Model-based reactive power response control method for wind power converters

Abstract

This invention relates to a model-based reactive power response control method for wind power converters. It involves real-time acquisition of PPC point voltage, converter output current, and power reference values ​​to accurately calculate the current d-q axis current and voltage, as well as the d-q axis reference current for the next moment. Subsequently, fuzzy inference is used to process this current data, intelligently generating an adaptive weighting coefficient. This adaptive weighting coefficient can be dynamically adjusted based on current tracking error and then applied to the cost function evaluation of the predicted d-q current set. This guides the controller to select the optimal switching state from multiple candidate states, thereby optimizing reactive power response performance. In this way, the tracking accuracy and dynamic response speed of the wind power converter to grid signals can be significantly improved, thus greatly enhancing the grid-connected adaptability and overall operational reliability of the wind farm.

Description

Technical Field

[0001] This invention relates to the field of intelligent control technology, and in particular to a reactive power response control method for wind power converters based on model prediction. Background Technology

[0002] With the increasing global demand for renewable energy, wind power has become an important component of modern power systems. To ensure the safe and stable grid connection of wind farms and to actively respond to grid operation requirements, wind power converters must possess precise and rapid reactive power regulation capabilities. Especially under transient conditions such as grid voltage fluctuations and fault ride-throughs, the converter's ability to promptly provide or absorb reactive power is crucial for maintaining grid voltage stability and system operational reliability, and is also key to meeting increasingly stringent grid connection guidelines. Therefore, developing advanced and reliable reactive power response control methods for wind power converters is one of the key technologies for promoting the large-scale development and application of wind power.

[0003] Currently, Model Predictive Control (MPC) is widely used in reactive power response control of wind power converters due to its excellent dynamic performance and multi-objective optimization potential. However, existing fixed-parameter-based MPC schemes have revealed significant technical shortcomings in practice. MPC control performance is highly dependent on the accuracy of the predictive model, but the actual operating environment is complex and variable. Grid impedance exhibits significant time-varying characteristics; for example, the switching of transmission lines or the start-up and shutdown of large loads can cause instantaneous changes in grid impedance. This mismatch between the model and the actual system makes fixed-parameter MPC prone to current overshoot or large steady-state errors when rapid reactive power response is required. Consequently, it cannot accurately track reactive power commands, severely impacting control accuracy and dynamic response speed. Summary of the Invention

[0004] The present invention aims to solve at least one of the problems existing in the prior art and provide a reactive power response control method for wind power converters based on model prediction.

[0005] This invention provides a model-predictive reactive power response control method for wind power converters, comprising: Obtain the instantaneous values ​​of the three-phase AC voltage at the PPC point, the instantaneous values ​​of the three-phase AC current at the converter outlet, the current active power reference value, and the current reactive power reference value; Based on the instantaneous values ​​of the three-phase AC voltage at point PPC, the instantaneous values ​​of the three-phase AC current at the converter outlet, the current active power reference value, and the current reactive power reference value, determine the dq-axis current at the current moment, the dq-axis reference current at the next moment, and the dq-axis voltage at the current moment. Fuzzy inference is performed on the current dq-axis current and the current dq-axis reference current to obtain adaptive weighting coefficients; Perform current prediction under multiple candidate states for the current dq-axis current and current dq-axis voltage at the current moment to obtain the predicted dq current set; Based on the dq-axis reference current and adaptive weight coefficients at the next time step, the cost function based on adaptive weights is evaluated on the predicted dq current set to obtain the total cost set of all candidate states. The current optimal state is determined based on the total value set of all candidate states.

[0006] Furthermore, based on the instantaneous values ​​of the three-phase AC voltage at the PPC point, the instantaneous values ​​of the three-phase AC current at the converter outlet, the current active power reference value, and the current reactive power reference value, the dq-axis current at the current moment, the dq-axis reference current at the next moment, and the dq-axis voltage at the current moment are determined, including: Input the instantaneous values ​​of the three-phase AC voltage at point PPC and the instantaneous values ​​of the three-phase AC current at the converter outlet into the phase-locked loop to obtain the grid voltage synchronization phase angle at the current moment; Based on the grid voltage synchronization phase angle at the current moment, the instantaneous values ​​of the three-phase AC voltage at point PPC and the instantaneous values ​​of the three-phase AC current at the converter outlet are transformed into the dq coordinate system to obtain the dq-axis voltage and the dq-axis current at the current moment. Based on the current active power reference value, the current reactive power reference value, and the current dq-axis voltage, determine the dq-axis reference current for the next moment.

[0007] Furthermore, based on the current grid voltage synchronization phase angle, the instantaneous values ​​of the three-phase AC voltage at point PPC and the instantaneous values ​​of the three-phase AC current at the converter outlet are transformed into dq coordinates to obtain the current dq-axis voltage and current. This includes transforming the instantaneous values ​​of the three-phase AC voltage at point PPC and the instantaneous values ​​of the three-phase AC current at the converter outlet into dq coordinates using the following formula: ; in, This refers to the instantaneous value of the three-phase AC voltage at the PPC point or the instantaneous value of the three-phase AC current at the converter outlet. This represents the current dq-axis voltage or the current dq-axis current. This represents the current phase angle of the grid voltage synchronization.

