A Method for Parameter Identification of Unmanned Surface Vessel Response Model Based on Improved Particle Swarm Optimization Algorithm

Abstract

This invention proposes a parameter identification method for unmanned surface vessel (USV) response models based on an improved particle swarm optimization (PSO) algorithm. The method identifies the bow motion response model and velocity response model of USVs. Compared to other complex models, this method uses easily acquired rudder angles and rotational speeds as inputs, rather than forces and torques, and has fewer model parameters, making it more practical in engineering applications. This invention employs an improved PSO algorithm that incorporates genetic principles. It introduces crossover and Gaussian white noise perturbation mutation mechanisms into the PSO algorithm, and adaptively adjusts the mutation probability with the number of stagnant generations. This avoids the PSO algorithm getting trapped in local optima during identification, improves the algorithm's success rate, and solves the premature convergence problem in parameter identification of USV response models with no prior knowledge and a large and unknown parameter range.

Description

Technical Field

[0001] This invention belongs to the field of unmanned surface vessel (USV) model parameter identification. Specifically, it relates to a method for identifying USV response model parameters based on an improved particle swarm optimization algorithm. Technical background:

[0002] Establishing accurate mathematical models of unmanned surface vessels (USVs) is fundamental to studying their motion characteristics and control problems. The most crucial step in USV modeling is determining the values ​​of each parameter within the model. Due to the complexity of USV systems, it is difficult to accurately model the mechanisms of different types of USVs using uniform methods and empirical formulas. Therefore, it is often necessary to estimate the mathematical model parameters using real-world test data. System identification provides an effective solution for establishing mathematical models for different USVs. Mathematical models of USVs are mainly classified into three types: holistic, discrete, and response-based. Among these, USV response models are widely used in research on USV maneuverability, heading control, and trajectory control.

[0003] Particle swarm optimization (PSO) is an optimization algorithm that starts from random values ​​and iteratively searches for the optimal solution. Due to its ease of implementation, high accuracy, and fast convergence, it has found applications in system identification. However, PSO requires a pre-defined search space. If prior knowledge of the model parameters is lacking before identification, and the approximate range of the parameters cannot be determined, the search space needs to be set large enough to ensure it includes the true model parameters. However, an excessively large search space can lead to slow identification speeds and even cause the algorithm to get trapped in local optima, resulting in premature convergence and failure to reach the global optimum. While initial values ​​or approximate parameter ranges can often be obtained using least squares methods, this increases the complexity of the identification process. Therefore, the aforementioned problems need to be addressed when applying PSO to identify the motion model parameters of unmanned surface vessels. Summary of the Invention

[0004] The purpose of this invention is to provide a method for identifying unmanned surface vessel (USV) response model parameters based on an improved particle swarm optimization algorithm, in order to solve the problem of identifying USV response model parameters when there is no prior knowledge and the parameter range is large and unknown.

[0005] The technical solution of this invention is: a method for identifying parameters of an unmanned surface vessel response model based on an improved particle swarm optimization algorithm, comprising:

[0006] Conduct specific unmanned vessel identification and navigation tests under ideal environmental conditions, and collect the input and output test data required for model identification;

[0007] Establish unmanned surface vessel (USV) response models, including USV bow turning motion response models and velocity response models;

[0008] The unmanned vessel's bow turning motion response model and velocity response model were organized, and the identification error index was determined.

[0009] Using the experimental data, parameter identification was performed on the unmanned vessel's bow-turning motion response model and velocity response model, respectively, and the parameter identification results of the unmanned vessel response model were obtained.

[0010] Preferably, the input and output test data required for the model identification include:

[0011] Maneuvering tests were conducted, and the collected data was used for the identification of the bow motion response model.

[0012] Using the finite difference method, the bow angular acceleration at each sampling moment is calculated based on the bow angular velocity r and rudder angle δ collected during the maneuverability test. Rate of change of bow angular acceleration Rudder angle change rate

[0013] A speed navigation test was conducted, and the collected data was used for speed response model identification. The speed navigation test procedure was as follows: before the test began, the unmanned surface vessel maintained a constant speed and straight-line navigation. Then, sinusoidal steering commands and sinusoidally varying speed control commands were sent. At the same time, the bow angular velocity r, rudder angle δ, speed V, and acceleration of the unmanned surface vessel were recorded at a fixed sampling period. The test ended after a number of sinusoidal cycles of steering commands and speed control commands, with the rotational speed n being constant.

