New energy unit black box model control parameter identification method based on improved PSO algorithm

Abstract

A new energy unit black box model control parameter identification method based on an improved PSO algorithm specifically comprises the steps that an impedance model of a new energy unit is constructed, and a transfer function of the new energy unit is deduced based on the impedance model; acquiring frequency domain response characteristics based on the transfer function; wherein the frequency domain response characteristics are obtained by calculating the Bode diagram of the transfer function; and fitting the frequency domain response characteristics based on an improved PSO algorithm to identify control parameters. The improved PSO algorithm comprises the following steps: allocating different dynamic weight values to frequency points of different frequency bands; grouping the parameters to be identified according to the parameter sensitivity and setting different learning rates for different groups; and constructing a composite fitness function containing an amplitude error, a phase error and a parameter grouping error for optimization. According to the method, a dynamic frequency weight distribution strategy is introduced, interference of high-band switching noise is suppressed, high-sensitivity parameters and low-sensitivity parameters are treated differently, effective decoupling of strong coupling parameters is achieved, and interference of secondary parameter updating on core parameters is avoided.

Description

Technical Field

[0001] This invention belongs to the field of model parameter identification technology, specifically relating to a method for optimizing the parameters of a structured model of a wind turbine under all operating conditions. Background Technology

[0002] In recent years, the large-scale grid connection of new energy units, represented by wind power and photovoltaics, has led to increasingly complex interactions between their dynamic characteristics and the power grid, posing new challenges to the stable operation of the power system. Establishing accurate models of new energy units and obtaining their actual control parameters is crucial for power grid stability analysis, oscillation mechanism research, and the formulation of suppression strategies. However, due to the high nonlinearity and black-box characteristics of the control systems of new energy units, their impedance model parameters exhibit significant strong coupling and frequency domain sensitivity. In existing technologies, the least squares method and the PSO algorithm are widely used for parameter identification. Patent CN118232765A uses the least squares method to fit the model to actual data, aiming to minimize the sum of squared errors to solve for the parameters to be identified. However, it is highly sensitive to the initial value setting of the parameters, and when facing complex nonlinear problems with multiple local extrema, it is prone to getting trapped in local optima, resulting in poor robustness of the identification results and difficulty in guaranteeing global convergence. Patent CN106296461B uses the PSO algorithm to complete parameter identification, a method with excellent capabilities in handling high-dimensional, nonlinear problems. However, high-frequency Bode plots are susceptible to noise interference from power electronic switches, forming a "noise masking zone." When dealing with amplitude / phase frequency errors, they struggle to differentiate the weighting of different frequency bands' influence on system characteristics, typically treating all frequency points equally and ignoring differences in frequency band sensitivity. Furthermore, the algorithm's exploration of the parameter space is often uniform, failing to implement differentiated learning strategies based on the actual influence of parameters (such as their contribution to low-frequency stability), resulting in difficulties in effectively decoupling coupled parameters. Therefore, a new method is urgently needed that can overcome the shortcomings of existing algorithms and rapidly, accurately, and robustly identify control parameters of black-box models of new energy generating units even in high-noise environments. Summary of the Invention

[0003] This invention aims to solve the problems of existing parameter identification methods (such as gradient-based algorithms and standard PSO algorithms) when facing high-frequency noise interference, strong parameter coupling, and frequency domain sensitivity differences exhibited by the impedance model of new energy units, such as sensitivity to initial values, susceptibility to local optima, and insufficient fitting accuracy.

[0004] To achieve the above objectives, this invention addresses the specific technical challenges of model parameter identification by designing an improved PSO algorithm. This algorithm is highly efficient and fast, and can effectively identify control parameters under black-box conditions.

