Wind power plant voltage control method considering influence of power grid voltage supporting capability

Abstract

The invention relates to a wind power plant voltage control method considering the influence of a power grid voltage supporting capability. The method comprises the following steps: acquiring a wind power plant reactive power output time sequence and a wind power plant active power change time sequence by using a wind turbine generator data acquisition and monitoring system based on DBSCAN spatial clustering; estimating a grid-connected point voltage prediction value at each moment, and calculating a deviation between the grid-connected point voltage prediction value at each moment and the grid-connected point voltage reference value by using a deep residual shrinkage network algorithm; on the basis, adjusting and controlling reactive power output and active power change of the wind power plant by utilizing rotor kinetic energy based on adaptive clutter suppression, so that the deviation between a grid-connected point voltage predicted value and a grid-connected point voltage reference value is minimum; according to the invention, automatic control of wind power fluctuation on grid-connected point voltage can be finally realized; the method has the advantages of improving the power grid stability, improving the power grid reliability and improving the power grid voltage regulation capability.

Description

Technical Field

[0001] The present invention belongs to the technical field of wind farm voltage control, and in particular relates to a wind farm voltage control method taking into account the influence of grid voltage support capability. Background Art

[0002] With the widespread application of wind energy as a clean and renewable energy source, the grid-connected operation of large-scale wind farms has become an important part of the modern power system. However, due to the uneven distribution of wind energy resources, a large number of large-scale wind farms are usually built in areas far away from the load center and connected to the grid terminal. The connection between this type of wind farm and the grid system is weak, lacking strong and effective voltage support from the system side, which is called a weakly connected sending-end system. In a weakly connected sending-end system, when the wind power fluctuates significantly in a short period of time, it is easy to cause drastic fluctuations in the voltage at the grid connection point. Such voltage fluctuations may lead to unstable operation of the grid and even trigger chain disconnection accidents, posing a serious threat to the safety and stability of the grid. Therefore, solving the voltage control problem of this type of system has become a severe challenge faced by the grid-connected operation of large-scale wind power. Therefore, it is very necessary to provide a wind farm voltage control method that takes into account the influence of the grid voltage support capacity to improve grid stability, grid reliability, and grid voltage regulation capability. Summary of the Invention

[0003] The purpose of the present invention is to overcome the deficiencies of the prior art and to provide a wind farm voltage control method that takes into account the influence of the grid voltage support capability, thereby improving grid stability, grid reliability, and grid voltage regulation capability.

[0004] The object of the present invention is achieved by providing a method for controlling wind farm voltage taking into account the influence of grid voltage support capability, the method comprising the following steps:

[0005] Step 1: Use the wind turbine data acquisition and monitoring system with a density-based spatial clustering DBSCAN model with noise to obtain the wind farm reactive output time series and the wind farm active power change time series;

[0006] Step 2: Estimate the predicted grid connection point voltage at each moment. Under the constraints that the grid connection point voltage operates between the grid connection point voltage lower limit and the grid connection point voltage upper limit, and the wind farm reactive output is between the wind farm reactive output lower limit and the wind farm reactive output upper limit, use the deep residual shrinkage network algorithm to calculate the deviation between the grid connection point voltage predicted value at each moment and the grid connection point voltage reference value.

[0007] Step 3: Use rotor kinetic energy regulation based on adaptive clutter suppression to control the reactive output of the wind farm and the change of the active power of the wind farm so that the deviation between the grid connection point voltage prediction value and the grid connection point voltage reference value is minimized.

[0008] The step 1 specifically includes the following steps:

[0009] Step 1.1: Conduct preliminary rule-based screening: Delete obviously abnormal data based on the following rules: ① Wind speed is not greater than 0; ② Active power of wind turbine is not greater than 0; ③ Main shaft speed of wind turbine is not greater than 0;

[0010] Step 1.2: After the initial screening, DBSCAN clustering is performed: To eliminate the impact of the different dimensions of wind speed and power, the data is first Z-Score regularized according to the following formula: In the formula, x represents the original data to be standardized; μ represents the mean value of the original data; σ represents the standard deviation of the original data; z represents the result after standardization;

[0011] Step 1.3: Substitute the candidate parameter combinations into the DBSCAN model to obtain the clustered results, and use the clustered data to calculate the prediction error e pn and classification accuracy a c , arrange the parameter combinations in the order of increasing prediction error, and select the parameter combination corresponding to the first maximum value of classification accuracy as the optimal clustering parameter;

[0012] Step 1.4: Use the optimized data acquisition and monitoring system to obtain the wind farm reactive output time series and wind farm active power change time series.

[0013] In step 2, the layered model predictive control method is used to estimate the grid connection point voltage prediction value at each moment. Specifically, in the grid-connected adaptive regulation layer, the voltage at the wind farm grid connection point is easily affected by the power fluctuation on the electric field side. The voltage deviation ΔV caused by the power changes ΔP and ΔQ is: In the formula, the variable Indicates the active voltage sensitivity coefficient of the grid connection point; variable Indicates the reactive voltage sensitivity coefficient of the grid connection point; ① The greater the active output of the wind farm, the more likely it is to cause reactive power shortage. When optimizing the active output, the maximum active power prediction value should be determined first. And the upper and lower limits of the voltage regulation at the corresponding time and And the predicted voltage at the grid connection point ② In the case of reactive power shortage, in order to achieve the upper or lower limit of the limit voltage regulation and the grid voltage regulation instruction The control goal is to minimize the deviation. This layer aims to determine the maximum value of the active power prediction. The active power output that needs to be adjusted is established based on the objective function as follows: minF1=||ΔV pre || 2 , where ΔV pre The upper or lower limit of the limit voltage regulation ③ The maximum value of active power prediction At , the upper and lower limits of the wind farm’s reactive output can be obtained from the PQ function relationship of the wind turbine: Among them, f Q is the reactive power output function of the wind farm; the expressions for the upper and lower limits of the grid-connected voltage are: in, is the grid connection point voltage function.

