Mechanism model parameter iterative optimization method based on data and neural network

Abstract

The invention discloses a mechanism model parameter iterative optimization method based on data and a neural network, and the method comprises the steps: constructing a mechanism model based on existing experience, analyzing and selecting key optimization parameters of the model, and constructing an initial parameter value model as a to-be-optimized model; secondly, collecting state data such as a flight path of the aircraft, and preprocessing the state data; thirdly, a parameter iterative optimization model is constructed through intelligent optimization algorithms such as a neural network algorithm, training is conducted through the preprocessed data, the simulation result of the mechanism model is made to be gradually close to actual data, and finally the optimized aircraft mechanism model is obtained. According to the method, the advantages of mechanism modeling and neural network modeling are fused, and mechanism model iterative optimization under mechanism and data hybrid driving is realized.

Description

Technical Field

[0001] The invention relates to a method for iterative optimization of mechanism model parameters based on data and neural networks, and belongs to the technical field of simulation modeling. Background Art

[0002] At present, there is a great practical demand for multi-granularity and multi-level high-precision modeling in the fields of aviation, aerospace, ships, weapons, military equipment, etc. How to make full use of data resources and promote the development of high-fidelity modeling technology has attracted more and more attention and concern from the defense and military industry and military leaders, and it is also the focus and difficulty of the vast number of scientific researchers and engineering and technical personnel.

[0003] The iterative optimization method of mechanism model parameters based on data and neural networks is a concrete implementation of data and mechanism hybrid simulation modeling. It is also a good combination of artificial intelligence and simulation modeling. It can effectively improve the applicability and intelligence of simulation, and has important engineering value for simulation model construction.

[0004] The Chinese patent discloses a method for identifying aerodynamic parameters of controlled missiles and arrows based on Elman recursive neural networks. The Elman recursive neural network is applied to the identification of aerodynamic parameters of gliding guided projectiles in the controlled flight phase. According to the model and sample data, the Elman neural network model is constructed: the number of input layer nodes, the number of hidden layer neurons, the activation function, and the number of output layer nodes of the Elman neural network model are determined to adjust the neural network structure, and the error function is solved according to the actual situation of the gliding missile, and the network weights are updated. The disadvantage is that the iterative optimization process of the mechanism model parameters cannot take into account special states such as the highest point and the farthest point of the aircraft.

[0005] The Chinese patent discloses a method for identifying aerodynamic parameters of an aircraft based on a physical information neural network. The method establishes a dynamic model and an aerodynamic model of the aircraft to be identified, designs a control input signal, uses the model in step 1 to perform a simulated flight test to obtain flight data, inputs flight data, sets the loss function as the mean square error of the test data, and trains a deep neural network between time and aircraft state quantities; based on the deep neural network, sets the parameters to be identified, uses an automatic differentiator, adds a differential equation residual term to the loss function, and continues to train the deep neural network until the loss function converges or reaches a set threshold. The disadvantage is that the model cannot consider the impact of long-period parameters, including the maximum range and maximum flight distance of the aircraft.

[0006] A Chinese patent discloses a method and device for modeling aircraft dynamics based on a NARX neural network. The method includes: obtaining input and output data of an aircraft dynamics model based on Simulink simulation, and preprocessing the data to generate training data and test data; using a NARX neural network to train and test the aircraft dynamics model; and establishing an aircraft longitudinal dynamics model based on the trained and tested NARX neural network structure. The patent realizes the prediction of aircraft angle of attack and pitch rate parameters under input conditions such as random elevator and throttle lever. Its shortcoming is that it only predicts the aircraft angle of attack and pitch angle parameters, but does not include the aircraft thrust parameters.

[0007] Through the research and analysis of existing patent technologies, we can find that the main problems are: the parameter iteration types are single, focusing more on aerodynamic parameters; the iterative optimization parameters are mostly the parameters of the aircraft model itself, while the performance parameters such as the farthest aircraft distance and maximum flight altitude of the aircraft are not iteratively optimized. This adds a lot of repetitive work to model developers and prolongs the system development time. Summary of the invention

[0008] The technical problem solved by the present invention is: to overcome the shortcomings of the prior art and propose a method for iterative optimization of mechanism model parameters based on data and neural networks. On the basis of mechanism modeling, the data is combined with the mechanism model to achieve iterative optimization of model parameters and improve the accuracy and adaptability of the model.

