A parameter identification method for ship steering gear model based on WBRB-PSO

Abstract

The present invention relates to a method for identifying the parameters of a ship steering gear model based on WBRB-PSO. The method first provides a ship steering gear model and collects the command rudder angle and actual rudder angle data of the steering gear. Then, the parameters such as the speed, position, inertia weight of each particle in the particle swarm optimization algorithm are initialized, and the search range of the parameters to be identified is set. Secondly, a weighted confidence rule base reasoning structure is introduced, and an estimated value of the inertia weight increment is obtained through confidence reasoning, and the inertia weight is updated. Subsequently, based on the inertia weight of each particle that has been obtained, the speed and position of each particle are updated in combination with the global optimal value and the individual optimal value of the particle. Finally, the fitness value of the particle is calculated in combination with the collected data and the optimal position of the particle, and the global optimal value and the individual optimal value of the particle are updated, and the process is iterated in sequence until the stopping condition is met, thereby realizing the identification of the transfer function model parameters. The present invention effectively improves the identification accuracy of the ship steering gear model parameters.

Description

Technical Field

[0001] The present invention relates to a ship steering gear model parameter identification method based on WBRB-PSO, and belongs to the field of ship parameter intelligent identification. Background Art

[0002] Establishing an accurate servo model is crucial for studying the dynamic characteristics of servo systems. However, due to the complex structure of servos, accurate determination of model parameters through methods such as mechanism analysis is difficult. Therefore, it is necessary to develop effective system identification strategies and conduct dynamic performance analysis of servos. System identification involves optimizing the model parameters of the identified system within a pre-set model according to optimization criteria to find the best fit to the data. Numerous methods have been proposed for servo parameter identification, including the least squares method, frequency characteristics method, and correlation function method. While these classic algorithms can improve the identification accuracy of model parameters to a certain extent, they often suffer from limitations in terms of optimization speed and accuracy.

[0003] In recent years, with the continuous development and maturity of intelligent control theory, an increasing number of intelligent optimization algorithms, such as evolutionary algorithms and support vector machines, have been gradually applied to parameter identification. As a type of evolutionary algorithm, the particle swarm optimization algorithm (PSO) can effectively solve complex nonlinear problems while ensuring that the model has strong robustness. In addition, compared with other evolutionary algorithms, the particle swarm optimization algorithm (PSO) has the advantages of simple algorithmic rules, easy implementation, fast convergence speed, and fewer adjustable parameters, making it easier to obtain the optimal solution to the optimization problem. At the same time, an improvement scheme is proposed to address the problem of improper selection of particle inertia weights in traditional particle swarm algorithms, which causes the optimization results to easily fall into local optimality. By introducing a weighted belief rule base (WBRB) structure, the real-time update of the particle inertia weights is achieved, which effectively improves the problem of being prone to falling into local optimality to a certain extent and effectively improves the identification accuracy of the ship steering gear model parameters. Summary of the Invention

[0004] Aiming at the shortcomings of existing system identification methods, the present invention proposes a ship steering gear model parameter identification method based on WBRB-PSO. Based on the standard particle swarm optimization algorithm (PSO), this method introduces a weighted belief rule base (WBRB) structure, which realizes the online real-time adjustment of the inertia weight w of each particle and the estimation of the parameters to be identified, effectively improving the identification accuracy of the ship steering gear model parameters.

[0005] The WBRB-PSO-based ship steering gear model parameter identification method proposed in this invention includes the following steps:

[0006] (1) When no load is applied, the ship steering gear model can be expressed as a transfer function:

[0007]

[0008] In formula (1), G represents the transfer function of the servo, δ E is the command rudder angle, δ is the actual rudder angle, K E and T E They represent the control gain and time constant of the servo respectively;

[0009] (2) The command rudder angle δ of the servo is collected in real time through the rudder angle sensor E and the actual rudder angle δ signal, set the sampling period T = 0.02s, sample K samples in total, take K ≥ 300, and convert δ E (k),δ(k) is expressed as the sample set S={[δ E (k),δ(k)]|k=1,2,...,K}, where [δ E (k),δ(k)] is a sample vector;