[0014] The hierarchical model predictive control method includes a reactive power coordination and allocation layer and a tracking control layer. The reactive power coordination and allocation layer prediction model is constructed based on the dynamic process of wind turbine reactive power response. The control input of the prediction model is the reactive power change ΔQ(k) of the wind farm at time k, and the state variable is the measured voltage V at the grid connection point. PCC , the measured voltage of the collection busbar V CB And the reactive output Q of the static VAR generator SVG S , the disturbance information is the active power change of the wind farm, and the prediction model between the control input and the state variable output is: Where x(k) is as follows: Where x(k) represents the predicted change sequence of the grid connection point, the collection bus, and the reactive power output of the SVG at the next H time points starting from time k; y(k) represents the predicted value of the reactive power output of the grid connection point, the collection bus, and the SVG at time k; A, B, C, and D are coefficient matrices. The voltage control of the wind farm aims to minimize the deviation between the node voltage and the corresponding reference value, while also ensuring that the SVG has a sufficient reactive power reserve. Based on this, the objective function of the reactive power coordination and allocation layer is constructed as follows: Where ΔV PCC , ΔV CB , ΔQ S are the deviations between the grid connection point voltage, the busbar voltage, and the SVG reactive output and the corresponding reference values; W PCC 、W CB 、W S is the weight coefficient corresponding to each deviation, satisfying W PCC >W CB >W S ; H represents the prediction time domain; considering the influence of active power changes, corresponding to H time points in a control cycle, the expression of the grid connection point voltage prediction value under the combined action of active power and reactive power of the wind farm is: Where, It represents the reactive response coefficient of the wind turbine at the Hth time point in a reactive command cycle; Indicates the predicted value of the grid connection point voltage at the prediction start time.

[0015] The voltage deviation of the grid connection point and the collection bus in the reactive power coordination and distribution layer prediction model satisfies:

[0016] Where ΔP(k) is the change in active power of the wind farm at time k; ΔQ(k) is the change in reactive power of the wind farm at time k; variable Indicates the busbar active voltage sensitivity coefficient; variable Indicates the reactive voltage sensitivity coefficient of the collection bus; is the bus reference voltage; the objective function of the reactive power coordination distribution layer prediction model satisfies the following constraints: Where, They are the upper and lower limits of the grid connection point voltage respectively; are the upper and lower limits of the bus voltage respectively; Q max , Q min They are the upper and lower limits of reactive output of wind farms respectively.

[0017] The tracking control layer of the hierarchical model predictive control method is specifically as follows: selecting the stator and rotor d-axis and q-axis current components [i sd i sq i rd i rq ] T As the state variable x, the rotor d-axis and q-axis voltage components [u rd u rq ] T As the control input u r , stator d-axis and q-axis voltage components [u sd u sq ] T As an external measurable disturbance u s , after discretization, the fan state space expression is as follows: Where A1, B1, C1, and D1 are coefficient matrices. The wind turbine operation mainly tracks the active and reactive output instructions. After power decoupling, the active and reactive power can be represented by the rotor dq axis current components, respectively. Based on this, the rotor current reference output y can be obtained. ref , and then, with the goal of minimizing both the control output and the control action change amplitude, the objective function of the unit control is established as follows: Where W y is the control output weight coefficient; W u The incremental weight coefficient for control input; y p (k+m|k) is the predicted control output of the mth step; Δu r (k+m|k) is the control input increment; M represents the prediction time domain.

[0018] The deep residual shrinkage network algorithm in step 2 is specifically: using soft thresholding to achieve noise reduction in signal processing, and the expression of the soft threshold function is as follows: Among them, d in is the input; d out is the output; τ is the threshold, which is a constant positive number.

[0019] According to the aerodynamic principle in step 3, the mechanical power captured by the wind wheel is: Where ρ is the air density; R is the radius of the fan blade; v is the wind speed; C p (λ, β) is the wind energy utilization coefficient, which is a nonlinear function of the tip speed ratio λ and the pitch angle β: Where ω is the mechanical angular velocity of the wind turbine rotor. When the wind turbine is operating below the rated wind speed, in order to improve the utilization of wind energy, the pitch angle β is adjusted to 0°. When the wind speed changes, the generator rotor speed is changed to keep λ at the optimal tip speed ratio λ. opt , which can make the wind energy utilization coefficient C p Constant at the maximum value C p,max At this time, the maximum wind power P that the wind turbine can capture max for: Therefore, the optimal output power P of the wind turbine characterized by the speed is opt It can be expressed as: Where k opt is the optimal power proportional coefficient.

[0020] The second-order filter introduced in step 3 adopts a two-pole-one-zero architecture, and its transfer function expression can be written as: Where a, b, and c are filter parameters. After the second-order filter is introduced, the wind turbine output active reference value P ref2 becomes: Similarly, the transfer function from wind speed to rotation speed can be derived as: Where J is the moment of inertia of the system including the wind rotor and generator; ω0 is the mechanical angular velocity of the rotor at the typical optimal power point selected as the reference operating point; based on the second-order filter relationship between output power and the cube of the speed, the wind speed to power transfer function can be derived as:

[0021] In step 3, in order to minimize the deviation between the grid connection point voltage prediction value and the grid connection point voltage reference value, the filter parameters need to be adjusted online based on fuzzy control. The improved second-order filtering method uses a fuzzy logic controller to correct the filter coefficient online. The fuzzy logic controller does not rely on the precise mathematical model of the controlled object and can simulate human operating experience to adjust the parameters online.

[0022] Since wind speed is difficult to measure accurately, a large wind speed disturbance event is defined by measuring the speed based on the consistency of the rotational speed and wind speed change trends: when the rotational speed increases at a large acceleration for a period of time, it is judged that a large wind speed disturbance increase event has occurred; similarly, it can be judged that a large wind speed disturbance decreases; in actual control, the speed difference between adjacent sampling moments is usually used as the input acceleration signal; here, it is necessary to emphasize the asymmetry of the rotor kinetic energy control: when the filter is introduced, it will cause speed overshoot, and when the wind speed increases significantly, it is hoped that the speed will be fully increased to store kinetic energy. Therefore, the hysteresis comparator output is 1; a significant decrease in wind speed is a dangerous signal, and there is a risk of excessive deceleration and instability. Therefore, the hysteresis loop returns a value of 0.