[0008] Furthermore, based on the current active power reference value, the current reactive power reference value, and the current dq-axis voltage, the dq-axis reference current for the next moment is determined, including: determining the dq-axis reference current for the next moment using the following formula: ; ; in, This is the current active power reference value. This is the current reactive power reference value. This represents the d-axis voltage in the dq-axis voltage at the current moment. Let d be the d-axis reference current in the dq-axis reference current at the next moment. This is the q-axis reference current in the dq-axis reference current at the next moment.

[0009] Furthermore, fuzzy inference is performed on the current dq-axis current and the current dq-axis reference current to obtain adaptive weighting coefficients, including: Calculate the original error between the current dq-axis current and the current dq-axis reference current to obtain the d-axis error and q-axis error; The d-axis error and q-axis error are aggregated by direction-sensitive weighted aggregation to obtain the aggregated error; The aggregation error is input into the fuzzy logic system to obtain the adaptive weight coefficients.

[0010] Furthermore, direction-sensitive error weighting aggregation is performed on the d-axis error and q-axis error to obtain the aggregated error, including: direction-sensitive error weighting aggregation of the d-axis error and q-axis error using the following formula: ; ; in, It is the direction-sensitive factor at the current moment; It is a preset, positive-zero directional sensitivity coefficient; It is a very small positive number; It is the aggregation error at the current moment. This represents the d-axis error at the current moment. This represents the q-axis error at the current moment.

[0011] Furthermore, inputting the aggregation error into the fuzzy logic system to obtain the adaptive weight coefficients includes: the fuzzy logic system processing the aggregation error using the following formula to obtain the adaptive weight coefficients: ; in, This represents a complete fuzzy logic mapping function. This represents the adaptive weighting coefficient at the current moment.

[0012] Furthermore, current predictions are performed on the current dq-axis current and voltage at the current moment under multiple candidate states to obtain a set of predicted dq currents, including: A discretized system model corresponding to the topology of the wind power converter is used as the prediction model; In each control cycle, all possible switching states of the wind power converter are traversed as candidate states. For each candidate state, the current dq-axis current and current dq-axis voltage are combined with the output voltage vector determined by the candidate state and substituted into the prediction model to calculate the corresponding dq-axis current at the next moment. The dq-axis currents at the next moment corresponding to all candidate states form the predicted dq current set.

[0013] Furthermore, based on the dq-axis reference current and adaptive weight coefficients at the next time step, a cost function based on adaptive weights is performed on the predicted dq-current set to obtain the total value set of all candidate states, including: For the dq-axis current at the next moment corresponding to each candidate state in the predicted dq current set, calculate the deviation between it and the dq-axis reference current at the next moment to obtain the current tracking error. Calculate the cost associated with switching losses for each candidate state to obtain the switching losses; For each candidate state, the corresponding current tracking error and the switching loss after adaptive weighting coefficient adjustment are weighted and summed to obtain the corresponding total value. The total values ​​corresponding to each candidate state form the total value set.

[0014] Furthermore, based on the total value set of all candidate states, the current optimal state is determined, including: The minimum total agency value is identified as the total agency value that has the smallest value in the total agency value set. Determine the candidate states corresponding to the minimum total value; The candidate state corresponding to the minimum total value is taken as the current optimal state.

[0015] Compared with existing technologies, this invention provides a model-based reactive power response control method for wind power converters. It accurately calculates the current dq-axis current and voltage, and the dq-axis reference current at the next moment by real-time acquisition of PPC point voltage, converter output current, and power reference values. Subsequently, fuzzy inference is used to process this current data, intelligently generating an adaptive weighting coefficient. This adaptive weighting coefficient can be dynamically adjusted according to the current tracking error and then applied to the cost function evaluation of the predicted dq-axis current set. This guides the controller to select the optimal switching state from multiple candidate states, thereby optimizing reactive power response performance. In this way, the tracking accuracy and dynamic response speed of the wind power converter to grid signals can be significantly improved, thus greatly enhancing the grid-connected adaptability and overall operational reliability of the wind farm. Attached Figure Description

[0016] One or more embodiments are illustrated by way of example with the corresponding pictures in the accompanying drawings. These illustrations do not constitute a limitation on the embodiments. Elements with the same reference numerals in the drawings are denoted as similar elements. Unless otherwise stated, the figures in the drawings are not to be limited by scale.

[0017] Figure 1 This is a flowchart of a model-predictive reactive power response control method for wind power converters according to an embodiment of the present invention; Figure 2 This is a schematic diagram of data flow for a model-based reactive power converter reactive power response control method according to an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the various embodiments of the present invention will be described in detail below with reference to the accompanying drawings. However, those skilled in the art will understand that many technical details are presented in the various embodiments of the present invention to facilitate a better understanding of the invention. However, the technical solutions claimed in the present invention can be implemented even without these technical details and with various variations and modifications based on the following embodiments. The division of the various embodiments below is for ease of description and should not constitute any limitation on the specific implementation of the present invention. The various embodiments can be combined with and referenced by each other without contradiction.

