Data assimilation fused cascade type aircraft model parameter identification method and system, and medium

Abstract

The invention provides a cascaded aircraft model parameter identification method and system fused with data assimilation. The method comprises the steps that S1, a six-degree-of-freedom hydrodynamic model including surging, swaying, heaving, rolling, pitching and yawing is constructed for a marine aircraft; s2, motion response data, including the attitude, the speed and the acceleration, of the marine vehicle are obtained through real boat experiments and / or numerical simulation; s3, performing parameter identification on each degree of freedom of the six-degree-of-freedom hydrodynamic model to obtain a hydrodynamic parameter preliminary estimation value of each degree of freedom; and S4, carrying out multi-degree-of-freedom joint parameter identification on the six-degree-of-freedom hydrodynamic model of the marine vehicle based on the preliminary estimated values of the hydrodynamic parameters of all degrees of freedom to obtain a final parameter identification result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ocean vehicle maneuvering and control, in particular, to a cascaded vehicle model parameter identification method and system fusing data assimilation, more particularly, to an ocean vehicle six-degree-of-freedom hydrodynamic model parameter identification method and system based on a data assimilation mechanism and a cascaded strategy. BACKGROUND

[0002] Ocean science is an observation-based discipline, and ocean vehicles play a key role in it, such as water sampling, front tracking, etc. In order to more accurately and stably perform these cutting-edge scientific tasks, it is required that the ocean vehicle has good maneuverability, which requires an accurate ocean vehicle hydrodynamic model, especially the numerous parameters therein. Therefore, the hydrodynamic model parameter identification of the ocean vehicle is a difficult problem in the field of ocean engineering in recent years. In practical applications, traditional parameter identification algorithms, such as the least squares method, are easily affected by noise, leading to overfitting; black-box identification algorithms such as machine learning have high accuracy, but their model interpretability is insufficient. In addition, these parameter identification methods all take the minimization of single-step prediction error as the training target, and the identification results may lead to error accumulation when applied to long-term simulation or multi-step prediction, causing the multi-step prediction results to diverge. Therefore, considering the shortcomings of existing algorithms, it is necessary to develop an ocean vehicle hydrodynamic parameter identification algorithm and system that has clear physical meaning and can better adapt to multi-step prediction.

[0003] Patent document CN112487555A (application number: 202011366190.0) discloses a dimensionless resistance coefficient identification method for a water-air amphibious submersible. First, a dynamic model of the water-air amphibious submersible in the water under the condition of no interference is established, then the dynamic model of the water-air amphibious submersible in the sinking and floating state is discretized to obtain an identification model of the dimensionless resistance coefficient in the dynamic model of the water-air amphibious submersible in the sinking and floating state, and finally, the observation data for a period of time is used to solve the identification model by using the least squares method to obtain the estimated value of the dimensionless resistance coefficient in the dynamic model of the water-air amphibious submersible in the sinking and floating state.

[0004] Patent document CN113704678A (application number: 202110886408.3) discloses a forgetting factor least squares model parameter identification method based on full rank decomposition, which makes the underexcitation data satisfy the excitation condition again after dimensionality reduction to ensure the convergence of the identification results of tracking time-varying parameters and disturbances. By setting a dead zone to determine the dimensionality reduction method, the influence of disturbances on the identification process is weakened. The ship data under long-time underexcitation and ocean environment disturbance conditions is used for ship heading model parameter identification. By comparing the identification result curves and heading angle velocity prediction error curves of the FFLS and FRDLS algorithm identification ship maneuvering model parameters, it can be found that the full rank decomposition least squares method can effectively suppress the drift and divergence problems in the parameter identification process under data underexcitation and ocean environment disturbance, and improve the accuracy of parameter identification, while having the advantages of small calculation amount, strong real-time performance, and simple implementation.

[0005] Patent document CN117634136A (application number: 202311350079.6) discloses an underwater robot vertical plane motion model parameter identification method, comprising: S1, establishing a vertical plane motion model of an underwater robot according to the motion characteristics of the underwater robot; S2, discretizing the vertical plane motion model to obtain a discretized vertical plane motion model; S3, performing difference processing on the discretized vertical plane motion model, and obtaining a parameter identification model based on the difference-processed vertical plane motion model; S4, performing parameter identification based on a multi-innovation least squares method and the parameter identification model to identify model parameters; and S5, verifying the identified model parameters using the discretized vertical plane motion model, and outputting the identification result if the error is less than or equal to a preset error, or returning to S2 if the error is greater than the preset error.

[0006] In the prior art method, a traditional parameter identification algorithm such as least squares is usually used for model identification of marine vehicle motion, which is easily affected by noise and has low precision; and only the motion of part of the degrees of freedom of the marine vehicle is considered, and the modeling is not comprehensive enough. SUMMARY

[0007] In view of the defects in the prior art, the purpose of the present application is to provide a cascade type vehicle model parameter identification method and system fusing data assimilation.