[0023] Beneficial effects of the present invention: The present invention is a wind farm voltage control method taking into account the influence of the voltage support capacity of the power grid. During use, the present invention obtains the wind farm reactive output time series and the wind farm active power change time series through the wind turbine data acquisition and monitoring system based on the improved DBSCAN, adopts the prediction error and classification accuracy to select the key clustering parameters neighborhood radius and the minimum number of neighborhood sample points, and the parameter selection process is fully automated, realizing the effective identification of abnormal data of the wind turbine data acquisition and monitoring system; under the voltage support mechanism between the wind farm and the power grid, the upper and lower limits of the reactive power of the wind farm are determined, the voltage support capacity of the external power grid is effectively considered, and the connection and coordination relationship between the voltage support capacity of wind power and the power grid is clarified; on this basis, the predicted wind power The rate is well incorporated into the wind farm voltage control based on hierarchical model predictive control, and the deviation between the grid connection point voltage prediction value and the reference value at each moment is calculated by the deep residual shrinkage network algorithm, thereby improving the stability and effectiveness of the wind power grid connection point voltage; finally, the rotor kinetic energy regulation based on adaptive clutter suppression controls the wind farm reactive output and the wind farm active power change, and the improved second-order filtering algorithm is used to monitor and analyze the state and voltage change of the power grid, so as to adjust the operating parameters of the wind farm in time, thereby more effectively responding to the power grid demand and maintaining voltage stability when the power grid load changes, thereby realizing effective control of the wind power grid connection point voltage and automatic control of the wind power fluctuation on the grid connection point voltage; the present invention has the advantages of improving the stability of the power grid, improving the reliability of the power grid, and improving the voltage regulation capability of the power grid. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Flowchart of the present invention.

[0025] Figure 2 This is a flow chart of abnormal data identification in the wind turbine data acquisition and monitoring system of the present invention.

[0026] Figure 3 This is a flow chart of fan control based on hierarchical model prediction according to the present invention.

[0027] Figure 4 This is a basic module structure diagram of the deep residual shrinkage network of the present invention. DETAILED DESCRIPTION

[0028] The present invention will be further described below with reference to the accompanying drawings.

[0029] Example 1

[0030] like Figure 1 As shown, a method for controlling wind farm voltage taking into account the influence of grid voltage support capability is provided, the method comprising the following steps:

[0031] Step 1: Use the wind turbine data acquisition and monitoring system with a density-based spatial clustering DBSCAN model with noise to obtain the wind farm reactive output time series and the wind farm active power change time series;

[0032] Step 2: Estimate the predicted grid connection point voltage at each moment. Under the constraints that the grid connection point voltage operates between the grid connection point voltage lower limit and the grid connection point voltage upper limit, and the wind farm reactive output is between the wind farm reactive output lower limit and the wind farm reactive output upper limit, use the deep residual shrinkage network algorithm to calculate the deviation between the grid connection point voltage predicted value at each moment and the grid connection point voltage reference value.

[0033] Step 3: Use rotor kinetic energy regulation based on adaptive clutter suppression to control the reactive output of the wind farm and the change of the active power of the wind farm so that the deviation between the grid connection point voltage prediction value and the grid connection point voltage reference value is minimized.

[0034] The present invention is a wind farm voltage control method taking into account the influence of the voltage support capacity of the power grid. In use, the present invention obtains the wind farm reactive output time series and the wind farm active power change time series through the wind turbine data acquisition and monitoring system based on the improved DBSCAN, adopts the prediction error and classification accuracy to select the key clustering parameters neighborhood radius and the minimum number of neighborhood sample points, and the parameter selection process is fully automated, realizing the effective identification of abnormal data of the wind turbine data acquisition and monitoring system; under the voltage support mechanism between the wind farm and the power grid, the upper and lower limits of the reactive power of the wind farm are determined, the voltage support capacity of the external power grid is effectively considered, and the connection and coordination relationship between the voltage support capacity of wind power and the power grid is clarified; on this basis, the predicted wind power is well It is incorporated into the wind farm voltage control based on hierarchical model predictive control, and the deep residual shrinkage network algorithm is used to calculate the deviation between the grid connection point voltage prediction value and the reference value at each moment, thereby improving the stability and effectiveness of the wind power grid connection point voltage; finally, the rotor kinetic energy regulation based on adaptive noise suppression controls the wind farm reactive output and the wind farm active power change, and the improved second-order filtering algorithm is used to monitor and analyze the state and voltage change of the power grid, so as to adjust the operating parameters of the wind farm in time, thereby more effectively responding to the power grid demand and maintaining voltage stability when the power grid load changes, thereby realizing effective control of the wind power grid connection point voltage and automatic control of the wind power fluctuation on the grid connection point voltage; the present invention has the advantages of improving the stability of the power grid, improving the reliability of the power grid, and improving the voltage regulation capability of the power grid.

[0035] Example 2

[0036] like Figure 1-4 As shown, a method for controlling wind farm voltage taking into account the influence of grid voltage support capability is provided, the method comprising the following steps:

[0037] Step 1: Use the wind turbine data acquisition and monitoring system with a noise-based density-based spatial clustering DBSCAN model to obtain the wind farm reactive power output time series and the wind farm active power variation time series. The wind turbine data acquisition and monitoring system based on the improved DBSCAN can effectively identify the large number of abnormal data records contained in the original wind turbine data acquisition and monitoring system, thereby improving the accuracy of the collected data of the wind farm reactive power output time series and the wind farm active power variation time series.