[0019] This invention proposes a reactive power response control method for wind power converters based on model prediction. Figure 1 This is a flowchart of a model-predictive reactive power response control method for wind power converters according to an embodiment of the present invention. Figure 2 This is a schematic diagram of data flow in a model-predictive reactive power response control method for wind power converters according to an embodiment of the present invention. Figure 1 and Figure 2 According to an embodiment of the present invention, a model-based reactive power response control method for wind power converters includes the following steps: S1, acquiring the instantaneous values ​​of the three-phase AC voltage at the PPC point, the instantaneous values ​​of the three-phase AC current at the converter outlet, the current active power reference value, and the current reactive power reference value; S2, based on the instantaneous values ​​of the three-phase AC voltage at the PPC point, the instantaneous values ​​of the three-phase AC current at the converter outlet, the current active power reference value, and the current reactive power reference value, determining the dq-axis current at the current moment, the dq-axis reference current at the next moment, and the current dq-axis current at the next moment. S3, perform fuzzy inference on the current dq-axis current and the current dq-axis reference current to obtain adaptive weight coefficients; S4, perform current prediction under multiple candidate states on the current dq-axis current and the current dq-axis voltage to obtain a set of predicted dq currents; S5, based on the next dq-axis reference current and adaptive weight coefficients, evaluate the set of predicted dq currents using a cost function based on adaptive weights to obtain a set of total cost values ​​for all candidate states; S6, based on the set of total cost values ​​for all candidate states, determine the current optimal state.

[0020] Specifically, S1 involves acquiring the instantaneous values ​​of the three-phase AC voltage at the PPC point, the instantaneous values ​​of the three-phase AC current at the converter outlet, the current active power reference value, and the current reactive power reference value. To achieve precise and rapid control of the reactive power response of the wind power converter, it is essential to first comprehensively and accurately perceive the current operating state and the desired control objectives. As described in the background section, wind power generation, as a crucial component of modern power systems, requires its converters to possess precise and rapid reactive power regulation capabilities to ensure the safe and stable grid connection of wind farms and to actively respond to grid operation requirements. Especially under transient conditions such as grid voltage fluctuations and fault ride-through, the converter's ability to promptly provide or absorb reactive power is crucial for maintaining grid voltage stability and system operational reliability. The performance of model predictive control highly depends on the accurate acquisition of the current system state and reference commands. These instantaneous and reference values ​​constitute the direct basis for control law calculations and are the starting point for predictive model establishment, error assessment, and optimization decisions. Without accurate acquisition of these core data, any subsequent complex control strategies will be unable to be effectively implemented due to a lack of reliable input.

[0021] The instantaneous three-phase AC voltage at the PPC point refers to the instantaneous value of the three-phase AC voltage monitored in real time at the wind power converter's grid connection point (PPC point, i.e., the point of common coupling). This data reflects the real-time voltage status on the grid side and is crucial for the converter to sense grid voltage fluctuations, provide voltage support, and regulate reactive power. The instantaneous three-phase AC current at the converter output refers to the instantaneous value of the three-phase AC current flowing from the wind power converter's output side to the grid. This data reflects the real-time active and reactive power output of the converter to the grid and is the core feedback signal for controlling the converter's output. The current active power reference value and the current reactive power reference value refer to the active power target and reactive power target that the wind power converter needs to achieve within the current control cycle, respectively. These reference values ​​are usually given by the upper-level grid dispatch or wind farm management system based on grid demand, wind turbine operating status, and other factors, and are the basis for generating the dq-axis current reference value.

[0022] In its specific implementation, S1 first converts physical quantities into digital signals that the controller can process through sensors and a data acquisition system. Specifically, for the instantaneous values ​​of the three-phase AC voltage at the PPC point and the instantaneous values ​​of the three-phase AC current at the converter outlet, high-precision voltage transformers (PTs) and current transformers (CTs) are typically used for real-time measurement. The measured analog signals are then converted into digital signals by an analog-to-digital converter (ADC) and transmitted to a digital controller (such as a DSP or FPGA) for processing. For the current active power reference value and reactive power reference value, these values ​​are typically received from the upper-level control system through a communication interface (such as Modbus, Ethernet, etc.) or calculated in real-time by the converter's own power control strategy and directly input into the control algorithm as target values.