[0008] According to the cascade type vehicle model parameter identification method fusing data assimilation provided by the present application, the following steps are included: Step S1: A six-degree-of-freedom hydrodynamic model including surge, sway, heave, roll, pitch and yaw is constructed for a marine vehicle; Step S2: Obtain marine vehicle attitude, speed and acceleration motion response data through real ship experiment and / or numerical simulation; Step S3: Identify the parameters for each degree of freedom of the six-degree-of-freedom hydrodynamic model to obtain preliminary estimates of the hydrodynamic parameters for each degree of freedom; Step S4: Based on the preliminary estimates of the hydrodynamic parameters of each degree of freedom, perform multi-degree-of-freedom joint parameter identification on the six-degree-of-freedom hydrodynamic model of the marine vehicle to obtain the final parameter identification results.

[0009] Preferably, the six-degree-of-freedom hydrodynamic model includes:

[0010] Two right-handed coordinate systems are established: the geodetic coordinate system {I} and the body coordinate system {B} of the ocean vehicle. The body coordinate system {B} x The axis points towards the bow of the ship. y The axis points to the starboard side. z The axis points vertically downwards; The pose of a marine vehicle in the geodetic coordinate system {I} is defined as follows: ,in, The three-dimensional spatial position of the marine vehicle in the geodetic coordinate system {I}; This refers to the roll angle; The pitch angle; To raise the bow angle; The velocity of an ocean vehicle in the body coordinate system {B} is defined as: ;in, For the body coordinate system {B} The linear velocity of the shaft; For orbital coordinate system Angular velocity of the axis; In the body coordinate system {B}, let the position of the center of gravity of the ocean vehicle be denoted as . The position of the center of buoyancy is Let the mass of the ocean-going vehicle be denoted as . Then its gravity is ,in, It is the acceleration due to gravity; Let the moments of inertia of the ocean-going vehicle about the three coordinate axes be: The product of inertia is: ; The vertical rudder angle of an ocean-going vehicle. The horizontal rudder angle of a marine vehicle; The coefficients on the right side of the equation are the hydrodynamic parameters that need to be identified, and are collectively denoted as a parameter vector. ; parameter vector Based on the structure of the six-degree-of-freedom model equations, it is divided into six subsets of degrees of freedom: ; All parameters is represented as ; The continuous six-degree-of-freedom hydrodynamic model is converted into a discrete six-degree-of-freedom hydrodynamic model, and the derivative term is discretized using the forward Euler method:

[0011] wherein, is a sampling time interval, , is the k+1 time; is the k time; Then the discretized derivative term is substituted into the hydrodynamic model to obtain a discrete hydrodynamic model.

[0012] Preferably, the step S2 comprises: Collecting the motion response data of the marine vehicle, assuming that the data is equally sampled, denoted as a sampling interval h , and the sampling time point is: , then: , N is the data length; the collected data is the speed and angular velocity of the marine vehicle;

[0013] The collected motion response data of the marine vehicle is denoised to obtain processed motion response data of the marine vehicle. Acceleration data is calculated based on the processed motion response data of the marine vehicle.

[0014] Preferably, the step S3 comprises: Step S3.1: generating a prediction value at the current time according to the discrete hydrodynamic model and using the prediction value at the previous time; wherein the single-degree-of-freedom identification generates a prediction value of the current degree of freedom using the prediction value of the current degree of freedom at the previous time and the measured values of other degrees of freedom; Step S3.2: constructing a loss function for the error between the prediction value and the measured value, and minimizing the loss function through an optimization algorithm to obtain preliminary estimated values of the hydrodynamic parameters of each degree of freedom; Step S3.3: linearly weighting the prior prediction value and the measured value to obtain a posterior prediction value; the posterior prediction value is used for the next iteration to obtain the prediction value at the next time.

[0015] Preferably, step S3.2 comprises: constructing a loss function for the surge degree of freedom based on the measured value and the prediction value . .