[0038] In this embodiment, the present invention uses a wind turbine data acquisition and monitoring system based on DBSCAN spatial clustering to obtain the reactive output time series and the active power change time series of the wind farm; starting from the characteristics of the wind speed-power scatter plot, the present invention uses the prediction error and classification accuracy to select the key clustering parameters neighborhood radius and the minimum number of neighborhood sample points, avoiding the subjectivity of manually determining the clustering parameters, and the parameter selection process can be fully automated, while ensuring that abnormal data is eliminated, retaining as much normal data as possible, such as Figure 2 As shown, it specifically includes the following sub-steps:

[0039] Step 101: Perform preliminary rule-based screening: In theory, when the proportion of normal monitoring data in the total monitoring data is much greater than the proportion of abnormal data, the DBSCAN clustering-based anomaly identification method can easily find "normal" patterns in the data and thus eliminate outliers. The proportion of various types of abnormal data in wind turbine monitoring data is not low, especially the bottom accumulation type, which is adjacent to the normal data area. It is difficult to directly use the outlier detection method on all monitoring data to process this type of accumulation data. Therefore, it is necessary to filter out obviously abnormal data to reduce the impact of this data on the accuracy of subsequent clustering-based outlier identification.

[0040] Delete obviously abnormal data based on the following rules: ① Wind speed is not greater than 0; ② Active power of wind turbine is not greater than 0; ③ Main shaft speed of wind turbine is not greater than 0;

[0041] Step 102: After the initial screening, DBSCAN clustering is performed: To eliminate the impact of the different dimensions of wind speed and power, the data is first Z-Score regularized according to the following formula: In the formula, x represents the original data to be standardized; μ represents the mean value of the original data; σ represents the standard deviation of the original data; z represents the result after standardization;

[0042] Step 103: Substitute the candidate parameter combinations into the DBSCAN model to obtain the clustered results, and use the clustered data to calculate the prediction error e pn and classification accuracy a c , arrange the parameter combinations in the order of increasing prediction error, and select the parameter combination corresponding to the first maximum value of classification accuracy as the optimal clustering parameter;

[0043] Step 104: Utilize the optimized data acquisition and monitoring system to obtain the wind farm reactive output time series and the wind farm active power variation time series.

[0044] The indicators for selecting the optimal clustering parameters in the present invention are as follows: ① Prediction error e pn: Train a regression model. Both the training set and the test set are sample points clustered into normal categories by DBSCAN. The input is the wind speed of the sample point, and the output is the corresponding power. pn is the prediction error of the prediction model; the normal operation monitoring data of the wind turbine are more closely distributed around the ideal power curve, indicating that there is a specific mapping relationship between the wind speed and power of the normal data; the closer the data in the normal category is to the ideal power characteristic curve, the higher the proportion of normal data in the normal category, and the wind speed-power mapping relationship of the samples in the normal category is more "similar". Under the same regression model, the prediction error e pn will be smaller;

[0045] ②Classification accuracy a c : Train a classification model, where the input is wind speed and power, and the output is the category of the data. During training, it is assumed that the category after DBSCAN clustering (normal or abnormal) is the "true" label of the data. c is the classification accuracy of the classification model; when the clustering result of the clustering model is more accurate, it means that the “real” label of the data is also more accurate. When the classification model is the same, the classification accuracy a c It will also be higher.

[0046] Step 2: Estimate the predicted grid connection point voltage at each moment. Under the constraints that the grid connection point voltage operates between the grid connection point voltage lower limit and the grid connection point voltage upper limit, and the wind farm reactive output is between the wind farm reactive output lower limit and the wind farm reactive output upper limit, use the deep residual shrinkage network algorithm to calculate the deviation between the grid connection point voltage predicted value at each moment and the grid connection point voltage reference value.

[0047] In this embodiment, the present invention uses a hierarchical model predictive control method to estimate the grid connection point voltage prediction value at each moment. In the grid-connected adaptive regulation layer, the voltage at the wind farm grid connection point is easily affected by the power fluctuation on the electric field side. The voltage deviation ΔV caused by the power changes ΔP and ΔQ is: In the formula, the variable Indicates the active voltage sensitivity coefficient of the grid connection point; variable Indicates the reactive voltage sensitivity coefficient of the grid connection point;

[0048] The greater the active power output of a wind farm, the more likely it is to cause reactive power shortages. When optimizing the active power output, the maximum active power prediction value should be determined first. And the upper and lower limits of the voltage regulation at the corresponding time and And the predicted voltage at the grid connection point

[0049] In the case of reactive power shortage, in order to achieve the upper or lower limit of the voltage regulation and the voltage regulation instruction of the grid connection point The control goal is to minimize the deviation. This layer aims to determine the maximum value of the active power prediction. The active power output that needs to be adjusted is established based on the objective function as follows: minF1=||ΔV pre || 2 , where ΔV pre The upper or lower limit of the limit voltage regulation Deviation between

[0050] At the maximum active power prediction At , the upper and lower limits of the wind farm’s reactive output can be obtained from the PQ function relationship of the wind turbine: Among them, f Q is the reactive power output function of the wind farm; the expressions for the upper and lower limits of the grid-connected voltage are: in, is the grid connection point voltage function.

[0051] The hierarchical model prediction control method of the present invention includes a reactive coordination distribution layer and a tracking control layer, wherein the reactive coordination distribution layer prediction model is constructed based on the dynamic process of the wind turbine reactive response. The control input of the prediction model is the reactive change ΔQ(k) of the wind farm at time k, and the state variable is the measured voltage V at the grid connection point. PCC , the measured voltage of the collection busbar V CB And the reactive output Q of the static VAR generator SVG S , the disturbance information is the active power change of the wind farm, and the prediction model between the control input and the state variable output is: Where x(k) is as follows: Where x(k) represents the predicted change sequence of the grid connection point, the collection bus, and the reactive power output of the SVG at the next H time points starting from time k; x(k+1) represents the predicted change sequence of the grid connection point, the collection bus, and the reactive power output of the SVG at the next H time points starting from time k+1; y(k) represents the predicted value of the grid connection point, the collection bus, and the reactive power output of the SVG at time k; A, B, C, and D are coefficient matrices, as follows: C=[I 3×3 0 … 0] 3×(3·H) , Where, The reactive response coefficient of the wind farm at each smaller time point in a control cycle; the variable Indicates the reactive voltage sensitivity coefficient of the grid connection point; variable Indicates the busbar reactive voltage sensitivity coefficient; variable Indicates the reactive voltage sensitivity coefficient of the grid connection point relative to the SVG node;