[0010] (3) The command rudder angle δ collected in step (2) is E The actual rudder angle δ signal is used as the input of the identification model, and the output of the identification model is the steering gear model parameter (K E ,T E ), construct the servo model parameters K E and T E The WBRB-PSO identification model is:

[0011] (3-1) Based on the particle swarm optimization algorithm (PSO), the servo model parameters are identified, and the parameter identification problem is transformed into a parameter optimization problem;

[0012] (3-2) The weighted belief rule base (WBRB) reasoning model is introduced to obtain the inertia weight of particles in the PSO algorithm through belief reasoning;

[0013] (3-3) After obtaining the updated position and velocity of the particle according to steps (3-1) and (3-2), the fitness value of the particle is calculated, and its individual optimal value and global optimal value are updated;

[0014] (3-4) Repeat steps (3-1) to (3-3) until the maximum number of iterations is reached, stop the optimization, and finally obtain the parameter K to be identified. E and T E The optimal value of .

[0015] The proposed method for identifying ship steering gear model parameters based on WBRB-PSO determines the parameters to be identified and the input and output characteristics of the identification model based on the identification model of the ship steering gear parameters. It initializes parameters such as particle velocity, position, and inertia weight, and sets a search range for the parameters to be identified. It updates the velocity and position of each particle based on the global optimal value and the particle's individual optimal value. It constructs a weighted belief rule inference model (WBRB) for inertia weight, obtains an estimate of the inertia weight increment through belief reasoning, and updates the particle's inertia weight. It calculates the particle's fitness value, updates the global optimal value and the particle's individual optimal value, and iterates until the stopping condition is met, achieving transfer function model parameter identification. This method combines the traditional particle swarm optimization algorithm with the belief rule inference method to construct a weighted belief rule inference model (WBRB) for inertia weight, obtains an estimate of the inertia weight increment through belief reasoning, and updates the inertia weight online in real time, effectively improving the accuracy of parameter identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a flowchart of the program of the method of the present invention;

[0017] Figure 2 This is the structural diagram of the ship steering gear model;

[0018] Figure 3 This is a simulation diagram of the ship steering gear parameter identification results. DETAILED DESCRIPTION

[0019] The WBRB-PSO-based ship steering gear model parameter identification method proposed in this invention has a flow chart as shown in the following figure: Figure 1 As shown, it includes the following steps:

[0020] (1) When no load is applied, the ship steering gear model can be expressed as a transfer function:

[0021]

[0022] In formula (1), G represents the transfer function of the servo, δ E is the command rudder angle, δ is the actual rudder angle, K E and T E They represent the control gain and time constant of the servo respectively.

[0023] (2) The command rudder angle δ of the servo is collected in real time through the rudder angle sensor E and the actual rudder angle δ signal, set the sampling period T = 0.02s, sample K samples in total, take K ≥ 300, and convert δ E (k),δ(k) is expressed as the sample set S={[δ E (k),δ(k)]|k=1,2,...,K}, where [δE (k),δ(k)] is a sample vector.

[0024] (3) The command rudder angle δ collected in step (2) is E The actual rudder angle δ signal is used as the input of the identification model, and the output of the identification model is the steering gear model parameter (K E ,T E ), construct the servo model parameters K E and T E WBRB-PSO identification model; specifically:

[0025] (3-1) Based on the particle swarm optimization algorithm (PSO), the servo model parameters are identified, and then the parameter identification problem is transformed into a parameter optimization problem.

[0026] (3-2) The weighted belief rule base (WBRB) reasoning model is introduced to obtain the inertia weight of particles in the PSO algorithm through belief reasoning.

[0027] (3-3) After obtaining the updated position and velocity of the particle according to steps (3-1) and (3-2), the fitness value of the particle is calculated, and its individual optimal value and global optimal value are updated.

[0028] (3-4) Repeat steps (3-1) to (3-3) until the maximum number of iterations is reached, stop the optimization, and finally obtain the parameter K to be identified. E and T E The optimal value of .