[0016] Preferably, step S3.2 includes: The NMG optimization algorithm is obtained based on a hybrid optimization algorithm of Nelder-Mead simplex method and gradient descent method, and the loss function is minimized using the NMG optimization algorithm. The NMG optimization algorithm includes: Based on the initial point Obtaining the dimensions of the optimization problem D Then generate randomly D Initial point ;These( D +1) points form a simplex; in each iteration, perform the following steps: Will( D +1) points according to their function values Sort by size from smallest to largest, satisfying:

[0017] Then, the classic gradient descent method is used to attempt to optimize the optimal point. :

[0018] in, For learning rate, For the function in The derivative at point; Calculate the function value at the gradient descent point and compare it with the worst point. If the gradient descent point is better than the worst point, replace the worst point and re-sort the function values ​​in descending order. Calculate and exclude the worst case The center of mass of the simplex:

[0019] Reflection: the worst point Regarding the center of mass m reflection:

[0020] in, It is the reflection point. It is the reflection coefficient; Calculate the function value of the reflection point and compare it with the function value of the optimal point. Compare; if the reflection point is better than the optimal point, then perform the expansion operation:

[0021] in, It is an extension point. It is the expansion coefficient; if the expansion point is better than the reflection point: If the worst-case scenario is not found, the extended point is used to replace the worst-case point, and the process proceeds to the next iteration; otherwise, the reflected point is used to replace the worst-case point, and the process proceeds to the next iteration. If the reflection point is worse than the optimal point, but better than the second worst point: If the worst-case scenario is not found, the reflection point is used to replace the worst-case point, and the process proceeds to the next iteration. If the reflection point is worse than the second worst point, a contraction operation is performed, including: If the reflection point is worse than the second worst point, but better than the worst point: Then, the external contraction operation is performed:

[0022] If the reflection point is worse than the worst point, then perform an inward contraction operation:

[0023] in, They are all contraction points. This is the contraction coefficient; if the contracted point is better than the reflection point and the worst point, then the contracted point replaces the worst point, and the process proceeds to the next iteration; otherwise, a backtracking operation is performed, shrinking all vertices of the simplex towards the optimal point:

[0024] in, It is the backoff point corresponding to each vertex of the simplex. It is the backoff coefficient; the backoff point and the optimal point are used to form a new simplex, and then the process proceeds to the next iteration.

[0025] When the convergence condition is met, the iteration terminates, and the optimization result is used as the preliminary identification result.

[0026] Preferably, step S3.3 includes: introducing measured values ​​through a data assimilation mechanism to enable parameter identification to perceive the impact of environmental disturbances on prediction errors. u In terms of direction, the specific formula is as follows:

[0027] in, The posterior predicted value from the previous time step. The velocity vector of the motion at the previous moment. For the aforementioned u The parameter vector of degrees of freedom, The prior prediction value at the current moment. It is the weighting factor of the measured value in the data assimilation mechanism. The measured value at the current moment. This is the posterior prediction value at the current time.

[0028] Preferably, step S4 includes: Based on the preliminary estimates of hydrodynamic parameters for each degree of freedom, a method of minimizing prediction error and multi-step iterative prediction is used to find the parameter combination that minimizes the error between the predicted and measured values, which is then used as the final identification result. For integrating the errors between predicted and measured values ​​across multiple degrees of freedom, various loss functions are employed, which can be freely selected according to actual needs, including:

[0029] Then, the NMG optimization algorithm is used to optimize the loss function of multiple degrees of freedom, and the optimization result is used as the final hydrodynamic parameter identification result of the marine vehicle.

[0030] A cascaded vehicle model parameter identification system based on fused data assimilation, according to the present invention, includes: Module M1: Constructs a six-degree-of-freedom hydrodynamic model for marine vehicles, including sway, roll, heave, pitch, pitch and yaw. Module M2: Acquires motion response data of marine vehicles, including attitude, velocity, and acceleration, through methods including live-fire experiments and / or numerical simulations; Module M3: Identifies parameters for each degree of freedom in the six-degree-of-freedom hydrodynamic model and obtains preliminary estimates of the hydrodynamic parameters for each degree of freedom; Module M4: Based on the preliminary estimates of hydrodynamic parameters for each degree of freedom, multi-degree-of-freedom joint parameter identification is performed on the six-degree-of-freedom hydrodynamic model of the marine vehicle to obtain the final parameter identification results.

[0031] According to a computer-readable storage medium provided by the present invention, the computer-readable storage medium stores instructions that, when executed by a processor, implement the various steps of the cascaded aircraft model parameter identification method described above for data assimilation.

[0032] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention adopts a parameter identification method based on multi-step iterative prediction with data assimilation to minimize prediction error, making the identified parameters more suitable for long-term prediction, improving model robustness, effectively reducing the cumulative error of multi-step prediction, and significantly improving the effect of parameter identification results on multi-step motion prediction tasks of marine vehicles. 2. This invention adopts a cascaded parameter identification strategy, which transforms the complex and coupled multi-degree-of-freedom parameter identification problem into first identifying the parameters of each single degree of freedom separately, and then performing refined parameter identification based on the preliminary identification results. This strategy effectively reduces the difficulty of multi-degree-of-freedom parameter identification problems and improves the accuracy of parameter identification results. 3. This invention is based on a hybrid algorithm of Nelder-Mead simplex method and gradient descent method, namely NMG optimization algorithm. By using gradient descent method to assist in improving the simplex, it can combine the global optimization capability of simplex method and the local optimization capability of gradient descent method, so that the optimization algorithm can find the minimum point of loss function more quickly and accurately, thereby improving the speed and accuracy of hydrodynamic parameter identification of marine vehicles. 4. This invention designs adjustable data assimilation weights and optimization algorithm hyperparameters, which can be flexibly adjusted according to user needs, providing users with customized parameter identification functions. It realizes efficient and accurate identification of hydrodynamic parameters and motion prediction of marine vehicles, providing strong support for the field of marine science research. Attached Figure Description