[0052] Wind farm voltage control aims to minimize the deviation between node voltage and the corresponding reference value, while also ensuring that the SVG has sufficient reactive power reserves. Based on this, the objective function of the reactive power coordination and distribution layer is constructed: Where ΔV PCC , ΔV CB , ΔQ S are the deviations between the grid connection point voltage, the busbar voltage, and the SVG reactive output and the corresponding reference values; W PCC 、W CB 、W S is the weight coefficient corresponding to each deviation, satisfying W PCC >W CB >W S ; H represents the prediction time domain; the voltage deviation of the grid connection point and the collection bus satisfies:

[0053] Where ΔP(k) is the change in active power of the wind farm at time k; ΔQ(k) is the change in reactive power of the wind farm at time k; variable Indicates the busbar active voltage sensitivity coefficient; variable Indicates the reactive voltage sensitivity coefficient of the collection bus; is the bus reference voltage;

[0054] The objective function satisfies the following constraints: Where, They are the upper and lower limits of the grid connection point voltage respectively; are the upper and lower limits of the bus voltage respectively; Q max , Q min are the upper and lower limits of reactive output of wind farms respectively;

[0055] Considering the influence of active power variation, corresponding to H time points in a control cycle, the expression of the grid connection point voltage prediction value under the combined effect of wind farm active power and reactive power is: Where, It represents the reactive response coefficient of the wind turbine at the Hth time point in a reactive command cycle; Indicates the predicted value of the grid connection point voltage at the prediction start time.

[0056] The tracking control layer of the hierarchical model predictive control method of the present invention selects the stator and rotor d-axis and q-axis current components [i sd i sq i rd i rq ] T As the state variable x, the rotor d-axis and q-axis voltage components [u rd u rq ] T As the control input ur , stator d-axis and q-axis voltage components [u sd u sq ] T As an external measurable disturbance u s ;

[0057] After discretization, the state space expression of the fan is as follows: Where A1, B1, C1, and D1 are coefficient matrices, as follows:

[0058] ω1=pω m ±ω2, Among them, R s Represents stator resistance; R r Indicates rotor resistance; L s Represents stator inductance; L r Indicates rotor inductance; L m represents the mutual inductance between the stator and rotor; ω1 represents the power frequency of 50Hz; ω m represents the rotor mechanical speed; ω2 represents the rotor current angular frequency; p represents the number of pole pairs of the doubly fed machine; t represents the sampling period;

[0059] The wind turbine operation mainly tracks the active and reactive output instructions. After power decoupling, the active and reactive power can be represented by the rotor dq axis current components respectively, based on which the rotor current reference output y can be obtained. ref , and then, with the goal of minimizing both the control output and the control action change amplitude, the objective function of the unit control is established as follows: Where W y is the control output weight coefficient; W u The incremental weight coefficient for control input; y p (k+m|k) is the predicted control output of the mth step; Δu r (k+m|k) is the control input increment; M represents the prediction time domain.

[0060] Combining the above state equation and objective function, the specific fan control process is as follows: Figure 3 As shown in the figure, through vertical stratification, adaptive adjustments to active power output are made in a timely manner over long time scales, addressing potential reactive power shortages in advance. Furthermore, the reactive power capacity of wind turbines is fully utilized through coordinated reactive power distribution, and the reactive power exchange between wind turbines and SVGs enables the SVGs to reserve more rapid reactive power reserves for potential voltage fluctuations. Finally, based on feedback data and reference data, and based on real-time tracking control, the normal and stable operation of the wind farm's grid-connected voltage is guaranteed even when the grid-side support is weak.

[0061] The deep residual shrinkage network in the present invention is essentially a network integrated with a deep residual network, an attention mechanism, and a soft threshold function. The deep residual network stacks multiple residual unit modules in the main part of the network to deepen the number of network layers. When the deep residual network performs model training based on backpropagation, its loss can not only be backpropagated layer by layer through convolutional layers, but also can be more conveniently backpropagated through identity mapping, making it easier to train and obtain a better model.

[0062] Soft thresholding is a commonly used noise reduction method in signal processing. The expression of the soft threshold function is as follows: Among them, d in is the input; d out is the output; τ is the threshold, which is a constant positive number. Its basic principle is to first decompose the input signal, then filter the decomposed signal through the soft threshold function, and finally reconstruct the signal. As can be seen from the above formula, it sets the features with absolute values ​​lower than τ to 0, and adjusts other features towards 0, that is, achieving a "shrinkage" effect.

[0063] The threshold τ in the soft threshold function is crucial, as its value directly impacts the noise reduction results. The deep residual shrinkage network incorporates an attention mechanism to automatically set the threshold, avoiding the uncertainty associated with manual threshold setting. The attention mechanism focuses on key local information, automatically learning a set of weights through a small subnetwork to weight each feature channel, thereby enhancing useful information while suppressing redundant information.

[0064] The residual shrinkage module is the core of the deep residual shrinkage network, which retains the identity mapping in the residual module and adds an attention module and soft threshold after the two convolutional hidden layers. Its structure is as follows Figure 4 As shown in the figure, W and H are the width and height of the feature map respectively; C in 、C out are the number of input feature map channels and the number of output feature map channels, respectively; the threshold τ obtained by the subnet is a set of vectors whose number of elements is equal to the number of feature map channels C2 output by the second convolutional hidden layer; multiplying each element by the feature map of the corresponding channel is to apply different sizes of attention to each feature channel, and then shrinking and denoising is performed through the soft threshold function; stacking the convolution pooling layer and the residual shrinkage module forms a deep residual shrinkage network.

[0065] Step 3: Use rotor kinetic energy regulation based on adaptive clutter suppression to control the reactive output of the wind farm and the change of the active power of the wind farm so that the deviation between the grid connection point voltage prediction value and the grid connection point voltage reference value is minimized.

