CN118137899A - PMSWG parameter identification method based on particle swarm optimization - Google Patents

Abstract

The invention discloses a PMSWG parameter identification method based on a particle swarm algorithm, which updates the speed and displacement of each particle by adopting a random inertia weight and an asynchronously-changed learning factor, and adjusts the inertia weight by utilizing the characteristics of the random variable, so that the algorithm jumps out of a local optimal state faster, the diversity of the population is maintained, the global searching performance of the algorithm is improved, and the parameter identification precision is improved. Meanwhile, the asynchronous variation learning factors dynamically adjust individual and group learning factors, so that the global searching capability is enhanced, the particle optimizing speed is accelerated, and the searching precision and the convergence speed are improved. The invention also adopts a voltage compensation method to compensate the voltage drop of the voltage source type inverter, ensures that the voltage output by the inverter is consistent with the given voltage, avoids voltage errors, improves the parameter identification precision, and solves the problem that the voltage drop of the conventional voltage source type inverter reduces the parameter identification precision of the permanent magnet synchronous wind driven generator.

Description

PMSWG parameter identification method based on particle swarm optimization

Technical Field

The invention relates to the technical field of permanent magnet synchronous wind power generator parameter identification, in particular to a PMSWG parameter identification method based on a particle swarm algorithm.

Background

The permanent magnet synchronous wind driven generator (PMSWG) has the advantages of high efficiency, high power density, simple structure and the like, is a nonlinear, strong-coupling and multivariable complex system, and has motor parameters which change in the working operation, thereby influencing the control performance and the working efficiency of the motor and further influencing the stability and the reliability of the system. Therefore, it is important to identify the motor parameters online.

In CN202110366193.2, an on-line parameter identification method of a permanent magnet synchronous wind driven generator comprises the following steps: randomly giving parameter initial values of stator resistance, dq axis inductance and permanent magnet flux linkage of the permanent magnet synchronous wind driven generator; collecting an electrical signal, and calculating the fitness value of the output of the adjustable model and the initial parameter; determining the current individual optimal parameter value and the group optimal parameter value, and updating the parameter values; obtaining dq axis stator current corresponding to the new parameter; judging whether to accept new parameters, and continuously updating the optimal values of the individual and group parameters; and performing rapid tempering annealing operation and outputting optimal parameters. However, the voltage source type inverter has a nonlinear factor, resulting in deviation of an ideal voltage from an actual voltage, which reduces identification accuracy.

The voltage deviation is mainly caused by voltage fluctuation due to existence of dead time, and in the parameter identification method of the permanent magnet synchronous wind driven generator based on the particle swarm algorithm of CN202310064387.6, the dead time of an inverter in the permanent magnet synchronous wind driven generator is compensated by adopting a time compensation method, and the speed and the displacement of each particle are updated by adopting an anti-predatory particle swarm algorithm. In the patent, inertial weights and learning factors are used, and in most particle swarm algorithm improvement, the inertial weights generally adopt a linearly decreasing strategy, and the initial inertial weights of the strategy can take larger values favorable for global search, but the algorithm search efficiency is low; a small value is obtained later to facilitate the convergence of the acceleration algorithm, but the algorithm is easily trapped in a locally optimal situation. Meanwhile, in the traditional particle swarm algorithm, particles lack of diversity in the later period of optimizing are easy to converge to local extremum prematurely. In addition, although the dead time is considered to adopt the anti-predation particle swarm algorithm to improve the identification accuracy of the identification method, the influence of the nonlinear factor is difficult to be accurately and effectively reflected only by the compensation of the dead time, and a certain defect still exists in the identification accuracy. In the prior art, the distortion voltage, the resistance and the flux linkage are estimated in real time by using the EKF, so that the distortion voltage can be compensated in real time, but the dynamic performance is poor; an adaptive neural network algorithm is also provided, and the compensation of nonlinear factors of the inverter is realized through a voltage feedforward compensation mode, but the robustness is not high; the Kalman filter observer is also constructed for nonlinear detection of the inverter, and high-precision compensation and the like are difficult to realize due to strong coupling of motor parameters and inverter parameters.

Disclosure of Invention

The invention aims to solve the main technical problem that the parameter identification precision of a permanent magnet synchronous wind driven generator is reduced by voltage drop of a voltage source type inverter in the prior art, and provides a PMSWG parameter identification method based on a particle swarm algorithm.