[0014] Preferably, the test procedure for the maneuverability test is as follows: before the test begins, the unmanned vessel maintains a constant speed and straight-line navigation state, then sends a sinusoidal steering command, and at the same time records the bow angular velocity r, rudder angle δ, speed V, and rotational speed n of the unmanned vessel at a fixed sampling period, maintains this speed, and ends the test after several sinusoidal cycles of steering commands.

[0015] Preferably, the unmanned vessel's bow-turning motion response model is organized, and the identification error indicators are determined, including:

[0016] The unmanned vessel's bow-turning motion response model is represented as:

[0017] Y r =A r H r

[0018] in

[0019]

[0020]

[0021] The parameters that need to be identified in the model are the gain coefficient K, time constants T1, T2, T3, and nonlinear term coefficient α; r is the bow angular velocity. For bow angular acceleration, Here, δ is the rate of change of bow angular acceleration, and δ is the rudder angle. The rate of change of rudder angle;

[0022] Error index for identifying the bow motion response model:

[0023]

[0024] Among them, J r The error index for identifying the bow motion model is represented as a minimum objective function related to the parameters of the model to be identified, i.e., the fitness function; N r Y represents the number of sampling points for the bow maneuverability test data. ri For the i-th test sample in the steering maneuverability test, (This is the actual output value.) This is the estimated output value for the i-th test sample in the turning motion experiment.

[0025] Preferably, the unmanned vessel's bow-turning motion response model is organized, and the identification error indicators are determined, including:

[0026] The velocity response model is expressed as:

[0027] Y v =A v H v

[0028] in,

[0029] A v =[a vv a rr a δδ a nn a nv ],

[0030] H v =[-V 2 -r 2 -V 2 δ 2 n 2 nV] T ,

[0031] The parameter that needs to be identified in the model is the drag coefficient α. vv Centrifugal force coefficient a rr rudder force coefficient a δδ propeller force coefficient a nn and a nv V represents the speed of the unmanned vessel moving in the horizontal plane. Let r be the acceleration, r be the heading angular velocity, δ be the rudder angle, and n be the rotational speed;

[0032] Velocity response model identification error index:

[0033]

[0034] Among them, J v N is the fitness value of the velocity model identification error index, which is the velocity model identification error index. v Y represents the number of sampling points for speed navigation test data. vi This represents the actual output value of the i-th test sample in the speed navigation test. This is the estimated output value for the i-th test sample in the speed navigation test.

[0035] Preferably, an improved particle swarm optimization algorithm incorporating genetic principles is used to identify parameters for the unmanned vessel's bow-turning motion response model and velocity response model.

[0036] Preferably, the improved particle swarm optimization algorithm incorporating genetic concepts includes:

[0037] S1. The problem of model identification is transformed into minimizing the identification error index of the steering motion response model and the velocity response model; and the particles are encoded as X = (K, T1+T2, T1T2, T3, α) and X = (a vv ,a rr ,a δδ ,a nn ,a nv The fitness function is expressed as f(X);

[0038] S2, Initialize the motion range of the parameters;

[0039] S3. Calculate the initial fitness value of each particle in the population, and the optimal position pbest for each particle. i Initialize to the current position and find the current global best position gbest for the entire swarm; at the same time, find the initial fitness value of each particle in the swarm and the optimal fitness value of the swarm.

[0040] S4. Update the velocity and position of the particles, generate a new population, and perform out-of-bounds checks on the velocity and position of the particles.

[0041] S5. Based on the fitness value f(X) of each particle at its current position. i Update the individual optimal position and the population optimal position;

[0042] S6. Based on the hybridization probability p cSelect a portion of the particles and place them into a hybridization pool. Randomly select two parent particles from the pool and hybridize them to produce the same number of offspring particles. Calculate the position and velocity of the offspring particles. If the fitness value of the offspring is better than that of the parent, the hybridized offspring particles replace the parent particles to update the individual optimal position and the population optimal position. Otherwise, the parent particles are used.

[0043] S7. For each particle in the population, with mutation probability p n Perform perturbation and mutation, compare the mutated fitness value with the population's optimal fitness value. If the mutated fitness value is better, replace the original particles in the population with the mutated particles to update the individual's optimal position and the population's optimal position; otherwise, the original particles remain unchanged.

[0044] S8. Update the upper and lower bounds of the motion velocity parameter, and update the value of the cumulative stall algebra N;

[0045] S9: Check the termination condition. If it is met, stop the search, output the model parameter identification results, and proceed to the next generation particle processing to return to step S4. The termination condition is reaching the maximum number of generations or the fitness value is less than the preset computational precision μ.