[0023] Specifically, S2 determines the dq-axis current at the current moment, the dq-axis reference current at the next moment, and the dq-axis voltage at the current moment based on the instantaneous values ​​of the three-phase AC voltage at the PPC point, the instantaneous values ​​of the three-phase AC current at the converter outlet, the current active power reference value, and the current reactive power reference value. To achieve independent and decoupled control of the active and reactive power of the wind power converter, the three-phase AC system is transformed into a synchronously rotating dq coordinate system for analysis and control. It should be understood that in a three-phase stationary coordinate system, AC quantities are time-varying, and their active and reactive components are coupled together, making direct control difficult. Through dq coordinate transformation, AC quantities are converted into DC components, decoupling the control of active power (usually related to the d-axis current) and reactive power (usually related to the q-axis current), greatly simplifying the design and implementation of the controller. Furthermore, determining the dq-axis reference current at the next moment is the basis for error calculation and cost function evaluation in model predictive control, providing a clear control objective for subsequent prediction and optimization.

[0024] In this context, the dq-axis current / voltage refers to the components of three-phase AC current or voltage transformed from the stationary abc coordinate system to the synchronously rotating dq coordinate system using the Park transformation. The d-axis is typically aligned with the grid voltage vector, while the q-axis is orthogonal to the d-axis. During stable operation, these components are DC values, facilitating control.

[0025] In the specific implementation of S2, firstly, the instantaneous values ​​of the three-phase AC voltage at point PPC and the instantaneous values ​​of the three-phase AC current at the converter outlet are input into the phase-locked loop (PLL) to obtain the current grid voltage synchronization phase angle. The PLL is a control system used to detect the phase and frequency of the grid voltage and generate a phase angle signal synchronized with the grid voltage. This phase angle signal is crucial for the transformation from the abc stationary coordinate system to the dq coordinate system. During this process, the instantaneous values ​​of the three-phase AC voltage at point PPC and the instantaneous values ​​of the three-phase AC current at the converter outlet, obtained in the previous step, are input into the PLL. The PLL continuously outputs the current grid voltage synchronization phase angle by tracking the phase of the grid voltage in real time. This synchronization phase angle serves as the reference for the subsequent dq coordinate transformation.

[0026] Next, based on the current grid voltage synchronization phase angle, the instantaneous values ​​of the three-phase AC voltage at point PPC and the instantaneous values ​​of the three-phase AC current at the converter outlet are transformed into dq coordinates to obtain the current dq-axis voltage and current. Specifically, the dq-axis transformation of the instantaneous values ​​of the three-phase AC voltage at point PPC and the instantaneous values ​​of the three-phase AC current at the converter outlet is performed using the following formula: ; in, This refers to the instantaneous value of the three-phase AC voltage at the PPC point or the instantaneous value of the three-phase AC current at the converter outlet. This represents the current dq-axis voltage or the current dq-axis current. This represents the current phase angle of the grid voltage synchronization.

[0027] Furthermore, based on the current active power reference value, the current reactive power reference value, and the current dq-axis voltage, the dq-axis reference current for the next moment is determined. Specifically, the dq-axis reference current for the next moment is determined using the following formula: ; ; in, This is the current active power reference value. This is the current reactive power reference value. This represents the d-axis voltage in the dq-axis voltage at the current moment. Let d be the d-axis reference current in the dq-axis reference current at the next moment. This is the q-axis reference current in the dq-axis reference current at the next moment.

[0028] Specifically, in S3, fuzzy inference is performed on the current dq-axis current and the current dq-axis reference current to obtain adaptive weight coefficients. It should be understood that in model-based wind power converter control, adaptively adjusting the weight λ of the cost function to balance dynamic response and switching losses is an effective optimization approach. However, if the weight generation mechanism relies solely on the maximum scalar value of the d-axis or q-axis current error, a new technical problem arises. This stems from the fact that this error assessment method is scenario-blind; it decouples the d-axis and q-axis errors into two independent scalars for comparison, completely ignoring the inherent physical meaning of the error vector they collectively constitute. In specific wind power grid-connected scenarios, the q-axis current is directly related to the ability to support grid voltage and is a high-priority task for ensuring grid stability, especially during grid voltage dips and other faults, where its speed and accuracy of response are crucial. In contrast, the active power associated with the d-axis current typically has a lower control priority during transient periods. The original mechanism treats d-axis errors and q-axis errors as equivalent, failing to distinguish between the high risk of failing to complete critical tasks and the low risk of failing to complete secondary tasks. This results in an inaccurate assessment of the overall system operational risk. This ignorance of the error vector direction and the underlying control priority prevents the generated adaptive weight λ from achieving an optimal match with the actual operating risk. Consequently, it may exhibit slow response due to a conservative strategy when dynamic responsiveness is most needed, or increase switching losses due to unnecessary aggressiveness when the system is stable.

[0029] To address the aforementioned technical deficiencies, this technical solution proposes a weighted aggregated error generation mechanism based on error vector direction and priority. Its core lies in no longer using simple scalar errors, but instead constructing an aggregated error index that can dynamically sense the error direction and reflect the control priority, thereby driving the generation of adaptive weights.

[0030] In specific implementation, step S3 first involves calculating the original error between the current dq-axis current and the current dq-axis reference current to obtain the d-axis error and q-axis error. This step aims to obtain the most fundamental deviation between the controller's control objective and the actual system state at the current moment. Specifically, the reference value and measured value of the dq-axis current at the current moment k are obtained, and the original tracking errors of the d-axis and q-axis are calculated by subtracting the corresponding components.