[0009] The technical solution of the present invention is:

[0010] A method for iterative optimization of mechanism model parameters based on data and neural network, comprising:

[0011] S1: Analyze and select key optimization parameters of the existing mechanism model of the aircraft;

[0012] S2: Collect historical data of aircraft flight trajectory status, pre-process the historical data, and remove abnormal data;

[0013] S3: Construct a parameter iteration optimization model, where the model input is the state of the aircraft at a certain moment, including but not limited to the flight time, position state deviation, velocity state deviation, acceleration state deviation, and heading angle state deviation; the model output is the key optimization parameters of the aircraft mechanism model; the aircraft mechanism model using the current key optimization parameters is used to calculate the estimated state value at the next moment, and the estimated state value is subtracted from the actual state value at the next moment in the actual trajectory to construct a loss function;

[0014] S4: Use the preprocessed historical data to train the parameter iteration optimization model so that the simulation results of the mechanism model gradually approach the actual data, and obtain the optimized aircraft mechanism model.

[0015] Furthermore, in step S1, the existing mechanism model of the aircraft is analyzed and key optimization parameters of the model are selected, specifically:

[0016] S11: Classify the parameters in the existing mechanism model of the aircraft, and extract multiple parameters in each category that have the greatest impact on the flight trajectory;

[0017] S12: Select the flight trajectory error as the objective function to measure the deviation between the simulated trajectory and the actual trajectory; perturb each parameter extracted in step S11, record the changes in the flight trajectory, evaluate the influence of each parameter on the objective function, and identify the multiple parameters that are most critical to the output of the mechanism model as the key optimization parameters.

[0018] Furthermore, in step S11, various parameters in the existing mechanism model of the aircraft are classified, including but not limited to aerodynamic parameters, inertial parameters, control parameters, and guidance parameters;

[0019] Extract multiple parameters in each category that have the greatest impact on the flight trajectory:

[0020] Among the aerodynamic parameters, those that have the greatest impact on flight trajectory include lift coefficient, drag coefficient, and side force coefficient;

[0021] Among the inertial parameters, the ones that have the greatest impact on the flight trajectory include the total mass and moment of inertia of the aircraft;

[0022] Among the control parameters, the ones that have the greatest impact on the flight trajectory include rudder deflection and thrust;

[0023] Among the guidance parameters, the parameters that have the greatest impact on the flight trajectory include the aircraft's guidance method and guidance rate; among them, the guidance method is the method of controlling the aircraft to fly along a predetermined trajectory.

[0024] Furthermore, in step S12, the method for identifying multiple parameters that are most critical to the output of the mechanism model is:

[0025] First, the baseline values ​​of all input parameters in the mechanism model are determined, and the objective function values ​​when all parameters are at the baseline values ​​are calculated and recorded; the OAT method is used to change only one of the input parameters at a time, and the change in the objective function after the parameter change is calculated;

[0026] The sensitivity of each parameter is evaluated by the change in the objective function after each parameter is changed. All parameters are sorted according to the sensitivity index to identify multiple parameters that have a great impact on the objective function.

[0027] Furthermore, in step S2, historical data of the aircraft's flight trajectory status are collected, including parameters under different flight conditions and at different time points, including position, velocity, acceleration, heading angle and attitude angle; the above historical data are preprocessed, and the preprocessing operations are: first, outlier detection and filtering are performed, and abnormal data are cleaned and corrected; then state quantity differentiation and normalization processing are performed.

[0028] Further, in step S3, the loss function performs a subtraction process on the estimated state value at the next moment obtained by calculating the aircraft mechanism model and the actual state value at the next moment in the actual trajectory to obtain the loss function value, which is specifically:

[0029] The current position coordinates of the aircraft are [p x(k) ,p y(k) ,p z(k) ], the position coordinates at the next moment estimated by the mechanism model are [p x(k+1) ,p y(k+1) ,p z(k+1) ], the actual coordinates at the next moment are [p x(k+1) ,p y(k+1) ,p z(k+1) ], and the difference is obtained to get the deviation between the estimated position and the actual position:

[0030] [p x(k+1) -p x(k+1) ,p y(k+1) -p y(k+1) ,p z(k+1) -p z(k+1) ]

[0031] Calculate the absolute distance as the loss function value:

[0032]

[0033] The loss function is dimensionless, and the updated loss function is:

[0034]

[0035] In the formula, It is a dimensionless factor, which makes all items of the loss function dimensionless and takes 120% to 150% of the maximum value of the distance between two adjacent moments.

[0036] Furthermore, in step S3, a parameter iteration optimization model is constructed through a BP neural network.

[0037] Furthermore, the Adam adaptive learning rate neural network optimization method is adopted to adjust the learning rate of each parameter by tracking the first-order moment and second-order moment of the gradient, so that the learning rate can be adaptively adjusted in each training step, and the iterative optimization of parameters driven by the hybrid mechanism and data can be realized, thereby improving the training efficiency and convergence speed.

[0038] Furthermore, the accuracy of the optimized aircraft mechanism model is verified:

[0039] The preprocessed historical data is divided into training data and test data; the training data is used to train the parameter iterative optimization model;

[0040] Simulation experiments are carried out based on the optimized aircraft mechanism model, and the simulation results are compared with the test data to evaluate the accuracy of the aircraft mechanism model and obtain the model deviation, which is used to correct the aircraft mechanism model.