The aim of the invention is realized by the following technical scheme:

A PMSWG parameter identification method based on a particle swarm algorithm comprises the following steps:

s1, collecting data signals and initializing parameters of the data signals;

s2, calculating random inertia weight omega _r and asynchronously-changed learning factors;

S3, updating the speed and displacement of each particle according to the random inertia weight omega _r and the learning factor c _ac1、c_ac2;

s4, compensating the tube voltage drop of the inverter by adopting a voltage compensation method;

s5, calculating fitness function values of the particles, and updating individual optimal and group optimal information of the particles;

s6, judging whether a termination condition is met:

If the condition is met, outputting the best individual history position P _best and the best group history position G _best;

if the condition is not met, returning to the step S2 to continue operation.

Further, the data signals in S1 are voltage, current and angular velocity electrical signals when the sets i _d = -2 and i _d = 0.

Further, the parameters in S1 include a population size N, an inertia weight ω, an individual learning factor c ₁, and a population learning factor c ₂.

Further, the random inertial weight ω _r in S2 is expressed as:

ω_r＝ω_min+(ω_max-ω_min)×rand()+σ×randn()

Where ω _min is the minimum value of the random inertial weight, ω _max is the maximum value of the random inertial weight, rand () is a random number subject to a uniform distribution of [0,1], rand () is a random number subject to a normal distribution, and σ is a measure of the degree of deviation of the inertial weight from its mathematical expectation.

The invention adopts a random inertia weight strategy, adjusts the inertia weight by utilizing the characteristics of random variables, can lead the algorithm to jump out of a local optimal state faster, is beneficial to maintaining the diversity of the population and improving the global searching performance of the algorithm. The randomness enables particles to take larger or smaller weight values at the initial stage of operation and to take smaller or larger weight values at the later stage of calculation. When the particles are near the optimal particles, the randomly distributed inertia weights can generate relatively small values, so that the convergence speed of the algorithm is favorably increased; if the random inertia weight takes a larger value, the larger value is obtained when the adaptive function is calculated, the value is worse than the optimal value, the larger inertia weight is eliminated, and the algorithm regenerates a new inertia weight value. If the particles are farther from the optimal particles, the random inertial weights have the opportunity to produce larger values, which also contributes to faster convergence rates of the algorithm. If the value of the random inertia weight is smaller, the obtained adaptive function value is also worse than the optimal value, the smaller inertia weight is eliminated, and the inertia weight value of the algorithm is regenerated. The weight values are generated by using a random distribution method, the inertia weights can take good values in the later period, and the algorithm is not easy to adapt to function value stagnation.

Further, the asynchronously varying learning factor in S2 is expressed as:

Where c _1ini and c _2ini respectively represent initial values of the learning factor c ₁、c₂, c _1fin and c _2fin respectively represent final values of the learning factor c ₁、c₂, T represents the current iteration number, and T _max represents the maximum iteration number.

According to the invention, the learning factors can be regulated, and the particles are subjected to large-scale searching in the initial searching stage, so that high-quality particles with better diversity are obtained, and the interference of local extremum is avoided as far as possible. On the basis of introducing a random inertia weight strategy, an asynchronous variation learning factor strategy is introduced to improve a particle swarm algorithm. In the particle swarm algorithm, an individual's "self-cognition" ability is controlled by an individual learning factor c ₁, while an individual's "social cognition" ability is controlled by a population learning factor c ₂. Different change strategies are adopted to optimize the two learning factors, so that the particle swarm algorithm takes a larger value c ₁ and a smaller value c ₂ in the early stage of searching, so that the particles learn more to self-optimization and learn less to social optimization, and the global searching capability of the particles is enhanced; and in the later stage of searching, the value of c ₁ is smaller and the value of c ₂ is larger, so that the particles learn more to the social optimum and learn less to the self optimum, and the method is favorable for quickly converging to the global optimum solution.

Further, the updating of each particle velocity and displacement in S2 is expressed as:

further, the expression for compensating the tube voltage drop of the inverter in S4 is:

Where V _dc represents the dc bus voltage, and V _ce and V _df represent the triode saturation voltage drop and the diode forward voltage drop, respectively.

Further, the updating step of the optimal information in S5 includes:

S51, comparing the fitness value of each particle with the fitness value of the best position P _best of the history of the individual, and keeping the solution with the minimum fitness value to update P _best;

S52, comparing the fitness value of each particle with the fitness value of the best position G _best of the group history, and reserving a solution with the minimum fitness value for updating G _best.

