Doubly-fed fan low voltage ride through cooperative control method based on power prediction

Abstract

The invention discloses a double-fed fan low voltage ride through cooperative control method based on power prediction, and belongs to the technical field of wind power generation. The method comprises the following steps: inputting data; performing power prediction based on a CNN-LSTM (Convolution and Long Short Term Memory Neural Network); converter capacity dynamic distribution calculation; the invention relates to an improved direct power control strategy design. According to the method, the CNN-LSTM power prediction model is introduced, so that the problems that traditional direct power control depends on real-time power feedback, a control instruction lags, and consequently, the voltage of a direct-current bus is unstable and reactive power support is delayed during low-voltage ride-through are solved. Through the predicted active power curve, the capacity distribution strategy of the converter is adjusted in advance, so that the reactive power support response is accelerated. Through the adaptive hysteresis control technology, the switching frequency is reduced in the low-voltage ride-through period to protect the device, so that the low-voltage ride-through operation capability of the doubly-fed fan is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and particularly to a coordinated control method for low-voltage ride-through of a doubly-fed wind turbine based on power prediction. Background Art

[0002] As a mainstream wind power generation unit, the doubly-fed wind turbine has been widely applied in the field of new energy power generation due to its economic advantages such as small converter capacity, active / reactive power decoupling, and flexible control. However, its grid-connected topology with the stator winding directly connected to the grid makes the unit extremely sensitive to grid faults. When the grid voltage drops suddenly, it is very easy to cause the wind turbine to trip off the grid, seriously threatening the safe operation of the power system. Therefore, it is particularly important to improve its fault ride-through ability.

[0003] The traditional direct power control strategy relies on a real-time power feedback mechanism. Although it can achieve accurate power tracking under steady-state conditions, during grid faults, power mutations will cause control commands to lag, leading to problems such as instability of the DC bus voltage and a sharp increase in the converter switching frequency. These problems indicate the technical limitations of existing control strategies in fault ride-through scenarios. Summary of the Invention

[0004] The purpose of the present invention is to solve the defects existing in the prior art, and to propose a coordinated control method for low-voltage ride-through of a doubly-fed wind turbine based on power prediction to improve the fault ride-through ability of the doubly-fed wind turbine.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions:

[0006] In a first aspect, the present invention provides a coordinated control method for low-voltage ride-through of a doubly-fed wind turbine based on power prediction, including the following steps:

[0007] S1. Extract data inputs from two dimensions of space and time;

[0008] S2. Perform power prediction on the doubly-fed wind turbine based on a convolutional and long short-term memory network CNN-LSTM model to obtain a power prediction result;

[0009] S3. Use the result of the power prediction as a decoupling reference for dynamic calculation of converter capacity allocation;

[0010] S4. Design an improved direct power control strategy.

[0011] The data inputs are divided into two-dimensional inputs. The spatial dimension inputs meteorological spatial parameters of the current wind speed, temperature, and air pressure, and the time dimension inputs parameters of the wind turbine speed, DC bus voltage, and historical power curve;

[0012] The power prediction is carried out by introducing a CNN-LSTM model. The CNN network extracts the spatial dimension, and the LSTM network extracts the time dimension. Then, the final predicted active power is output through a fully connected layer, which is used as a decoupling benchmark for the dynamic allocation calculation of the converter capacity; a convolutional and long short-term memory neural network CNN-LSTM is constructed for power prediction, and the construction is as follows:

[0013]

[0014] In the formula, represents the predicted value of wind power at time t; p t-k represents historical power data; w t represents meteorological data;

[0015] The CNN-LSTM model consists of two parts: First, the input data passes through the CNN network to achieve the extraction and dimensionality reduction of data features; the data processed by the CNN network is input into the LSTM network. The LSTM network adjusts its own parameters through continuous iterative training of the data for dynamic modeling, and finally outputs the power prediction value through the fully connected layer.