[0041] Furthermore, in step S4, the preprocessed historical data is used to train the parameter iteration optimization model so that the simulation results of the mechanism model gradually approach the actual data. After completing the iterative optimization, the parameter iteration optimization model outputs the key optimization parameters of the aircraft corresponding to the current input; if the parameter is a constant, the output of the parameter is a constant value; if the parameter is a time-varying quantity, the parameter will change with different inputs, and the parameter iteration optimization model parameters corresponding to different times are obtained.

[0042] The advantages of the present invention compared with the prior art are:

[0043] (1) The present invention introduces a neural network algorithm to iteratively optimize the parameters of the mechanism model, thereby solving the problems of low efficiency, low intelligence level, and low model parameter accuracy in the existing simulation modeling process. It integrates the advantages of mechanism modeling and neural network modeling to achieve iterative optimization of the mechanism model under the hybrid drive of mechanism and data.

[0044] (2) The present invention selects the parameters to be optimized of the mechanism model through sensitivity analysis, thereby solving the problem of lack of effective basis for selecting the parameters to be identified in the model. All parameters are sorted according to the sensitivity index to select the parameters to be optimized that have the greatest impact on the objective function.

[0045] (3) The present invention solves the problem that direct fitting of neural networks is difficult to converge by constructing a loss function that combines the estimated value of the mechanism model with the actual historical data value, and realizes iterative optimization of neural network parameters under the constraints of the mechanism model. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0047] Figure 1 This is a flow chart of a method for iterative optimization of mechanism model parameters based on data and neural networks according to an embodiment of the present invention;

[0048] Figure 2 This is a schematic diagram of the motion equation of an aircraft according to an embodiment of the present invention;

[0049] Figure 3 This is a flow chart of aircraft trajectory data processing according to an embodiment of the present invention;

[0050] Figure 4 Schematic diagram of the BP neural network iterative optimization model according to an embodiment of the present invention. DETAILED DESCRIPTION

[0051] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0052] The present invention proposes an iterative optimization method for mechanism model parameters based on data and neural network, which is divided into two layers: model layer and data layer, respectively performing construction of iterative optimization model and processing of model data, and jointly completing iterative optimization of mechanism model.

[0053] Taking the aircraft model as an example, the process of this method is as follows Figure 1 As shown, the specific steps include:

[0054] S1: Build a mechanism model based on existing experience, analyze and select key optimization parameters of the model, and build an initial parameter value model as the model to be optimized.

[0055] S10: Model object determination

[0056] The model object is a six-degree-of-freedom aircraft model, which is a mathematical model that describes the motion state of the aircraft in three-dimensional space. Figure 2 As shown, the three translational degrees of freedom (displacement along the X, Y, and Z axes) and three rotational degrees of freedom (rotation around the X, Y, and Z axes) of the aircraft are covered.

[0057] Based on the principles of Newtonian mechanics, the vector equation for the motion of the center of mass of an aircraft in inertial space is:

[0058] ma i =F+mg

[0059] What we are usually concerned about is the motion of the aircraft relative to the earth. The speed of the center of mass of the aircraft relative to the earth, that is, the relative speed, is expressed as V k Indicates; and the relative acceleration is dV k / dt. Since the earth has an angular velocity of rotation, the earth coordinate system S e It is not an inertial reference frame, so the angular velocity w appears in the relative motion equation. e Inertia force caused.

[0060] The absolute acceleration of the aircraft is a i Equal to the relative acceleration dV k / dt, involved acceleration a e and the Columbia acceleration a c The sum is:

[0061] a i =dV k / dt+a e +a c

[0062] The involved acceleration is:

[0063] a e =ω E ×(ω E ×r)

[0064] The G-force acceleration is:

[0065] a c =2ω E ×V k

[0066] The gravitational acceleration caused by the spherical earth can be expressed as:

[0067] g=-(μ / r 3 )r

[0068] The dynamic equation of the relative motion of the aircraft to the ground can be obtained:

[0069]

[0070] Its expression in the aircraft track coordinate system is:

[0071]

[0072] The kinematic equation of the center of mass under the assumption of a spherical earth is as follows:

[0073]

[0074] The dynamic equation of the aircraft's rotational motion is:

[0075]

[0076] The absolute angular velocity ω of the aircraft should be equal to the sum of the following three parts: the angular velocity of the earth's rotation ω E , S u Relative to S e The angular velocity ω ue and S b Relative to S u The angular velocity ω bu .

[0077] Establish the rotational kinematics differential equation, that is, the rate of change of relative attitude angle dφ / dt, dθ / dt and dψ / dt and the relative angular velocity component ω rxb ,ω ryb and ω rzb relationship.