[0046] Preferably, the initialization parameters include the local learning factor c1, the global learning factor c2, the maximum number of generations G, and the current number of generations t=1. Within the parameter range, the position and velocity matrices of m particles in the initial population are randomly generated, and the position of the i-th particle in the entire d-dimensional search space is represented by X. i =(x i1 ,x i2 ,...,x id The speed is expressed as V. i =(v i1 ,v i2 ,...,v id ).

[0047] Preferably, updating the optimal position of the individual and the optimal position of the population includes:

[0048] Set the current position fitness value f(X) of each particle. i Comparing f(X) with its own fitness extreme value, if f(X) i If the current position is better than the individual's fitness extreme value, then the current position is taken as the individual's optimal position; otherwise, the individual's historical optimal position is used.

[0049] The updated individual optimal fitness value of each particle is compared with the population optimal fitness value. If the individual optimal fitness value is better than the population optimal fitness value, the population optimal position is replaced by the individual optimal position; otherwise, the population optimal position remains unchanged.

[0050] Preferably, the position X′ of the offspring particlei 、X′ j and velocity V′ i V′ j The calculation formula is as follows:

[0051] X′ i =p·X i +(1-p)·X j

[0052] X′ j =p·X j +(1-p)·X i

[0053]

[0054]

[0055] Where p is a random number distributed on [0,1], and V i V j For the two parent particles X i X j The speed.

[0056] Preferably, the perturbed and mutated particles are:

[0057] x″ iq =gbest q ·(0.5+σ)

[0058] x″ iq For the q-th dimension position component of the i-th particle after mutation, gbest q The q-th dimension coordinates represent the optimal position of the population; σ is a random number generated by Gaussian white noise.

[0059] For particles that exceed the boundary, mutation is performed in the following manner:

[0060]

[0061] are the upper and lower bounds of the q-th dimension position component of the particle, respectively, and λ is a random number distributed between [0,1].

[0062] Preferably, the mutation probability p n Adaptively adjust according to the following formula:

[0063]

[0064] in, κ represents the maximum and minimum values ​​of the mutation probability. pγ is a constant coefficient, N is the number of generations of stagnation in fitness value, that is, the cumulative number of cycles in which the extreme value of fitness of the population changes less than a sufficiently small normal number γ, and γ is a positive number selected with the same order of magnitude as the preset fitness function calculation precision μ, or can be customized during the identification process.

[0065] Preferably, the upper and lower bounds of the update parameter motion speed are updated to update the value of the cumulative stagnation algebra N;

[0066]

[0067]

[0068] These are the maximum and minimum values ​​of the upper limit of the q-th dimension velocity component of the particle, respectively, both positive numbers. μ is the preset precision of the fitness evaluation value, and κ is the maximum and minimum values ​​of the upper limit of the upper limit of the q-th dimension velocity component of the particle. v γ is a constant coefficient, γ is a sufficiently small positive number, and γ is a positive number selected with the same order of magnitude as the preset fitness function operation precision μ, or can be customized during the identification process.

[0069] The beneficial effects of this invention;

[0070] 1. This invention identifies the response model of an unmanned surface vessel (USV), including the bow motion response model and the velocity response model. Existing USV response model identification methods typically only consider the maneuverability of the USV, that is, only the bow motion model is identified. This invention not only considers the maneuverability characteristics of the USV, but also the velocity characteristics. This model has fewer parameters. Compared with other complex models, the model's input is easily acquired rudder angle and speed, rather than force and torque. Moreover, the model has fewer parameters. Therefore, this model is more practical in research fields and engineering practices such as USV motion simulation and USV motion controller design.

[0071] 2. This invention directly employs an improved particle swarm optimization (PSO) algorithm incorporating genetic principles for parameter identification of unmanned surface vessel (USV) response models. It introduces a crossover and Gaussian white noise perturbation mutation mechanism, adaptively adjusting the mutation probability with the algorithm's stagnant generation. This avoids the PSO algorithm getting trapped in local optima during identification, improving the algorithm's success rate and solving the premature convergence problem in parameter identification of USV response models with no prior knowledge and a large, unknown parameter range. No prior knowledge of the model parameters is required before identification, nor is it necessary to use other methods for preliminary parameter identification.

[0072] 3. This invention employs an improved particle swarm optimization algorithm that incorporates genetic principles, enabling adaptive adjustment of the upper and lower bounds of particle motion speed. This allows it to adapt to the different requirements of global and local search capabilities at different stages of the identification process. Compared to the standard particle swarm optimization algorithm, it improves the identification speed when the particle search space is too large, balancing search speed and search granularity, and can quickly obtain high-precision identification results of unmanned surface vessel response model parameters.