[0005] The technical solution of this invention is as follows:

[0006] Step 1: Construct the impedance model of the new energy unit, and derive its transfer function based on the impedance model;

[0007] Step 2: Obtain the frequency domain response characteristics based on the transfer function; wherein, the frequency domain response characteristics are obtained by calculating the Bode plot of the transfer function;

[0008] Step 3: Based on the improved PSO algorithm, using the frequency domain response characteristics (Bode plot) of the impedance model as the fitting target, identify the unknown control parameters in the model; wherein, the improved PSO algorithm includes: assigning different dynamic weight values ​​to frequency points in different frequency bands; grouping the parameters to be identified according to parameter sensitivity and setting different learning rates for different groups; and constructing a model including amplitude error E mag Phase error E phase And parameter grouping error E param The composite fitness function is optimized.

[0009] Preferably, the method of assigning different dynamic weight values ​​to frequency points of different frequency bands specifically involves: assigning a first weight value to the low-frequency band, assigning a second weight value to the mid-frequency band, and assigning a third weight value to the high-frequency band to suppress the influence of high-frequency noise, wherein the first weight value is greater than the second weight value, and the second weight value is greater than the third weight value.

[0010] Preferably, the low-frequency band is 1-50Hz, the mid-frequency band is 0-500Hz, and the high-frequency band is greater than 500Hz; the dynamic frequency weighting allocation function w(f) is:

[0011] .

[0012] Preferably, the step of grouping the parameters to be identified based on parameter sensitivity and setting different learning rates for different groups specifically involves: classifying key parameters that involve the dominant mode and have high sensitivity into the core parameter group, i.e., group N1, and classifying parameters that affect the secondary dominant mode or have low sensitivity into the auxiliary parameter group, i.e., group N2; setting a first cognitive factor C1 for group N1 and a second cognitive factor C2 for group N2, wherein C1 is greater than C2.

[0013] Preferably, the velocity update process of the improved PSO algorithm employs dynamically adjusted inertia weights ω. (t) .

[0014] Preferably, the improved PSO algorithm's position update process includes a boundary damping strategy to prevent particles from oscillating near the boundary and improve system stability.

[0015] Preferably, the composite fitness function Fitness is specifically:

[0016] Fitness = 0.4E mag+0.4E phase +E param；

[0017] Preferably, the amplitude-frequency error E of the improved PSO algorithm is constructed. mag The calculation method is as follows:

[0018]

[0019] Among them, Z est (f) represents the estimated impedance value, Z real (f) represents the true impedance value, which is obtained through impedance scanning.

[0020] Constructing the phase frequency error E of the improved PSO algorithm phase The calculation method is as follows:

[0021]

[0022] Among them, unwrap is the phase unwinding operation, which is mainly used to correct periodic jumps of 360° integer multiples that may exist in the impedance phase data due to measurement or calculation reasons, to ensure that the phase difference reflects the real physical deviation rather than the illusion caused by the periodic jump, Z phase angle;

[0023] Constructing the parameter grouping error E of the improved PSO algorithm param The calculation method is as follows:

[0024]

[0025] Where, x j The current parameter value, gbest j Let be the optimal position of the group in dimension j.

[0026] Preferably, the construction of the impedance model of the new energy unit, and the derivation of its transfer function based on the impedance model, specifically involves: establishing the closed-loop transfer function Z on the rotor side of the converter. d,RSC The converter grid-side closed-loop transfer function Z d,GSC Phase-locked loop coupling impedance Z PLL Cross-coupling impedance Z cross Generator body impedance Z gen Crowbar circuit impedance Z Protection Grid connection interface impedance Z Grid Filtering and Coupling Terms Z Filter and dynamic coupling time constant T coupling The impedance model Z of the doubly-fed asynchronous wind turbine is derived from the function. DFIG .

[0027] This invention discloses a method for identifying control parameters of a black-box model of a new energy generating unit based on an improved PSO algorithm, significantly improving identification accuracy and robustness. By introducing a dynamic frequency weight allocation strategy, interference from high-frequency switching noise is effectively suppressed, while the fitting weight of the low-frequency band, which plays a decisive role in system stability, is strengthened, making the identification results more reflective of the true dominant dynamics of the system. The parameter grouping learning rate strategy adopted treats high and low sensitivity parameters differently based on global sensitivity analysis, achieving effective decoupling of strongly coupled parameters and avoiding interference from secondary parameter updates to core parameters. Furthermore, the method provided by this invention has universality and stability. The composite fitness function integrates amplitude-frequency error, phase-frequency error, and grouping error, ensuring that the identified parameters are consistent with the real system in both amplitude and phase, improving the physical rationality of the results. This method has low sensitivity to initial values ​​and good global convergence, overcoming the shortcomings of traditional gradient-based algorithms that are prone to getting trapped in local optima, providing a more reliable and practical model parameter foundation for grid stability analysis and oscillation suppression. Attached Figure Description