[0031] Next, the d-axis and q-axis errors are aggregated using a direction-sensitive weighted aggregation method to obtain the aggregated error. It should be understood that the original error components themselves do not contain priority information. In a preferred embodiment of this invention, through a single mathematical construction, the expert experience inherent in the error direction (i.e., the q-axis error is more critical than the d-axis error) is explicitly and non-linearly injected into the error metric. Specifically, a direction-sensitive factor for the current time k is first constructed. This direction sensitivity factor uses the ratio of the squares of the d-axis error to the q-axis error to quantitatively assess the degree to which the error vector deviates from the high-priority q-axis. Then, this direction sensitivity factor is multiplied by the true Euclidean norm (i.e., the magnitude) of the error vector to obtain the final aggregated error at the current time k. This process can be expressed by the following formula: ; ; in, It is the direction-sensitive factor at the current moment; It is a preset, positive-zero directional sensitivity coefficient used to define the degree of importance attached to q-axis error; It is a very small positive number used to avoid the denominator being zero and to ensure the numerical stability of the calculation; It is the final aggregated error at the current moment. This represents the d-axis error at the current moment. This represents the q-axis error at the current moment. Specifically, It is no longer a simple measure of error magnitude, but an intelligent system risk signal. In scenarios where the power grid experiences voltage dips and urgently needs reactive power support, the q-axis error... It will increase significantly at this time, at which point the direction sensitivity factor It will quickly approach its maximum value of 1+ This will reduce the aggregation error. Nonlinear amplification immediately indicates that the system is in a high-risk state. Conversely, in steady-state operation or when only active power disturbances occur, the q-axis error... Very small, direction-sensitive factor Approaching 1, aggregation error This will faithfully reflect the actual physical magnitude of the error. In this way, the system has the ability to automatically sense and quantify operational risks under different operating conditions.

[0032] Then, the aggregation error is input into the fuzzy logic system to obtain the adaptive weight coefficients. That is, the abstract risk signal perceived in the preceding steps is transformed into specific parameters that can directly control the behavior of the controller. Specifically, the... The input is a fuzzy logic system. This system performs fuzzification, inference, and defuzzification operations based on the rule that higher risk corresponds to lower weight, ultimately outputting a smooth, continuous adaptive weight λ. This process is expressed by the formula: ; in, This represents a complete fuzzy logic mapping function. This represents the adaptive weighting coefficient at the current moment. This step establishes a bridge from risk perception to control decision-making. Its purpose and effect is to produce the final weight λ, which accurately reflects the system's immediate needs: when the aggregation error of high-risk signals... When inputting, a very small adaptive weight λ is output, thereby reducing the penalty for the number of switching operations in the cost function. The ultimate goal is to unleash the full dynamic potential of the controller to cope with emergencies. When the risk signal is very small, a larger adaptive weight λ is output, which aims to strengthen the constraint on the switching frequency in order to achieve the steady-state operation goal of reducing losses and improving power quality.

[0033] In this way, the entire model predictive control system evolves from a scenario-blind error tracker into an intelligent decision-making entity with risk perception capabilities, fundamentally resolving the sharp contradiction between dynamic response speed and steady-state operating efficiency in traditional solutions. Specifically, when the power grid requires emergency support, this mechanism can identify the priority of q-axis errors and drive the controller to perform reactive power compensation at the fastest speed by amplifying the risk signal, ensuring the safety and compliance of grid connection and achieving the technical goal of ensuring grid stability. After the system enters steady state, the mechanism can smoothly switch the control strategy to a high-efficiency mode, reducing converter losses and improving output current harmonics by suppressing unnecessary switching actions, thus achieving the technical goal of improving power quality and operating economy. This ability to automatically switch between motion and economic modes based on operating condition risks achieves global optimization of control performance.

[0034] Specifically, in S4, current prediction under multiple candidate states is performed on the current and voltage of the dq-axis at the current moment to obtain a set of predicted dq currents. It should be understood that the performance of Model Predictive Control (MPC) is highly dependent on the accuracy of the predictive model and its ability to predict the future state of the system within a finite number of control cycles. As described in the background section, traditional fixed-parameter-based MPC often suffers from a mismatch between the model and the actual system when performing rapid reactive power response due to the time-varying characteristics of the grid impedance in actual operation (such as transmission line switching and large load start-up and shutdown). This mismatch produces significant current overshoot or steady-state error, thus failing to accurately track reactive power commands. By predicting the current dq-axis current and voltage under multiple candidate states, the converter controller can simulate the impact of all possible switching states on the dq-axis current at the next moment in each control cycle. This forward-looking prediction allows the controller to select the optimal action from a series of prediction results, thereby effectively responding to changes in grid parameters, reducing current tracking errors, and improving dynamic response speed. This overcomes the limitations of fixed-parameter models and achieves more accurate tracking of reactive power commands.