Further, the fitness function in S5 is defined as:

Wherein R _s is a stator resistor, ψ _f is a permanent magnet flux linkage, L _d、L_q is an AC-DC axis inductance, i _d、i_q is an AC-DC axis current, Is the output of the adjustable model.

Further, the termination condition is that the maximum iteration number or the best position searched by the particle swarm meets a preset minimum fitness function value.

Compared with the prior art, the beneficial effects are that:

The invention adopts random inertia weight, adjusts the inertia weight by utilizing the characteristic of random variable, enables the algorithm to jump out of local optimal state faster, maintains diversity of population, improves global searching performance of the algorithm, and improves parameter identification precision. Meanwhile, the invention adopts asynchronous variation learning factors to dynamically adjust individual and group learning factors, enhances global searching capability, accelerates particle optimizing speed, and improves searching precision and convergence speed. In addition, the invention also adopts a voltage compensation method to compensate the voltage drop of the voltage source type inverter, ensures that the voltage output by the inverter is consistent with the given voltage, avoids voltage errors and improves the parameter identification precision.

Drawings

FIG. 1 is a flowchart of a method for providing PMSWG parameter identification based on a particle swarm algorithm;

fig. 2 is a schematic diagram of a current path of one phase of a PWM inverter.

FIG. 3 is a graph of the effect of tube pressure drop.

FIG. 4 is a graph illustrating the identification of stator resistance under normal conditions.

Fig. 5 is a graph of permanent magnet flux linkage identification under a general operating condition.

Fig. 6 is a graph of the identification of the inductance of the ac-dc axis under a general condition.

FIG. 7 is a graph of stator resistance identification for dynamic conditions.

Fig. 8 is a graph of permanent magnet flux linkage identification in the dynamic case.

Fig. 9 is a graph of the ac-dc axis inductance identification in the dynamic case.

Detailed Description

The present invention is further illustrated and described below with reference to examples, which are not intended to be limiting in any way.

Example 1

The embodiment provides a PMSWG parameter identification method based on a particle swarm algorithm, which comprises the following steps:

S1, collecting voltage, current and angular velocity electric signals when i _d = -2 and i _d =0, and initializing parameters such as population scale N, inertia weight omega, individual learning factor c ₁, population learning factor c ₂ and the like.

S2, calculating random inertia weight omega _r:

ω_r＝ω_min+(ω_max-ω_min)×rand()+σ×randn() (1)

Where ω _min is the minimum value of the random inertial weight, ω _max is the maximum value of the random inertial weight, rand () is a random number subject to a uniform distribution of [0,1], rand () is a random number subject to a normal distribution, and σ is a measure of the degree of deviation of the inertial weight from its mathematical expectation.

S3, calculating a learning factor of asynchronous change, wherein the learning factor is as follows:

Where c _1ini and c _2ini respectively represent initial values of the learning factor c ₁、c₂, c _1fin and c _2fin respectively represent final values of the learning factor c ₁、c₂, T represents the current iteration number, and T _max represents the maximum iteration number.

S4, updating the speed and displacement of each particle by a random inertia weight omega _r and an asynchronously-changed learning factor c _ac1、c_ac2:

Where v _i and x _i represent the velocity and position of particle i, P _best is the individual best solution, G _best is the population best solution, and r ₁ and r ₂ are random numbers between 0 and 1.

S5, compensating the voltage drop of the inverter in the permanent magnet synchronous wind driven generator by adopting a voltage compensation method, wherein the voltage compensation method is utilized to enable the actual value of the voltage to be the same as the given value, and the expression for compensating the voltage drop of the inverter is as follows:

Where V _dc represents the dc bus voltage, and V _ce and V _df represent the triode saturation voltage drop and the diode forward voltage drop, respectively.

S6, defining a fitness function as follows:

Wherein R _s is a stator resistor, ψ _f is a permanent magnet flux linkage, L _d、L_q is an AC-DC axis inductance, i _d、i_q is an AC-DC axis current, Is the output of the adjustable model;

and calculating the fitness function value of each particle.

S7, updating and recording particle individual optimal and group optimal information;

S71, comparing the fitness value of each particle with the fitness value of the best position P _best of the history of the individual, and reserving a solution with the minimum fitness value for updating P _best;

S72, comparing the fitness value of each particle with the fitness value of the best position G _best of the group history, and reserving a solution with the minimum fitness value for updating G _best.