[0033] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Fig. 1 This is a flowchart of a parameter identification method for a six-DOF hydrodynamic model of a marine vehicle based on a data assimilation mechanism and a cascading strategy.

[0034] Fig. 2 This is a schematic diagram of the data assimilation mechanism in a multi-step iterative prediction model.

[0035] Fig. 3 The flowchart shows the NMG optimization algorithm, a hybrid optimization algorithm based on the Nelder-Mead simplex method and gradient descent method. Detailed Implementation

[0036] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0037] Example 1 The method for identifying the hydrodynamic model of a visual marine submersible provided by the present invention, such as Figs. 1 to 3 As shown, it includes: Step 1: Establish a six-degree-of-freedom hydrodynamic model, including: establishing nonlinear motion equations for the marine vehicle that include six degrees of freedom: sway, roll, heave, pitch, pitch and yaw. The hydrodynamic coefficients in the equations are used as parameters to be identified. Step 2: Collect motion state data, including: obtaining motion response data such as attitude, speed, and acceleration of the marine vehicle through actual boat tests or numerical simulations.

[0038] Step 3: Single-degree-of-freedom parameter identification, including: parameter identification for each degree of freedom of the six-degree-of-freedom hydrodynamic model of the ocean vehicle, and obtaining preliminary estimates of the hydrodynamic parameters for each degree of freedom.

[0039] Step 4: Multi-degree-of-freedom joint parameter identification. Using the single-degree-of-freedom parameter identification results as initial values, multi-degree-of-freedom joint parameter identification is performed on the six-degree-of-freedom hydrodynamic model of the marine vehicle to obtain the final parameter identification results.

[0040] Specifically, step 1 includes: Establish two right-handed coordinate systems: the geodetic coordinate system {I} and the body coordinate system {B} of the ocean vehicle.

[0041] The body coordinate system {B} x The axis points towards the bow of the ship. y The axis points to the starboard side. z The axis points vertically downwards.

[0042] The pose of a marine vehicle in the geodetic coordinate system {I} is defined as follows: ,in, The three-dimensional spatial position of the marine vehicle in the geodetic coordinate system {I}; This refers to the roll angle; The pitch angle; To raise the bow angle; The velocity of an ocean vehicle in the body coordinate system {B} is defined as: ;in, For the body coordinate system {B} The linear velocity of the shaft; For orbital coordinate system Angular velocity of the axis; In the body coordinate system {B}, let the position of the center of gravity of the ocean vehicle be denoted as . The position of the center of buoyancy is The origin of the body coordinate system {B} is generally defined at the center of buoyancy of the ocean vehicle. Let the mass of the ocean-going vehicle be... Then its gravity is ,in Let g be the acceleration due to gravity. Let the magnitude of the buoyancy force acting on the ocean-going vessel in the water be denoted as . Assuming the ocean-going vehicle is neutrally buoyant, then Let the moments of inertia of the ocean-going vehicle about the three coordinate axes be: The product of inertia is: .

[0043] The six-degree-of-freedom hydrodynamic model of an ocean-going vehicle can be described as follows:

[0044] in, Let represent the acceleration and angular acceleration of the marine vehicle, respectively; the term on the right-hand side of the equation represents the total hydrodynamic force acting on the marine vehicle. For ease of modeling, it is assumed that the marine vehicle is symmetric about the three coordinate axes, therefore its product of inertia is zero. The model can be simplified to:

[0045] Furthermore, assuming the ocean-going vehicle is a rotating body, then Assuming the center of gravity is directly below the center of buoyancy, then The model can be simplified to:

[0046] Expanding the hydrodynamic force on the right side of the equation, the model is as follows:

[0047] in, The vertical rudder angle of an ocean-going vehicle. The horizontal rudder angle of a marine vehicle; The coefficients on the right side of the equation are the hydrodynamic parameters that need to be identified, and are collectively denoted as parameter vectors. This parameter vector can be divided into six subsets of degrees of freedom based on the structure of the six-degree-of-freedom model equations: Among them, with u A subset of parameters for directional degrees of freedom For example, it contains a total of 8 parameters: It can be represented in vector form:

[0048] The parameter set structure for the remaining degrees of freedom is similar, with all parameters... It can be represented as .