[0066] In this embodiment, the present invention utilizes rotor kinetic energy regulation based on adaptive clutter suppression to control wind farm reactive output and wind farm active power variation, thereby minimizing the deviation between the grid connection point voltage prediction value and the grid connection point voltage reference value. An improved second-order filtering algorithm is applied to monitor and analyze the state of the power grid and voltage variations, so as to timely adjust the operating parameters of the wind farm. By optimizing the reactive output and active power variation of the wind farm, the system can more effectively respond to the needs of the power grid and maintain voltage stability when the grid load changes. The kinetic energy of the wind farm is used to balance the power demand of the power grid, thereby minimizing voltage deviation and improving the operating efficiency and reliability of the power grid.

[0067] According to the aerodynamic principles of the present invention, the mechanical power captured by the wind wheel is: Where ρ is the air density; R is the radius of the fan blade; v is the wind speed; C p (λ, β) is the wind energy utilization coefficient, which is a nonlinear function of the tip speed ratio λ and the pitch angle β: Where ω is the mechanical angular velocity of the wind turbine rotor;

[0068] When the wind turbine is operating below the rated wind speed, in order to improve the utilization rate of wind energy, the pitch angle β is adjusted to 0°, and the generator rotor speed is changed when the wind speed changes, so that λ is maintained at the optimal tip speed ratio λ. opt , which can make the wind energy utilization coefficient C p Constant at the maximum value C p,max At this time, the maximum wind power P that the wind turbine can capture max for: Therefore, the optimal output power P of the wind turbine characterized by the speed is opt It can be expressed as: Where k opt is the optimal power proportional coefficient, which is a constant related only to the wind turbine parameters.

[0069] The second-order filter introduced in the present invention adopts a two-pole-one-zero architecture, and its transfer function expression can be written as: Where a, b, and c are filter parameters;

[0070] After the introduction of the second-order filter, the wind turbine output active reference value P ref2 becomes: Similarly, the transfer function from wind speed to rotation speed can be derived as: Where J is the system moment of inertia including the wind wheel and generator; ω0 is the mechanical angular velocity of the rotor at the typical optimal power point selected as the reference operating point;

[0071] According to the second-order filtering relationship between output power and the cube of speed, the transfer function from wind speed to power can be derived as: According to the second-order filter G2nd The amplitude-frequency characteristics of (s) itself have three transition frequencies: ω1 = b, ω2 = ab / c, ω3 = a; the configuration principles of filter parameters a, b, and c are as follows:

[0072] ① To fully demonstrate the discrimination of the frequency band selection of the second-order filtering method, the denominator corner frequencies must satisfy b<<a. The b value needs to be small to attenuate power fluctuations in the low-frequency band. However, a parameter b that is too small will lead to excessive attenuation of the overall output power amplitude, resulting in reduced wind turbine benefits. Therefore, it is also necessary to balance the wind energy capture efficiency, and finally b=0.05 is selected.

[0073] ②ω3=a is the final turning frequency point, which is responsible for filtering out high-frequency fluctuations. Therefore, a should not be too large and generally needs to be fixed. If the goal is to enhance the power smoothing effect above 0.5Hz, the value of a should be 0.5×2π≈3.14;

[0074] ③ A certain bandwidth needs to be maintained between the turning frequency points ω2 = ab / c and ω3 = a to ensure the tracking accuracy of the wind speed input trend, thereby ensuring the wind energy capture efficiency while dynamically smoothing the power; the parameter c is defined as a proportional relationship with a + b: c = K c (a+b), where K c is the proportional coefficient, and its domain range is [0.3,0.5].

[0075] In order to minimize the deviation between the grid connection point voltage prediction value and the grid connection point voltage reference value, the filter parameters need to be adjusted online based on fuzzy control. The basic principle of rotor kinetic energy smoothing fluctuations is that when the wind speed increases, the rotor speed increases sufficiently to store kinetic energy, and when the wind speed decreases, the rotor decelerates to release energy for compensation. Therefore, according to the fluctuation of wind speed amplitude, parameters with stronger smoothing effect can be used when the wind speed is greatly disturbed, while parameters with weaker smoothing effect can be used for control when the wind speed disturbance is not large. The improved second-order filtering method uses a fuzzy logic controller to perform online correction of the filter coefficient. The fuzzy logic controller does not rely on the precise mathematical model of the controlled object and can simulate human operating experience for online parameter adjustment. The characteristics of fuzzy control are as follows:

[0076] ① When no major disturbance occurs, K c Maintain the initial value; when a large disturbance is detected in the wind speed, that is, when the hysteresis signal is 1, the parameter K c Adjust to 0.3 to allow the rotor to fully accelerate and store kinetic energy;

[0077] ② When a large disturbance of wind speed is detected, conservative parameters are used, that is, K c Corrected back to the initial value 0.5;

[0078] ③ When the rotor speed is about to reach the upper limit of operation, the output is appropriately reduced according to the current hysteresis signal; when in the process of large disturbance rising, K is lowered. c When in the rising process without large disturbance, it returns to the initial value of 0.5 to avoid the electromagnetic power mutation caused by speed limit exceeding.

[0079] Since wind speed is difficult to measure accurately, a large wind speed disturbance event is defined by measuring the speed based on the consistency of the rotational speed and wind speed change trends: when the rotational speed increases at a large acceleration for a period of time, it is judged that a large wind speed disturbance increase event has occurred; similarly, it can be judged that a large wind speed disturbance decreases; in actual control, the speed difference between adjacent sampling moments is usually used as the input acceleration signal, among which the threshold for judging acceleration and duration can be selected according to experience and actual control requirements; here, it is necessary to emphasize the asymmetry of rotor kinetic energy control: when the filter is introduced, it will cause speed overshoot. When the wind speed increases significantly, it is hoped that the speed will be fully increased to store kinetic energy. Therefore, the hysteresis comparator output is 1; and a significant decrease in wind speed is a dangerous signal, and there is a risk of excessive deceleration and instability. Therefore, the hysteresis loop returns a value of 0.