S8, judging whether a termination condition is met, and judging whether the maximum iteration times or the best position searched by the particle swarm meet a preset minimum fitness function value or not:

If the condition is met, outputting P _best、G_best, and ending the algorithm operation;

Otherwise, the method returns to the step 2 to calculate the random inertia weight omega _r to continue operation.

Example 2

The embodiment provides a voltage compensation method for compensating the pipe voltage drop of an inverter in a permanent magnet synchronous wind driven generator.

Taking a phase a as an example, as shown in a current path of fig. 2, forward voltage drops of the switching transistor and the diode are determined by types of power electronic devices and magnitudes of currents, current directions and gate signals determine that four current paths exist, output voltage errors caused by voltage drops are different for different conducting states, and influences of the current directions on the output voltages are determined by the voltage drops as shown in fig. 3. Defining the current output direction as the positive direction, when ix >0, the diode of the lower arm is conducted when the gate signal Gp of the upper arm is 0, and the current path is shown as a path A; when the gate signal Gp of the upper arm is 1, the diode of the upper arm is turned on, and the current is in the path B. When ix <0, the gate signal Gn of the lower arm is 1, the switching triode of the lower arm is conducted, and the current is a path C; when the gate signal Gn of the lower arm is 0, the diode of the upper arm is turned on, and the path D is set.

The invention compensates the voltage drop of the inverter tube, ensures that the actual voltage output by the inverter to the motor is consistent with the ideal value of the given reference voltage, avoids the occurrence of voltage errors, and ensures that the system can run safely and stably. The method enables the actual value of the voltage to be the same as the given value, reduces the calculated error, thereby improving the accuracy of PMSWG parameter identification, stabilizing the voltage waveform and enabling the real-time identification model to be more accurate.

Example 3

The embodiment provides a simulation experiment example of PMSWG parameter identification method based on a particle swarm algorithm.

The embodiment is set to be speed closed-loop control, the rotating speed of a given motor is 1000r/min, the rated torque is 10N.m, the population number is 30, the iteration number is the ratio of the running time to the sampling time, the learning factor c ₁＝2.5-0.5,c₂ =1.0-2.25, the inertia weight omega=0.8, the running time of a simulation system is 0.4s, and the sampling time is 1e-6s. The effect of the PMSWG parameter identification method based on the particle swarm algorithm and the traditional particle swarm algorithm under the general working condition and the dynamic condition is detected by the method, and the detection result is as follows:

(1) General working conditions: operating at torque 10n.m and rotational speed 1000r/min and taking the average as the final output value, the detection results are shown in table 1 below:

TABLE 1

The stator resistance identification curve shown in fig. 4 is analyzed from two aspects of identification error and convergence time: the convergence time of the traditional particle swarm algorithm is 0.35s, and the identification error is 4.8851%; the convergence time of the method is 0.09s, and the identification error is 1.0752%. The method disclosed by the invention is closer to the reference value curve, has smaller identification error, and has the advantages of short convergence time and high identification precision.

As shown in fig. 5, the identification value of the resistance of the conventional particle swarm algorithm is about 5.4187% different from the reference value, and the convergence time is 0.36s; the identification error of the method is reduced by 2.6820 percent, and the convergence time is shortened by 0.26s.

As shown in fig. 6, the identification error of the conventional particle swarm algorithm resistance is 5.1333% according to the alternating-direct axis inductance curve; the identification error of the method is only 1.1000%.

(2) Dynamic conditions: in 0.1 seconds, the torque increased from 10n.m to 13n.m, the speed increased from 1000r/min to 1300r/min, and the average value was taken as the final output value, and the detection results are shown in table 2 below:

TABLE 2

Parameters (parameters)	Traditional particle swarm algorithm	The invention is that
			Stator resistor/omega	1.1064	1.0477
Error/%	6.9295	1.2564
			Stator AC/DC axis inductance/mH	1.3744	1.3009
Error/%	7.0238	1.3004
			Permanent magnet flux linkage Wb	0.1896	0.1796
Error/%	6.9977	1.3544

As shown in the identification curve of the stator resistance in fig. 7, the conventional particle swarm algorithm has large fluctuation of the identification curve, the identification error is 6.9295%, and compared with the general case, the identification error is increased by 2.0444%; the identification error of the method is only 1.2564%, compared with the common situation, the identification error is increased by 0.1812%, and the overall identification accuracy is good.