[0029] The specific steps of step (3-1) are as follows:

[0030] (3-1-1) Initialize the various parameters in the particle swarm optimization algorithm. Specifically, define the target search space dimension N (that is, the number of parameters to be identified, satisfying N = 2), and define the particle swarm as S p ={p i |i=1,2,...,M}(50≤M≤500), maximum number of iterations H max (50≤H max ≤200), initialize particle p i Position X i , speed V i , individual optimal value P i , and initialize the global optimal value P of the particle swarm g , initial inertia weight w and target search space [x n,min ,x n,max ], where p i represents the i-th particle in the particle swarm, M represents the total number of particles, and x n,min and xn,max They represent the upper and lower boundaries of the n-dimensional target search space, that is, the upper and lower boundaries of the n-th parameter to be identified, and K E ∈[x 1,min ,x 1,max ],T E ∈[x 2,min ,x 2,max ],n=1,2.

[0031] (3-1-2) Define the process h(1≤h <H max ) iterations, the global optimal value The corresponding fitness value is Particle p i The inertia weight is Location is Speed ​​is Individual optimal value And its corresponding fitness value is in and (n=1,2) respectively represent the number of iterations when the particle p is h. i The position of particle p in the n-th dimension target search space i The historical individual optimal position and the global optimal position among the M individual optimal positions satisfy When the h+1th iteration is performed, particle p i Location and speed It can be updated according to the following formula:

[0032]

[0033]

[0034] In formula (2), r l , r2 are random numbers uniformly distributed in the interval [0,1], c1 and c2 are the positive acceleration constants of the particles, satisfying 0≤c1,c2≤4. The positions and velocities of M particles can be updated using equations (2) and (3); Indicates that the particle p in the h+1th iteration process i The inertia weight can be obtained by introducing the weighted belief rule base (WBRB) reasoning model.

[0035] For ease of understanding, here are steps (3-1-1) to (3-1-2) as an example. The process is as follows:

[0036] Initialize the parameters of the particle swarm, define the target search space dimension N = 2, and the particle swarm is S p ={p i|i=1,2,...,M}, M=100, maximum number of iterations H max =50, initial inertia weight w = 0.7, target search space K E ∈[0,1],T E ∈[10,50]; and initialize the position and velocity of each particle. Due to too many parameters, only the position and velocity parameters of particle p1 after initialization are listed here: position X1 = [0,0], velocity V1 = [0.129,0.855].

[0037] After h=1 iterations, the parameters of particle p1 are: positive acceleration constant c1=c2=0.6; r1=r2=0.5; inertia weight is Speed ​​is Location is The individual optimal value is and its corresponding fitness value Global optimal value And its corresponding fitness value is

[0038] From formula (2) and formula (3), we can see that the speed and position update process of particle p1 in the second iteration is as follows:

[0039]

[0040]

[0041]

[0042]

[0043] The specific steps of step (3-2) are as follows:

[0044] (3-2-1) A confidence rule base containing L rules is constructed to describe the complex nonlinear relationship between the input variables f1(h+1,i-1),f2(h+1,i-1) and the output variable Δw(h+1,i-1), where Δw(h+1,i-1) represents the increment of the inertia weight. The lth rule in the established confidence rule base is:

[0045]

[0046] In formula (4), f1(h+1,i-1) represents the particle p in the h+1th iteration process. i-1 Inertia weight f2(h+1,i-1) is defined as particle p i-1 The best performance evaluation value NCBPE( ),satisfy where Q min , Q max Respectively represent the upper and lower boundary values ​​of the particle fitness value; represents the reference level of the θth input variable in the lth rule, The elements satisfy where K θ R represents the number of reference levels of the θth input variable, totaling L = K1 × K2 rules; l The consequent attribute is D={D j |j=1,2,...,J}, satisfying D1 <D2<…<D J ρ 1,l ,ρ 2,l ,...,ρ J,l To assign to D1,D2,...,D J Confidence, satisfying