[0035] In this context, "multiple candidate states" refers to the voltage output states corresponding to all possible switching combinations of the converter within each control cycle. For a three-phase two-level voltage source converter, there are eight theoretical switching states, each corresponding to a unique output voltage vector. These voltage vectors act on the converter outlet, producing different current responses, hence the term "multiple candidate states," which form the basis for prediction. Current prediction utilizes the converter's mathematical model and current measurements (dq-axis current and voltage) to calculate and predict the dq-axis current value of the converter output at the next sampling time, based on different candidate states. This is crucial for MPC to achieve feedforward control and optimal selection. The predicted dq current set refers to the set of predicted d-axis and q-axis currents for the next time step or some future time step for all possible multiple candidate states. This set contains all potential future current responses and serves as the input for subsequent cost function evaluation.

[0036] S4 includes: using a discretized system model corresponding to the topology of the wind power converter as the prediction model; in each control cycle, traversing all possible switching states of the wind power converter as candidate states; for each candidate state, substituting the current dq-axis current and current dq-axis voltage with the output voltage vector determined by the candidate state into the prediction model to calculate the corresponding dq-axis current at the next moment, and the dq-axis currents at the next moment corresponding to all candidate states form the predicted dq current set.

[0037] In other words, in the specific implementation of S4, the prediction model is first determined. In a cluster example, a discretized system model corresponding to the wind power converter topology can be used. For a voltage source converter, its dynamic behavior in the dq synchronous rotating coordinate system is usually described by a set of differential equations, which characterize the relationship between the rate of change of the d-axis and q-axis currents and the converter's inductance, resistance, grid angular frequency, converter output voltage, and grid voltage. After discretizing these continuous differential equations, they can be used to predict current changes within a sampling period.

[0038] Next, candidate states are enumerated. In each control cycle, the controller iterates through all possible switching states of the wind power converter (for example, a three-phase two-level converter has eight voltage vectors corresponding to its switching states). These switching states determine the converter's output voltage vector in the next sampling cycle.

[0039] Then, current prediction is performed for each candidate state. For each candidate state, the measured dq-axis current at the current time (time k) and the dq-axis voltage obtained from the previous step are used as inputs. The output voltage vector determined by the candidate state is then substituted into the discretized prediction model. Using this prediction model, the dq-axis current at the next time (time k+1) is calculated and estimated. By repeating the above prediction process, the predicted dq-axis current values ​​for all candidate states are combined to obtain the predicted dq-axis current set.

[0040] Specifically, in step S5, based on the dq-axis reference current at the next moment and the adaptive weighting coefficient, a cost function evaluation based on adaptive weights is performed on the predicted dq current set to obtain the total cost set of all candidate states. It should be understood that, facing complex operating conditions such as time-varying grid impedance and dynamic changes in reactive power response priority, the controller must achieve the optimal balance between rapid response and operational efficiency. Traditional cost functions often use fixed weights, failing to dynamically adjust the control emphasis according to the actual operating state, leading to a sharp contradiction between dynamic performance and steady-state operating efficiency. For example, when the grid requires emergency support, the system may respond slowly due to excessive penalty for switching losses; while when the system is stable, it may unnecessarily increase switching losses due to excessive aggression. Therefore, in the technical solution of this invention, by introducing adaptive weighting coefficients, the cost function can intelligently perceive operational risks and control priorities, thereby accurately reflecting the optimal performance requirements under the current operating condition when evaluating different candidate states, avoiding the scenario blindness problem, and ensuring that the final decision aligns to the actual grid operation requirements and the converter's own protection strategy to the greatest extent possible.

[0041] The cost function typically consists of multiple objective terms, such as current tracking error, switching frequency, and voltage deviation. A quantified cost value is obtained by weighted summation of these objective terms. The objective of the cost function is to minimize this cost value. The total cost value set refers to the set of quantified values ​​obtained by calculating the cost function for each predicted candidate state. This set contains the evaluation results for every possible control action.

[0042] S5 includes: for each candidate state in the predicted dq current set, calculating the deviation between the current at the next moment and the reference current at the next moment to obtain the current tracking error; calculating the cost related to switching losses for each candidate state to obtain the switching loss; for each candidate state, weighting and summing the current tracking error and the switching loss after adaptive weighting coefficient adjustment to obtain the corresponding total cost value, and the total cost values ​​corresponding to each candidate state form the total cost value set.

[0043] In other words, in the specific implementation of S5, firstly, a cost function is constructed. The cost function typically contains at least two main terms: a current tracking error term and a switching loss term. The current tracking error term quantifies the deviation between the predicted current and the reference current, with the goal of making the converter output as close as possible to the commanded value. The switching loss term quantifies the frequency or number of converter switching operations, with the goal of reducing unnecessary switching operations to lower energy loss and extend device lifespan.

[0044] Next, the current tracking error is calculated. For the predicted dq-axis current corresponding to each candidate state in the predicted dq-axis current set, i.e. the dq-axis current at the next moment, the deviation between it and the dq-axis reference current at the next moment is first calculated. This is usually done by calculating the sum of the squares of the differences between the two. This deviation reflects how close the predicted current is to the expected current.