[0073] 4. When the particle search space is too large, the method of the present invention can avoid premature convergence of the particle swarm algorithm in the identification process, improve the algorithm's optimization rate and identification speed, and take into account both search speed and search granularity, so as to quickly obtain high-precision identification results of unmanned vessel response model parameters. Attached Figure Description

[0074] Figure 1 This is a step diagram of the unmanned surface vessel response model parameter identification method of the present invention;

[0075] Figure 2 This is a flowchart of the improved particle swarm optimization algorithm that incorporates genetic concepts, as described in this invention.

[0076] Figure 3 This is a parameter identification curve of the bow motion response model in an embodiment of the present invention;

[0077] Figure 4 This is a graph showing the identification of velocity response model parameters in an embodiment of the present invention.

[0078] Figure 5 This is a comparison chart of fitness values ​​during the parameter identification process of the bow motion response model using the method embodiment of the present invention and the standard particle swarm optimization method.

[0079] Figure 6 This is a comparison chart of fitness values ​​during the identification of velocity response model parameters using the method of the present invention and the standard particle swarm optimization method. Detailed Implementation

[0080] To better understand the present invention, the technical solutions of the present invention will be described in detail below with reference to the accompanying drawings of the embodiments of the present invention:

[0081] This invention discloses a method for identifying parameters of unmanned surface vessel response models based on an improved particle swarm optimization algorithm. Figure 1 The following is a flowchart illustrating the steps of the unmanned surface vessel response model parameter identification method of the present invention.

[0082] Step S1: Conduct a specific unmanned vessel identification navigation test under ideal environmental conditions and collect the input and output test data required for model identification.

[0083] In one specific embodiment, step S1 specifically includes:

[0084] Step S1.1: First, a maneuverability test is conducted, and the collected data is used for the identification of the bow motion response model. The test procedure is as follows: Before the test begins, the unmanned surface vessel maintains a constant speed and straight-line navigation state. Then, a sinusoidal steering command is sent, and at the same time, the bow angular velocity r, rudder angle δ, speed V, and rotational speed n of the unmanned surface vessel are recorded at a fixed sampling period h. The speed is maintained, and the test ends after several sinusoidal cycles of steering commands.

[0085] In this example, during the maneuvering test, the unmanned surface vessel travels at a speed of 15 knots, and the steering commands are generated according to the following formula:

[0086]

[0087] The sampling period h = 0.02 seconds, and the sampling time is 1000 seconds.

[0088] Step S1.2: Using the finite difference method, calculate the bow angular acceleration at each sampling moment based on the bow angular velocity r and rudder angle δ collected from the maneuverability test. Rate of change of bow angular acceleration Rudder angle change rate

[0089]

[0090]

[0091] Step S1.3: Conduct a speed navigation test, and collect the data for speed response model identification. The test procedure is as follows: Before the test begins, the unmanned surface vessel maintains a constant speed and straight-line navigation state. Then, sinusoidal steering commands and sinusoidally varying speed control commands are sent. Simultaneously, the bow angular velocity r, rudder angle δ, speed V, and acceleration of the unmanned surface vessel are recorded at a fixed sampling period h. The test ended after a number of sinusoidal cycles of steering commands and speed control commands, with the rotational speed n being constant.

[0092] In this example, the steering commands during the speed navigation test are generated according to the following formula:

[0093]

[0094] Speed ​​commands are generated according to the following formula:

[0095]

[0096] The sampling interval h = 0.02 seconds, and the sampling time is 1000 seconds.

[0097] Step S2: Establish the unmanned vessel response model, including the unmanned vessel bow turning motion response model and velocity response model.

[0098] In one specific embodiment, the unmanned vessel's bow-turning motion response model in step S2 is as follows:

[0099]

[0100] The velocity response model is:

[0101]

[0102] Where V is the velocity of the unmanned vessel in the horizontal plane, r is the bow angular velocity, and δ is the rudder angle. Let n be the rate of change of rudder angle, n be the rotational speed, K be the gain coefficient, T1, T2, and T3 be the time constants, and α be the coefficient of the nonlinear term. vv a is the drag coefficient. rr a is the centrifugal force coefficient. δδ a is the rudder force coefficient. nn a nv This is the propeller force coefficient.

[0103] Step S3: Further refine the unmanned vessel's bow-turning motion response model and velocity response model, and determine the identification error index. The error index can be expressed as a minimum objective function related to the model parameters to be identified, i.e., the fitness function. In the method of this invention, the particle encoding is related to the model parameters to be identified, and the fitness of an individual particle is directly taken as the corresponding objective function value. Calculating the current fitness value of the particle can measure the quality of the particle's position.

