Converter station nonlinear cooperative control method and system optimized by big data

Abstract

The application discloses a large data optimized nonlinear cooperative control method and system of a converter station, and relates to the technical field of converter station control. The method comprises the following steps: obtaining real-time voltage and real-time current of a target converter station grid-connected point, and calculating real-time power grid impedance and impedance fluctuation rate; identifying the current power grid working condition type according to the impedance fluctuation rate; when the working condition is identified as a disturbance working condition, taking the real-time power grid impedance as input, and obtaining preliminary virtual inertia parameters and preliminary damping parameters through a double-branch cooperative regression network; based on the real-time power grid impedance and the impedance fluctuation rate, performing nonlinear correction on the preliminary virtual inertia parameters and the preliminary damping parameters to obtain target virtual inertia parameters and target damping parameters, generating control output, performing amplitude limiting processing through a saturation function to obtain nonlinear cooperative control instructions, and driving the converter station to perform flexible synchronous output. The application realizes real-time adaptive optimization of virtual inertia and damping parameters under weak power grid working conditions, and suppresses transient overshoot, oscillation and harmonics.

Description

Technical Field

[0001] This invention relates to the field of converter station control technology, specifically to a nonlinear collaborative control method and system for converter stations optimized by big data. Background Technology

[0002] Under the condition of large-scale time-varying equivalent impedance in weak power grids, the modular multilevel converter, as the core equipment for grid connection of new energy, directly affects the frequency support capability and transient stability of the system due to the adaptability of virtual inertia and damping parameters in its virtual synchronous machine control.

[0003] Traditional methods typically employ offline-tuned fixed parameters or impedance-segmented lookup table switching, making it difficult to achieve real-time adaptive optimization of time-varying impedance. Fixed parameters cannot be dynamically adjusted with real-time impedance changes, easily leading to insufficient frequency support, transient overshoot, or oscillations; segmented lookup table switching relies on manually preset thresholds, resulting in a limited number of parameter sets and potential control jumps at the boundaries. Furthermore, existing methods lack the ability to automatically identify weak grid conditions and power surges, easily causing harmonic degradation and reduced robustness during severe impedance fluctuations or power surges. Therefore, a nonlinear collaborative control method for converter stations that can sense impedance changes in real time, automatically identify operating condition types, and adaptively optimize control parameters is urgently needed. Summary of the Invention

[0004] This invention addresses the technical problems in existing technologies, such as the inability to dynamically adjust virtual inertia and damping parameters in real time according to grid impedance, the lack of automatic identification capability for operating conditions, the tendency for segmented table lookup switching to cause control jumps, and the limited coverage of operating conditions. It provides a big data-optimized nonlinear collaborative control method and system for converter stations.

[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 big data-optimized nonlinear collaborative control method for converter stations, comprising: The real-time voltage and real-time current of the target converter station's grid connection point are obtained, the real-time grid impedance is calculated, and the impedance fluctuation rate of the real-time grid impedance relative to the historical average impedance is calculated. The current power grid operating condition type is identified based on the impedance fluctuation rate; When a disturbance condition is identified, the real-time grid impedance is used as input, and forward calculation is performed through a pre-trained dual-branch cooperative regression network to output preliminary virtual inertia parameters and preliminary damping parameters. Based on the real-time grid impedance and impedance fluctuation rate, nonlinear corrections are performed on the preliminary virtual inertia parameters and the preliminary damping parameters to obtain the target virtual inertia parameters and the target damping parameters. A control output is generated based on the target virtual inertia parameter and the target damping parameter, and the control output is limited by a saturation function to obtain a nonlinear cooperative control command, which drives the target converter station to perform flexible synchronous output of the grid connection point voltage.

[0006] Secondly, this invention provides a big data-optimized nonlinear collaborative control system for converter stations, comprising: The real-time data acquisition module is used to acquire the real-time voltage and real-time current at the grid connection point of the target converter station, calculate the real-time grid impedance, and calculate the impedance fluctuation rate of the real-time grid impedance relative to the historical average impedance. Operating condition identification module, used to identify the current power grid operating condition type based on the impedance fluctuation rate; The dual-branch cooperative regression calculation module is used to perform forward calculations through a pre-trained dual-branch cooperative regression network when a disturbance condition is identified, taking the real-time grid impedance as input, and outputting preliminary virtual inertia parameters and preliminary damping parameters. The nonlinear correction module is used to perform nonlinear correction on the preliminary virtual inertia parameters and the preliminary damping parameters based on the real-time grid impedance and impedance fluctuation rate, so as to obtain the target virtual inertia parameters and the target damping parameters. The control output and limiting module is used to generate a control output based on the target virtual inertia parameter and the target damping parameter, and to limit the control output through a saturation function to obtain a nonlinear cooperative control command, which drives the target converter station to perform flexible synchronous output of the grid connection point voltage.

[0007] The beneficial effects of this invention are: Compared to existing technologies, this invention first achieves dynamic sensing and quantification of time-varying impedance in weak power grids by real-time calculation of grid impedance and impedance fluctuation rate. Secondly, it automatically identifies steady-state, weak power grid, and power surge conditions based on impedance fluctuation rate, solving the problem of traditional methods lacking the ability to distinguish operating conditions. Thirdly, under disturbance conditions, using real-time impedance as input, preliminary parameters are output through a dual-branch cooperative regression network, and nonlinear correction is performed using impedance fluctuation rate, enabling the virtual inertia and damping parameters to adaptively match with impedance changes, avoiding control jumps caused by segmented table lookups. Finally, nonlinear cooperative control commands are generated through saturation function amplitude limiting processing to drive the converter station's flexible synchronous output. This invention suppresses transient overshoot, oscillations, and harmonics, improving frequency support capability and system robustness under weak power grid conditions. Attached Figure Description

[0008] Figure 1 A flowchart illustrating the big data-optimized nonlinear collaborative control method for converter stations provided by this invention; Figure 2A schematic diagram of the structure of the big data-optimized nonlinear collaborative control system for converter stations provided by the present invention.

[0009] In the attached diagram, the components represented by each number are as follows: Real-time data acquisition module 11, working condition identification module 12, dual-branch collaborative regression calculation module 13, nonlinear correction module 14, control output and limiting module 15. Detailed Implementation

[0010] Example 1, as Figure 1 As shown, this embodiment of the invention provides a big data-optimized nonlinear cooperative control method for converter stations, including: S10: Obtain the real-time voltage and real-time current at the grid connection point of the target converter station, calculate the real-time grid impedance, and calculate the impedance fluctuation rate of the real-time grid impedance relative to the historical average impedance. First, real-time voltage and current at the grid connection point of the target converter station are acquired. A converter station is a crucial facility in a power system used to convert AC to DC power, with its core equipment being a modular multilevel converter. In renewable energy grid-connected scenarios, such as offshore wind or solar power plants, the power generation system typically first converts AC to DC through rectification for long-distance transmission. Upon reaching the receiving-end converter station, the DC is then converted back to AC by an inverter and fed into the AC grid. Under weak grid conditions, the grid's equivalent impedance exhibits nonlinear, fast-changing, and wide-range fluctuation characteristics, making traditional fixed-parameter virtual synchronous machine control and sliding mode control ill-suited to real-time impedance changes. Therefore, it is necessary to collect voltage and current data in real-time at the grid connection point to provide a foundation for subsequent impedance calculation and operating condition identification.

[0011] The real-time grid impedance is calculated by collecting real-time voltage and current data. Grid impedance reflects the electrical distance and connection strength between the converter station's grid connection point and the AC system. Under weak grid conditions, the impedance value is large and fluctuates drastically; under strong grid conditions, the impedance value is small and relatively stable. Calculating the real-time grid impedance allows for a quantitative reflection of the changing trend of the electrical distance between the converter station's grid connection point and the AC system, providing crucial state input for subsequent operating condition identification and adaptive adjustment of virtual inertia and damping parameters.

[0012] Furthermore, the impedance fluctuation rate of the real-time grid impedance relative to the historical average impedance is calculated. The historical average impedance is the weighted arithmetic mean of the historical grid impedance sequences recorded within a past sliding time window, representing the recent overall average level of the grid impedance. The impedance fluctuation rate reflects the degree of deviation of the current grid impedance from the historical average level. A larger impedance fluctuation rate indicates more drastic changes in grid impedance, and the more likely the system is under weak grid or power surge conditions. This impedance fluctuation rate is a key input parameter for subsequent operating condition identification, used to distinguish between steady-state and disturbance conditions, and further distinguish between weak grid and power surge conditions within the disturbance condition.

[0013] Specifically, by using real-time impedance sensing and impedance fluctuation rate calculation, this step can dynamically capture the changing trend of time-varying impedance in weak power grids, providing a quantitative basis for the adaptive adjustment of virtual inertia and damping parameters.