[0047] (3-2-2) Obtain the sample data generated during the h+1th iteration f(h+1,i-1)=[f1(h+1,i-1),f2(h And there is Calculate f θ The matching degree of (h,i) relative to its reference value a θ,q :

[0048]

[0049] (3-2-3) Calculate the matching degree of the sample data f(h+1,i-1) relative to its reference value, and fuse the L rules activated by f(h+1,i-1) to obtain the confidence level of the output reference value corresponding to the input sample data f(h+1,i-1) = [f1(h+1,i-1),f2(h+1,i-1)]

[0050]

[0051]

[0052] In formula (6), the activation weight w l ∈[0,1], is the relative attribute weight, satisfying δ θ represents the variable importance factor; Enter f for rule l θ (h+1,i-1) and its reference value degree of matching; Indicates the weight of the lth rule.

[0053] (3-2-5) According to step (3-2-4), the increment of inertia weight Δw(h+1,i-1) can be obtained, and then the particle p in the h+1th iteration process can be obtained. i Inertia weight

[0054]

[0055]

[0056] For ease of understanding, the reasoning process of Δw(2,1) is illustrated here. The confidence rule base is a model with two inputs and one output, and the input and output reference values ​​of the model are set as follows:

[0057] Table 1 Semantic values ​​and reference values ​​of input and output

[0058]

[0059] The fitness value of particle p1 in the second iteration By formula Converted into the current best performance evaluation value of particle p1 where Q min , Q max Respectively represent the upper and lower boundary values ​​of the particle fitness value, and Q min =0, Q max =30, the conversion process is as follows:

[0060]

[0061] The inertia weight of particle p1 will be obtained and the best performance evaluation value into input sample data The corresponding reference value intervals are [0.65,0.75] and [0,2] respectively.

[0062] From formula (5), we can see that the sample input f1(2,1) and A 1,4 and A 1,5 The matching degree α 1,4 =0.4092,α 1,5 =0.5908, input f2(2,1) and A 2,1 and A 2,2 The matching degree α 2,1 =0.9973,α 2,2 =0.0027.

[0063] According to formula (7), the activation weight of each rule can be obtained as shown in the following table:

[0064] Table 2 Activation weight of each rule

[0065]

[0066] All rules are fused according to formulas (6) and (7), and the fusion results are as follows:

[0067] Table 3 Confidence of output reference values

[0068]

[0069]

[0070] The increment of inertia weight can be estimated from the known fusion result using formula (8):

[0071] Δw(2,1)=(-0.3×0.1807)+(-0.2×0.5772)+(-0.1×0.1264)+(0×0.0579)+(0.1×0.0289)+(0.2×0.0289)+(0.3×0)

[0072] =-0.1736

[0073] According to formula (9), we can get:

[0074] It should be noted here that the inertia weight value of particle p1 in the h+1th iteration is the inertia weight of particle M in the hth iteration. It is calculated by adding the estimated value of the inertia weight increment Δw(h,M), that is, in the hth iteration, the inertia weight of particle M is With the best performance evaluation value As the input of the weighted belief rule reasoning model (WBRB), the estimated value of the inertia weight increment Δw(h,M) is obtained through belief reasoning, and then the formula The inertia weight value of particle p1 in the h+1th iteration process can be calculated.

[0075] The specific steps of step (3-3) are as follows:

[0076] (3-3-1) For particle p i , its fitness value The calculation is as follows:

[0077]

[0078] In formula (11), Indicates that during the h+1th iteration, particle p i The location is When the parameter to be identified is Estimated rudder angle:

[0079]

[0080] (3-3-2) Combine the particles p obtained in step (3-1) and step (3-3-1) i The fitness values ​​corresponding to the individual optimal values ​​after the hth and h+1th iterations respectively and Compare and Take the particle p corresponding to the minimum fitness value between the two i The position of the h+1 generation particle p i The individual optimal value of

[0081] (3-3-3) According to step (3-3-2), the fitness value corresponding to the individual optimal value of the h+1th generation M particles can be obtained Combined with the global optimal value known in step (3-1) The corresponding fitness value is Pick The position of the particle corresponding to the minimum fitness value is taken as the global optimal value of the h+1 generation

[0082] The specific steps of the steps (3-4) are as follows:

[0083] After obtaining the global optimal value of the h+1 generation Then, repeat steps (3-1) to (3-3) to obtain the global optimal value after the h+2 generation update. By analogy, the global optimal value is continuously updated through recursive iteration until the maximum number of iterations is H max When , stop updating, and finally get the global optimal value of the group history, that is, the parameter K to be identified E and T E The optimal value of .