[0104] In one specific embodiment, in step S3, the bow motion response model and velocity response model are obtained as follows:

[0105]

[0106]

[0107] The unmanned vessel's bow-turning motion response model can be expressed as:

[0108] Y r =A r H r

[0109] in,

[0110]

[0111]

[0112] The parameters to be identified in the model are K, T1, T2, T3, and α. The parameters of the steering response model can also be represented in dimensionless form as K′, T′1, T′2, T′3, and α′ during identification. The identification error index for the steering motion response model is:

[0113]

[0114] Among them, J r Fitness values ​​identified for the bow motion model; N r Y represents the number of sampling points for the bow maneuverability test data. ri For the i-th test sample in the steering maneuverability test, (This is the actual output value.) This is the estimated output value for the i-th test sample in the turning motion experiment.

[0115] The velocity response model can be represented as follows:

[0116] Y v =A v H v

[0117] in,

[0118] A v =[a vv a rr a δδ a nn a nv ],

[0119] H v =[-V 2 -r 2 -V 2 δ 2 n 2 nV] T ,

[0120] The parameter that needs to be identified in the model is a. vv a rr a δδ a nn a nv The velocity response model identification error index is set as follows:

[0121]

[0122] Among them, J v The fitness value N identified for the velocity model v Y represents the number of sampling points for speed navigation test data. vi This represents the actual output value of the i-th test sample in the speed navigation test. This is the estimated output value for the i-th test sample in the speed navigation test.

[0123] Step S4: An improved particle swarm optimization algorithm incorporating genetic principles is used to identify the parameters of the unmanned vessel's bow-turning motion response model and velocity response model, respectively, to obtain the parameter identification results of the unmanned vessel response model.

[0124] In one specific embodiment, step S4, as follows: Figure 2 The specific steps shown include:

[0125] Step S4.1: During the identification of the steering motion response model, for the problem... To find the minimum value, the particle can be encoded as X = (K, T1 + T2, T1T2, T3, α). When identifying the velocity response model, for the problem... To find the minimum value, the particle can be encoded as X = (a vv ,a rr ,a δδ ,a nn ,a nv The fitness function is expressed as f(X). In the problem of identifying the steering motion response model, the fitness function is taken as f(X) = J. r In the velocity response model identification problem, the fitness function is f(X) = J. v (X).

[0126] Step S4.2: Initialize the parameter range, including the local learning factor c1, the global learning factor c2, the maximum number of generations G, and the current number of generations t=1. Randomly generate the position and velocity matrices of m particles in the initial population within the parameter range. The position of the i-th particle in the entire d-dimensional search space is represented by X. i =(x i1 ,x i2 ,...,x id The speed is expressed as V. i =(v i1 ,v i2 ,...,v id In this example, c1 = 1.3, c2 = 1.3, G = 2000, and m = 2000.

[0127] Step S4.3: Calculate the initial fitness value of each particle in the population, and the optimal position pbest for each individual particle. i Initialize to the current position and find the current global best position gbest for the entire group.

[0128] Step S4.4: Update the particle velocity and position, generate a new population, and perform boundary checks on the particle velocity and position. Particles update their velocity and position using the following formula:

[0129] v iq (t+1)=ω(t)v iq (t)+c1r1[pbest iq (t)-x iq (t)]+c2r2[gbest q (t)-x iq (t)]

[0130] x iq (t+1)=x iq (t)+v iq (t+1)

[0131] In the formula: v iq (t) represents the q-th dimension velocity component of the i-th particle when it evolves to generation t, x iq (t) represents the q-th position component of the i-th particle when it evolves to generation t, pbest iq (t) represents the optimal position of the i-th particle in the q-th dimension when it evolves to generation t. q ω(t) represents the q-th dimension component of the optimal position of the entire population in generation t, where i = 1, 2, ..., m, q = 1, 2, ..., d, t = 1, 2, ..., G, r1 and r2 are random numbers distributed between [0, 1], and ω(t) represents the inertia weight. The weight is updated using a linear decreasing weight method, and the formula is:

[0132]

[0133] Where, ω max ω min Let be the maximum and minimum values ​​of ω(t), respectively. In this example, ω max =0.9, ω min =0.1.