[0014] Specifically, the real-time voltage and current at the target converter station's grid connection point are obtained, the real-time grid impedance is calculated, and the impedance fluctuation rate of the real-time grid impedance relative to the historical average impedance is calculated, including: The three-phase real-time voltage waveform and three-phase real-time current waveform of the target converter station grid connection point are collected at a fixed sampling frequency, and the real-time voltage effective value and real-time current effective value of each phase are calculated respectively. The ratio of the real-time effective value of voltage to the real-time effective value of current in the same phase is taken as the phase impedance of the corresponding phase, and the average value of the three phase impedances is taken as the real-time grid impedance. Extract the historical power grid impedance sequence recorded within a past sliding time window, assign decreasing weight coefficients according to the time from recent to distant, and perform a weighted arithmetic mean on the historical power grid impedance sequence to obtain the historical average impedance; The impedance fluctuation rate is calculated based on the real-time grid impedance and the historical average impedance.

[0015] First, the three-phase real-time voltage and current waveforms at the grid connection point of the target converter station are acquired at a fixed sampling frequency, and the real-time RMS voltage and current values ​​for each phase are calculated. The fixed sampling frequency is determined based on the Nyquist theorem, the converter switching frequency, and the control bandwidth. For example, it can be set to 10 kHz to ensure both harmonic acquisition accuracy and control real-time requirements.

[0016] The three-phase real-time voltage waveform refers to the instantaneous sequence of voltage values ​​for phases A, B, and C at the grid connection point of the target converter station over time. Three voltage values ​​are recorded at each sampling moment to characterize the instantaneous amplitude and phase information of the voltage at the grid connection point. Similarly, the three-phase real-time current waveform refers to the instantaneous sequence of current values ​​for phases A, B, and C at the grid connection point of the target converter station over time. Three current values ​​are recorded at each sampling moment to characterize the instantaneous amplitude and phase information of the current at the grid connection point. By synchronously acquiring the three-phase voltage and current waveforms, complete electrical quantity information of the converter station's grid connection point can be obtained.

[0017] Then, the real-time RMS values ​​of voltage and current are calculated. Specifically, the RMS voltage value is calculated as follows: the instantaneous values ​​of a phase voltage collected within a complete cycle are squared and summed. The sum is then divided by the number of sampling points within that cycle, and the square root of the quotient is taken. The RMS current value is calculated in the same way: the instantaneous values ​​of a phase current collected within a complete cycle are squared and summed. The sum is then divided by the number of sampling points within that cycle, and the square root of the quotient is taken.

[0018] Furthermore, the ratio of the real-time effective voltage value to the real-time effective current value of the same phase is taken as the phase impedance of the corresponding phase, and the average value of the three phase impedances is taken as the real-time grid impedance. Specifically, this step of taking the average value of the three-phase phase impedances can offset single-phase measurement errors and three-phase imbalance disturbances, and obtain a more overall representative real-time grid impedance.

[0019] Furthermore, historical grid impedance sequences recorded within a past sliding time window are extracted. These sequences are then weighted with decreasing weights from most recent to oldest data points, and a weighted arithmetic mean is calculated to obtain the historical average impedance. The sliding time window refers to a fixed-length continuous time interval tracing back from the current moment, used to define the data range for calculating the historical average impedance. The length of the sliding time window is determined based on the time-varying characteristics of the grid impedance; for example, it can be set from 10 seconds to 5 minutes to cover a sufficient number of power frequency cycles and filter out instantaneous disturbances, while simultaneously tracking the slow changing trend of the impedance.

[0020] Optionally, the weighting coefficients can be implemented by decreasing the weighting coefficients from most recent to oldest as follows: Assume the sliding time window contains N historical grid impedance values, arranged in chronological order from oldest to newest as Z1, Z2, ..., Z... N Z N The impedance value at the most recent moment. Assign a weighting factor w to each impedance value. i The weighting coefficients satisfy the condition from Z1 to Z. N The weights follow an increasing pattern, and the sum of all weight coefficients is 1. For example, an exponential decay method can be used to set the weight coefficients, letting w... iThis is equal to e raised to the power of α, multiplied by (iN), and then divided by the normalization factor, where α is the decay rate constant; the larger α is, the greater the recent weight. In a simplified approach, the weight coefficients can be set as an arithmetic sequence, for example, w... i This equals 2i divided by N multiplied by (N+1), making the weight coefficients change from Z1 to Z. N The impedance increases linearly. Finally, each impedance value is multiplied by its corresponding weighting coefficient and summed to obtain the weighted arithmetic mean, i.e., the historical average impedance. Through this weighting method, the historical average impedance can more sensitively reflect recent trends in grid impedance while smoothing out long-term historical interference.

[0021] Furthermore, impedance fluctuation rate is calculated based on real-time grid impedance and historical average impedance. The formula for calculating impedance fluctuation rate is: impedance fluctuation rate equals the absolute value of the difference between real-time grid impedance and historical average impedance, divided by the historical average impedance. Impedance fluctuation rate reflects the degree of deviation of the current grid impedance from the historical average level. A larger impedance fluctuation rate indicates more drastic changes in grid impedance, and the more likely the system is under weak grid or power surge conditions. This impedance fluctuation rate is a key input parameter for subsequent operating condition identification, used to distinguish between steady-state and disturbance conditions.

[0022] S20: Identify the current power grid operating condition type based on the impedance fluctuation rate; Furthermore, the current power grid operating condition type is identified based on the calculated impedance fluctuation rate. Power grid operating condition types include steady-state operating conditions and disturbance operating conditions, with disturbance operating conditions including at least weak grid operating conditions and power surge operating conditions. Impedance fluctuation rate reflects the degree of deviation of the current power grid impedance from the historical average level and is a key quantitative indicator for distinguishing different operating conditions. Under steady-state operating conditions, the power grid impedance is relatively stable, and the impedance fluctuation rate is small; when the power grid impedance changes significantly, the impedance fluctuation rate increases, indicating that the system has entered a disturbance operating condition.

[0023] Specifically, identifying the current power grid operating condition type based on the impedance fluctuation rate includes: The power grid operating condition types include steady-state operating conditions and disturbance operating conditions, wherein the disturbance operating conditions include at least weak power grid operating conditions and power surge operating conditions; Multiple historical impedance fluctuation rates were continuously calculated during the steady-state operation of the target converter station, and the arithmetic mean and standard deviation of the multiple historical impedance fluctuation rates were calculated. The sum of the arithmetic mean and three times the standard deviation is configured as a first threshold, and the sum of the arithmetic mean and one standard deviation is configured as a second threshold; When the impedance fluctuation rate calculated in real time is greater than the first threshold, it is identified as a weak power grid condition. When the impedance fluctuation rate obtained by real-time calculation is greater than the second threshold and less than or equal to the first threshold, it is identified as a power sudden change condition. When the impedance fluctuation rate obtained from real-time calculation is less than or equal to the second threshold, it is identified as a steady-state operating condition.

[0024] Power grid operating conditions include steady-state conditions and disturbance conditions. Disturbance conditions include at least weak grid conditions and power surge conditions. Specifically, steady-state conditions correspond to normal operation with relatively stable grid impedance; weak grid conditions correspond to a state where grid impedance is large and fluctuates violently, and the system connection strength is weak; power surge conditions correspond to a state where grid impedance fluctuates to a moderate degree due to sudden changes in load or power generation.

[0025] First, multiple historical impedance fluctuation rates were continuously calculated during the steady-state operation of the target converter station, and the arithmetic mean and standard deviation of the historical impedance fluctuation rates were calculated. The steady-state operation period refers to the time interval during which the system is in a stable operating state. The impedance fluctuation rate data collected during this period represents the statistical characteristics within the normal fluctuation range of the system. The arithmetic mean reflects the central trend of the impedance fluctuation rate, and the standard deviation reflects the dispersion of the impedance fluctuation rate.

[0026] The steps for determining the steady-state operating segment include: Calculate the absolute value of the deviation between each historical grid impedance and the historical average impedance from the historical grid impedance sequence extracted within the sliding time window; Calculate the standard deviation of the historical power grid impedance sequence; Starting from the current moment, trace back and record each control cycle that continuously satisfies the condition that the absolute value of the deviation is less than the standard deviation, and determine the time interval covered by each control cycle as the steady-state operating period.

[0027] First, from the historical grid impedance sequence extracted within the sliding time window, the absolute value of the deviation between each historical grid impedance and the historical average impedance is calculated. The historical grid impedance sequence records the grid impedance values ​​corresponding to each control cycle within the sliding time window, and the historical average impedance is the weighted arithmetic mean of this sequence. The absolute value of the deviation is equal to the absolute value of the difference between the historical grid impedance and the historical average impedance, and is used to quantify the degree of deviation of the impedance value at each sampling time from the average level.