[0045] Next, the switching loss term is calculated. For each candidate state, a cost associated with the switching loss is also calculated. This could be a measure of the number of switches required to switch from the current switching state to that candidate state, or a metric for a certain switching frequency.

[0046] Subsequently, the adaptive weighting coefficients obtained in the previous step are integrated into the cost function to balance current tracking performance and switching losses. When the adaptive weighting coefficients are small, the penalty for the number of switching operations in the cost function is reduced, thus prioritizing dynamic response. Conversely, when the adaptive weighting coefficients are large, the cost function strengthens the constraint on the switching frequency to prioritize reducing losses and improving power quality. This adaptive weighting method allows the cost function to dynamically adjust the relative emphasis on current tracking accuracy and switching losses.

[0047] Finally, the calculated current tracking error term and the adaptively weighted switching loss term are weighted and summed to obtain the total cost of the candidate state. This process is repeated for all candidate states in the predicted dq current set to obtain the total cost set of all candidate states.

[0048] Specifically, in step S6, the current optimal state is determined based on the total cost set of all candidate states. By identifying the optimal state from the total cost set of all candidate states, the controller can transform the theoretical optimization result into actual control commands, thereby guiding the converter to perform practical operations that maximize the balance between current tracking accuracy, switching losses, and dynamic response speed. The current optimal state refers to the switching state with the lowest total cost value among all possible switching combinations of the converter within the current control cycle, after evaluation using the cost function. This state represents the best control action the converter should take at the current moment, based on the set control objectives and constraints. Selecting this state means that the converter will execute the corresponding switching mode, thereby achieving optimal current output and power response in the next control cycle.

[0049] S6 includes: identifying the minimum total value in the total value set as the minimum total value; determining the candidate state corresponding to the minimum total value; and taking the candidate state corresponding to the minimum total value as the current optimal state.

[0050] In other words, when S6 is implemented, firstly, the controller compares each value in the total value set of all candidate states one by one, with the goal of identifying the total value with the smallest value in the set. Next, once the minimum value is determined, the controller will trace back to the specific candidate state corresponding to this minimum cost value. This identified switching state is then determined as the current optimal state.

[0051] Furthermore, the wind power converter is controlled to operate in the switching mode specified by the current optimal state. This process is repeated in each sampling period, realizing real-time dynamic adjustment of the converter output current.

[0052] In summary, the model-based reactive power response control method for wind power converters according to embodiments of the present invention is explained. It accurately calculates the current dq-axis current and voltage, and the dq-axis reference current at the next moment by real-time acquisition of PPC point voltage, converter output current, and power reference values. Subsequently, fuzzy inference is used to process these current data to intelligently generate an adaptive weighting coefficient. This adaptive weighting coefficient can be dynamically adjusted according to the current tracking error and then applied to the cost function evaluation of the predicted dq-axis current set to guide the controller in selecting the optimal switching state from multiple candidate states, thereby optimizing reactive power response performance. In this way, the tracking accuracy and dynamic response speed of the wind power converter to the grid signal can be significantly improved, thereby greatly enhancing the grid-connected adaptability and overall operational reliability of the wind farm.

[0053] Those skilled in the art will understand that the above embodiments are specific implementations of the present invention, and in practical applications, various changes can be made in form and detail without departing from the spirit and scope of the present invention.

Claims

1. A reactive power response control method for wind power converters based on model prediction, characterized in that, include: Obtain the instantaneous values ​​of the three-phase AC voltage at the PPC point, the instantaneous values ​​of the three-phase AC current at the converter outlet, the current active power reference value, and the current reactive power reference value; Based on the instantaneous values ​​of the three-phase AC voltage at point PPC, the instantaneous values ​​of the three-phase AC current at the converter outlet, the current active power reference value, and the current reactive power reference value, determine the dq-axis current at the current moment, the dq-axis reference current at the next moment, and the dq-axis voltage at the current moment. Fuzzy inference is performed on the current dq-axis current and the current dq-axis reference current to obtain adaptive weighting coefficients; Perform current prediction under multiple candidate states for the current dq-axis current and current dq-axis voltage at the current moment to obtain the predicted dq current set; Based on the dq-axis reference current and adaptive weight coefficients at the next time step, the cost function based on adaptive weights is evaluated on the predicted dq current set to obtain the total cost set of all candidate states. The current optimal state is determined based on the total value set of all candidate states.

2. The reactive power response control method for wind power converters based on model prediction according to claim 1, characterized in that, Based on the instantaneous values ​​of the three-phase AC voltage at point PPC, the instantaneous values ​​of the three-phase AC current at the converter outlet, the current active power reference value, and the current reactive power reference value, determine the dq-axis current at the current moment, the dq-axis reference current at the next moment, and the dq-axis voltage at the current moment, including: Input the instantaneous values ​​of the three-phase AC voltage at point PPC and the instantaneous values ​​of the three-phase AC current at the converter outlet into the phase-locked loop to obtain the grid voltage synchronization phase angle at the current moment; Based on the grid voltage synchronization phase angle at the current moment, the instantaneous values ​​of the three-phase AC voltage at point PPC and the instantaneous values ​​of the three-phase AC current at the converter outlet are transformed into the dq coordinate system to obtain the dq-axis voltage and the dq-axis current at the current moment. Based on the current active power reference value, the current reactive power reference value, and the current dq-axis voltage, determine the dq-axis reference current for the next moment.