As shown in the permanent magnet flux linkage identification curve in fig. 8, the identification accuracy of the conventional particle swarm algorithm is greatly influenced by system fluctuation caused by the change of working conditions, and the flux linkage identification accuracy is reduced by 1.5790%; compared with the common situation, the identification error of the method is 1.3544%, and the flux linkage identification precision is reduced by 0.2597%.

As shown in the alternating-direct axis inductance identification curve in fig. 9, the response time of the conventional particle swarm algorithm is long, 42s, and the identification error is 7.0238%; the response time of the method is 16s, the identification error is 1.3004%, the dynamic performance is better, and the method is more suitable for being applied to the field of high-precision control.

It is to be understood that the above examples of the present invention are provided by way of illustration only and not by way of limitation of the embodiments of the present invention. Other variations or modifications of the above teachings will be apparent to those of ordinary skill in the art. It is not necessary here nor is it exhaustive of all embodiments. Any modification, equivalent replacement, improvement, etc. which come within the spirit and principles of the invention are desired to be protected by the following claims.

Claims (10)

1. A PMSWG parameter identification method based on a particle swarm algorithm is characterized by comprising the following steps:

s1, collecting data signals and initializing parameters of the data signals;

s2, calculating random inertia weight omega _r and asynchronously-changed learning factors;

S3, updating the speed and displacement of each particle according to the random inertia weight omega _r and the learning factor c _ac1、c_ac2;

s4, compensating the tube voltage drop of the inverter by adopting a voltage compensation method;

s5, calculating fitness function values of the particles, and updating individual optimal and group optimal information of the particles;

s6, judging whether a termination condition is met:

If the condition is met, outputting the best individual history position P _best and the best group history position G _best;

if the condition is not met, returning to the step S2 to continue operation.

2. The method according to claim 1, wherein the data signals in S1 are voltage, current and angular velocity signals when i _d = -2 and i _d = 0.

3. The method according to claim 1, wherein the parameters in S1 include population size N, inertial weight ω, individual learning factor c ₁ and population learning factor c ₂.

4. The method for identifying PMSWG parameters based on the particle swarm algorithm according to claim 1, wherein the random inertial weight ω _r in S2 is expressed as:

ω_r＝ω_min+(ω_max-ω_min)×rand()+σ×randn()

Where ω _min is the minimum value of the random inertial weight, ω _max is the maximum value of the random inertial weight, rand () is a random number subject to a uniform distribution of [0,1], rand () is a random number subject to a normal distribution, and σ is a measure of the degree of deviation of the inertial weight from its mathematical expectation.

5. The method for identifying PMSWG parameters based on a particle swarm algorithm according to claim 1, wherein the learning factor of the asynchronous variation in S2 is expressed as:

Where c _1ini and c _2ini respectively represent initial values of the learning factor c ₁、c₂, c _1fin and c _2fin respectively represent final values of the learning factor c ₁、c₂, T represents the current iteration number, and T _max represents the maximum iteration number.

6. The method for identifying parameters of PMSWG based on a particle swarm algorithm according to claim 1, wherein the updating of the velocity and the displacement of each particle in S2 is expressed as:

7. the method for identifying parameters PMSWG based on a particle swarm algorithm according to claim 1, wherein the expression for compensating the voltage drop of the inverter in S4 is:

Where V _dc represents the dc bus voltage, and V _ce and V _df represent the triode saturation voltage drop and the diode forward voltage drop, respectively.

8. The method for identifying parameters of PMSWG based on a particle swarm algorithm according to claim 1, wherein the updating step of the optimal information in S5 includes:

S51, comparing the fitness value of each particle with the fitness value of the best position P _best of the history of the individual, and keeping the solution with the minimum fitness value to update P _best;

S52, comparing the fitness value of each particle with the fitness value of the best position G _best of the group history, and reserving a solution with the minimum fitness value for updating G _best.

9. The method for identifying PMSWG parameters based on a particle swarm algorithm according to claim 1, wherein the fitness function in S5 is defined as:

Wherein R _s is a stator resistor, ψ _f is a permanent magnet flux linkage, L _d、L_q is an AC-DC axis inductance, i _d、i_q is an AC-DC axis current, Is the output of the adjustable model.

10. The method for identifying PMSWG parameters based on a particle swarm algorithm according to claim 1, wherein the termination condition is that a maximum number of iterations or a best position searched by the particle swarm satisfies a preset minimum fitness function value.