[0084] For ease of understanding, steps (3-3) to (3-4) are described here as an example. The process is as follows:

[0085] From step (3-1), we can know that the updated speed and position of particle p1 are: According to formulas (10) and (11), the fitness value of M (M = 100) particles after the second iteration can be calculated: According to step (3-3), the fitness value of particle p1 is obtained Compared with the fitness value corresponding to the individual optimal value of the particle after the h=1th iteration obtained in step (3-1), From step (3-3-1), we can see that the individual optimal value of particle p1 is the fitness value The corresponding position, that is

[0086] According to step (3-3-3), the fitness value of M particles after h=2 iterations is compared with the fitness value corresponding to the global optimal value after h=1 iterations. For comparison, the particle position corresponding to the minimum fitness value is the global optimal value after the h=2th iteration. The minimum fitness value is Then the global optimal value after h=2 iterations is the fitness value The corresponding particle position is The global optimal value is continuously updated through recursive iteration until the maximum number of iterations is H max When , stop updating, and finally get the global optimal value of the group history That is, the identification result is K E =0.3669 and T E =10.

[0087] The following describes in detail the embodiments of the method of the present invention with reference to the accompanying drawings:

[0088] The core of the present invention is to construct a WBRB-PSO identification model for ship steering gear model parameters, describing the nonlinear relationship between model input (command rudder angle and actual rudder angle signals) and model output (steering gear model parameters). First, a ship steering gear model is given, and the command rudder angle and actual rudder angle data of the steering gear are collected for use in the identification model. Then, the particle swarm optimization (PSO) algorithm is initialized with parameters such as the velocity, position, and inertia weight of each particle, and a search range for the parameters to be identified is set. Secondly, a weighted belief rule base (WBRB) reasoning structure is introduced to obtain an estimated value of the inertia weight increment through belief reasoning, and the inertia weight is updated. Subsequently, based on the acquired inertia weight of each particle, the velocity and position of each particle are updated in combination with the global optimal value and the individual optimal value of the particle. Finally, the fitness value of the particle is calculated based on the collected data and the optimal position of the particle, and the global optimal value and the individual optimal value of the particle are updated. The iteration is carried out in sequence until the stopping condition is met, thereby achieving the transfer function model parameter identification.

[0089] The following describes in detail the various steps of this embodiment in conjunction with the SM40BL model marine steering gear.

[0090] 1. Provide the parameter model of the ship's steering gear

[0091] Depend on Figure 2 It can be seen that the ship steering system consists of a control driver, a servo motor, a reduction gear and a sensor. Usually, without load, the ship steering model can be simplified to the form of a transfer function:

[0092]

[0093] From the transfer function, we can see that the parameter to be identified is K E and T E ; From the two parameters to be identified, we know that the target search space N = 2, and set the target search boundary to K E ∈[0,1],T E =[10,50];

[0094] 2. Acquisition and feature processing of experimental data

[0095] The rudder angle sensor is used to collect the rudder angle δ of the steering gear on the ONRT warship model in real time. E and the actual rudder angle δ signal, and use it as the input of the identification model; the parameters to be identified (K E , T E ) is the output of the identification model; set the sampling period T = 0.02s, sample K = 376 groups of sample data in total, and set δ E (k), δ(k) is expressed as a sample set S = {[δ E (k),δ(k)]|k=1,2,...,K}, where [δ E (k),δ(k)] is a sample vector, and (δ E ,δ)∈[-35°,35°];