[0078]

[0079] S11: Determination of key optimization parameters of the model

[0080] 1) Parameter classification

[0081] First, the model parameters are divided into the following categories, and the parameters that have the greatest impact on the flight trajectory are selected, including aerodynamic parameters (lift coefficient, drag coefficient, side force coefficient, etc.), inertial parameters (mass distribution, moment of inertia, etc.), control parameters (rudder deflection angle, thrust, etc.), and guidance parameters (guidance method, guidance rate, etc.).

[0082] Aerodynamic parameters: Aerodynamic parameters are key factors that describe the behavior of an aircraft in an airflow, including the lift coefficient, drag coefficient, and side force coefficient. The lift coefficient represents the lift generated by the aircraft in the vertical direction and depends on the angle of attack and airfoil design; the drag coefficient describes the air resistance encountered by the aircraft in the forward direction and is affected by the surface smoothness and angle of attack; the side force coefficient reflects the side force of the air on the aircraft in the lateral direction and affects the lateral stability of the aircraft. These parameters are crucial in aircraft modeling and simulation, as they affect flight trajectory, fuel efficiency, and flight stability.

[0083] Inertial parameters: Inertial parameters describe the mass distribution and rotational characteristics of the aircraft, including the total mass and moment of inertia of the aircraft. The total mass affects the gravity and acceleration response of the aircraft, while the moment of inertia measures the resistance of the aircraft to rotation on each axis. The mass distribution determines the location of the center of gravity of the aircraft, while the moment of inertia affects the attitude stability and control response of the aircraft. Accurate understanding of these inertial parameters can help optimize the control and stability of the aircraft, ensuring that it performs well in various flight conditions.

[0084] Control parameters: Control parameters are key factors that determine the attitude and movement of the aircraft, including steering and thrust. Steering refers to the angle of deflection of the aircraft's control surfaces (such as elevators, rudders, and ailerons). These deflections control the aircraft's pitch, yaw, and roll attitude by changing the direction of the airflow. Thrust refers to the thrust strength generated by the engine, which directly affects the acceleration and speed of the aircraft. By accurately adjusting these control parameters, stable control of the aircraft and the target flight trajectory can be achieved.

[0085] Guidance parameters: are the core elements of the aircraft navigation and control system, involving two key aspects: the aircraft's guidance method and guidance rate. Guidance method refers to the method of controlling the aircraft to fly along a predetermined trajectory, which can be achieved through a variety of technologies, such as guidance (PN), and pure tracking. Different guidance methods are suitable for different mission requirements and can affect the aircraft's path planning and ability to ultimately reach the target. Guidance rate refers to the speed at which the aircraft follows the target or adjusts its heading, affecting its responsiveness and the accuracy of path adjustment. A high guidance rate can increase the aircraft's response speed, enabling it to respond more quickly to changes in the external environment and changes in the target's position, thereby improving the aircraft's stability and flexibility. When designing and adjusting guidance parameters, it is necessary to find a balance between response speed and stability. By optimizing these guidance parameters, the aircraft's tracking capability and navigation accuracy can be significantly improved, ensuring that it can fly along the expected route and accurately fly to the designated target.

[0086] 2) Sensitivity analysis

[0087] Select the flight trajectory error as the objective function to measure the deviation between the simulated trajectory and the actual trajectory. Perturb each parameter within a reasonable range and record the changes in the flight trajectory. Evaluate and select the key optimization parameters.

[0088] In the vehicle trajectory analysis, the flight trajectory error is selected as the objective function to measure the deviation between the simulated trajectory and the actual trajectory. Each parameter is perturbed within a reasonable range, and the changes in the flight trajectory are recorded to evaluate the influence of each parameter on the objective function. In this way, the parameters that are most critical to the model output can be effectively identified, so that resources can be focused on optimizing these parameters.

[0089] In order to conduct a detailed analysis, the OAT method (One-at-a-Time) is used, that is, only one parameter is changed at a time to observe its effect on the objective function. First, the baseline values ​​of all input parameters in the model are determined. These baseline values ​​are usually based on actual data, literature values, or expert experience. For each input parameter, keep other parameters fixed, only change the current parameter, and record the change in the objective function.

[0090] Calculate and record the objective function value F0 when all parameters are at baseline values, perturb each input parameter in turn, and calculate the change ΔF of the objective function after the perturbation. The perturbation method can be to increase or decrease the parameter value. The specific methods include: linear perturbation, increasing or decreasing the parameter value by 10%; range perturbation, multiple perturbations within the allowable range of the parameter, such as increasing or decreasing the parameter value by 5%, 10%, 15%, etc.

[0091] By calculating the change in the objective function ΔF i To evaluate the sensitivity of each parameter. The sensitivity index can be expressed by the change in the objective function, as shown in the following formula:

[0092]

[0093] All parameters are sorted according to the sensitivity index to identify the parameters that have the greatest impact on the objective function. At the same time, the parameters are screened based on traditional experience, and the obtained parameters are the key optimization objects.