[0028] Secondly, calculate the standard deviation of the historical power grid impedance sequence. The standard deviation reflects the degree of dispersion of each impedance value in the historical power grid impedance sequence relative to the average value. The larger the standard deviation, the more severe the impedance fluctuation; the smaller the standard deviation, the more concentrated the impedance.

[0029] Furthermore, starting from the current moment, the system traces backwards, recording each control cycle that consecutively satisfies the condition that the absolute value of the deviation is less than the standard deviation. The time interval covered by each control cycle is defined as the steady-state operating period. Specifically, starting from the control cycle corresponding to the current moment, the system checks backwards one by one whether the absolute value of the deviation in each control cycle is less than the standard deviation. When the first control cycle that does not meet the condition is encountered, the tracing stops, and the time interval covered by all consecutive control cycles that meet the condition is defined as the steady-state operating period. During this period, the grid impedance fluctuation amplitude is small, and the system is in a relatively stable operating state. Therefore, the impedance fluctuation rate data collected during this period is representative and can be used to calculate the arithmetic mean and standard deviation of historical impedance fluctuation rates, providing a statistical basis for configuring subsequent operating condition identification thresholds.

[0030] Using the above method, the determination of the steady-state operating period does not require manually setting a fixed time length, but is adaptively determined based on the fluctuation characteristics of actual data, which improves the accuracy and adaptability of operating condition identification.

[0031] Furthermore, the sum of the arithmetic mean of the calculated historical impedance volatility and three times the standard deviation is set as the first threshold, and the sum of the arithmetic mean and one standard deviation is set as the second threshold.

[0032] The first threshold is configured based on the three-standard-deviation principle in statistics: under a normal distribution, approximately 99.7% of normal data fall within the range of the mean plus or minus three standard deviations. Therefore, setting the sum of the mean and three standard deviations as the first threshold can effectively distinguish extreme abnormal operating conditions, i.e., weak power grid operating conditions, ensuring an extremely low false alarm rate. The second threshold is configured based on the one-standard-deviation principle: approximately 68% of the data falls within the range of the mean plus or minus one standard deviation. Exceeding this range indicates a significant deviation in the data, which can be used to identify moderate disturbances such as power surges.

[0033] Specifically, when the impedance fluctuation rate calculated in real time is greater than the first threshold, it indicates that the current impedance fluctuation level is far beyond the normal range, and the system is in an extreme abnormal state, identified as a weak power grid condition. When the impedance fluctuation rate calculated in real time is greater than the second threshold but less than or equal to the first threshold, it indicates that the current impedance fluctuation level significantly deviates from the normal level but has not reached an extreme level, identified as a power surge condition. When the impedance fluctuation rate calculated in real time is less than or equal to the second threshold, it indicates that the current impedance fluctuation level is within the normal range, identified as a steady-state condition.

[0034] In summary, the operating condition identification method based on statistical thresholds described above can automatically and accurately distinguish between steady-state operating conditions, weak grid operating conditions, and power surge operating conditions, providing differentiated control strategies for different operating conditions. For example, under steady-state operating conditions, existing control parameters can be maintained to reduce computational overhead; under disturbance operating conditions, a dual-branch cooperative regression network is triggered to perform real-time optimization calculations of virtual inertia and damping parameters, achieving adaptive matching of control parameters to time-varying impedance. The starting point of this step is to solve the problem of traditional methods lacking automatic operating condition identification capabilities, ensuring that subsequent parameter optimization is performed only for disturbance operating conditions, avoiding unnecessary computational overhead under steady-state operating conditions.

[0035] S30: When a disturbance condition is identified, the real-time grid impedance is used as input, and forward calculation is performed through a pre-trained dual-branch cooperative regression network to output preliminary virtual inertia parameters and preliminary damping parameters. Furthermore, when the current grid operating condition is identified as a disturbance condition, i.e., a weak grid condition or a power surge condition, based on the impedance fluctuation rate, it is necessary to optimize the virtual inertia parameters and damping parameters in real time to adapt to the impact of time-varying grid impedance on the converter station control performance. The virtual inertia parameter is a control coefficient simulating the rotational inertia of a synchronous generator, used to characterize the converter station's ability to provide inertial support when the frequency changes; a larger value results in a smaller frequency change rate but a slower response speed. The damping parameter is a control coefficient simulating the damping winding of a synchronous generator, used to suppress power oscillations; a larger value results in faster oscillation decay but may introduce overshoot. At this point, using the calculated real-time grid impedance as input, a pre-trained bi-branch cooperative regression network is used for forward calculation, outputting preliminary virtual inertia parameters and preliminary damping parameters.

[0036] The dual-branch cooperative regression network is a deep learning-based big data optimization model. Its network architecture includes a shared feature extraction layer, a virtual inertia output branch, and a damping parameter output branch. The input of the shared feature extraction layer receives real-time grid impedance and extracts deep features from the impedance data through a multi-layer fully connected network. The output of the shared feature extraction layer is connected to the inputs of the virtual inertia output branch and the damping parameter output branch, respectively, with each branch performing regression prediction independently. The output of the virtual inertia output branch outputs the predicted values ​​of the virtual inertia parameters, and the output of the damping parameter output branch outputs the predicted values ​​of the damping parameters. This dual-branch cooperative regression network is pre-trained using a large amount of historical data. The training samples include historical grid impedance sequences under different weak grid conditions and different power surge conditions, as well as the labels for the optimal virtual inertia parameter and optimal damping parameter at each time step.

[0037] Specifically, this pre-trained bi-branch collaborative regression network can quickly and accurately output the appropriate preliminary virtual inertia parameters and preliminary damping parameters based on the real-time grid impedance under disturbance conditions, avoiding the limitations of offline tuning of fixed parameters or segmented table lookup switching in traditional methods.

[0038] The pre-training steps of the dual-branch collaborative regression network include: Historical grid impedance sequences under different weak grid conditions and different power surge conditions are collected. The real-time grid impedance at each moment is used as the input state, and the corresponding optimal virtual inertia parameter and optimal damping parameter are used as the output label to construct a sample state-label pair set. A network architecture for a dual-branch collaborative regression network is constructed, wherein the network architecture includes a shared feature extraction layer, a virtual inertia output branch, and a damping parameter output branch; The input terminal of the shared feature extraction layer is used to receive the real-time grid impedance. The output terminal of the shared feature extraction layer is connected to the input terminals of the virtual inertia output branch and the damping parameter output branch, respectively. The output terminal of the virtual inertia output branch is used to output the predicted value of the virtual inertia parameter, and the output terminal of the damping parameter output branch is used to output the predicted value of the damping parameter. The sample state-label pair set is used as training data, and the sum of the mean square error between the predicted value and the label value of the virtual inertia output branch and the mean square error between the predicted value and the label value of the damping parameter output branch is used as the loss function. The network weights of the network architecture are jointly updated using the backpropagation algorithm until the loss function converges. The trained network architecture is then used as the dual-branch collaborative regression network.

[0039] First, historical grid impedance sequences under different weak grid conditions and different power surge conditions are collected. The real-time grid impedance at each moment is used as the input state, and the corresponding optimal virtual inertia parameter and optimal damping parameter are used as output labels to construct a sample state-label pair set. This sample data comes from various disturbance conditions recorded by the converter station in actual operation or simulation, covering different scenarios from weak grids to power surges, to ensure that the model can learn the complex mapping relationship between grid impedance and control parameters.

[0040] Among them, the optimal virtual inertia parameter and the optimal damping parameter are the ideal combination of control parameters that can minimize the voltage overshoot, the number of oscillations and the frequency deviation at the converter station connection point under specific grid impedance conditions.

[0041] Specifically, the steps for obtaining the optimal virtual inertia parameter and the optimal damping parameter include: For each historical grid impedance sequence, extract the virtual inertia parameters, damping parameters, grid connection point voltage overshoot, oscillation number and frequency deviation recorded in multiple time windows during the actual operation of the converter station in the corresponding period; The overshoot, oscillation count, and frequency deviation of the grid connection point voltage extracted from all time windows are normalized respectively, and the weighted sum of the normalized overshoot, oscillation count, and frequency deviation under each time window is calculated. The virtual inertia parameters and damping parameters corresponding to the time window with the smallest weighted sum are determined as the optimal virtual inertia parameters and optimal damping parameters corresponding to the current historical power grid impedance sequence.