[0134] Step S4.5: Update the individual optimal position and the population optimal position. Set the current position fitness value f(X) of each particle... i Comparing f(X) with its own fitness extreme value, if f(X) i If the current position is better than the individual fitness extreme value, then the current position is taken as the individual's optimal position; otherwise, the individual's historical optimal position is used. The individual optimal fitness value of each particle (i.e., the optimal position pbest generated by each particle from the initial to the current iteration number) is used. iThe corresponding fitness value, i.e., the extreme fitness value of each particle after the current iteration number, as mentioned in the previous paragraph, is compared with the population optimal fitness value (the fitness value corresponding to the global optimal position gbest of the entire population after the last iteration search (the population optimal fitness value obtained by substituting gbest into the fitness function). If the individual's optimal fitness value is better than the population optimal fitness value, then the individual's optimal position replaces the population optimal position; otherwise, the population optimal position remains unchanged.

[0135] Step S4.6: In this example, the crossover probability p c =0.2, based on the crossover probability p c Select a portion of the particles and place them into a hybridization pool. Randomly select two parent particles X from the pool. i X j Hybridization produces the same number of offspring particles, with the offspring particles positioned at X′. i 、X′ j and velocity V′ i V′ j The calculation formula is as follows:

[0136] X′ i =p·X i +(1-p)·X j

[0137] X′ j =p·X j +(1-p)·X i

[0138]

[0139]

[0140] Where p is a random number distributed in [0,1]. If the fitness value of the offspring is better than that of the parent, the offspring particles after hybridization are used to replace the parent particles, and the optimal position of the individual and the optimal position of the population are updated according to the method in step S4.5. Otherwise, the parent particles are used.

[0141] Step S4.7: For each particle in the population, with mutation probability p n After perturbation and mutation, the mutated particle is:

[0142] x″ iq =gbest q ·(0.5+σ)

[0143] x″ iq For the q-th dimension position component of the i-th particle after mutation, gbest qLet q be the coordinates of the optimal position in the population. σ is a random Gaussian white noise with a specified intensity of 0 dBW. For particles that exceed the boundary, mutation is performed as follows:

[0144]

[0145] These are the upper and lower bounds of the q-th dimension position component of the particle, respectively. In this example, we take... λ is a random number distributed between [0,1].

[0146] Mutation probability p n Adaptive adjustment:

[0147]

[0148] in, These are the maximum and minimum values ​​of the mutation probability; in this example, we take... N is the number of generations in which the fitness value stagnates, i.e., the cumulative number of periods in which the extreme change in the population fitness value is less than a sufficiently small constant γ, and κ p This is a constant coefficient. The fitness value after mutation is compared with the optimal fitness value of the population; if the fitness value after mutation is better, then X″ is used. i Replace X i Update the individual's best position pbest i The optimal position of the population is gbest; otherwise, the original particle remains unchanged.

[0149] Step S4.8: Update the upper and lower bounds of the motion velocity parameter, and update the value of the cumulative stagnation algebra N.

[0150]

[0151]

[0152] These are the maximum and minimum values ​​of the upper limit of the q-th dimension velocity component of the particle, respectively, both of which are positive numbers. In this example, we take... μ is the preset precision for the fitness evaluation value, κ is the preset precision for the fitness evaluation value. v γ is a constant coefficient, and γ is a sufficiently small positive number.

[0153] Step S4.9: Check the termination condition. If it is met, stop the search and output the model parameter identification results; otherwise, t = t + 1, and return to step S4.4. The termination condition is reaching the maximum number of generations or the fitness value being less than the preset computational precision μ.

[0154] Example

[0155] In this embodiment, the "Yulong" ocean-going training vessel of Dalian Maritime University was used as a prototype. A simulation experiment of unmanned vessel model identification navigation was conducted, and the collected data was used for model identification. The true values ​​of the unmanned vessel response model parameters and the parameter identification results are shown in Table 1. The parameters of the bow-turning motion model are expressed in dimensionless form. It can be seen that the identification results are close to the true values, indicating high identification accuracy.

[0156] Table 1

[0157] <![CDATA[T1′+T′2]]> 23.944 24.0332 <![CDATA[T1′T′2]]> 9.238 9.2592 <![CDATA[T3′]]> 10.064 10.0662 K′ 7.926 7.9560 α′ 30 30.1286 <![CDATA[a vv ]]> <![CDATA[1.3586×10 -4 ]]> <![CDATA[1.3586×10 -4 ]]> <![CDATA[a rr ]]> 101.520 101.5200 <![CDATA[a δδ ]]> <![CDATA[1.6658×10 -3 ]]> <![CDATA[1.6659×10 -3 ]]> <![CDATA[a nn ]]> <![CDATA[1.4042×10 -2 ]]> <![CDATA[1.4042×10 -2 ]]> <![CDATA[a nv ]]> <![CDATA[5.9167×10 -4 ]]> <![CDATA[5.9175×10 -4 ]]>

[0158] Figure 3 , Figure 4 To identify the parameter identification curves of the unmanned vessel's turning motion response model and velocity response model parameters using the method embodiments of the present invention, Figure 5 , Figure 6 This paper compares the fitness values ​​of the proposed method with those of the standard particle swarm optimization method during the parameter identification process of the bow motion response model and velocity response model. It is evident that the proposed method improves the identification convergence speed compared to the standard particle swarm optimization method.