[0028] Appendix Figure 1 This refers to the parameter settings for the doubly fed wind turbine topology and the derived impedance formula in this embodiment of the invention.

[0029] Appendix Figure 2 The actual impedance scan curve of the doubly fed fan at 100-1000Hz according to an embodiment of the present invention.

[0030] Appendix Figure 3 The Bode plot is obtained by substituting the actual parameters of the doubly fed fan (1-1000Hz) according to an embodiment of the present invention.

[0031] Appendix Figure 4 The actual parameters of the 1-1000Hz frequency were substituted into the Bode plot fitting parameters for the doubly fed fan according to the embodiment of the present invention. Detailed Implementation

[0032] To make the technical solution of the present invention easier to understand, a method for optimizing the full-condition parameters of a structured model of a wind turbine generator disclosed in the present invention will now be clearly and completely described in conjunction with embodiments and accompanying drawings.

[0033] The method disclosed in this invention is applicable to various types of new energy generating units, and its impedance model refers to the equivalent frequency domain impedance viewed from the unit's grid connection point (PCC). This impedance characteristic is mainly determined by the control system of the unit's grid-connected converter. For example... Figure 1 As shown, this embodiment takes a typical doubly fed induction generator (DFIG) as an example to explain in detail the process of constructing the impedance model.

[0034] Step 100: Confirm the control parameters to be identified, specifically including: clarifying the specific type of the control parameters to be identified. The control parameters to be identified are key parameters of the interface between the converter control system of the new energy unit and the power grid, including at least the current loop proportional parameter, the current loop time constant, the phase-locked loop proportional control parameter, and the phase-locked loop time constant.

[0035] Step 110: Derive the impedance model of the doubly-fed asynchronous wind turbine, specifically including:

[0036] Establish the closed-loop transfer function Z on the rotor side of the converter. d,RSC :

[0037]

[0038] Among them, its v d,ref i is the reference value for the d-axis voltage. d,ref K is the reference value for the d-axis current. P-RSC T is the proportional parameter of the rotor-side current loop. i-RSC Rotor-side current loop time constant, R r With L r For the inherent impedance of the rotor winding, the PI controller K p K is a proportional parameter. i Digital delay control for integral parameters T s This is the delay time.

[0039] Establish the closed-loop transfer function Z of the converter on the grid side. d,GSC :

[0040]

[0041] Among them, K P_GSC T is the proportional parameter of the grid-side current loop. i_GSC T is the time constant of the grid-side current loop. s,G For the delay time, R g With L g The inherent impedance of the grid-side converter. , The cutoff frequency, The damping ratio is denoted as .

[0042] Establish phase-locked loop coupling impedance Z PLL :

[0043]

[0044] Among them, V base I is the voltage reference value. base K is the current reference value. P_PLL T is the proportional control parameter for the phase-locked loop.i_PLL The voltage reference value and the current reference value are both rated values, which represent the phase-locked loop time constant.

[0045] Establish cross-coupling impedance Z cross :

[0046]

[0047] in, The synchronous angular frequency of the generator. This is the actual angular frequency of the power grid. , This is to decouple the error gain from the time constant.

[0048] Establish the generator body impedance Z gen Considering the impact of RSC control on stator-rotor coupling in DFIG:

[0049] Among them, R s With L s For stator resistance and leakage inductance, L m This is the excitation inductance of the generator.

[0050] Establish the Crowbar circuit impedance Z Protection :

[0051]

[0052] Among them, R Crowbar T represents the Crowbar resistance value. Crowbar K is the time constant of the Crowbar action. LVRT T LVRT The low-voltage ride-through control gain and time constant are used.