[0042] First, for each historical grid impedance sequence, virtual inertia parameters, damping parameters, grid-connected voltage overshoot, oscillation frequency, and frequency deviation are extracted from multiple time windows recorded during the actual operation of the converter station within the corresponding time period. Specifically, data within each window is extracted from the converter station's historical database in a sliding manner according to fixed time windows, for example, extracting a time window every 10 seconds. Each window records the virtual inertia and damping parameters actually used during that time period, as well as the grid-connected voltage overshoot, oscillation frequency, and frequency deviation generated by the system under those parameters. Among them, the voltage overshoot reflects the maximum magnitude of the voltage deviation from the steady-state value during the transient process, the oscillation frequency reflects the speed at which the system recovers stability after a disturbance, and the frequency deviation reflects the converter station's ability to support the grid frequency. Through sliding window extraction, multiple sets of system performance indicators under different combinations of control parameters can be obtained.

[0043] Secondly, the overshoot, oscillation frequency, and frequency deviation of the grid-connected point voltage extracted from all time windows are normalized, and the weighted sum of the normalized overshoot, oscillation frequency, and frequency deviation for each time window is calculated. Optionally, the normalization process uses the minimum-maximum normalization method to map each performance index to the interval between 0 and 1, eliminating the differences in dimensions and numerical ranges between different indices.

[0044] Specifically, for each performance indicator, the minimum and maximum values ​​are calculated across all time windows. Then, the indicator value for each window is normalized by dividing the original value by the minimum value using the formula (original value minus minimum value). After normalization, the values ​​for all three indicators range from 0 to 1. Subsequently, weighting coefficients are assigned based on the importance the business places on each performance aspect. For example, voltage overshoot is weighted at 0.4, oscillation frequency at 0.3, and frequency deviation at 0.3. The normalized values ​​of the three indicators are then multiplied by their respective weights and summed to obtain the overall performance score for that time window.

[0045] Finally, the virtual inertia and damping parameters corresponding to the time window with the smallest weighted sum are determined as the optimal virtual inertia and damping parameters for the current historical grid impedance sequence. A smaller weighted sum indicates better overall performance in terms of voltage overshoot, oscillation frequency, and frequency deviation under the control parameters corresponding to that window. Therefore, selecting the window with the smallest overall performance score among multiple time windows yields the optimal combination of virtual inertia and damping parameters under the current grid impedance conditions.

[0046] Using the above method, each historical grid impedance sequence can be labeled with its corresponding optimal control parameter, constructing a set of sample state-label pairs for supervised training of the two-branch cooperative regression network. These sample state-label pairs enable the two-branch cooperative regression network to learn the end-to-end mapping relationship from grid impedance to optimal control parameters.

[0047] Furthermore, a dual-branch cooperative regression network architecture is constructed, comprising a shared feature extraction layer, a virtual inertia output branch, and a damping parameter output branch. The input of the shared feature extraction layer is used to receive the real-time grid impedance, and the output of the shared feature extraction layer is connected to the inputs of the virtual inertia output branch and the damping parameter output branch, respectively. The output of the virtual inertia output branch is used to output the predicted value of the virtual inertia parameter, and the output of the damping parameter output branch is used to output the predicted value of the damping parameter.

[0048] For example, the shared feature extraction layer employs a three-layer fully connected network structure, with 64, 128, and 64 neurons in each layer, followed by a batch normalization layer and a ReLU activation function, to extract high-dimensional feature representations from the input real-time grid impedance. The virtual inertia output branch uses a two-layer fully connected network structure, with the first layer containing 32 neurons and the second layer containing 1 neuron, outputting predicted virtual inertia parameters. The damping parameter output branch uses the same two-layer fully connected network structure, with the first layer containing 32 neurons and the second layer containing 1 neuron, outputting predicted damping parameters. Both output branches use linear activation functions, directly outputting continuous parameter values.

[0049] Furthermore, a set of sample state-label pairs is used as training data, and the sum of the mean squared errors of the predicted values ​​and label values ​​of the virtual inertia output branch and the damping parameter output branch is used as the loss function. Specifically, the loss function is defined as the sum of the mean squared errors of the two branches, that is, the loss equals the square of the difference between the predicted value and label value of the virtual inertia branch plus the square of the difference between the predicted value and label value of the damping branch. This loss function can simultaneously optimize the prediction accuracy of both branches, ensuring that both parameters can be accurately fitted.

[0050] The network weights of the network architecture are jointly updated using the backpropagation algorithm until the loss function converges. Optionally, during training, the Adam optimizer is used, with an initial learning rate of 0.001, a batch size of 32, and 200 training epochs. After each training epoch, the loss value is evaluated using a validation set. Training stops when the validation set loss no longer decreases for 10 consecutive epochs, and the optimal model weights are saved. The trained network architecture is then used as a two-branch collaborative regression network.

[0051] Through the pre-training described above, the dual-branch cooperative regression network can quickly output preliminary virtual inertia parameters and preliminary damping parameters adapted to the current operating conditions based on the input real-time grid impedance, providing good initial values ​​for subsequent nonlinear correction.

[0052] Finally, when a disturbance condition is identified, the real-time grid impedance obtained from real-time calculation is used as input, and forward calculation is performed through a pre-trained dual-branch cooperative regression network to output preliminary virtual inertia parameters and preliminary damping parameters. This provides basic control parameters based on big data learning for subsequent nonlinear correction, enabling the final control parameters to quickly approach the optimal value under the current impedance conditions.

[0053] S40: Based on the real-time grid impedance and impedance fluctuation rate, perform nonlinear correction on the preliminary virtual inertia parameter and the preliminary damping parameter to obtain the target virtual inertia parameter and the target damping parameter; Furthermore, after obtaining the preliminary virtual inertia parameters and preliminary damping parameters through the dual-branch cooperative regression network, it is necessary to perform nonlinear correction on the two parameters based on the real-time grid impedance and impedance fluctuation rate to obtain the final target virtual inertia parameters and target damping parameters.

[0054] Specifically, since the initial parameters output by the dual-branch cooperative regression network are based on benchmark values ​​learned from large datasets, but the network is trained using historical sample data, its predictions may have some bias for extreme impedance values ​​or rapid fluctuations occurring in the real-time power grid. Therefore, it is necessary to dynamically adjust the initial virtual inertia parameters and initial damping parameters through nonlinear correction to make them more accurately adapt to the current power grid state.

[0055] The basic principle of nonlinear correction is as follows: the greater the deviation of the real-time grid impedance from the historical average level, or the more severe the impedance fluctuation, the worse the current grid operating conditions. This necessitates strengthening the response of the control parameters, i.e., appropriately increasing the virtual inertia parameter to provide stronger frequency support, and simultaneously appropriately increasing the damping parameter to suppress oscillations. The correction coefficients are designed using nonlinear functions to reflect the nonlinear response characteristics of the control parameters to impedance changes.

[0056] Specifically, based on the real-time grid impedance and impedance fluctuation rate, nonlinear corrections are performed on the preliminary virtual inertia parameters and the preliminary damping parameters to obtain the target virtual inertia parameters and the target damping parameters, including: Calculate the ratio of the real-time grid impedance to the historical average impedance, and use the square root of the ratio as the first correction factor; Calculate the sum of 1 and the impedance fluctuation rate, and use it as the first intermediate value; The natural logarithm of the first intermediate value plus 1 is used as the second correction coefficient; Multiply the preliminary virtual inertia parameter by the first correction coefficient to obtain the target virtual inertia parameter; The target damping parameter is obtained by multiplying the preliminary damping parameter by the second correction coefficient.

[0057] First, the ratio of the real-time grid impedance to the historical average impedance is calculated, and the square root of this ratio is used as the first correction factor. The ratio reflects the degree of deviation of the current grid impedance from the recent average level; a larger ratio indicates a more severe weak grid. The square root of this ratio is used as the first correction factor, which increases with the impedance ratio, but the rate of increase gradually slows down to avoid excessively large virtual inertia parameters under extremely high impedance, which could lead to a slow system response. This first correction factor is used to dynamically adjust the virtual inertia parameter; the higher the impedance, the stronger the inertia, thereby improving the frequency support capability under weak grid conditions.

[0058] Secondly, the sum of 1 and the impedance volatility is calculated as the first intermediate value. The impedance volatility reflects the degree of fluctuation of the current grid impedance relative to the historical average level; the higher the volatility, the more the system is under disturbance. Adding 1 to the impedance volatility maps the range of impedance volatility from 0 to positive infinity to the range of 1 to positive infinity, serving as the input domain for the logarithmic function and making subsequent logarithmic operations mathematically valid.