[0049] To facilitate computer processing, the continuous hydrodynamic model described above needs to be converted into a discrete hydrodynamic model. The key lies in discretizing the derivative terms (acceleration and angular acceleration) in the model. Generally, the forward Euler method is used for discretization.

[0050] in, The sampling time interval, i.e. k +1 time and k Time interval: ; Then, by substituting the discretized derivative terms into the hydrodynamic model, the discrete hydrodynamic model can be obtained.

[0051] Specifically, step 2 includes: We collect motion information from marine vehicles, assuming the data is sampled at equal intervals, and denote the sampling interval as . h The sampling time points are: Then the following condition is met: , N The data length is specified. The collected data consists of the speed and angular velocity of the ocean vehicle, which are stored as follows:

[0052] The raw motion data was denoised using a low-pass filter to remove high-frequency measurement noise. Then, a Savitzky-Golay filter was used for further denoising, and the acceleration data was obtained by automatic differentiation.

[0053] Specifically, step 3 includes: For each degree of freedom, a method of minimizing prediction error is used to find the parameter combination that minimizes the error between the predicted and measured values ​​as the initial identification result. To avoid overfitting and improve the generalization and robustness of the identification results, this invention uses multi-step iterative prediction to generate predicted values. That is, based on the discrete hydrodynamic model described above, the predicted value from the previous moment is used to generate the current predicted value. Since the velocity derivative of each degree of freedom is not only related to the velocity of the current degree of freedom, but also to the velocity and acceleration of other degrees of freedom, the single degree of freedom identification in step 3 uses the predicted value of the current degree of freedom from the previous moment and the measured values ​​of other degrees of freedom to generate the predicted value of the current degree of freedom. For the error between the predicted and measured values, this invention uses multiple loss functions, which can be freely selected according to actual needs. Each degree of freedom component can be selected from different loss function types to cope with different noise environments and robustness requirements, using the measured value of the oscillation velocity. and predicted value For example, its loss function includes, but is not limited to:

[0054] In practical applications, the most suitable loss function type can be selected based on factors such as the proportion of outliers in the data to ensure the effectiveness of the identification results. The aforementioned loss function based on multi-step prediction can effectively prevent the parameter identification algorithm from falling into overfitting; however, multi-step iterative prediction accumulates errors, affecting the identification results. Therefore, this invention also integrates a data assimilation mechanism, that is, by introducing measured values, the parameter identification algorithm can perceive the impact of environmental interference and other factors on the prediction error, in order to... u Taking direction as an example, the specific formula is as follows:

[0055] Among them, the original predicted value We now use the model's prior predictions, which are then linearly weighted with the measured values ​​to obtain the posterior predictions, used for the next iteration. It is the weighting factor of the measured value in the data assimilation mechanism. This multi-step iterative prediction method, which integrates the data assimilation mechanism, can effectively avoid the cumulative deviation of multi-step prediction errors from the true value. Using it in the loss function mentioned above can enhance the robustness of the parameter identification algorithm.

[0056] To minimize the loss function, this invention proposes a novel hybrid optimization algorithm based on the Nelder-Mead simplex method and gradient descent method: the NMG optimization algorithm.

[0057] Specifically, the optimization algorithm module includes: Based on the initial point Obtaining the dimensions of the optimization problem D Then generate randomly D Initial point .These( D +1) points form a simplex. In each iteration, the following steps are performed: Will( D +1) points according to their function values Sort by size from smallest to largest, satisfying:

[0058] Then, the classic gradient descent method is used to attempt to optimize the optimal point. :

[0059] in For learning rate, For the function in The derivative at point .

[0060] Calculate the function value at the gradient descent point and compare it with the worst point. If the gradient descent point is better than the worst point, replace the worst point and re-sort the function values ​​in descending order.

[0061] Calculate and exclude the worst case The center of mass of the simplex:

[0062] Reflection: the worst point Regarding the center of mass m reflection:

[0063] in, It is the reflection point. It is the reflection coefficient.

[0064] Calculate the function value of the reflection point and compare it with the function value of the optimal point. Comparison. If the reflection point is better than the optimal point, then perform the expansion operation:

[0065] in It is an extension point. It is the expansion coefficient. If the expansion point is better than the reflection point: If the worst-case scenario is correct, replace the worst-case point with the extended point and proceed to the next iteration; otherwise, replace the worst-case point with the reflected point and proceed to the next iteration.

[0066] If the reflection point is worse than the optimal point, but better than the second worst point: If the worst-case scenario is not found, the reflection point is used to replace the worst-case point, and the process proceeds to the next iteration.