[0094] S2: Collect the status data of the aircraft such as the flight trajectory, and pre-process it to become training data.

[0095] S20: Aircraft trajectory data collection

[0096] The trajectory data includes the position, speed, acceleration, and heading angle information of the aircraft at different time points. The trajectory data of the aircraft under different initial conditions (including the launch point, target point, initial flight direction, etc.) are collected to obtain multiple trajectory data, which are then further processed to obtain training data. This information can be collected through a variety of high-precision measurement equipment, including airborne sensors, global positioning systems (GPS), inertial navigation systems (INS), etc. Airborne sensors usually include accelerometers, gyroscopes, magnetometers, etc., which are used to monitor the dynamic changes of aircraft in real time; GPS provides accurate position and speed data; the inertial navigation system supplements the data of other sensors by measuring angular velocity and acceleration to ensure accurate positioning and navigation when GPS signals are insufficient.

[0097] S21: Aircraft trajectory data processing

[0098] Since the aircraft trajectory data may be affected by many factors, such as sensor errors, communication failures, or environmental interference, these factors may cause outliers or noise in the data. For example, the data may show abrupt speed changes, unreasonable displacements, or discontinuous trajectory paths. If these outliers are not processed, they will have a negative impact on subsequent data analysis and model optimization. Therefore, it is necessary to first perform outlier detection on the trajectory data to identify possible discontinuous areas or erroneous data, such as Figure 3 shown.

[0099] In order to further improve the data quality, filtering technology is often used to smooth the trajectory data. Filtering is a method to improve data quality by removing noise or interference signals. Common filtering techniques include sliding average filtering and Kalman filtering. The sliding average filter smoothes short-term fluctuations by calculating the local average of the data points, which is suitable for dealing with uniform noise; while the Kalman filter is a recursive filter that can effectively estimate the state in a dynamic system and is suitable for dealing with noise with dynamic characteristics.

[0100] Once abnormal data is identified, data cleaning and correction must be performed. Specific operations include eliminating erroneous data and using interpolation to fill missing values. Interpolation is a method of estimating missing data points through mathematical methods, which can effectively smooth data and maintain data continuity. In addition to interpolation, trends and patterns in historical data can also be used to correct data to ensure the rationality and consistency of the data. After data processing, further data verification is performed to confirm that the processed data can truly reflect the motion status of the aircraft.

[0101] At this point, continuous and smooth aircraft trajectory data is obtained, which is further processed into training data and test data for neural networks. For any trajectory data, the trajectory data is sliced ​​according to a fixed time step Δt to obtain state information at different times.

[0102] [x0,x1,x2,...,x k ,...,x f ]

[0103] In the formula, x represents the state of the aircraft at a certain moment, including position, velocity, acceleration, and heading angle information, forming a high-dimensional vector. The subscript represents the time series, where t0 refers to the time when the flight starts, and the time interval between each two moments is Δt, that is, t k+1 -t k =Δt,t f is the last moment of the flight trajectory. k The state quantity at the moment is represented by x k Therefore, the high-dimensional vector matrix can be used to describe the state quantity in a certain aircraft trajectory.

[0104] The state quantity is differentially processed, that is, the state quantity at the next moment is subtracted from the state quantity at the previous moment to obtain the new state differential data, which is expressed as follows:

[0105] [Δx1,Δx2,...,Δx k ,...,Δx f ]=[x1-x0,x2-x1,...,x k -x k-1 ,...,x f -x f-1 ]

[0106] Finally, the data is normalized. The main purpose of normalization is to adjust data of different scales to a relatively consistent range, usually scaling the eigenvalues ​​to between [-1,1]. This prevents some features from dominating the model update due to their large magnitude during the training process, and ensures that the model has the same sensitivity when processing each feature. Using the Min-MaxScaling normalization method, for a given eigenvalue Δx, the calculation formula for its scaled value Δx' is

[0107]

[0108] In addition, the denormalization method is:

[0109] Δx=Δx′·(max(Δx)-min(Δx))+min(Δx)

[0110] S22: Classification of aircraft training data and test data

[0111] Training data is a dataset used to build and train a neural network model. Through training data, the model can learn the dynamic characteristics and behavior patterns of the aircraft. The training data should have a wide coverage, covering a variety of possible flight conditions and scenarios, so that the model can learn to handle diverse situations. In this method, the training data is the differential data of the aircraft state after normalization processing as described above, that is, [Δx1', Δx2', ..., Δx f '].

[0112] Test data is a set of data independent of training data, which is used to evaluate the performance of the model and verify its performance on new data. Test data is usually data that has never been seen by the model, which can effectively test the generalization ability and prediction accuracy of the model. The test data should be consistent with the source of the training data, but should not be included in the training process to ensure that the evaluation results of the model are not affected by the training process. By analyzing the test data, we can understand the performance of the model in actual applications and evaluate its robustness and accuracy under different flight conditions.