[0059] Furthermore, the natural logarithm of the first intermediate value plus 1 is used as the second correction coefficient. Specifically, the second correction coefficient equals ln(1 + impedance volatility) + 1. This second correction coefficient grows logarithmically with increasing impedance volatility; that is, the correction coefficient grows faster when the volatility is low and grows more gradually when the volatility is high. This logarithmic growth characteristic ensures that the damping parameter grows gradually with impedance volatility, preventing excessive damping at high volatility from causing sluggish system response or conservative control. This second correction coefficient is used to dynamically adjust the damping parameter, appropriately increasing the damping with increasing volatility to suppress transient oscillations.

[0060] Finally, the initial virtual inertia parameter is multiplied by the first correction factor to obtain the target virtual inertia parameter. When the real-time grid impedance is greater than the historical average level, the first correction factor is greater than 1, and the target virtual inertia parameter is greater than the initial value, thus enhancing inertia support; when the real-time grid impedance is less than the historical average level, the first correction factor is less than 1, and the target virtual inertia parameter is less than the initial value, thus avoiding excessive inertia that could lead to a slow response.

[0061] Simultaneously, the initial damping parameter is multiplied by the second correction coefficient to obtain the target damping parameter. When the impedance fluctuation rate is large, the second correction coefficient is greater than 1, and the target damping parameter is greater than the initial value, thus enhancing damping to suppress oscillations; when the impedance fluctuation rate is small, the second correction coefficient is close to 1, and the target damping parameter is close to the initial value, maintaining the original damping characteristics.

[0062] Through the aforementioned nonlinear corrections, the target virtual inertia parameters and target damping parameters can be adaptively adjusted according to changes in real-time grid impedance and impedance fluctuation rate, providing stronger inertia support and damping capabilities under weak grid conditions, rapidly suppressing transient oscillations under power surge conditions, and avoiding over-correction under steady-state conditions that could lead to overly conservative control.

[0063] Specifically, the square root design of the first correction factor appropriately enhances the virtual inertia under weak grid conditions, improving frequency support capability; the logarithmic design of the second correction factor ensures that damping increases gradually with volatility, avoiding excessive damping under high volatility. These two correction factors act on different control parameters, achieving multi-dimensional adaptive matching of time-varying impedance.

[0064] S50: Generate a control output based on the target virtual inertia parameter and the target damping parameter, and perform amplitude limiting processing on the control output through a saturation function to obtain a nonlinear cooperative control command, driving the target converter station to execute the flexible synchronous output of the grid connection point voltage.

[0065] Finally, the target virtual inertia parameters and target damping parameters obtained from the preceding steps are used to generate the control output. This control output, based on the virtual synchronous machine control principle, simulates the electromechanical transient characteristics of a synchronous generator, converting active power deviation and reactive power deviation into the phase angle and amplitude of the modulation voltage, thereby generating a three-phase modulation voltage signal. Specifically, the control output process includes two stages: an active power loop and a reactive power loop. The active power loop uses the target virtual inertia parameters, target damping parameters, and active power deviation as inputs to solve the synchronous generator oscillation equation and obtain the modulation voltage phase angle. The reactive power loop uses the rated voltage amplitude and reactive power deviation as inputs to calculate the modulation voltage amplitude. The outputs of both stages are combined to form the three-phase modulation voltage signal, which serves as the control output.

[0066] Specifically, generating a control output based on the target virtual inertia parameters and the target damping parameters includes: The actual active power and reference active power of the target converter station are obtained, and the active power deviation is calculated. The reference active power is the active power target value issued by the power grid dispatch center to the target converter station. Obtain the actual reactive power and reference reactive power of the target converter station, and calculate the reactive power deviation, wherein the reference reactive power is the reactive power target value issued by the power grid dispatch center to the target converter station; Substitute the target virtual inertia parameter, the target damping parameter, and the active power deviation into the synchronous generator swing equation to calculate the angular frequency deviation, and then perform an integral operation on the angular frequency deviation to obtain the modulation voltage phase angle. Based on the nominal voltage of the target converter station's grid connection point, the rated voltage amplitude is obtained, and combined with the reactive power deviation, the modulation voltage amplitude is calculated. The modulation voltage amplitude is positively correlated with the rated voltage amplitude and negatively correlated with the reactive power deviation. Based on the modulation voltage phase angle and the modulation voltage amplitude, a three-phase modulation voltage signal is synthesized and used as a control output. The three-phase modulation voltage signal includes A-phase modulation voltage, B-phase modulation voltage and C-phase modulation voltage.

[0067] First, the actual active power and reference active power of the target converter station are obtained, and the active power deviation is calculated. The actual active power is the active power value collected in real time by a power measurement device installed at the converter station's grid connection point, while the reference active power is the active power target value issued by the power grid dispatch center to the target converter station. The active power deviation equals the reference active power minus the actual active power. When the actual active power is lower than the reference value, the deviation is positive, requiring an increase in power output; when the actual active power is higher than the reference value, the deviation is negative, requiring a decrease in power output.

[0068] Secondly, the actual reactive power and reference reactive power of the target converter station are obtained, and the reactive power deviation is calculated. The actual reactive power is the reactive power value collected in real time by a power measurement device, while the reference reactive power is the reactive power target value issued by the power grid dispatch center to the target converter station. The reactive power deviation equals the reference reactive power minus the actual reactive power. When the actual reactive power is lower than the reference value, the deviation is positive, requiring an increase in reactive power output; when the actual reactive power is higher than the reference value, the deviation is negative, requiring a decrease in reactive power output.

[0069] Furthermore, the target virtual inertia parameters, target damping parameters, and active power deviation are substituted into the synchronous generator swing equation to calculate the angular frequency deviation. The modulation voltage phase angle is then obtained by integrating the angular frequency deviation. Specifically, the synchronous generator swing equation is used to simulate the rotor motion characteristics of a traditional synchronous generator, enabling the converter station to exhibit inertial and damping behavior similar to that of a synchronous generator in its dynamic response. The purpose of this equation is to convert the active power deviation into an angular frequency deviation, and then obtain the modulation voltage phase angle through integration, thereby achieving closed-loop control of the active power. Its standard form of the discretized iterative calculation formula is as follows: Let the control period be The current time is The previous moment was Angular frequency deviation The calculation formula is: in, The target virtual inertia parameter, expressed in kilograms per square meter, characterizes the magnitude of the converter station's inertia in resisting frequency changes. The target damping parameter, expressed in Newton-meter-second per radian, characterizes the damping capability of the converter station to suppress power oscillations. The active power deviation at the current moment is equal to the reference active power minus the actual active power, and the unit is watts. This represents the angular frequency deviation at the previous moment, expressed in radians per second.

[0070] This iterative formula uses Euler's forward method to discretize and solve the oscillation equation. The physical meaning is: active power deviation. As driving torque, it overcomes the damping term. The effect, divided by virtual inertia The rate of change of the angular frequency deviation is then obtained and added to the angular frequency deviation of the previous moment to obtain the angular frequency deviation of the current moment.

[0071] Obtain angular frequency deviation Then, the phase angle of the modulation voltage is obtained through integration. : in, This represents the modulation voltage phase angle at the previous moment, in radians. This integral formula expresses the cumulative effect of angular frequency deviation, meaning the phase angle changes linearly with time; the larger the frequency deviation, the faster the phase angle changes. The final obtained modulation voltage phase angle... Used to synthesize three-phase modulated voltage signals to control the direction and magnitude of active power transmission. When the actual active power is lower than the reference value, When the value is positive, the phase angle leads, and the active power output of the converter station increases; when the actual active power is higher than the reference value, A negative value indicates a lagging phase angle, resulting in a reduction in the active power output of the converter station. Through the above discretized iterative calculation, the modulation voltage phase angle can be updated in real time within each control cycle, achieving rapid closed-loop regulation of active power.

[0072] Furthermore, based on the nominal voltage of the target converter station's grid connection point, the rated voltage amplitude is obtained, and combined with the reactive power deviation, the modulation voltage amplitude is calculated. The rated voltage amplitude is the peak phase voltage corresponding to the nominal voltage of the converter station's grid connection point, serving as a reference value for reactive power voltage regulation, and is measured in volts. The modulation voltage amplitude is the actual voltage amplitude used to synthesize the modulation wave after reactive power deviation regulation, also measured in volts. This value determines the magnitude of the converter station's output voltage, thereby controlling the exchange of reactive power.

[0073] Specifically, the formula for calculating the rated voltage amplitude is as follows: in, This refers to the nominal line voltage RMS value. The nominal line voltage RMS value is the rated voltage level of the converter station's grid connection point. For example, the nominal line voltage RMS value of a 10 kV power grid is 10,000 volts. The square root of 2 is used to convert the RMS value to the peak value, and the square root of 3 is used to convert the line voltage to the phase voltage. Taking a 10 kV power grid as an example, the rated voltage amplitude is equal to 10,000 multiplied by 1.414 divided by 1.732, which is approximately equal to 8165 volts.