[0016] In a possible implementation manner, the dynamic allocation calculation of the converter capacity is specifically implemented as follows:

[0017] The total capacity of the converter is a fixed value, and its active power and reactive power need to satisfy:

[0018]

[0019] In the formula, S max represents the maximum apparent power, that is, the capacity of the converter; P represents the active power, and Q represents the reactive power;

[0020] Predict the active power sequence in the next 3 cycles (about 15 ms) through the CNN-LSTM model, and calculate the current available reactive capacity margin according to the predicted value as:

[0021]

[0022] In the formula, Q a represents the current reactive capacity margin; P p represents the predicted value of the active power;

[0023] According to the grid voltage dip depth and the reactive capacity margin, dynamically set the reference value of the reactive power as:

[0024] Q ref = K·ΔU·Q a ;

[0025] In the formula, Q refIt represents the dynamic reactive power reference value; K is the voltage support coefficient; ΔU represents the voltage drop rate monitored in real time.

[0026] In a possible implementation, the specific steps of the improved direct power control strategy are as follows:

[0027] S41. Construct the direct power control model of the doubly-fed wind turbine as follows:

[0028] The flux linkage equation and voltage equation in the rotor α r β r coordinate system are:

[0029]

[0030] In the formula, L, u, i represent flux linkage, inductance, voltage, and current respectively; R represents resistance, and L m represents mutual inductance; ω r represents the rotor electrical angular velocity; the superscript r represents the rotor α r β r coordinate system; the subscripts s and r represent the stator and rotor respectively;

[0031] The active power absorbed by the stator from the power grid and the reactive power generated are respectively:

[0032]

[0033] The direct power control model is obtained by differentiation as:

[0034]

[0035] In the formula, ω1 is the grid synchronous angular velocity; σ is the leakage magnetic coefficient, and θ is the angle difference between the rotor and stator flux linkages;

[0036] S42. Design the direct power control strategy based on adaptive hysteresis control as follows:

[0037] The hysteresis comparator accepts the comparison result and issues a corresponding voltage vector selection signal to control the on and off of the switching tubes, thereby controlling the working state of the converter. The control laws of the active and reactive power hysteresis comparators are:

[0038]

[0039] In the formula, S p = 1, S p = 0, and S p = -1 represent that the active power needs to increase, remain unchanged, or decrease respectively; similarly, S Q = 1, S Q = 0, and S Q= 0 represents that the reactive power needs to increase, remain unchanged, or decrease respectively; represent the reference values of active power and reactive power respectively;

[0040] Introduce adaptive hysteresis control and dynamically adjust the hysteresis width to be:

[0041] H = H0·(1 + α·ΔU);

[0042] In the formula, H0 is the hysteresis width during normal grid connection operation, and α is the fault sensitivity coefficient;

[0043] Through predictive weight allocation, the hybrid active power reference value is generated as:

[0044] P ref = β·P p + (1 - β)·P real ;

[0045] In the formula, β is the predictive confidence weight, which is adjusted online according to the prediction error:

[0046]

[0047] In the formula, γ is the sensitivity coefficient, P p is the predicted active power value, and P real is the actual active power value.

[0048] In a possible implementation, the hysteresis comparator selects the signal to control the on and off of the switch tube by accepting the comparison result and sending out the corresponding voltage vector. Description of the Drawings

[0049] The drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation to the present invention.

[0050] Figure 1 is a flow block diagram of a coordinated control method for low-voltage ride-through of a doubly-fed wind turbine based on power prediction proposed by the present invention;

[0051] Figure 2 is a structural block diagram of an improved direct power control based on adaptive hysteresis control proposed by the present invention. Detailed Embodiments

[0052] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the drawings in the embodiments of the present invention.

[0053] An embodiment of the present invention provides a coordinated control method for low-voltage ride-through of a doubly-fed wind turbine based on power prediction. This method includes the following processes: First, a CNN-LSTM power prediction model is constructed. By inputting data information in two dimensions of space and time, the convolutional neural network is used to extract the spatial features of multi-node electrical parameters, and the long short-term memory network is combined to capture the dynamic time-series characteristics. After being fused by the fully connected layer, the final active power prediction value is output. Then, the predicted active power value is used as the decoupling reference for the dynamic allocation calculation of the converter capacity, and the current reactive power release margin is calculated according to the prediction value, so as to support the grid voltage to the greatest extent during the fault. Finally, an adaptive hysteresis control is introduced to establish a functional relationship between the hysteresis width and the voltage drop depth. By dynamically adjusting the hysteresis width, the switching frequency is reduced when the grid voltage drops deeply to protect the converter devices. In addition, in order to improve the prediction accuracy, a hybrid active power reference value is generated through weight allocation to improve the accuracy of the reference value. It should be clear that the above embodiments are only partial embodiments of the present application, and other unlisted embodiments are also within the protection scope of the present application.