[0096] 3. Establish the identification model of particle swarm algorithm

[0097] (3.1) Generate particle population and initialize

[0098] Randomly generate M = 100 particles as the initial population of the identification model, and the maximum number of iterations H max =50, and initialize the particle parameters such as particle position, velocity, inertia weight, etc. The parameters of particle p1 after initialization are: X1 = [0, 0], V1 = [0.129, 0.855], inertia weight w1 = 0.7, acceleration constant c1 = c2 = 0.6; r1 = r2 = 0.5;

[0099] (3.2) Update the particle speed

[0100] After h=1 iterations, the parameters of particle p1 are: positive acceleration constant c1=c2=0.6; r1=r2=0.5; inertia weight is Speed ​​is Location The individual optimal value is and its corresponding fitness value Global optimal value , and its corresponding fitness value is

[0101] According to formula (2) and formula (3) of the present invention, the speed and position update process of particle p1 during the h=2th iteration is as follows:

[0102]

[0103]

[0104]

[0105]

[0106] The estimated value of the inertia weight w can be obtained by constructing a weighted belief rule reasoning (WBRB) method. Taking the acquisition process as an example, the specific process is as follows:

[0107] (a) Select reference values ​​of confidence rule characteristic parameters and construct a confidence matrix table

[0108] According to step (3-2) of the present invention, the result reference value set D = {-0.3, -0.2, -0.1, 0, 0.1, 0.2, 0.3} of the increment of inertia weight is set, where the input f1 is the inertia weight of the particle swarm algorithm parameter The input reference value is A1 = {0.35, 0.45, 0.55, 0.65, 0.75, 0.85, 0.95}; input f2 is the current optimal performance value of the particle Its input reference value set A2 = {0, 2, 4, 6, 8, 10}, and the specific settings of the reference values ​​(semantic values) of each input variable and output variable are shown in Table 4.

[0109] Table 4 Reference values ​​of inputs and outputs

[0110]

[0111] Furthermore, the lth rule in the confidence rule base can be given as:

[0112]

[0113] A total of L=7×6=42 rules can be generated.

[0114] (b) Estimated output calculated by the confidence rule base

[0115] The fitness value of particle p1 in the h=2th iteration By formula The best performance evaluation value converted into particles where Q min, Q max Respectively represent the upper and lower boundary values ​​of the particle fitness value, and Q min =0, Q max =30, the conversion process is as follows:

[0116]

[0117] The obtained inertia weight and best performance evaluation value of particle p1 in the h=2th iteration process are converted into the input sample data f(2,1)=[f1(2,1), The corresponding reference value intervals are [0.65,0.75] and [0,2] respectively.

[0118] From formula (5), we can see that the sample input f1(2,1) and A 1,4 and A 1,5 The matching degree α 1,4 =0.4092,α 1,5 =0.5908, input f2(2,1) and A 2,1 and A 2,2 The matching degree α 2,1 =0.9973,α 2,2 =0.0027.

[0119] According to formula (7) of the present invention, the activation weight of each rule can be obtained as shown in the following table:

[0120] Table 5 Activation weight of each rule

[0121]

[0122] The estimated output Δw(2,1) of the confidence rule base, i.e., the increment of the inertia weight, is calculated through steps (3-2-4) and (3-2-5) of the method of the present invention. All rules are integrated according to formulas (6) and (7), and the calculation can be obtained: Then the estimated value of the increment of inertia weight from formula (8) is:

[0123] Δw(2,1)=(-0.3×0.1807)+(-0.2×0.5772)+(-0.1×0.1264)+(0×0.0579)+(0.1×0.0289)+(0.2×0.0289)+(0.3×0)

[0124] =-0.1736

[0125] Then the inertia weight is given by formula (9): The values ​​are:

[0126] (3.3) Update the individual optimal value and global optimal value of the particle

[0127] According to step (3-3-1) of the present invention, the position of the updated particle p1 is brought into formula (10) and formula (11) to calculate the particle fitness value The fitness value of particle p1 is obtained Compared with the fitness value corresponding to the individual optimal value of the particle after the h=1th iteration obtained in step (3-1), From step (3-3-1), we can see that the individual optimal value of particle p1 is the fitness value The corresponding position, that is By analogy, the fitness value of 100 particles after the second iteration can be calculated according to formulas (10) to (11): Then, according to the individual optimal value updating method given in step (3-3-2), the individual optimal values ​​of the M particles are updated; then, according to step (3-3-3), the fitness values ​​of the M particles after the h=2th iteration are compared with the fitness values ​​corresponding to the global optimal value after the h=1th iteration. For comparison, the particle position corresponding to the minimum fitness value is the global optimal value after the h=2th iteration. The minimum fitness value is Then the global optimal value after h=2 iterations is the fitness value The corresponding particle position is .