[0113] The data is divided into training data and test data. The training data is used to evaluate the fit of the optimized model, and the test data is used to evaluate its generalization ability. The division ratio is 80% for training data and 20% for test data. This division ensures that the training data is rich enough to enable the model to learn fully, and the test data is independent enough to effectively verify the performance of the model. In order to further verify the reliability of the model, the cross-validation method can also be used, that is, the data set is divided into multiple subsets, and different subsets are used alternately as training data and test data to comprehensively evaluate the performance of the model on different data.

[0114] S3: Construct parameter iteration optimization model through intelligent optimization algorithms such as neural network algorithms

[0115] 1) BP neural network model construction

[0116] BP neural network is a widely used feedforward artificial neural network model with a multi-layer structure, such as Figure 4 As shown in Figure 1, it includes input layer, hidden layer and output layer. By learning the training data set, the BP neural network can adjust the weights and biases in the network to achieve modeling of the complex nonlinear relationship between input and output.

[0117] The training process of BP neural network includes two stages: forward propagation and back propagation. In the forward propagation stage, the input data is passed from the input layer to the output layer through the hidden layer, and the loss function is calculated to measure the difference between the model output and the actual target. In the back propagation stage, the error is passed back from the output layer to the input layer through the hidden layer, the gradient is calculated, and the neuron weights are updated by the gradient descent method to reduce the loss function. Through multiple iterative training, the network performance gradually converges to a satisfactory level.

[0118] The input of the neural network is the state difference component Δx' of the aircraft at a certain moment, which is specifically expressed as

[0119] x input =[t,Δp,ΔV,Δa,Δχ]

[0120] Where t is time, Δp is the position state deviation between two adjacent moments, ΔV is the velocity state deviation between two adjacent moments, Δa is the acceleration state deviation between two adjacent moments, and Δχ is the heading state deviation between two adjacent moments.

[0121] The output is the key optimization parameters of the aircraft [B1, B2, ..., B n ] corresponds to the parameter value [b1,b2,...,b n ], expressed as

[0122] y output=[b1,b2,...,b n ]

[0123] After outputting the key optimization parameters, substitute the parameters into the mechanism model, perform single-step integration and obtain the estimated state value at the next moment, and subtract it from the actual state true value at the next moment to obtain the loss function (the specific loss function will be introduced in detail later). After completing the iterative optimization, the output of the neural network will be the key parameter of the aircraft corresponding to the current state. If the parameter is a constant, the output of the quantity is a constant value. If the parameter is a time-varying quantity, the quantity will change with different inputs, and the model parameters corresponding to different moments will be obtained.

[0124] The activation function uses the Relu activation function, and its equation is as follows:

[0125]

[0126] The calculation process of the aircraft dynamics model is still completed by the mechanism model, and the calculated results are transmitted to the output layer of the neural network. The parameters to be optimized will be included in the neural network model as feature quantities, and iterative optimization will be achieved through the subsequent constructed loss function.

[0127] 2) Loss function construction

[0128] The loss function obtains the loss function value by subtracting the estimated value of the next moment state calculated according to the aircraft mechanism model from the actual value of the next moment state in the actual trajectory.

[0129] The current position coordinates of the aircraft are [p x(k) ,p y(k) ,p z(k) ], the position coordinates at the next moment estimated by the mechanism model are [p x(k+1) ,p y(k+1) ,p z(k+1) ], the actual coordinates at the next moment are [p x(k+1) ,p y(k+1) ,p z(k+1) ]. The difference between the estimated position and the actual position is obtained by subtracting

[0130] [p x(k+1) -p x(k+1) ,p y(k+1) -p y(k+1) ,p z(k+1) -p z(k+1) ]

[0131] And calculate the absolute distance as the loss function

[0132]

[0133] In order to better balance the loss function, the loss function is dimensionless, and the updated loss function is

[0134]

[0135] In the formula, It is a dimensionless factor, which makes all items of the loss function dimensionless and takes 120% to 150% of the maximum value of the distance between two adjacent moments.

[0136] S4: Through iterative parameter optimization driven by the hybrid mechanism and data, the simulation results of the mechanism model gradually approach the actual data, and finally the optimized aircraft mechanism model is obtained.

[0137] S40: Intelligent iterative optimization of model parameters

[0138] The Adam adaptive learning rate neural network optimization method is used to adjust the learning rate of each parameter by tracking the first-order moment (mean) and second-order moment (variance) of the gradient. This method enables the learning rate to be adaptively adjusted in each training step, thereby improving training efficiency and convergence speed.