[0054] Figure 1 As shown in the flowchart of a coordinated control method for low-voltage ride-through of a doubly-fed wind turbine based on power prediction proposed by the present invention, Figure 1 the method includes the following steps:

[0055] Step S1: Extract data inputs from two dimensions of space and time;

[0056] The data input is divided into two-dimensional inputs. The spatial dimension inputs the meteorological spatial parameters of the current wind speed, temperature, and air pressure, and the time dimension inputs the parameters of the wind turbine speed, DC bus voltage, and historical power curve;

[0057] Step S2: Based on the convolutional and long short-term memory neural network CNN-LSTM model, perform power prediction on the doubly-fed wind turbine to obtain the power prediction result;

[0058] Power prediction is achieved by introducing the CNN-LSTM model. The CNN network extracts the spatial dimension, and the LSTM network extracts the time dimension. Then, the final predicted active power is output through the fully connected layer and used as the decoupling reference for the dynamic allocation calculation of the converter capacity. A convolutional and long short-term memory neural network CNN-LSTM is constructed for power prediction, and the construction is as follows:

[0059]

[0060] In the formula, represents the predicted value of the wind power at time t; p t-k represents the historical power data; w t represents the meteorological data;

[0061] The CNN-LSTM model consists of two parts: First, the input data passes through the CNN network to extract and reduce the dimensionality of the data features; the data processed by the CNN network is input into the LSTM network. The LSTM network adjusts its own parameters through continuous iterative training of the data for dynamic modeling, and finally outputs the power prediction value through the fully connected layer.

[0062] Step S3: Use the power prediction result as the decoupling reference for dynamic calculation of the converter capacity allocation;

[0063] The specific implementation of the dynamic calculation of the converter capacity allocation is as follows:

[0064] The total capacity of the converter is a fixed value, and its active power and reactive power need to satisfy:

[0065]

[0066] In the formula, S max represents the maximum apparent power, that is, the capacity of the converter; P represents the active power, and Q represents the reactive power;

[0067] Predict the active power sequence in the next 3 cycles (about 15 ms) through the CNN-LSTM power prediction model, and calculate the current available reactive capacity margin according to the predicted value as:

[0068]

[0069] In the formula, Q a represents the current reactive capacity margin; P p represents the predicted value of the active power;

[0070] According to the grid voltage dip depth and the reactive capacity margin, dynamically set the reference value of the reactive power as:

[0071] Q ref = K·ΔU·Q a ;

[0072] In the formula, Q ref represents the dynamic reactive power reference value; K is the voltage support coefficient; ΔU represents the real-time monitored voltage dip rate;

[0073] Dynamically allocate the converter capacity by calculating the converter reactive power release margin in real time to achieve intelligent limiting of the active power and maximizing the regulation of the reactive power output.

[0074] Step S4: Design an improved direct power control strategy;

[0075] The specific steps for designing the improved direct power control strategy are as follows:

[0076] Step S41: Construct the direct power control model of the doubly-fed wind turbine as follows:

[0077] Rotor α r β r The flux linkage equation and voltage equation in the coordinate system are as follows:

[0078]

[0079] Wherein, L, u, and i respectively represent flux linkage, inductance, voltage, and current; R represents resistance, and L m represents mutual inductance; ω r represents the electrical angular velocity of the rotor; the superscript r represents the rotor α r β r coordinate system; the subscripts s and r respectively represent the stator and the rotor;

[0080] The active power absorbed by the stator from the power grid and the reactive power generated are respectively:

[0081]

[0082] The direct power control model obtained by differentiation is:

[0083]

[0084] Wherein, ω1 is the synchronous angular velocity of the power grid; σ is the leakage magnetic coefficient, and θ is the position angle difference between the rotor and the stator flux linkage;

[0085] Step S42: The direct power control strategy based on adaptive hysteresis control is designed as follows:

[0086] The hysteresis comparator accepts the comparison result and issues a corresponding voltage vector selection signal to control the on and off of the switching tube, thereby controlling the working state of the converter. The control rules of the active and reactive power hysteresis comparators are:

