interference feature groups, and the five second predicted interference feature groups, the first suppression score is obtained through analysis and processing. According to the first optimization granularity, the five first PIR suppression parameters are iteratively optimized until convergence, and the five optimal PIR suppression parameters with the largest suppression scores are obtained.

[0080] Specifically, based on the first merging stability, the first optimization granularity is calculated, including: Obtain the preset optimization granularity; Based on the first merging stability, a granularity adjustment coefficient is configured, and the preset optimized granularity is adjusted and calculated to obtain the first optimized granularity. The magnitude of the merging stability and the magnitude of the granularity adjustment coefficient are negatively correlated.

[0081] Specifically, based on the five first PIR suppression parameters, five first predicted interference feature groups, and five second predicted interference feature groups, a first suppression score is obtained through analysis and processing, including: The PIR suppression prediction agent is invoked, wherein the PIR suppression prediction agent is constructed based on machine learning, using a set of sample interference feature groups and a set of sample PIR suppression parameters as training data, and using a set of labeled sample suppression scores as supervision labels for training. The five first PIR suppression parameters are combined with the five first predicted interference feature groups and the five second predicted interference feature groups, respectively, and input into the PIR suppression prediction agent. The mean of the output results is calculated to obtain the first suppression score.

[0082] The circulating current suppression module 14 is specifically used for: Circulation suppression control is performed based on five optimal PIR control parameters, and the parameters are updated in real time.

Claims

1. A six-phase MMC circulating current suppression method based on PIR control, characterized in that, The method includes: The six operating parameters of the six bridge arms in the six-phase motor are continuously monitored. When any phase open circuit fault exists, the five interfering operating parameter sequences of the other five bridge arms are obtained. According to the moving window, the five interference operation parameter sequences are extracted and respectively input into the five corresponding interference feature prediction agents to obtain five first predicted interference feature groups. Further analysis of the five predicted disturbance feature groups was conducted, the combined stability was calculated, and an optimization strategy was configured to optimize PIR control, resulting in five optimal PIR control parameters. Circulation suppression control is performed based on five optimal PIR control parameters, and the parameters are updated in real time.

2. The six-phase MMC circulating current suppression method based on PIR control according to claim 1, characterized in that, Six operating parameters of the six bridge arms in a six-phase motor are continuously monitored. When any phase open circuit fault occurs, five interfering operating parameter sequences of the other five bridge arms are obtained, including: During the operation of the six-phase MMC motor, six operating parameters of the six bridge arms are continuously monitored; When any operating parameter is interrupted, it is determined that there is a single-phase interruption fault. The operating parameters of the other five bridge arms are screened and continuously recorded to obtain five interference operating parameter sequences.

3. The six-phase MMC circulating current suppression method based on PIR control according to claim 1, characterized in that, According to the moving window, the five interference operation parameter sequences are input into the corresponding five interference feature prediction agents respectively, resulting in five first predicted interference features, including: Among the six pre-configured interference feature prediction agents, select and call five interference feature prediction agents from the five normally operating bridge arms. According to the preset window length, extract five first interference operation parameter groups from five interference operation parameter sequences, input them into the corresponding five interference feature prediction intelligent bodies, and output five first predicted interference feature groups.

4. The six-phase MMC circulating current suppression method based on PIR control according to claim 3, characterized in that, The configuration steps for the six perturbation feature prediction agents include: Based on the single-phase fault record data of the six-phase motor during its historical operation, six sets of sample interference operating parameters were collected, and the corresponding sample interference feature groups were obtained and labeled to obtain six sets of sample interference feature groups. Each sample interference feature group includes circulating current DC offset and circulating current second harmonic. Based on machine learning, a predictive agent for six perturbation features is constructed. The training data is a set of six sample interference parameters, and the supervision label is a set of six sample interference features. The six interference feature prediction agents are trained and supervised and adjusted respectively until the training converges and the configuration is completed.

5. The six-phase MMC circulating current suppression method based on PIR control according to claim 1, characterized in that, Continue analyzing the five predicted interference feature groups and calculate the merge stability, including: Continue to extract five second interference operation parameter groups from the five interference operation parameter sequences according to the preset window length, input them into the corresponding five interference feature prediction intelligent bodies, output five predicted interference feature groups, and obtain five second predicted interference feature groups. The merging stability is calculated based on five first predicted interference feature groups and five second predicted interference feature groups.

6. The six-phase MMC circulating current suppression method based on PIR control according to claim 5, characterized in that, The methods for calculating merge stability include: The first merging stability is obtained by calculating the similarity between the sum of the five first predicted interference feature groups and the sum of the five second predicted interference feature groups.

7. The six-phase MMC circulating current suppression method based on PIR control according to claim 6, characterized in that, The PIR control was optimized using a configuration optimization strategy, resulting in five optimal PIR control parameters, including: Calculate the first optimization granularity based on the first merging stability; Based on the PIR controller, five first PIR suppression parameters are generated for the five bridge arms that are operating normally. Based on the five first PIR suppression parameters, the five first predicted interference feature groups, and the five second predicted interference feature groups, the first suppression score is obtained through analysis and processing. According to the first optimization granularity, the five first PIR suppression parameters are iteratively optimized until convergence, and the five optimal PIR suppression parameters with the largest suppression scores are obtained.

8. The six-phase MMC circulating current suppression method based on PIR control according to claim 7, characterized in that, Based on the first merging stability, calculate the first optimization granularity of the configuration, including: Obtain the preset optimization granularity; Based on the first merging stability, a granularity adjustment coefficient is configured, and the preset optimized granularity is adjusted and calculated to obtain the first optimized granularity. The magnitude of the merging stability and the magnitude of the granularity adjustment coefficient are negatively correlated.

9. The six-phase MMC circulating current suppression method based on PIR control according to claim 7, characterized in that, Based on the five first PIR suppression parameters, five first predicted interference feature groups, and five second predicted interference feature groups, the first suppression score is obtained through analysis and processing, including: The PIR suppression prediction agent is invoked, wherein the PIR suppression prediction agent is constructed based on machine learning, using a set of sample interference feature groups and a set of sample PIR suppression parameters as training data, and using a set of labeled sample suppression scores as supervision labels for training. The five first PIR suppression parameters are combined with the five first predicted interference feature groups and the five second predicted interference feature groups, respectively, and input into the PIR suppression prediction agent. The mean of the output results is calculated to obtain the first suppression score.

10. A six-phase MMC circulating current suppression system based on PIR control, characterized in that, The method for implementing the six-phase MMC circulating current suppression method based on PIR control as described in any one of claims 1-9 includes: The monitoring module is used to continuously monitor six operating parameters of the six bridge arms in the six-phase motor. When any phase open circuit fault exists, it filters out the five interfering operating parameter sequences of the other five bridge arms. The feature prediction module is used to extract the five interference running parameter sequences according to the moving window and input them into the corresponding five interference feature prediction agents to obtain five first predicted interference feature groups. The parameter optimization module is used to further analyze the five predicted interference feature groups, calculate the combined stability, and configure the optimization strategy to optimize the PIR control, thereby obtaining five optimal PIR control parameters. The circulating current suppression module is used to perform circulating current suppression control according to five optimal PIR control parameters and update them in real time.

Images