[0074] The formula for calculating the amplitude of the modulation voltage is: Among them, reactive power deviation It equals the reference reactive power minus the actual reactive power, and the unit is var. The positive proportionality coefficient, measured in volts per var, represents the sensitivity of the modulation voltage amplitude to reactive power deviation. When the reactive power deviation is positive (i.e., insufficient reactive power), the modulation voltage amplitude is greater than the rated voltage amplitude, and the converter station increases reactive power output. When the reactive power deviation is negative (i.e., excess reactive power), the modulation voltage amplitude is less than the rated voltage amplitude, and the converter station reduces reactive power output. Through the above calculations, the modulation voltage amplitude adaptively adjusts with the reactive power deviation, achieving closed-loop control of reactive voltage.

[0075] Finally, based on the modulation voltage phase angle and modulation voltage amplitude, a three-phase modulation voltage signal is synthesized and used as the control output. The three-phase modulation voltage signal includes the A-phase modulation voltage, the B-phase modulation voltage, and the C-phase modulation voltage. Specifically, based on the modulation voltage phase angle θ and the modulation voltage amplitude E, the three-phase modulation voltage signal is synthesized using the following formulas: A-phase modulation voltage = E × sin(θ); B-phase modulation voltage = E × sin(θ - 120°); C-phase modulation voltage = E × sin(θ + 120°).

[0076] The generated three-phase sinusoidal waveform is used as the control output to drive the converter to generate the corresponding grid connection point voltage, thereby achieving precise control of active and reactive power. In summary, through the above control output generation process, the target virtual inertia parameters and target damping parameters are effectively converted into actual voltage modulation signals, realizing the closed-loop execution of the virtual synchronous machine control strategy.

[0077] Furthermore, the control output is limited using a saturation function to obtain a nonlinear coordinated control command. The saturation function is used to limit the modulated voltage signal within a safe range, preventing overmodulation that could lead to converter output distortion or equipment damage.

[0078] Specifically, the control output is limited by a saturation function to obtain a nonlinear coordinated control command, which drives the target converter station to perform flexible synchronous output of the grid connection point voltage, including: The control output is limited by a saturation function, wherein the saturation function is defined as follows: For the input one-phase modulation voltage value, the absolute value of the modulation voltage value is compared with the rated voltage amplitude; If the absolute value of the modulation voltage is less than or equal to the rated voltage amplitude, then the output of the saturation function is equal to the modulation voltage itself; If the absolute value of the modulation voltage is greater than the rated voltage amplitude, then the output of the saturation function is equal to the rated voltage amplitude multiplied by the positive or negative sign of the modulation voltage. The A-phase modulation voltage, B-phase modulation voltage, and C-phase modulation voltage in the three-phase modulation voltage signal are respectively input into the saturation function, and the amplitude limiting is calculated in sequence to obtain the corresponding three amplitude limiting voltage values. The three amplitude limiting voltage values ​​are used as nonlinear cooperative control commands. The nonlinear coordinated control command is output to the modulator of the target converter station, driving the target converter station to generate the grid connection point voltage according to the limiting voltage value, thereby completing the flexible synchronous output.

[0079] First, the control output is limited using a saturation function. The saturation function is a nonlinear mapping function used to limit the input signal within preset upper and lower boundaries, preventing the control output from exceeding the converter's linear modulation range. Specifically, the saturation function is defined as follows: For a single-phase modulation voltage value, the absolute value of the modulation voltage is first taken and compared with the rated voltage amplitude. The rated voltage amplitude is the maximum voltage amplitude that the converter can linearly output; exceeding this value will enter the overmodulation region, leading to increased output voltage distortion and harmonics.

[0080] If the absolute value of the modulation voltage is less than or equal to the rated voltage amplitude, it indicates that the current modulation voltage is within the linear modulation range of the converter. In this case, the output of the saturation function is equal to the modulation voltage value itself, and no amplitude limiting is performed.

[0081] Furthermore, if the absolute value of the modulation voltage is greater than the rated voltage amplitude, it indicates that the current modulation voltage has exceeded the linear modulation range of the converter and needs to be limited. In this case, the saturation function output equals the rated voltage amplitude multiplied by the sign of the modulation voltage value. Specifically, when the modulation voltage is positive, the output is the positive rated voltage amplitude; when the modulation voltage is negative, the output is the negative rated voltage amplitude. This limiting method can cut off excessively high modulation voltages to boundary values ​​while preserving the polarity information of the original signal.

[0082] Finally, the modulation voltages of phase A, phase B, and phase C in the three-phase modulation voltage signals are input into the saturation function, and amplitude limiting calculations are performed sequentially to obtain the corresponding three amplitude limiting voltage values. These three amplitude limiting voltage values ​​are then used as nonlinear coordinated control commands. Since the three-phase modulation voltage signals are generated independently and may each be in a different amplitude limiting state, they need to be processed phase by phase.

[0083] Finally, the nonlinear coordinated control command is output to the modulator of the target converter station, driving the target converter station to generate the grid connection point voltage according to the limited voltage value, thus completing the flexible synchronous output. Specifically, after receiving the three-phase limited voltage value, the modulator generates the corresponding switching signal through pulse width modulation technology, controlling the on and off of the power devices inside the converter, so that the grid connection point voltage is output according to the limited modulated waveform.

[0084] In summary, by limiting the saturation function, the problem of modulation voltage exceeding the limit caused by excessive correction of virtual inertia or damping parameters can be effectively prevented, avoiding the converter from entering the over-modulation state, thereby ensuring the sinusoidal nature of the output voltage and power quality, and realizing flexible synchronous output of the grid connection point voltage.

[0085] In summary, the embodiments of this application have at least the following technical effects: This invention first achieves real-time sensing and quantitative characterization of time-varying impedance in weak grids by acquiring real-time voltage and current at the grid connection point, calculating real-time grid impedance, and then calculating impedance fluctuation rate. Second, it automatically identifies steady-state, weak grid, and power surge conditions based on impedance fluctuation rate, providing differentiated control strategies for different conditions. Third, under disturbance conditions, using real-time grid impedance as input, it outputs preliminary virtual inertia and damping parameters through a pre-trained bi-branch cooperative regression network, and performs nonlinear correction based on impedance fluctuation rate, achieving adaptive matching of control parameters to time-varying impedance. Finally, it generates control output based on target parameters and performs amplitude limiting processing using a saturation function to obtain nonlinear cooperative control commands to drive the converter station to execute flexible synchronous output.

[0086] This invention solves the technical problems in the prior art, such as the inability to dynamically adjust parameters with impedance in real time, lack of operating condition identification capability, and easy generation of control jumps. It suppresses transient overshoot, oscillation and harmonics, and improves frequency support capability and robustness under weak power grid conditions.

[0087] Example 2, as Figure 2 As shown, based on the same inventive concept as the big data-optimized converter station nonlinear collaborative control method provided in Embodiment 1, this embodiment of the invention also provides a big data-optimized converter station nonlinear collaborative control system, including: The real-time data acquisition module 11 is used to acquire the real-time voltage and real-time current of the target converter station grid connection point, calculate the real-time grid impedance, and calculate the impedance fluctuation rate of the real-time grid impedance relative to the historical average impedance. Operating condition identification module 12 is used to identify the current power grid operating condition type based on the impedance fluctuation rate; The dual-branch cooperative regression calculation module 13 is used to perform forward calculations through a pre-trained dual-branch cooperative regression network when a disturbance condition is identified, using the real-time grid impedance as input, and outputting preliminary virtual inertia parameters and preliminary damping parameters. Nonlinear correction module 14 is used to perform nonlinear correction on the preliminary virtual inertia parameter and the preliminary damping parameter based on the real-time power grid impedance and impedance fluctuation rate, so as to obtain the target virtual inertia parameter and the target damping parameter. The control output and limiting module 15 is used to generate a control output based on the target virtual inertia parameter and the target damping parameter, and to limit the control output through a saturation function to obtain a nonlinear cooperative control command, driving the target converter station to perform flexible synchronous output of the grid connection point voltage.

[0088] The real-time data acquisition module 11 is specifically used for: Obtain the real-time voltage and current at the target converter station's grid connection point, calculate the real-time grid impedance, and calculate the impedance fluctuation rate of the real-time grid impedance relative to the historical average impedance, including: The three-phase real-time voltage waveform and three-phase real-time current waveform of the target converter station grid connection point are collected at a fixed sampling frequency, and the real-time voltage effective value and real-time current effective value of each phase are calculated respectively. The ratio of the real-time effective value of voltage to the real-time effective value of current in the same phase is taken as the phase impedance of the corresponding phase, and the average value of the three phase impedances is taken as the real-time grid impedance. Extract the historical power grid impedance sequence recorded within a past sliding time window, assign decreasing weight coefficients according to the time from recent to distant, and perform a weighted arithmetic mean on the historical power grid impedance sequence to obtain the historical average impedance; The impedance fluctuation rate is calculated based on the real-time grid impedance and the historical average impedance.