[0087]

[0088] Wherein, S p = 1, S p = 0, and S p = -1 respectively represent that the active power needs to increase, remain unchanged, or decrease; similarly, S Q = 1, S Q = 0, and S Q = 0 respectively represent that the reactive power needs to increase, remain unchanged, or decrease; respectively represent the reference values of the active power and the reactive power;

[0089] Introduce adaptive hysteresis control, and dynamically adjust the hysteresis width to:

[0090] H = H0·(1 + α·ΔU);

[0091] In the formula, H0 is the hysteresis width during normal grid connection operation, and α is the fault sensitivity coefficient;

[0092] Through predictive weight allocation, the hybrid active power reference value is generated as follows:

[0093] P ref = β·P p +(1 - β)·P real ;

[0094] In the formula, β is the predictive confidence weight, which is adjusted online according to the prediction error:

[0095]

[0096] In the formula, γ is the sensitivity coefficient, P p is the predicted active power value, and P real is the actual active power value.

[0097] In the embodiment of the present invention, the hysteresis comparator controls the on - off of the signal control switch tube by accepting the comparison result and emitting the corresponding voltage vector.

[0098] Figure 2 Shown in the following is the structural block diagram of an improved direct power control based on adaptive hysteresis control proposed by the present invention. As Figure 2 shown, the predicted active power reference value is obtained through the convolution and long - short - term memory neural network CNN - LSTM power prediction model, and the reference value of the active power is generated through weight allocation. The calculation of the active power reference value is as follows:

[0099] P ref = β·P p +(1 - β)·P real ;

[0100] In the formula, β is the predictive confidence weight, which is adjusted online according to the prediction error:

[0101]

[0102] In the formula, γ is the sensitivity coefficient, P p is the predicted active power value, and P real is the actual active power value.

[0103] By calculating the reactive power release margin of the converter in real - time, the converter capacity is dynamically allocated, and according to the grid voltage dip depth and reactive power capacity margin, the reference value of the reactive power is dynamically set. The calculation of the reactive power reference value is as follows:

[0104]

[0105] In the formula, Q aIndicates the current reactive power margin; P p Represents the predicted value of active power.

[0106] Q ref = K·ΔU·Q a ;

[0107] In the formula, Q ref Indicates the dynamic reactive power reference value; K is the voltage support coefficient; ΔU represents the voltage drop rate monitored in real time.

[0108] Establish the functional relationship between the hysteresis width and the voltage drop depth. By dynamically adjusting the hysteresis width, the switching frequency is reduced during deep voltage dips in the power grid to protect the converter devices. The functional relationship is as follows:

[0109] H = H0·(1 + α·ΔU);

[0110] In the formula, H0 is the hysteresis width during normal grid connection operation, and α is the fault sensitivity coefficient.

[0111] Collect the grid voltage u sabc and i sabc Through coordinate transformation, obtain the grid voltage u sαβ and current i sαβ in the two-phase stationary coordinate system. Calculate the stator active power P s and reactive power Q s and compare them with the corresponding reference values and to obtain the power deviations ΔP s and ΔQ s Generate the power state signals S P and S Q through an adaptive hysteresis comparator; Input the real-time wind power, obtain the rotor angular velocity ω r according to the speed measuring device, and integrate to obtain the rotor angle Through the stator angle and the rotor angle Subtract to obtain the slip angle According to Judge the sector where it is located, and combine the power state signals S P 、S Q Select the appropriate switching vectors S a 、S b 、S c to control the conduction and cutoff of the switch tubes of the rotor-side converter, realize the support for the grid voltage and the suppression of the rotor overcurrent during faults, and improve the fault ride-through ability of the doubly-fed wind turbine.

Claims

1. A low voltage ride-through coordinated control method for a doubly-fed wind turbine based on power prediction, characterized in that: The following steps are involved: S1, extract data input from two dimensions: space and time; S2. Power prediction of the doubly-fed wind turbine is performed based on the convolution and long short-term memory neural network CNN-LSTM model to obtain the power prediction result; S3, using the power prediction result as a decoupling benchmark to perform dynamic allocation calculation of converter capacity; S4. Design an improved direct power control strategy.