[0067] If the reflection point is worse than the second worst point, then a contraction operation is performed, which has two possible scenarios: If the reflection point is worse than the second worst point, but better than the worst point: Then, the external contraction operation is performed:

[0068] If the reflection point is worse than the worst point, then perform an inward contraction operation:

[0069] in They are all contraction points. This is the contraction coefficient. If the contracted point is better than the reflection point and the worst point, then the contracted point replaces the worst point, and the iteration continues; otherwise, a backtracking operation is performed: all vertices of the simplex are contracted towards the optimal point.

[0070] in It is the backoff point corresponding to each vertex of the simplex. It is the backoff coefficient. A new simplex is formed using the backoff point and the optimum, and then the process proceeds to the next iteration.

[0071] When the convergence condition is met, the iteration terminates, and the optimization result is used as the preliminary identification result.

[0072] Specifically, step 4 includes: Based on the preliminary identification results obtained in step 3, the method of minimizing prediction error and multi-step iterative prediction is also used to find the parameter combination that minimizes the error between the predicted and measured values ​​as the final identification result. For integrating the error between predicted and measured values ​​with multiple degrees of freedom, this invention employs various loss functions, which can be freely selected according to actual needs, including:

[0073] Then, the same optimization algorithm as in step 3 is used to minimize the loss function, and the optimization result is used as the final identification result of the hydrodynamic parameters of the marine vehicle.

[0074] The present invention also provides a cascaded vehicle model parameter identification system with fused data assimilation. The cascaded vehicle model parameter identification system with fused data assimilation can be implemented by executing the process steps of the cascaded vehicle model parameter identification method with fused data assimilation. That is, those skilled in the art can understand the cascaded vehicle model parameter identification method with fused data assimilation as a preferred embodiment of the cascaded vehicle model parameter identification system with fused data assimilation.

[0075] Example 2 Example 2 is a preferred example of Example 1. The method for parameter identification of a cascaded aircraft model based on fused data assimilation provided by the present invention includes the following steps: Step 1: Use a six-degree-of-freedom hydrodynamic model to identify parameters of the ocean vehicle's navigation data; and perform filtering and noise reduction on the raw motion data.

[0076] Step 2: Define the loss function for a single degree of freedom, choosing the commonly used L2 norm as the loss function, with the oscillation velocity as the metric. For example: The weighting coefficients are set to That is, the weights of the prior predictions and the measured values ​​are consistent; with oscillation velocity For example: .

[0077] Step 3: Employ the NMG optimization algorithm. The hyperparameters of the optimization algorithm are selected as follows: Under these hyperparameter conditions, the formulas for calculating the reflection point, expansion point, contraction point, and regression point can be specified as follows:

[0078] The iterative convergence threshold is taken as The maximum number of iterations is set to After the iterations are terminated, preliminary parameter identification results are obtained.

[0079] Step 4: Define the loss function for multiple degrees of freedom, choosing the commonly used L2 norm as the loss function: The weighting coefficient will still be taken as That is, the weights of the prior predictions and the measured values ​​are consistent.

[0080] Step 5: Employ the NMG optimization algorithm. The hyperparameters of the optimization algorithm are still selected as follows: The iterative convergence threshold is set to... The maximum number of iterations is set to After the iterations are terminated, the final parameter identification result is obtained.

[0081] Example 3 Example 3 is a preferred example of Example 1. The method for parameter identification of a cascaded aircraft model based on fused data assimilation provided by the present invention includes the following steps: Step 1: Parameter identification is performed on the navigation data of the marine vehicle using a six-degree-of-freedom hydrodynamic model. The raw motion data is then filtered and denoised.

[0082] Step 2: Define a single-degree-of-freedom loss function, choosing the infinite norm as the loss function, with oscillation velocity as the metric. For example: The maximum deviation between the predicted and measured values ​​is used as the loss function. This is equivalent to adding a constraint band symmetrical about the measured values ​​to the predicted values, thereby gradually reducing the error band range during the optimization process. The weighting coefficients are set to... At this point, the posterior predicted value equals the measured value, so the next prior predicted value is directly generated from the measured value. This method of predictive value generation decouples the iterative relationship, allowing batch processing to be used to calculate the predicted value, greatly simplifying the calculation steps, improving computational efficiency, and achieving the goal of speed.

[0083] Step 3: Employ the NMG optimization algorithm. The hyperparameters of the optimization algorithm are selected as follows: Under these hyperparameter conditions, the formulas for calculating the reflection point, expansion point, contraction point, and regression point can be specified as follows:

[0084] The iterative convergence threshold is taken as The maximum number of iterations is set to After the iterations are terminated, preliminary parameter identification results are obtained.

[0085] Step 4: Define the loss function for multiple degrees of freedom, choosing the commonly used L2 norm as the loss function: The weighting coefficient will still be taken as That is, the weights of the prior predictions and the measured values ​​are consistent.