[0128] (3.4) Perform iterative optimization of particle swarm algorithm

[0129] After obtaining the global optimal value of the second generation After that, repeat the steps (3-1) to (3-3) of the method of the present invention to obtain the global optimal value after the third iteration. By analogy, the global optimal value is continuously updated through recursive iteration until the maximum number of iterations is H max When , stop updating, and finally get the global optimal value of the group history That is, the identification result is K E =0.3669, T E = 10. The reasoning method according to the present invention obtains more accurate steering gear model parameters, so that the minimum square error between the actual steering gear angle and the estimated steering gear angle is 0.1570, while the minimum square error corresponding to the identification result obtained by the improved PSO identification method (inertia weight linear decrease) is 0.2504. Figure 3The identification results of the two methods are shown in the figure. It can be seen from the figure that the method of the present invention is significantly better than the improved PSO identification method (linear decrease of inertia weight) in the identification of steering gear parameters, and can more accurately identify the steering gear model parameters.

Claims

1. A ship steering gear model parameter identification method based on WBRB-PSO, characterized by The method comprises the following steps: (1) When no load is applied, the ship steering gear model can be expressed as a transfer function: In formula (1), G represents the transfer function of the servo, δ E is the command rudder angle, δ is the actual rudder angle, K E and T E They represent the control gain and time constant of the servo respectively; (2) The command rudder angle δ of the servo is collected in real time through the rudder angle sensor E and the actual rudder angle δ signal, set the sampling period T = 0.02s, sample K samples in total, take K ≥ 300, and convert δ E (k),δ(k) is expressed as the sample set S={[δ E (k),δ(k)]|k=1,2,…,K}, where [δ E (k),δ(k)] is a sample vector; (3) The command rudder angle δ collected in step (2) is E The actual rudder angle δ signal is used as the input of the identification model, and the output of the identification model is the steering gear model parameter (K E ,T E ), construct the servo model parameters K E and T E WBRB-PSO identification model; (3-1) Based on the particle swarm optimization algorithm PSO, the servo model parameters are identified, and then the parameter identification problem is transformed into a parameter optimization problem; (3-2) Introducing the weighted belief rule base (WBRB) inference model, the inertia weight of particles in the PSO algorithm is obtained through belief reasoning; (3-3) After obtaining the updated position and velocity of the particle according to steps (3-1) and (3-2), the fitness value of the particle is calculated, and its individual optimal value and global optimal value are updated; (3-4) Repeat steps (3-1) to (3-3) until the maximum number of iterations is reached, stop the optimization, and finally obtain the parameter K to be identified. E and T E The optimal value of .

2. A ship steering gear model parameter identification method based on WBRB-PSO according to claim 1, characterized in that: The specific steps of step (3-1) are as follows: (3-1-1) Initialize the various parameters in the particle swarm optimization algorithm, specifically: Define the target search space dimension N and the particle swarm as Maximum number of iterations H max , initialize particles Position X i , speed V i , individual optimal value P i , and initialize the global optimal value P of the particle swarm g , initial inertia weight w and target search space [x n,min ,x n,max ],in represents the i-th particle in the particle swarm, M represents the total number of particles, and x n,min and x n,max They represent the upper and lower boundaries of the n-dimensional target search space, that is, the upper and lower boundaries of the n-th parameter to be identified, and K E ∈[x 1,min ,x 1,max ],T E ∈[x 2,min ,x 2,max ],n=1,2; (3-1-2) Define the global optimal value after h iterations The corresponding fitness value is particle The inertia weight is Location is Speed ​​is Individual optimal value And its corresponding fitness value is in and They represent the number of iterations when the particle Position, particle in the n-th dimension target search space The historical individual optimal position and the global optimal position among the M individual optimal positions satisfy n=1,2; when performing the h+1th iteration, the particle Location and speed V i h +1 Update according to the following formula: In formula (2), r l , r2 are random numbers uniformly distributed in the interval [0,1], c1 and c2 are the positive acceleration constants of the particles, satisfying 0≤c1,c2≤4. The positions and velocities of M particles can be updated using equations (2) and (3); Indicates that the particle is Inertia weight.