2. The low voltage ride through coordinated control method of a doubly fed wind turbine based on power prediction according to claim 1 is characterized in that: In step S1, data input is divided into two dimensions: the spatial dimension inputs the meteorological spatial parameters of the current wind speed, temperature, and air pressure; and the time dimension inputs the parameters of the fan speed, DC bus voltage, and historical power curve.

3. The low voltage ride through coordinated control method of a doubly fed wind turbine based on power prediction according to claim 1 is characterized in that: In step S2, the power prediction is performed by introducing a CNN-LSTM model, wherein the CNN network extracts the spatial dimension, and the LSTM network extracts the time dimension, and then the final predicted active power is output through a fully connected layer as a decoupling benchmark for dynamic capacity allocation calculation of the converter; a convolutional and long short-term memory neural network CNN-LSTM is constructed to perform power prediction, which is constructed as follows: In the formula, represents the predicted value of wind power at time t; p t-k Represents historical power data; w t Represents meteorological data; The CNN-LSTM model consists of two parts: first, the input data passes through the CNN network to extract and reduce the data features; the data processed by the CNN network is input into the LSTM network, and the LSTM network adjusts its own parameters through continuous iterative training of data to perform dynamic modeling, and finally outputs the power prediction value through the fully connected layer.

4. The low voltage ride through coordinated control method of a doubly fed wind turbine based on power prediction according to claim 1 is characterized in that: In step S3, the dynamic allocation calculation of the converter capacity is specifically implemented as follows: The total capacity of the converter is a fixed value, and its active power and reactive power must meet the following requirements: In the formula, S max Represents the maximum apparent power, i.e. the capacity of the converter; P represents active power, and Q represents reactive power; The CNN-LSTM model is used to predict the active power sequence of the next three cycles, and the reactive capacity margin that can be released is calculated based on the predicted value: In the formula, Q a Indicates the current reactive capacity margin; P p Represents the predicted value of active power; According to the grid voltage drop depth and reactive capacity margin, the reference value of reactive power is dynamically set as follows: Q ref =K·ΔU·Q a ; In the formula, Q ref Indicates the dynamic reactive power reference value; K is the voltage support coefficient; ΔU represents the voltage drop rate monitored in real time.

5. The low voltage ride through coordinated control method of a doubly fed wind turbine based on power prediction according to claim 1 is characterized in that: In step S4, the improved direct power control strategy is as follows: S41. Construct the direct power control model of the doubly-fed wind turbine as follows: Rotor α r β r The flux equation and voltage equation in the coordinate system are: In the formula, L, u, i represent flux, inductance, voltage and current respectively; R represents resistance, L m Represents mutual inductance; ω r Indicates the rotor electrical angular velocity; the superscript r represents the rotor α r β r Coordinate system; subscripts s and r represent stator and rotor respectively; The stator absorbs active power from the power grid and emits reactive power respectively: The direct power control model is obtained by differentiation: Where ω1 is the synchronous angular velocity of the power grid; σ is the magnetic flux leakage coefficient, and θ is the angular difference between the rotor and stator flux positions; S42, the direct power control strategy based on adaptive hysteresis control is designed as follows: The hysteresis comparator receives the comparison result and sends the corresponding voltage vector selection signal to control the on and off of the switch tube, thereby controlling the working state of the converter. The control law of the active and reactive power hysteresis comparator is: In the formula, S p =1, S p =0 and S p =-1 represents that the active power needs to be increased, unchanged or decreased; similarly, S Q =1, S Q =0 and S Q =0 means the reactive power needs to be increased, unchanged or decreased respectively; Represent the reference values ​​of active power and reactive power respectively; Adaptive hysteresis control is introduced to dynamically adjust the hysteresis width to: H = H0·(1+α·ΔU); Where H0 is the hysteresis loop width during normal grid-connected operation, and α is the fault sensitivity coefficient; By predicting the weight allocation, the hybrid active power reference value is generated as: P ref =β·P p +(1-β)·P real ; Where β is the prediction confidence weight, which is adjusted online according to the prediction error: Where γ is the sensitivity coefficient, P p To predict the active power value, P real is the actual value of active power.

6. The low voltage ride through coordinated control method of a doubly fed wind turbine based on power prediction according to claim 4 is characterized in that: The hysteresis comparator receives the comparison result and sends out the corresponding voltage vector to select the signal to control the on and off of the switch tube.

Images