[0053] Establish grid connection interface impedance Z Grid :

[0054]

[0055] Among them, R grid L is the equivalent resistance of the power grid. grid For the equivalent inductance of the power grid, C grid It is a parallel capacitor in the power grid.

[0056] Constructing the dynamic coupling time constant T coupling Functions:

[0057]

[0058] in, .

[0059] Based on the above, the complete impedance transfer function of the doubly fed wind turbine is derived, where the Z-factor in the numerator is... RSC and Z GSC The impedance of the converter on the machine side and the grid side is multiplied by the phase-locked loop impedance Z. PLL This reflects the modulation of the control system output by the PLL's dynamic characteristics, and the coupling impedance Z between the PLL and RSC and GSC. cross The summation is then added to the inherent impedance of the doubly-fed generator itself; the denominator is the correction for the interaction impedance between the wind turbine and external components, structurally forming a closed-loop transfer function of the feedback system; finally, the overall transfer function Z is obtained by multiplying the total by the dynamic coupling. DFIG :

[0060]

[0061] In this invention, the structure of the impedance model is considered known. Key control parameters in the model, such as the rotor-side current loop proportional parameter K, are... p_RSC Time constant T i_RSC Phase-locked loop parameter K p_PLL T i_PLL These parameters, such as [specific parameters], are considered unknown "black box" parameters due to their internal secrecy or difficulty in measurement. This invention utilizes an improved PSO algorithm to accurately identify these unknown parameters by matching them with measured frequency domain response curves.

[0062] Step 120: Obtain the frequency domain response characteristics based on the transfer function; wherein, the frequency domain response characteristics are obtained by calculating the Bode plot of the transfer function; an impedance scan is performed on a 5MW doubly-fed wind turbine. Figure 2 This is a 100-1000Hz actual impedance scan curve of a doubly-fed wind turbine according to an embodiment of the present invention. The parameters of a certain white-box 5MW doubly-fed wind turbine were substituted into the equivalent impedance formula for calculation. Figure 3 A Bode plot (1-1000Hz) of a doubly fed wind turbine according to an embodiment of the present invention.

[0063] Step 200: Initialize the particle swarm and parameter space, specifically including: defining the physical meaning of the particles, each particle corresponding to a set of control parameters to be identified determined in Step 100, and the j-th dimension position x of the particle. ij (t) represents the current value of the j-th control parameter to be identified for the i-th particle in the t-th iteration; physically feasible upper and lower limits x are set for each control parameter to be identified. minj x maxj The boundary damping distance Δb is set to prevent oscillations near the boundary during subsequent particle position updates; key operating parameters of the particle swarm are set, including a particle count of 80, a maximum number of iterations of 150, and an initial dynamic inertia weight.

[0064] Step 210: Define the improvement strategy:

[0065] To overcome high-frequency noise interference and emphasize key low-frequency dynamics, different dynamic weight values ​​are assigned to frequency points in different frequency bands: a first weight value of 5.0 is assigned to the low-frequency band, a second weight value of 2.0 is assigned to the mid-frequency band, and a third weight value of 0.3 is assigned to the high-frequency band. The dynamic frequency weight allocation function w(f) is defined as follows:

[0066]

[0067] Based on parameter sensitivity, the parameters to be identified are grouped and different learning rates are set for different groups. Specifically, key parameters involving the dominant mode and with high sensitivity are classified into the core parameter group, i.e., group N1, and assigned a higher cognitive factor C1=2.0. Parameters affecting the secondary dominant mode or with low sensitivity are classified into the auxiliary parameter group, i.e., group N2, and assigned a lower cognitive factor C2=1.8. The core basis for grouping is: the order and position of the parameter in the transfer function; the results of global sensitivity analysis; and the actual physical meaning and fault correlation. In this embodiment, a typical grouping is: group N1 (high influence group) includes rotor-side current loop proportional parameters, rotor-side current loop time constant, phase-locked loop proportional control parameters, phase-locked loop time constant, etc.; group N2 (low influence group) includes grid-side current loop proportional parameters, grid-side current loop time constant, etc.