[0086] Step 5: Employ the NMG optimization algorithm. The hyperparameters of the optimization algorithm are still selected as follows: The iterative convergence threshold is set to... The maximum number of iterations is set to After the iterations are terminated, the final parameter identification result is obtained. Those skilled in the art will understand that, in addition to implementing the system, apparatus, and their modules provided by this invention in purely computer-readable program code, the same program can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system, apparatus, and their modules provided by this invention can be considered a hardware component, and the modules included therein for implementing various programs can also be considered structures within the hardware component; alternatively, modules for implementing various functions can be considered both software programs implementing the method and structures within the hardware component.

[0087] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A method for parameter identification of a cascaded aircraft model with fused data assimilation, characterized in that, include: Step S1: Construct a six-degree-of-freedom hydrodynamic model for marine vehicles, including sway, roll, heave, pitch, pitch and yaw. Step S2: Obtain motion response data of the marine vehicle, including attitude, velocity, and acceleration, through methods including live-boat experiments and / or numerical simulations; Step S3: Identify the parameters for each degree of freedom of the six-degree-of-freedom hydrodynamic model to obtain preliminary estimates of the hydrodynamic parameters for each degree of freedom; Step S4: Based on the preliminary estimates of the hydrodynamic parameters of each degree of freedom, perform multi-degree-of-freedom joint parameter identification on the six-degree-of-freedom hydrodynamic model of the marine vehicle to obtain the final parameter identification results.

2. The method for parameter identification of cascaded aircraft models with fused data assimilation according to claim 1, characterized in that, The six-degree-of-freedom hydrodynamic model includes: Two right-handed coordinate systems are established: the geodetic coordinate system {I} and the body coordinate system {B} of the ocean vehicle. The body coordinate system {B} x The axis points towards the bow of the ship. y The axis points to the starboard side. z The axis points vertically downwards; The pose of a marine vehicle in the geodetic coordinate system {I} is defined as follows: ,in, The three-dimensional spatial position of the marine vehicle in the geodetic coordinate system {I}; This refers to the roll angle; The pitch angle; To raise the bow angle; The velocity of an ocean vehicle in the body coordinate system {B} is defined as: ;in, For the body coordinate system {B} The linear velocity of the shaft; For orbital coordinate system Angular velocity of the axis; In the body coordinate system {B}, let the position of the center of gravity of the ocean vehicle be denoted as . The position of the center of buoyancy is Let the mass of the ocean-going vehicle be denoted as . Then its gravity is ,in, It is the acceleration due to gravity; Let the moments of inertia of the ocean-going vehicle about the three coordinate axes be: The product of inertia is: ; The vertical rudder angle of an ocean-going vehicle. The horizontal rudder angle of a marine vehicle; The coefficients on the right side of the equation are the hydrodynamic parameters that need to be identified, and are collectively denoted as a parameter vector. ; parameter vector Based on the structure of the six-degree-of-freedom model equations, it is divided into six subsets of degrees of freedom: ; All parameters Represented as ; The continuous six-degree-of-freedom hydrodynamic model is transformed into a discrete six-degree-of-freedom hydrodynamic model, and the derivative terms are discretized using the forward Euler method: in, The sampling time interval, , This is time k+1; Let k be the time. Then, the discretized derivative terms are substituted into the hydrodynamic model to obtain the discrete hydrodynamic model.

3. The method for parameter identification of cascaded aircraft models with fused data assimilation according to claim 2, characterized in that, Step S2 includes: We collect motion response data from marine vehicles, assuming the data is sampled at equal intervals, and denote the sampling interval as . h The sampling time points are: Then the following condition is met: , N The data length is specified; the collected data consists of the speed and angular velocity of the marine vehicle. The collected motion response data of the marine vehicle is denoised to obtain the processed motion response data of the marine vehicle. Acceleration data is calculated based on the processed motion response data of the marine vehicle.

4. The method for parameter identification of cascaded aircraft models with fused data assimilation according to claim 3, characterized in that, Step S3 includes: Step S3.1: Based on the discrete hydrodynamic model, generate the predicted value for the current time using the predicted value from the previous moment; wherein, single degree of freedom identification uses the predicted value of the current degree of freedom from the previous moment and the measured values ​​of other degrees of freedom to generate the predicted value of the current degree of freedom. Step S3.2: Construct a loss function for the error between the predicted and measured values, and minimize the loss function through an optimization algorithm to obtain preliminary estimates of the hydrodynamic parameters for each degree of freedom; Step S3.3: Obtain the posterior prediction value by linearly weighting the prior prediction value and the measured value; the posterior prediction value is used for the next iteration to obtain the prediction value at the next time step.