[0139] Adam combines the idea of ​​momentum method, which helps the optimization algorithm to update gradients more stably by performing exponential weighted averaging of historical values ​​of gradients. Momentum method can smooth gradient changes and reduce oscillation, thereby accelerating convergence. Adam also incorporates the adaptive learning rate feature of RMSprop. RMSprop adjusts the learning rate by performing exponential weighted averaging of the square of the gradient of each parameter, which allows different learning rates to be used in the update of different parameters, helping to update at different speeds in different directions, thereby improving the efficiency and effectiveness of training.

[0140] S41: Model Accuracy Verification

[0141] 1) Model testing

[0142] After completing the iterative optimization of the aircraft model, conducting simulation experiments is a key step to verify the accuracy and performance of the model. By running the aircraft model in a simulation environment, its key flight status data, such as flight altitude, flight distance, time to reach maximum altitude, and time to reach the farthest point, are recorded. These simulation data are compared with the actual flight records to calculate their deviations. The deviation is to evaluate the accuracy of the model by comparing the difference between the simulation results and the actual values, ensuring that the model can accurately reflect the performance of the aircraft in the real environment. Through this comparative analysis, potential errors in the model can be identified and corrected, thereby improving the reliability and performance of the aircraft in actual applications.

[0143] 2) Qualitative analysis of errors

[0144] Calculate the deviation between the simulation results and the actual data. Quantitative statistics are achieved through mean square error (MSE).

[0145] Based on 20% of the test data, the actual values ​​of the flight altitude, flight distance, time to reach the maximum altitude, and time to reach the farthest point are compared with the model simulation values ​​to obtain the model deviation.

[0146] The present invention aims at the problems of nonlinearity and strong coupling in traditional mechanism modeling, which lead to complex mechanisms, high model development learning cost, low development efficiency, low intelligence level and inaccurate model parameters, and proposes a parameter iterative optimization method of mechanism model based on data and neural network. First, the key optimization parameters of the model are analyzed and selected, and the initial parameter value model is constructed as the model to be optimized based on existing research experience. The parameter iterative optimization model is constructed by intelligent optimization algorithms such as neural network algorithms, and state data such as aircraft flight trajectory are collected and preprocessed to become training data. Through the iterative optimization of parameters driven by the hybrid mechanism and data, the simulation results of the mechanism model gradually approach the actual data, and finally the optimized aircraft mechanism model is obtained. At the same time, the actual data is divided into training data and test data, and the fitting degree of the optimized model is evaluated by the training data, and its generalization ability is evaluated by the test data. The accuracy and credibility of the model are improved through the above evaluation and analysis optimization.

[0147] Through the application of the present invention, the modeling efficiency and modeling accuracy will be effectively improved, and a feasible solution will be provided for the accurate and rapid modeling of the mechanism model. Firstly, a general mechanism and data hybrid modeling method is proposed to solve the problem that the pure mechanism modeling is inaccurate and the pure neural network fitting is not easy to converge. The advantages of the two methods are integrated, and the optimization of the mechanism model is realized through the iterative optimization of parameters driven by the mechanism and data hybrid. Secondly, taking the modeling of a typical aircraft as an example, the problem of accurately modeling an aircraft when the aircraft parameters are unknown is solved. This method can be extended to various modeling and simulation fields such as aircraft, mobile platforms, and radars.

[0148] The above-described embodiments are only preferred specific implementations of the present invention. Common changes and substitutions made by those skilled in the art within the scope of the technical solution of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for iterative optimization of mechanism model parameters based on data and neural network, characterized in that: include: S1: Analyze and select key optimization parameters of the existing mechanism model of the aircraft; S2: Collect historical data of aircraft flight trajectory status, pre-process the historical data, and remove abnormal data; S3: Construct a parameter iteration optimization model, where the model input is the state of the aircraft at a certain moment, including but not limited to the flight time, position state deviation, velocity state deviation, acceleration state deviation, and heading angle state deviation; the model output is the key optimization parameters of the aircraft mechanism model; the aircraft mechanism model using the current key optimization parameters is used to calculate the estimated state value at the next moment, and the estimated state value is subtracted from the actual state value at the next moment in the actual trajectory to construct a loss function; S4: Use the preprocessed historical data to train the parameter iteration optimization model so that the simulation results of the mechanism model gradually approach the actual data, and obtain the optimized aircraft mechanism model.

2. The iterative optimization method for mechanism model parameters based on data and neural network according to claim 1, characterized in that: In step S1, the existing mechanism model of the aircraft is analyzed and key optimization parameters of the model are selected, specifically: S11: Classify the parameters in the existing mechanism model of the aircraft, and extract multiple parameters in each category that have the greatest impact on the flight trajectory; S12: Select the flight trajectory error as the objective function to measure the deviation between the simulated trajectory and the actual trajectory; perturb each parameter extracted in step S11, record the changes in the flight trajectory, evaluate the influence of each parameter on the objective function, and identify the multiple parameters that are most critical to the output of the mechanism model as the key optimization parameters.