[0159] The parts of this invention not described in detail are common knowledge to those skilled in the art.

Claims

1. A method for identifying parameters of an unmanned surface vessel response model based on an improved particle swarm optimization algorithm, characterized in that... include: Conduct specific unmanned vessel identification and navigation tests under ideal environmental conditions, and collect the input and output test data required for model identification; Establish unmanned surface vessel (USV) response models, including USV bow turning motion response models and velocity response models; The unmanned vessel's bow turning motion response model and velocity response model were organized, and the identification error index was determined. Using the experimental data, an improved particle swarm optimization algorithm incorporating genetic principles was employed to identify the parameters of the unmanned vessel's bow-turning motion response model and velocity response model, respectively, and the parameter identification results of the unmanned vessel response model were obtained. The improved particle swarm optimization algorithm incorporating genetic concepts includes: S1. The problem during model identification is transformed into minimizing the identification error index of the steering motion response model and the velocity response model; and the particles are encoded as... , The fitness function is expressed as ; S2, Initialize the motion range of the parameters; S3. Calculate the initial fitness value of each particle in the population and the optimal position of each individual particle. Initialize to the current position and find the current global optimal position for the entire group. Simultaneously, the initial fitness value of each particle in the swarm and the optimal fitness value of the swarm are calculated. S4. Update the velocity and position of the particles, generate a new population, and perform out-of-bounds checks on the velocity and position of the particles. S5. Based on the fitness value of each particle's current position. Update the individual optimal position and the population optimal position; S6. Based on the hybridization probability Select a portion of the particles and place them into a hybridization pool. Randomly select two parent particles from the pool and hybridize them to produce the same number of offspring particles. Calculate the position and velocity of the offspring particles. If the fitness value of the offspring is better than that of the parent, the hybridized offspring particles replace the parent particles to update the individual optimal position and the population optimal position. Otherwise, the parent particles are used. S7. For each particle in the population, use the mutation probability... Perform perturbation and mutation, compare the mutated fitness value with the population's optimal fitness value. If the mutated fitness value is better, replace the original particles in the population with the mutated particles to update the individual's optimal position and the population's optimal position; otherwise, the original particles remain unchanged. S8. Update the upper and lower bounds of the motion velocity parameter, and update the value of the cumulative stall algebra N; S9: Check the termination condition. If satisfied, stop the search, output the model parameter identification results, proceed to the next generation particle processing, and return to step S4. The termination condition is reaching the maximum number of generations or the fitness value being less than the preset computational precision. ; The perturbated and mutated particles are: For the q-th dimension position component of the i-th particle after mutation, The coordinates of the q-th dimension represent the optimal position in the population. The random number is Gaussian white noise; For particles that exceed the boundary, mutation is performed in the following manner: , These are the upper and lower bounds of the q-th dimension position component of the particle, respectively. These are random numbers distributed between [0,1]. Mutation probability Adaptively adjust according to the following formula: ， in, , These are the maximum and minimum values ​​of the mutation probability. Here, N is a constant coefficient, and N is the number of generations in which the fitness value stagnates, meaning the extreme change in the population fitness value is less than a sufficiently small normal number. The cumulative number of cycles, To select the appropriate fitness function calculation precision Positive numbers of the same order of magnitude, or custom adjustments made during the identification process.

2. The method for parameter identification of unmanned surface vessel response model based on an improved particle swarm optimization algorithm according to claim 1, characterized in that, The input and output test data required for the model identification include: Maneuverability tests were conducted, and the collected data was used for the identification of the bow motion response model. The differential method was used, based on the bow angular velocity r and rudder angle collected during the maneuverability test. The heading angular acceleration at each sampling time was calculated. Rate of change of bow angular acceleration rudder angle change rate ; A speed navigation test was conducted, and the collected data was used for speed response model identification. The speed navigation test procedure was as follows: before the test began, the unmanned surface vessel maintained a constant speed and straight-line navigation, then a sinusoidal steering command and a sinusoidally varying speed control command were sent, while the bow angular velocity r and rudder angle of the unmanned surface vessel were recorded at a fixed sampling period. Speed ​​V, acceleration The test ended after several sinusoidal cycles of steering commands and speed control commands, with the rotational speed n being constant.