[0089] The working condition identification module 12 is specifically used for: Identifying the current power grid operating condition type based on the impedance fluctuation rate includes: The power grid operating condition types include steady-state operating conditions and disturbance operating conditions, wherein the disturbance operating conditions include at least weak power grid operating conditions and power surge operating conditions; Multiple historical impedance fluctuation rates were continuously calculated during the steady-state operation of the target converter station, and the arithmetic mean and standard deviation of the multiple historical impedance fluctuation rates were calculated. The sum of the arithmetic mean and three times the standard deviation is configured as a first threshold, and the sum of the arithmetic mean and one standard deviation is configured as a second threshold; When the impedance fluctuation rate calculated in real time is greater than the first threshold, it is identified as a weak power grid condition. When the impedance fluctuation rate obtained by real-time calculation is greater than the second threshold and less than or equal to the first threshold, it is identified as a power sudden change condition. When the impedance fluctuation rate obtained from real-time calculation is less than or equal to the second threshold, it is identified as a steady-state operating condition.

[0090] The steps for determining the steady-state operating segment include: Calculate the absolute value of the deviation between each historical grid impedance and the historical average impedance from the historical grid impedance sequence extracted within the sliding time window; Calculate the standard deviation of the historical power grid impedance sequence; Starting from the current moment, trace back and record each control cycle that continuously satisfies the condition that the absolute value of the deviation is less than the standard deviation, and determine the time interval covered by each control cycle as the steady-state operating period.

[0091] Specifically, the dual-branch collaborative regression calculation module 13 is used for: When a disturbance condition is identified, the real-time grid impedance is used as input, and forward calculation is performed through a pre-trained dual-branch cooperative regression network to output preliminary virtual inertia parameters and preliminary damping parameters.

[0092] The pre-training steps of the dual-branch collaborative regression network include: Historical grid impedance sequences under different weak grid conditions and different power surge conditions are collected. The real-time grid impedance at each moment is used as the input state, and the corresponding optimal virtual inertia parameter and optimal damping parameter are used as the output label to construct a sample state-label pair set. A network architecture for a dual-branch collaborative regression network is constructed, wherein the network architecture includes a shared feature extraction layer, a virtual inertia output branch, and a damping parameter output branch; The input terminal of the shared feature extraction layer is used to receive the real-time grid impedance. The output terminal of the shared feature extraction layer is connected to the input terminals of the virtual inertia output branch and the damping parameter output branch, respectively. The output terminal of the virtual inertia output branch is used to output the predicted value of the virtual inertia parameter, and the output terminal of the damping parameter output branch is used to output the predicted value of the damping parameter. The sample state-label pair set is used as training data, and the sum of the mean square error between the predicted value and the label value of the virtual inertia output branch and the mean square error between the predicted value and the label value of the damping parameter output branch is used as the loss function. The network weights of the network architecture are jointly updated using the backpropagation algorithm until the loss function converges. The trained network architecture is then used as the dual-branch collaborative regression network.

[0093] The steps for obtaining the optimal virtual inertia parameter and the optimal damping parameter include: For each historical grid impedance sequence, extract the virtual inertia parameters, damping parameters, grid connection point voltage overshoot, oscillation number and frequency deviation recorded in multiple time windows during the actual operation of the converter station in the corresponding period; The overshoot, oscillation count, and frequency deviation of the grid connection point voltage extracted from all time windows are normalized respectively, and the weighted sum of the normalized overshoot, oscillation count, and frequency deviation under each time window is calculated. The virtual inertia parameters and damping parameters corresponding to the time window with the smallest weighted sum are determined as the optimal virtual inertia parameters and optimal damping parameters corresponding to the current historical power grid impedance sequence.

[0094] Specifically, the nonlinear correction module 14 is used for: Based on the real-time grid impedance and impedance fluctuation rate, nonlinear corrections are performed on the preliminary virtual inertia parameters and the preliminary damping parameters to obtain the target virtual inertia parameters and target damping parameters, including: Calculate the ratio of the real-time grid impedance to the historical average impedance, and use the square root of the ratio as the first correction factor; Calculate the sum of 1 and the impedance fluctuation rate, and use it as the first intermediate value; The natural logarithm of the first intermediate value plus 1 is used as the second correction coefficient; Multiply the preliminary virtual inertia parameter by the first correction coefficient to obtain the target virtual inertia parameter; The target damping parameter is obtained by multiplying the preliminary damping parameter by the second correction coefficient.

[0095] Specifically, the control output and limiting module 15 is used for: A control output is generated based on the target virtual inertia parameters and the target damping parameters, including: The actual active power and reference active power of the target converter station are obtained, and the active power deviation is calculated. The reference active power is the active power target value issued by the power grid dispatch center to the target converter station. Obtain the actual reactive power and reference reactive power of the target converter station, and calculate the reactive power deviation, wherein the reference reactive power is the reactive power target value issued by the power grid dispatch center to the target converter station; Substitute the target virtual inertia parameter, the target damping parameter, and the active power deviation into the synchronous generator swing equation to calculate the angular frequency deviation, and then perform an integral operation on the angular frequency deviation to obtain the modulation voltage phase angle. Based on the nominal voltage of the target converter station's grid connection point, the rated voltage amplitude is obtained, and combined with the reactive power deviation, the modulation voltage amplitude is calculated. The modulation voltage amplitude is positively correlated with the rated voltage amplitude and negatively correlated with the reactive power deviation. Based on the modulation voltage phase angle and the modulation voltage amplitude, a three-phase modulation voltage signal is synthesized and used as a control output. The three-phase modulation voltage signal includes A-phase modulation voltage, B-phase modulation voltage and C-phase modulation voltage.

[0096] Furthermore, the control output is limited by a saturation function to obtain a nonlinear cooperative control command, which drives the target converter station to perform flexible synchronous output of the grid connection point voltage, including: The control output is limited by a saturation function, wherein the saturation function is defined as follows: For the input one-phase modulation voltage value, the absolute value of the modulation voltage value is compared with the rated voltage amplitude; If the absolute value of the modulation voltage is less than or equal to the rated voltage amplitude, then the output of the saturation function is equal to the modulation voltage itself; If the absolute value of the modulation voltage is greater than the rated voltage amplitude, then the output of the saturation function is equal to the rated voltage amplitude multiplied by the positive or negative sign of the modulation voltage. The A-phase modulation voltage, B-phase modulation voltage, and C-phase modulation voltage in the three-phase modulation voltage signal are respectively input into the saturation function, and the amplitude limiting is calculated in sequence to obtain the corresponding three amplitude limiting voltage values. The three amplitude limiting voltage values ​​are used as nonlinear cooperative control commands. The nonlinear coordinated control command is output to the modulator of the target converter station, driving the target converter station to generate the grid connection point voltage according to the limiting voltage value, thereby completing the flexible synchronous output.

Claims

1. A big data-optimized nonlinear collaborative control method for converter stations, characterized in that, The method includes: The real-time voltage and real-time current of the target converter station's grid connection point are obtained, the real-time grid impedance is calculated, and the impedance fluctuation rate of the real-time grid impedance relative to the historical average impedance is calculated. The current power grid operating condition type is identified based on the impedance fluctuation rate; When a disturbance condition is identified, the real-time grid impedance is used as input, and forward calculation is performed through a pre-trained dual-branch cooperative regression network to output preliminary virtual inertia parameters and preliminary damping parameters. Based on the real-time grid impedance and impedance fluctuation rate, nonlinear corrections are performed on the preliminary virtual inertia parameters and the preliminary damping parameters to obtain the target virtual inertia parameters and the target damping parameters. A control output is generated based on the target virtual inertia parameter and the target damping parameter, and the control output is limited by a saturation function to obtain a nonlinear cooperative control command, which drives the target converter station to perform flexible synchronous output of the grid connection point voltage.