3. The reactive power response control method for wind power converters based on model prediction according to claim 2, characterized in that, Based on the current grid voltage synchronization phase angle, the instantaneous values ​​of the three-phase AC voltage at point PPC and the instantaneous values ​​of the three-phase AC current at the converter outlet are transformed into dq coordinates to obtain the current dq-axis voltage and current. This includes transforming the instantaneous values ​​of the three-phase AC voltage at point PPC and the instantaneous values ​​of the three-phase AC current at the converter outlet into dq coordinates using the following formulas: ； in, This refers to the instantaneous value of the three-phase AC voltage at the PPC point or the instantaneous value of the three-phase AC current at the converter outlet. This represents the current dq-axis voltage or the current dq-axis current. This represents the current phase angle of the grid voltage synchronization.

4. The reactive power response control method for wind power converters based on model prediction according to claim 2, characterized in that, Based on the current active power reference value, the current reactive power reference value, and the current dq-axis voltage, determine the dq-axis reference current for the next moment, including: determining the dq-axis reference current for the next moment using the following formula: ； ； in, This is the current active power reference value. This is the current reactive power reference value. This represents the d-axis voltage in the dq-axis voltage at the current moment. Let d be the d-axis reference current in the dq-axis reference current at the next moment. This is the q-axis reference current in the dq-axis reference current at the next moment.

5. The reactive power response control method for wind power converters based on model prediction according to claim 1, characterized in that, Fuzzy inference is performed on the current dq-axis current and the current dq-axis reference current to obtain adaptive weighting coefficients, including: Calculate the original error between the current dq-axis current and the current dq-axis reference current to obtain the d-axis error and q-axis error; The d-axis error and q-axis error are aggregated by direction-sensitive weighted aggregation to obtain the aggregated error; The aggregation error is input into the fuzzy logic system to obtain the adaptive weight coefficients.

6. The reactive power response control method for wind power converters based on model prediction according to claim 5, characterized in that, The d-axis and q-axis errors are aggregated using a direction-sensitive weighted aggregation method to obtain the aggregated error. This includes performing the direction-sensitive weighted aggregation of the d-axis and q-axis errors using the following formula: ； ； in, It is the direction-sensitive factor at the current moment; It is a preset, positive-zero directional sensitivity coefficient; It is a very small positive number; It is the aggregation error at the current moment. This represents the d-axis error at the current moment. This represents the q-axis error at the current moment.

7. The reactive power response control method for wind power converters based on model prediction according to claim 6, characterized in that, Inputting the aggregation error into the fuzzy logic system to obtain the adaptive weight coefficients includes: the fuzzy logic system processing the aggregation error using the following formula to obtain the adaptive weight coefficients: ； in, This represents a complete fuzzy logic mapping function. This represents the adaptive weighting coefficient at the current moment.

8. The reactive power response control method for wind power converters based on model prediction according to claim 1, characterized in that, Perform current prediction under multiple candidate states for the current dq-axis current and current dq-axis voltage at the current moment to obtain a set of predicted dq currents, including: A discretized system model corresponding to the topology of the wind power converter is used as the prediction model; In each control cycle, all possible switching states of the wind power converter are traversed as candidate states. For each candidate state, the current dq-axis current and current dq-axis voltage are combined with the output voltage vector determined by the candidate state and substituted into the prediction model to calculate the corresponding dq-axis current at the next moment. The dq-axis currents at the next moment corresponding to all candidate states form the predicted dq current set.

9. The reactive power response control method for wind power converters based on model prediction according to claim 1, characterized in that, Based on the dq-axis reference current and adaptive weight coefficients at the next time step, a cost function based on adaptive weights is used to evaluate the predicted dq-current set to obtain the total cost set of all candidate states, including: For the dq-axis current at the next moment corresponding to each candidate state in the predicted dq current set, calculate the deviation between it and the dq-axis reference current at the next moment to obtain the current tracking error. Calculate the cost associated with switching losses for each candidate state to obtain the switching losses; For each candidate state, the corresponding current tracking error and the switching loss after adaptive weighting coefficient adjustment are weighted and summed to obtain the corresponding total value. The total values ​​corresponding to each candidate state form the total value set.

10. The reactive power response control method for wind power converters based on model prediction according to claim 1, characterized in that, Based on the total value set of all candidate states, the current optimal state is determined, including: The minimum total agency value is identified as the total agency value that has the smallest value in the total agency value set. Determine the candidate states corresponding to the minimum total value; The candidate state corresponding to the minimum total value is taken as the current optimal state.

Images