[0068] Constructing the improved PSO algorithm speed update equation:

[0069]

[0070] in, Let be the velocity of particle i in the j-th dimension during the t-th iteration. This is a dynamic inertia weight, whose value is dynamically adjusted with the number of iterations according to a nonlinear exponential decay strategy. To address the nonlinear characteristics of parameter sensitivity, a high inertia weight is maintained in the initial stage, while the inertia weight coefficient decays with increasing iteration count. C2 is a hierarchical cognitive factor, with learning rates set independently for different parameters. r1 and r2 are uniformly distributed random numbers, and r1, r2 ~ U(0,1). The historical best position of particle i (in the j-th dimension). The optimal position for the group (in the j-th dimension).

[0071] Constructing the improved PSO algorithm position update equation:

[0072]

[0073] Among them, among them, Let i be the position of particle i in the j-th dimension during the t-th iteration. , Let be the upper and lower bounds of the j-th dimension parameter. To prevent particles from oscillating near the boundary and improve system stability, the number of particles is set to 80, and the maximum number of iterations is set to 150.

[0074] Step 220: Construct the fitness composite function:

[0075] Constructing the amplitude-frequency error of the PSO algorithm:

[0076]

[0077] Among them, Z est (f) represents the estimated impedance value, Z real (f) represents the actual impedance value;

[0078] Constructing the phase frequency error E of the improved PSO algorithm phase The calculation method is as follows:

[0079]

[0080] Among them, unwrap is the phase unwinding operation, which is mainly used to correct the periodic jumps of 360° integer multiples that may exist in the impedance phase data due to measurement or calculation reasons, to ensure that the phase difference reflects the real physical deviation rather than the illusion caused by the periodic jump, Z anti-phase angle;

[0081] Constructing the parameter grouping error E of the improved PSO algorithm param The calculation method is as follows:

[0082]

[0083] Where, x j The current parameter value, gbest j Let be the optimal position of the group in dimension j.

[0084] The fitness function of the improved PSO algorithm is: Fitness = 0.4E mag +0.4E phase +E param .

[0085] Step 230: Iterative optimization:

[0086] Velocity Update: An improved PSO algorithm is used for the velocity update equation, where the inertia weight... Starting from 0.97, it decays to 0.5 using a non-linear exponential strategy. The cognitive factor C is selected as either C1 or C2 based on the parameter's group (N1 or N2).

[0087] Position Update: A position update formula with boundary damping is used to prevent particles from oscillating at the search boundary. The fitness value of each particle at its new position is calculated. The optimal position of each particle is then updated. and global optimal position The algorithm terminates when the maximum number of iterations is reached or the fitness value meets the accuracy requirements. It outputs the global optimal position, which is the identified optimal set of control parameters.

[0088] by Figure 2 The Bode plot was used as the fitting target, and the improved PSO algorithm of this invention was used for identification. The identification results are as follows: Figure 4 As shown in Table 1.

[0089] Table 1 Identification Results

[0090]

[0091] Depend on Figure 4 It can be seen that the Bode plot calculated after parameter identification by the method of this invention is similar to the target Bode plot. Figure 2 The fitting effect is excellent, and the curves basically overlap.

[0092] Table 1 shows that for the N1 group of core parameters (K) p_RSC， T i_RSC The identification error was 0%; for N2 sets of parameters, the identification results were also within an acceptable range. This proves that the method of the present invention can prioritize the identification accuracy of key parameters, and the effect is significant.

[0093] It should be noted that for those skilled in the art, several improvements, substitutions, modifications and refinements can be made without departing from the principles and spirit of this invention, and these improvements, substitutions, modifications and refinements should also be considered within the scope of protection of this invention.