2. The big data-optimized nonlinear collaborative control method for converter stations according to claim 1, characterized in that, Obtain the real-time voltage and current at the target converter station's grid connection point, calculate the real-time grid impedance, and calculate the impedance fluctuation rate of the real-time grid impedance relative to the historical average impedance, including: The three-phase real-time voltage waveform and three-phase real-time current waveform of the target converter station grid connection point are collected at a fixed sampling frequency, and the real-time voltage effective value and real-time current effective value of each phase are calculated respectively. The ratio of the real-time effective value of voltage to the real-time effective value of current in the same phase is taken as the phase impedance of the corresponding phase, and the average value of the three phase impedances is taken as the real-time grid impedance. Extract the historical power grid impedance sequence recorded within a past sliding time window, assign decreasing weight coefficients according to the time from recent to distant, and perform a weighted arithmetic mean on the historical power grid impedance sequence to obtain the historical average impedance; The impedance fluctuation rate is calculated based on the real-time grid impedance and the historical average impedance.

3. The big data-optimized nonlinear collaborative control method for converter stations according to claim 1, characterized in that, Identifying the current power grid operating condition type based on the impedance fluctuation rate includes: The power grid operating condition types include steady-state operating conditions and disturbance operating conditions, wherein the disturbance operating conditions include at least weak power grid operating conditions and power surge operating conditions; Multiple historical impedance fluctuation rates were continuously calculated during the steady-state operation of the target converter station, and the arithmetic mean and standard deviation of the multiple historical impedance fluctuation rates were calculated. The sum of the arithmetic mean and three times the standard deviation is configured as a first threshold, and the sum of the arithmetic mean and one standard deviation is configured as a second threshold; When the impedance fluctuation rate calculated in real time is greater than the first threshold, it is identified as a weak power grid condition. When the impedance fluctuation rate obtained by real-time calculation is greater than the second threshold and less than or equal to the first threshold, it is identified as a power sudden change condition. When the impedance fluctuation rate obtained from real-time calculation is less than or equal to the second threshold, it is identified as a steady-state operating condition.

4. The big data-optimized nonlinear collaborative control method for converter stations according to claim 3, characterized in that, The steps for determining the steady-state operating segment include: Calculate the absolute value of the deviation between each historical grid impedance and the historical average impedance from the historical grid impedance sequence extracted within the sliding time window; Calculate the standard deviation of the historical power grid impedance sequence; Starting from the current moment, trace back and record each control cycle that continuously satisfies the condition that the absolute value of the deviation is less than the standard deviation, and determine the time interval covered by each control cycle as the steady-state operating period.

5. The big data-optimized nonlinear collaborative control method for converter stations according to claim 1, characterized in that, The pre-training steps of the dual-branch collaborative regression network include: Historical grid impedance sequences under different weak grid conditions and different power surge conditions are collected. The real-time grid impedance at each moment is used as the input state, and the corresponding optimal virtual inertia parameter and optimal damping parameter are used as the output label to construct a sample state-label pair set. A network architecture for a dual-branch collaborative regression network is constructed, wherein the network architecture includes a shared feature extraction layer, a virtual inertia output branch, and a damping parameter output branch; The input terminal of the shared feature extraction layer is used to receive the real-time grid impedance. The output terminal of the shared feature extraction layer is connected to the input terminals of the virtual inertia output branch and the damping parameter output branch, respectively. The output terminal of the virtual inertia output branch is used to output the predicted value of the virtual inertia parameter, and the output terminal of the damping parameter output branch is used to output the predicted value of the damping parameter. The sample state-label pair set is used as training data, and the sum of the mean square error between the predicted value and the label value of the virtual inertia output branch and the mean square error between the predicted value and the label value of the damping parameter output branch is used as the loss function. The network weights of the network architecture are jointly updated using the backpropagation algorithm until the loss function converges. The trained network architecture is then used as the dual-branch collaborative regression network.

6. The big data-optimized nonlinear collaborative control method for converter stations according to claim 5, characterized in that, The steps for obtaining the optimal virtual inertia parameters and optimal damping parameters include: For each historical grid impedance sequence, extract the virtual inertia parameters, damping parameters, grid connection point voltage overshoot, oscillation number and frequency deviation recorded in multiple time windows during the actual operation of the converter station in the corresponding period; The overshoot, oscillation count, and frequency deviation of the grid connection point voltage extracted from all time windows are normalized respectively, and the weighted sum of the normalized overshoot, oscillation count, and frequency deviation under each time window is calculated. The virtual inertia parameters and damping parameters corresponding to the time window with the smallest weighted sum are determined as the optimal virtual inertia parameters and optimal damping parameters corresponding to the current historical power grid impedance sequence.

7. The big data-optimized nonlinear collaborative control method for converter stations according to claim 1, characterized in that, Based on the real-time grid impedance and impedance fluctuation rate, nonlinear corrections are performed on the preliminary virtual inertia parameters and the preliminary damping parameters to obtain the target virtual inertia parameters and target damping parameters, including: Calculate the ratio of the real-time grid impedance to the historical average impedance, and use the square root of the ratio as the first correction factor; Calculate the sum of 1 and the impedance fluctuation rate, and use it as the first intermediate value; The natural logarithm of the first intermediate value plus 1 is used as the second correction coefficient; Multiply the preliminary virtual inertia parameter by the first correction coefficient to obtain the target virtual inertia parameter; The target damping parameter is obtained by multiplying the preliminary damping parameter by the second correction coefficient.

8. The big data-optimized nonlinear collaborative control method for converter stations according to claim 1, characterized in that, A control output is generated based on the target virtual inertia parameters and the target damping parameters, including: The actual active power and reference active power of the target converter station are obtained, and the active power deviation is calculated. The reference active power is the active power target value issued by the power grid dispatch center to the target converter station. Obtain the actual reactive power and reference reactive power of the target converter station, and calculate the reactive power deviation, wherein the reference reactive power is the reactive power target value issued by the power grid dispatch center to the target converter station; Substitute the target virtual inertia parameter, the target damping parameter, and the active power deviation into the synchronous generator swing equation to calculate the angular frequency deviation, and then perform an integral operation on the angular frequency deviation to obtain the modulation voltage phase angle. Based on the nominal voltage of the target converter station's grid connection point, the rated voltage amplitude is obtained, and combined with the reactive power deviation, the modulation voltage amplitude is calculated. The modulation voltage amplitude is positively correlated with the rated voltage amplitude and negatively correlated with the reactive power deviation. Based on the modulation voltage phase angle and the modulation voltage amplitude, a three-phase modulation voltage signal is synthesized and used as a control output. The three-phase modulation voltage signal includes A-phase modulation voltage, B-phase modulation voltage and C-phase modulation voltage.

9. The big data-optimized nonlinear collaborative control method for converter stations according to claim 1, characterized in that, By limiting the control output using a saturation function, a nonlinear cooperative control command is obtained to drive the target converter station to execute flexible synchronous output of the grid connection point voltage, including: The control output is limited by a saturation function, wherein the saturation function is defined as follows: For the input one-phase modulation voltage value, the absolute value of the modulation voltage value is compared with the rated voltage amplitude; If the absolute value of the modulation voltage is less than or equal to the rated voltage amplitude, then the output of the saturation function is equal to the modulation voltage itself; If the absolute value of the modulation voltage is greater than the rated voltage amplitude, then the output of the saturation function is equal to the rated voltage amplitude multiplied by the positive or negative sign of the modulation voltage. The A-phase modulation voltage, B-phase modulation voltage, and C-phase modulation voltage in the three-phase modulation voltage signal are respectively input into the saturation function, and the amplitude limiting is calculated in sequence to obtain the corresponding three amplitude limiting voltage values. The three amplitude limiting voltage values ​​are used as nonlinear cooperative control commands. The nonlinear coordinated control command is output to the modulator of the target converter station, driving the target converter station to generate the grid connection point voltage according to the limiting voltage value, thereby completing the flexible synchronous output.

10. A nonlinear collaborative control system for converter stations optimized by big data, characterized in that, The converter station nonlinear collaborative control method for implementing big data optimization as described in any one of claims 1-9 includes: The real-time data acquisition module is used to acquire the real-time voltage and real-time current at the grid connection point of the target converter station, calculate the real-time grid impedance, and calculate the impedance fluctuation rate of the real-time grid impedance relative to the historical average impedance. Operating condition identification module, used to identify the current power grid operating condition type based on the impedance fluctuation rate; The dual-branch cooperative regression calculation module is used to perform forward calculations through a pre-trained dual-branch cooperative regression network when a disturbance condition is identified, taking the real-time grid impedance as input, and outputting preliminary virtual inertia parameters and preliminary damping parameters. The nonlinear correction module is used to perform nonlinear correction on the preliminary virtual inertia parameters and the preliminary damping parameters based on the real-time grid impedance and impedance fluctuation rate, so as to obtain the target virtual inertia parameters and the target damping parameters. The control output and limiting module is used to generate a control output based on the target virtual inertia parameter and the target damping parameter, and to limit the control output through a saturation function to obtain a nonlinear cooperative control command, which drives the target converter station to perform flexible synchronous output of the grid connection point voltage.

Images