[0080] The present invention is a wind farm voltage control method taking into account the influence of the voltage support capacity of the power grid. In use, the present invention obtains the wind farm reactive output time series and the wind farm active power change time series through the wind turbine data acquisition and monitoring system based on the improved DBSCAN, adopts the prediction error and classification accuracy to select the key clustering parameters neighborhood radius and the minimum number of neighborhood sample points, and the parameter selection process is fully automated, realizing the effective identification of abnormal data of the wind turbine data acquisition and monitoring system; under the voltage support mechanism between the wind farm and the power grid, the upper and lower limits of the reactive power of the wind farm are determined, the voltage support capacity of the external power grid is effectively considered, and the connection and coordination relationship between the voltage support capacity of wind power and the power grid is clarified; on this basis, the predicted wind power is well It is incorporated into the wind farm voltage control based on hierarchical model predictive control, and the deep residual shrinkage network algorithm is used to calculate the deviation between the grid connection point voltage prediction value and the reference value at each moment, thereby improving the stability and effectiveness of the wind power grid connection point voltage; finally, the rotor kinetic energy regulation based on adaptive noise suppression controls the wind farm reactive output and the wind farm active power change, and the improved second-order filtering algorithm is used to monitor and analyze the state and voltage change of the power grid, so as to adjust the operating parameters of the wind farm in time, thereby more effectively responding to the power grid demand and maintaining voltage stability when the power grid load changes, thereby realizing effective control of the wind power grid connection point voltage and automatic control of the wind power fluctuation on the grid connection point voltage; the present invention has the advantages of improving the stability of the power grid, improving the reliability of the power grid, and improving the voltage regulation capability of the power grid.

Claims

1. A method for controlling wind farm voltage taking into account the influence of grid voltage support capability, characterized by: The method comprises the following steps: Step 1: Use the wind turbine data acquisition and monitoring system with a density-based spatial clustering DBSCAN model with noise to obtain the wind farm reactive output time series and the wind farm active power change time series; Step 2: Estimate the predicted grid connection point voltage at each moment. Under the constraints that the grid connection point voltage operates between the grid connection point voltage lower limit and the grid connection point voltage upper limit, and the wind farm reactive output is between the wind farm reactive output lower limit and the wind farm reactive output upper limit, use the deep residual shrinkage network algorithm to calculate the deviation between the grid connection point voltage predicted value at each moment and the grid connection point voltage reference value. Step 3: Use rotor kinetic energy regulation based on adaptive clutter suppression to control the reactive output of the wind farm and the change of the active power of the wind farm so that the deviation between the grid connection point voltage prediction value and the grid connection point voltage reference value is minimized.

2. The method for controlling wind farm voltage taking into account the influence of grid voltage support capability according to claim 1, characterized in that: The step 1 specifically includes the following steps: Step 1.1: Conduct preliminary rule-based screening: Delete obviously abnormal data based on the following rules: ① Wind speed is not greater than 0; ② Active power of wind turbine is not greater than 0; ③ Main shaft speed of wind turbine is not greater than 0; Step 1.2: After the initial screening, DBSCAN clustering is performed: To eliminate the impact of the different dimensions of wind speed and power, the data is first Z-Score regularized according to the following formula: In the formula, x represents the original data to be standardized; μ represents the mean value of the original data; σ represents the standard deviation of the original data; z represents the result after standardization; Step 1.3: Substitute the candidate parameter combinations into the DBSCAN model to obtain the clustered results, and use the clustered data to calculate the prediction error e pn and classification accuracy a c , arrange the parameter combinations in the order of increasing prediction error, and select the parameter combination corresponding to the first maximum value of classification accuracy as the optimal clustering parameter; Step 1.4: Use the optimized data acquisition and monitoring system to obtain the wind farm reactive output time series and wind farm active power change time series.

3. The method for controlling wind farm voltage taking into account the influence of grid voltage support capability according to claim 1, characterized in that: In step 2, the voltage prediction value of the grid connection point at each moment is estimated using the hierarchical model predictive control method. Specifically, in the grid-connected adaptive regulation layer, the voltage at the wind farm grid connection point is easily affected by the power fluctuation on the electric field side. The voltage deviation ΔV caused by the power changes ΔP and ΔQ is: In the formula, the variable Indicates the active voltage sensitivity coefficient of the grid connection point; variable Indicates the reactive voltage sensitivity coefficient of the grid connection point; ① The greater the active output of the wind farm, the more likely it is to cause reactive power shortage. When optimizing the active output, the maximum active power prediction value should be determined first. And the upper and lower limits of the voltage regulation at the corresponding time and And the predicted voltage at the grid connection point ② In the case of reactive power shortage, in order to achieve the upper or lower limit of the limit voltage regulation and the grid voltage regulation instruction The control goal is to minimize the deviation. This layer aims to determine the maximum value of the active power prediction. The active power output that needs to be adjusted is established based on the objective function as follows: min F1=||ΔV pre || 2 , where ΔV pre The upper or lower limit of the limit voltage regulation ③ The maximum value of active power prediction At , the upper and lower limits of the wind farm’s reactive output can be obtained from the PQ function relationship of the wind turbine: Among them, f Q is the reactive power output function of the wind farm; the expressions for the upper and lower limits of the grid-connected voltage are: in, is the grid connection point voltage function.

4. The method for controlling wind farm voltage taking into account the influence of grid voltage support capability according to claim 3, characterized in that: The hierarchical model predictive control method includes a reactive power coordination and allocation layer and a tracking control layer. The reactive power coordination and allocation layer prediction model is constructed based on the dynamic process of wind turbine reactive power response. The control input of the prediction model is the reactive power change ΔQ(k) of the wind farm at time k, and the state variable is the measured voltage V at the grid connection point. PCC , the measured voltage of the busbar V CB And the reactive output Q of the static VAR generator SVG S , the disturbance information is the active power change of the wind farm, and the prediction model between the control input and the state variable output is: Where x(k) is as follows: Where x(k) represents the predicted change sequence of the grid connection point, the collection bus, and the reactive output of the SVG at the next H time points starting from time k; x(k+1) represents the predicted change sequence of the grid connection point, the collection bus, and the reactive output of the SVG at the next H time points starting from time k+1; y(k) represents the predicted value of the reactive output of the grid connection point, the collection bus, and the SVG at time k; A, B, C, and D are coefficient matrices. The voltage control of the wind farm aims to minimize the deviation between the node voltage and the corresponding reference value, while also ensuring that the SVG has a large reactive reserve. Based on this, the objective function of the reactive coordination and allocation layer is constructed: Where ΔV PCC , ΔV CB , ΔQ S are the deviations between the grid connection point voltage, the busbar voltage, and the SVG reactive output and the corresponding reference values; W PCC 、W CB 、W S is the weight coefficient corresponding to each deviation, satisfying W PCC >W CB >W S ; H represents the prediction time domain; considering the influence of active power changes, corresponding to H time points in a control cycle, the expression of the grid connection point voltage prediction value under the combined action of wind farm active power and reactive power is: Where, It represents the reactive response coefficient of the wind turbine at the Hth time point in a reactive command cycle; Indicates the predicted value of the grid connection point voltage at the prediction start time.