3. A ship steering gear model parameter identification method based on WBRB-PSO according to claim 2, characterized in that: The specific steps of step (3-2) are as follows: (3-2-1) A confidence rule base containing L rules is constructed to describe the complex nonlinear relationship between the input variables f1(h+1,i-1),f2(h+1,i-1) and the output variable Δw(h+1,i-1), where Δw(h+1,i-1) represents the increment of the inertia weight. The lth rule in the established confidence rule base is: In formula (4), f1(h+1,i-1) represents the particle Inertia weight f2(h+1,i-1) is defined as particle The best performance evaluation value satisfy where Q min , Q max Respectively represent the upper and lower boundary values ​​of the particle fitness value; represents the reference level of the θth input variable in the lth rule, The elements satisfy where K θ R represents the number of reference levels of the θth input variable, totaling L = K1 × K2 rules; l The consequent attribute is D={D j |j=1,2,...,J}, satisfying D1 <D2<…<D J ρ 1,l ,ρ 2,l ,...,ρ J,l To assign to D1,D2,...,D J Confidence, satisfying (3-2-2) Obtain the sample data generated during the h+1th iteration And there is Calculate f θ The matching degree of (h,i) relative to its reference value: (3-2-3) Calculate the matching degree of the sample data f(h+1,i-1) relative to its reference value, and fuse the L rules activated by f(h+1,i-1) to obtain the confidence level of the output reference value corresponding to the input sample data f(h+1,i-1) = [f1(h+1,i-1),f2(h+1,i-1)] In formula (6), the activation weight ω l ∈[0,1], is the relative attribute weight, satisfying δ θ represents the variable importance factor; Enter f for rule l θ (h+1,i-1) and its reference value degree of matching; represents the weight of the lth rule; (3-2-5) According to step (3-2-4), the increment of inertia weight Δw(h+1,i-1) can be obtained, and then the particle in the h+1th iteration process can be obtained. Inertia weight 4. The method for identifying ship steering gear model parameters based on WBRB-PSO according to claim 3 is characterized in that: The specific steps of step (3-3) are as follows: (3-3-1) For particles Its fitness value The calculation is as follows: In formula (11), Indicates that during the h+1th iteration, the particle The location is When the parameter to be identified is Estimated rudder angle: (3-3-2) Particles obtained by combining step (3-1) and step (3-3-1) The fitness values ​​corresponding to the individual optimal values ​​after the hth and h+1th iterations respectively and Compare and Take the particle with the minimum fitness value between the two The position of the h+1 generation particle The individual optimal value P i h+1 ; (3-3-3) According to step (3-3-2), the fitness value corresponding to the individual optimal value of the h+1th generation M particles can be obtained Combined with the global optimal value known in step (3-1) The corresponding fitness value is Pick The position of the particle corresponding to the minimum fitness value is taken as the global optimal value of the h+1 generation 5. The method for identifying ship steering gear model parameters based on WBRB-PSO according to claim 4 is characterized in that: The specific steps of the steps (3-4) are as follows: After obtaining the global optimal value of the h+1 generation Then, repeat steps (3-1) to (3-3) to obtain the global optimal value after the h+2 generation update. By analogy, the global optimal value is continuously updated through recursive iteration until the maximum number of iterations is H max When , stop updating, and finally get the global optimal value of the group history, that is, the parameter K to be identified E and T E The optimal value of .

Images