3. The iterative optimization method for mechanism model parameters based on data and neural network according to claim 2, characterized in that: In step S11, various parameters in the existing mechanism model of the aircraft are classified, including but not limited to aerodynamic parameters, inertial parameters, control parameters, and guidance parameters; Extract multiple parameters in each category that have the greatest impact on the flight trajectory: Among the aerodynamic parameters, those that have the greatest impact on flight trajectory include lift coefficient, drag coefficient, and side force coefficient; Among the inertial parameters, the ones that have the greatest impact on the flight trajectory include the total mass and moment of inertia of the aircraft; Among the control parameters, the ones that have the greatest impact on the flight trajectory include rudder deflection and thrust; Among the guidance parameters, the parameters that have the greatest impact on the flight trajectory include the aircraft's guidance method and guidance rate; among them, the guidance method is the method of controlling the aircraft to fly along a predetermined trajectory.

4. The iterative optimization method for mechanism model parameters based on data and neural network according to claim 2, characterized in that: In step S12, the method for identifying multiple parameters that are most critical to the output of the mechanism model is: First, the baseline values ​​of all input parameters in the mechanism model are determined, and the objective function values ​​when all parameters are at the baseline values ​​are calculated and recorded; the OAT method is used to change only one of the input parameters at a time, and the change in the objective function after the parameter change is calculated; The sensitivity of each parameter is evaluated by the change in the objective function after each parameter is changed. All parameters are sorted according to the sensitivity index to identify multiple parameters that have a great impact on the objective function.

5. The iterative optimization method for mechanism model parameters based on data and neural network according to claim 1, characterized in that: In step S2, the historical data of the aircraft flight trajectory state is collected, including parameters under different flight conditions and at different time points, including position, speed, acceleration, heading angle and attitude angle; the above historical data is preprocessed, and the preprocessing operation is: first, outlier detection and filtering are performed, and the abnormal data is cleaned and corrected; then the state quantity difference and normalization processing are performed.

6. The iterative optimization method for mechanism model parameters based on data and neural network according to claim 1, characterized in that: In step S3, the loss function performs a subtraction process on the estimated state value at the next moment obtained by calculating the aircraft mechanism model and the actual state value at the next moment in the actual trajectory to obtain the loss function value, which is specifically: The current position coordinates of the aircraft are [p x(k) ,p y(k) ,p z(k) ], the position coordinates at the next moment estimated by the mechanism model are [p x(k+1) ,p y(k+1) ,p z(k+1) ], the actual coordinates at the next moment are [p x(k+1) ,p y(k+1) ,p z(k+1) ], and the difference is obtained to get the deviation between the estimated position and the actual position: [p x(k+1) -p x(k+1) ,p y(k+1) -p y(k+1) ,p z(k+1) -p z(k+1) ] Calculate the absolute distance as the loss function value: The loss function is dimensionless, and the updated loss function is: In the formula, It is a dimensionless factor, which makes all items of the loss function dimensionless and takes 120% to 150% of the maximum value of the distance between two adjacent moments.

7. The iterative optimization method for mechanism model parameters based on data and neural network according to claim 1, characterized in that: In step S3, a parameter iteration optimization model is constructed through a BP neural network.

8. The iterative optimization method for mechanism model parameters based on data and neural network according to claim 1, characterized in that: The Adam adaptive learning rate neural network optimization method is adopted to adjust the learning rate of each parameter by tracking the first-order moment and second-order moment of the gradient, so that the learning rate can be adaptively adjusted in each training step, and the iterative optimization of parameters driven by the hybrid mechanism and data can be realized, thereby improving the training efficiency and convergence speed.

9. The method for iterative optimization of mechanism model parameters based on data and neural network according to claim 1, characterized in that: Verify the accuracy of the optimized aircraft mechanism model: The preprocessed historical data is divided into training data and test data; the training data is used to train the parameter iterative optimization model; Simulation experiments are carried out based on the optimized aircraft mechanism model, and the simulation results are compared with the test data to evaluate the accuracy of the aircraft mechanism model and obtain the model deviation, which is used to correct the aircraft mechanism model.

10. The method for iterative optimization of mechanism model parameters based on data and neural network according to claim 1, characterized in that: In step S4, the parameter iteration optimization model is trained using the preprocessed historical data so that the simulation results of the mechanism model gradually approach the actual data. After the iterative optimization is completed, the parameter iteration optimization model outputs the key optimization parameters of the aircraft corresponding to the current input; If the parameter is a constant, the output of the parameter is the constant value; If the parameter is a time-varying quantity, the parameter will change with different inputs, and the parameter iterative optimization model parameters corresponding to different moments will be obtained.

Images