5. The method for controlling wind farm voltage taking into account the influence of grid voltage support capability according to claim 4, characterized in that: The voltage deviation of the grid connection point and the busbar in the reactive power coordination and distribution layer prediction model satisfies: Where ΔP(k) is the change in active power of the wind farm at time k; ΔQ(k) is the reactive power change of wind farm at time k; variable Indicates the busbar active voltage sensitivity coefficient; variable Indicates the reactive voltage sensitivity coefficient of the collection bus; is the bus reference voltage; the objective function of the reactive power coordination distribution layer prediction model satisfies the following constraints: Where, They are the upper and lower limits of the grid connection point voltage respectively; are the upper and lower limits of the bus voltage respectively; Q max , Q min They are the upper and lower limits of reactive output of wind farms respectively.

6. The method for controlling wind farm voltage taking into account the influence of grid voltage support capability according to claim 4, characterized in that: The tracking control layer of the hierarchical model predictive control method is specifically as follows: selecting the stator and rotor d-axis and q-axis current components [i sd i sq i rd i rq ] T As the state variable x, the rotor d-axis and q-axis voltage components [u rd u rq ] T As the control input u r , stator d-axis and q-axis voltage components [u sd u sq ] T As an external measurable disturbance u s , after discretization, the fan state space expression is as follows: Where A1, B1, C1, and D1 are coefficient matrices. The wind turbine operation mainly tracks the active and reactive output instructions. After power decoupling, the active and reactive power can be represented by the rotor dq axis current components, respectively. Based on this, the rotor current reference output y can be obtained. ref , and then, with the goal of minimizing both the control output and the control action change amplitude, the objective function of the unit control is established as follows: Where W y is the control output weight coefficient; W u The incremental weight coefficient for control input; y p (k+mk) is the predicted control output of the mth step; Δu r (k+mk) is the control input increment; M represents the prediction time domain.

7. The method for controlling wind farm voltage taking into account the influence of grid voltage support capability according to claim 1, characterized in that: The deep residual shrinkage network algorithm in step 2 is specifically: using soft thresholding to achieve noise reduction in signal processing, and the expression of the soft threshold function is as follows: Among them, d in is the input; d out is the output; τ is the threshold, which is a constant positive number.

8. The method for controlling wind farm voltage taking into account the influence of grid voltage support capability according to claim 1, characterized in that: According to the aerodynamic principle in step 3, the mechanical power captured by the wind wheel is: Where ρ is the air density; R is the radius of the fan blade; v is the wind speed; C p (λ, β) is the wind energy utilization coefficient, which is a nonlinear function of the tip speed ratio λ and the pitch angle β: Where ω is the mechanical angular velocity of the wind turbine rotor; When the wind turbine is operating below the rated wind speed, in order to improve the utilization rate of wind energy, the pitch angle β is adjusted to 0°, and the generator rotor speed is changed when the wind speed changes, so that λ is maintained at the optimal tip speed ratio λ. opt , which can make the wind energy utilization coefficient C p Constant at the maximum value C p,max At this time, the maximum wind power P that the wind turbine can capture max for: Therefore, the optimal output power P of the wind turbine characterized by the speed is opt It can be expressed as: Where k opt is the optimal power proportional coefficient.

9. The method for controlling wind farm voltage taking into account the influence of grid voltage support capability according to claim 8, characterized in that: The second-order filter introduced in step 3 adopts a two-pole-one-zero architecture, and its transfer function expression can be written as: Where a, b, and c are filter parameters. After the second-order filter is introduced, the wind turbine output active reference value P ref2 becomes: Similarly, the transfer function from wind speed to rotation speed can be derived as: Where J is the moment of inertia of the system including the wind rotor and generator; ω0 is the mechanical angular velocity of the rotor at the typical optimal power point selected as the reference operating point; based on the second-order filter relationship between output power and the cube of the speed, the wind speed to power transfer function can be derived as:

10. The method for controlling wind farm voltage taking into account the influence of grid voltage support capability according to claim 9, characterized in that: In step 3, in order to minimize the deviation between the grid connection point voltage prediction value and the grid connection point voltage reference value, the filter parameters need to be adjusted online based on fuzzy control. The improved second-order filtering method uses a fuzzy logic controller to correct the filter coefficient online. The fuzzy logic controller does not rely on the precise mathematical model of the controlled object and can simulate human operating experience to adjust the parameters online. Since wind speed is difficult to measure accurately, a large wind speed disturbance event is defined by measuring the speed based on the consistency of the speed and wind speed change trends. When the speed increases at a large acceleration for a period of time, it is judged that a large wind speed disturbance has occurred. Similarly, a large wind speed disturbance can be judged to have decreased. In actual control, the speed difference between adjacent sampling moments is usually used as the input acceleration signal. Here, it is important to emphasize the asymmetry of the rotor kinetic energy control: when the filter is introduced, the speed will overshoot. When the wind speed increases significantly, it is hoped that the speed will increase sufficiently to store kinetic energy. Therefore, the hysteresis comparator output is 1. A sharp drop in wind speed is a dangerous signal, and there is a risk of instability due to excessive deceleration. Therefore, the hysteresis loop returns to 0.

Images