5. The method for parameter identification of cascaded aircraft models with fused data assimilation according to claim 4, characterized in that, Step S3.2 includes: measurements based on oscillation velocity. and predicted value Construct a loss function for the oscillation degree of freedom; 。 6. The method for parameter identification of cascaded aircraft models based on fused data assimilation according to claim 4, characterized in that, Step S3.2 includes: The NMG optimization algorithm is obtained based on a hybrid optimization algorithm of Nelder-Mead simplex method and gradient descent method, and the loss function is minimized using the NMG optimization algorithm. The NMG optimization algorithm includes: Based on the initial point Obtaining the dimensions of the optimization problem D Then generate randomly D Initial point ;These( D +1) points form a simplex; in each iteration, perform the following steps: Will( D +1) points according to their function values Sort by size from smallest to largest, satisfying: Then, the classic gradient descent method is used to attempt to optimize the optimal point. : in, For learning rate, For the function in The derivative at point; Calculate the function value at the gradient descent point and compare it with the worst point. If the gradient descent point is better than the worst point, replace the worst point and re-sort the function values ​​in descending order. Calculate and exclude the worst case The center of mass of the simplex: Reflection: the worst point Regarding the center of mass m reflection: in, It is the reflection point. It is the reflection coefficient; Calculate the function value of the reflection point and compare it with the function value of the optimal point. Compare; if the reflection point is better than the optimal point, then perform the expansion operation: in, It is an extension point. It is the expansion coefficient; if the expansion point is better than the reflection point: If the worst-case scenario is not found, the extended point is used to replace the worst-case point, and the process proceeds to the next iteration; otherwise, the reflected point is used to replace the worst-case point, and the process proceeds to the next iteration. If the reflection point is worse than the optimal point, but better than the second worst point: If the worst-case scenario is not found, the reflection point is used to replace the worst-case point, and the process proceeds to the next iteration. If the reflection point is worse than the second worst point, a contraction operation is performed, including: If the reflection point is worse than the second worst point, but better than the worst point: Then, the external contraction operation is performed: If the reflection point is worse than the worst point, then perform an inward contraction operation: in, They are all contraction points. This is the contraction coefficient; if the contracted point is better than the reflection point and the worst point, then the contracted point replaces the worst point, and the process proceeds to the next iteration; otherwise, a backtracking operation is performed, shrinking all vertices of the simplex towards the optimal point: in, It is the backoff point corresponding to each vertex of the simplex. It is the backoff coefficient; the backoff point and the optimal point are used to form a new simplex, and then the process proceeds to the next iteration. When the convergence condition is met, the iteration terminates, and the optimization result is used as the preliminary identification result.

7. The method for parameter identification of cascaded aircraft models with fused data assimilation according to claim 4, characterized in that, Step S3.3 includes: introducing measured values ​​through a data assimilation mechanism to enable parameter identification to perceive the impact of environmental disturbances on prediction errors. u In terms of direction, the specific formula is as follows: in, The posterior predicted value from the previous time step. The velocity vector of the motion at the previous moment. For the aforementioned u The parameter vector of degrees of freedom, The prior prediction value at the current moment. It is the weighting factor of the measured value in the data assimilation mechanism. The measured value at the current moment. This is the posterior prediction value at the current time.

8. The method for parameter identification of cascaded aircraft models with fused data assimilation according to claim 4, characterized in that, Step S4 includes: Based on the preliminary estimates of hydrodynamic parameters for each degree of freedom, a method of minimizing prediction error and multi-step iterative prediction is used to find the parameter combination that minimizes the error between the predicted and measured values, which is then used as the final identification result. For integrating the errors between predicted and measured values ​​across multiple degrees of freedom, various loss functions are employed, which can be freely selected according to actual needs, including: Then, the NMG optimization algorithm is used to optimize the loss function of multiple degrees of freedom, and the optimization result is used as the final hydrodynamic parameter identification result of the marine vehicle.

9. A cascaded aircraft model parameter identification system that integrates data assimilation, characterized in that, include: Module M1: Constructs a six-degree-of-freedom hydrodynamic model for marine vehicles, including sway, roll, heave, pitch, pitch and yaw. Module M2: Acquires motion response data of marine vehicles, including attitude, velocity, and acceleration, through methods including live-fire experiments and / or numerical simulations; Module M3: Identifies parameters for each degree of freedom in the six-degree-of-freedom hydrodynamic model and obtains preliminary estimates of the hydrodynamic parameters for each degree of freedom; Module M4: Based on the preliminary estimates of hydrodynamic parameters for each degree of freedom, multi-degree-of-freedom joint parameter identification is performed on the six-degree-of-freedom hydrodynamic model of the marine vehicle to obtain the final parameter identification results.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the various steps of the cascaded vehicle model parameter identification method for fused data assimilation as described in any one of claims 1-8.

Images