3. The method for identifying parameters of an unmanned surface vessel response model based on an improved particle swarm optimization algorithm according to claim 2, characterized in that, The maneuverability test procedure is as follows: Before the test begins, the unmanned surface vessel maintains a constant speed and straight-line navigation state. Then, a sinusoidal steering command is sent, and the bow angular velocity r and rudder angle of the unmanned surface vessel are recorded at a fixed sampling period. Set the speed V and rotational speed n, maintain this speed, and end the test after several sinusoidal cycles of steering commands.

4. The method for identifying parameters of an unmanned surface vessel response model based on an improved particle swarm optimization algorithm according to claim 1, characterized in that, The unmanned surface vessel's bow turning motion response model was compiled, and the identification error indices were determined, including: The unmanned vessel's bow-turning motion response model is represented as: in , ， ， The parameters that need to be identified in the model are the gain coefficient K and the time constant. , , Nonlinear term coefficients ; r is the heading angular velocity, For bow angular acceleration, The rate of change of bow angular acceleration. As the rudder angle, The rate of change of rudder angle; Error index for identifying the bow motion response model: in, An error index is identified for the turning motion model. The error index is expressed as a minimum objective function related to the parameters of the model to be identified, i.e., the fitness function. The number of sampling points for the test data of the steering maneuverability test. For the i-th test sample in the steering maneuverability test, (This is the actual output value.) This is the estimated output value for the i-th test sample in the turning motion experiment.

5. The method for identifying parameters of an unmanned surface vessel response model based on an improved particle swarm optimization algorithm according to claim 4, characterized in that, The unmanned surface vessel's bow turning motion response model was compiled, and the identification error indices were determined, including: The velocity response model is expressed as: in, , ， ， The parameter that needs to be identified in the model is the drag coefficient. Centrifugal force coefficient rudder force coefficient propeller force coefficient and V represents the speed of the unmanned vessel moving in the horizontal plane. Let r be the acceleration and r be the heading angular velocity. Where n is the rudder angle and n is the rotational speed; Velocity response model identification error index: in, The fitness value of the velocity model identification is used as the error index for the velocity response model identification. The number of sampling points for speed sailing test data. This represents the actual output value of the i-th test sample in the speed navigation test. This is the estimated output value for the i-th test sample in the speed navigation test.

6. The method for identifying parameters of an unmanned surface vessel response model based on an improved particle swarm optimization algorithm according to claim 1, characterized in that, Initialize the motion range of the parameters, including the local learning factor. Global learning factor Given the maximum number of generations G, the current generation t=1, and randomly generated position and velocity matrices for m particles in the initial population within the parameter range, the position of the i-th particle in the entire d-dimensional search space is represented as... Speed ​​is expressed as .

7. The method for identifying parameters of an unmanned surface vessel response model based on an improved particle swarm optimization algorithm according to claim 1, characterized in that, The updated individual optimal position and population optimal position include: Set the current position fitness value of each particle Compared with its own fitness extreme value, if If the current position is better than the individual's fitness extreme value, then the current position is taken as the individual's optimal position; otherwise, the individual's historical optimal position is used. The updated individual optimal fitness value of each particle is compared with the population optimal fitness value. If the individual optimal fitness value is better than the population optimal fitness value, the population optimal position is replaced by the individual optimal position; otherwise, the population optimal position remains unchanged.

8. The method for parameter identification of unmanned surface vessel response model based on an improved particle swarm optimization algorithm according to claim 1, characterized in that, Position of offspring particles , and speed , The calculation formula is as follows: in, Let the random numbers be distributed in the interval [0,1]. , For two parent particles , The speed.

9. The method for identifying parameters of an unmanned surface vessel response model based on an improved particle swarm optimization algorithm according to claim 1, characterized in that, The upper and lower bounds of the update parameter motion speed are updated, and the value of the cumulative stagnation algebra N is updated. , These are the maximum and minimum values ​​of the upper limit of the q-th dimension velocity component of the particle, respectively, both of which are positive numbers. Preset the accuracy for fitness evaluation values. These are constant coefficients. For sufficiently small positive numbers, To select the appropriate fitness function calculation precision Positive numbers of the same order of magnitude, or those that can be customized during the identification process.

Images