Claims

1. A method for identifying control parameters of a black-box model of a new energy power unit based on an improved PSO algorithm, characterized in that, The method is specifically as follows: Step 1: Construct the impedance model of the new energy unit, and derive its transfer function based on the impedance model; Step 2: Obtain the frequency domain response characteristics based on the transfer function; wherein, the frequency domain response characteristics are obtained by calculating the Bode plot of the transfer function; Step 3: Fit the frequency domain response characteristics based on the improved PSO algorithm to identify the control parameters; wherein, the improved PSO algorithm includes: assigning different dynamic weight values ​​to frequency points in different frequency bands; grouping the parameters to be identified according to parameter sensitivity and setting different learning rates for different groups; and Construct a system including amplitude error E mag Phase error E phase And parameter grouping error E param The composite fitness function is optimized.

2. The method for identifying control parameters of a new energy unit black-box model based on an improved PSO algorithm as described in claim 1, characterized in that, The specific method of assigning different dynamic weight values ​​to frequency points of different frequency bands is as follows: assigning a first weight value to the low frequency band, assigning a second weight value to the mid frequency band, and assigning a third weight value to the high frequency band, wherein the first weight value is greater than the second weight value, and the second weight value is greater than the third weight value.

3. The method for identifying control parameters of a new energy unit black-box model based on an improved PSO algorithm as described in claim 2, characterized in that, The low-frequency band is 1-50Hz, the mid-frequency band is 0-500Hz, and the high-frequency band is greater than 500Hz; the dynamic frequency weighting allocation function w(f) is:

4. The method for identifying control parameters of a new energy unit black-box model based on an improved PSO algorithm as described in claim 1, characterized in that, The specific steps of grouping the parameters to be identified based on parameter sensitivity and setting different learning rates for different groups are as follows: key parameters that involve the dominant mode and have high sensitivity are classified into the core parameter group, i.e., group N1, and parameters that affect the secondary dominant mode or have low sensitivity are classified into the auxiliary parameter group, i.e., group N2; a first cognitive factor C1 is set for group N1, and a second cognitive factor C2 is set for group N2, wherein C1 is greater than C2.

5. The method for identifying control parameters of a new energy unit black-box model based on an improved PSO algorithm as described in claim 1, characterized in that, The improved PSO algorithm uses a dynamically adjusted inertia weight ω in its velocity update process. (t) .

6. The method for identifying control parameters of a new energy unit black-box model based on an improved PSO algorithm as described in claim 5, characterized in that, The inertial weight ω (t) The nonlinear exponential decay strategy is dynamically adjusted with the number of iterations t.

7. The method for identifying control parameters of a new energy unit black-box model based on an improved PSO algorithm as described in claim 1, characterized in that, The improved PSO algorithm's position update process includes a boundary damping strategy.

8. The method for identifying control parameters of a new energy unit black-box model based on an improved PSO algorithm as described in claim 1, characterized in that, The composite fitness function, Fitness, is specifically: Fitness=0.4E mag +0.4E phase +E param 。 9. The method for identifying control parameters of a new energy unit black-box model based on an improved PSO algorithm as described in claim 8, characterized in that, Constructing the amplitude-frequency error E of the improved PSO algorithm mag The calculation method is as follows: Among them, Z est (f) represents the estimated impedance value, Z real (f) represents the actual impedance value; Constructing the phase frequency error E of the improved PSO algorithm phase The calculation method is as follows: Where unwrap is the phase unwinding operation, and ∠Z is the impedance phase angle; Constructing the parameter grouping error E of the improved PSO algorithm param The calculation method is as follows: Where, x j The current parameter value, gbest j Let be the optimal position of the group in dimension j.

10. The method for identifying control parameters of a new energy unit black-box model based on an improved PSO algorithm as described in claim 1, characterized in that, The construction of the impedance model for the new energy unit, and the derivation of its transfer function based on the impedance model, specifically involves: establishing the closed-loop transfer function Z on the rotor side of the converter. d,RSC The converter grid-side closed-loop transfer function Z d,GSC Phase-locked loop coupling impedance Z PLL Cross-coupling impedance Z cross Generator body impedance Z gen Crowbar circuit impedance Z Protection Grid connection interface impedance Z Grid Filtering and Coupling Terms Z Filter and dynamic coupling time constant T coupling The impedance model Z of the doubly-fed asynchronous wind turbine is derived from the function. DFIG .

Images