Dynamic linearization dynamic modeling method and device for tank-liquid coupling of tank truck

Abstract

The invention relates to the technical field of dynamic modeling, in particular to a dynamic linearization dynamic modeling method and device for tank-liquid coupling of a tank truck, and the method comprises the steps: selecting a concerned degree of freedom from six spatial degrees of freedom of a tank body of the tank truck, and compiling an orthogonal working condition table; establishing a fine computational fluid dynamics model of the tank body, and inputting the orthogonal working condition table into the fine computational fluid dynamics model to establish a computational fluid dynamics (CFD) simulation database; establishing a nonlinear high-precision mechanical equivalent shaking model; selecting the most concerned degree of freedom to establish a linearized equivalent pendulum model; and calibrating a nonlinear high-precision mechanical equivalent shaking model and a linearized equivalent pendulum model by using a CFD simulation database to train a neural network so as to output the output characteristics in a prediction time domain step under various concerned degrees of freedom and various working conditions. Therefore, the problems that in the prior art, shaking of fluid in multiple modes cannot be expressed, and precision in long-time-domain prediction cannot be effectively guaranteed are solved.

Description

Technical Field

[0001] The present application relates to the technical field of dynamic modeling, and in particular to a method and device for dynamic linearization dynamic modeling of tank-liquid coupling of a liquid tank truck. Background Art

[0002] Liquid tanker trucks are common vehicles for transporting hazardous chemicals. Standards require them to be partially loaded, which makes the liquid prone to sloshing. Coupled with their high center of mass and heavy load, they have poor rollover stability after vehicle-liquid coupling, making them prone to rollover accidents that could lead to hazardous chemical leaks or even explosions. Therefore, in future intelligent automotive cyber-physical systems, manual tanker driving, which is purely for transportation purposes and highly dangerous and labor-intensive, will be replaced by autonomous driving. For autonomous tanker trucks, establishing a relatively accurate simplified model of the vehicle and liquid sloshing mechanism, designing an anti-rollover control algorithm based on the vehicle-liquid model that accounts for liquid sloshing, and achieving the measurement or estimation of the state quantities required for control are essential steps for achieving reliable autonomous driving of tanker trucks.

[0003] In related technologies, the sloshing of liquid in a tank truck is generally approximated as an equivalent simple pendulum model or an equivalent elliptical pendulum model for modeling. The simple pendulum or equivalent elliptical pendulum model usually assumes that the liquid sloshing is a small-amplitude oscillation that conforms to linear simple harmonic motion, and the liquid is equivalent to a point mass or a rigid body. The fixed length characteristic of the simple pendulum is used to constrain the motion trajectory of the fluid to a circular arc, thereby controlling the measurement or estimation of the required state quantity. Of course, there are also a few studies that use potential flow theory to establish fluid models.

[0004] However, the related technologies approximate the simple pendulum model or the elliptical pendulum model, whose own dynamic characteristics and geometric constraints limit the degrees of freedom of liquid sloshing, and cannot fully describe the complex fluid behavior in the real three-dimensional space. They can only simulate the two-dimensional sloshing of the liquid in the roll plane. Moreover, the simplified model established for real-time control is usually a linear model, which lacks the ability to model long-range dependencies. As a result, the sloshing of the fluid in multimodal conditions cannot be expressed and the accuracy in long-term predictions cannot be effectively guaranteed. These problems need to be improved urgently. Summary of the Invention

[0005] The present application provides a dynamic linearization dynamic modeling method and device for tank-liquid coupling of a liquid tank truck to solve the problems in related technologies such as the inability to express the sloshing of fluids in multimodal conditions and the inability to effectively guarantee the accuracy of long-term predictions.

[0006] The first aspect of the present application provides a dynamic linearized dynamic modeling method for tank-liquid coupling of a liquid tank truck, comprising the following steps: selecting at least one degree of freedom of interest from the six spatial degrees of freedom of the tank body of the liquid tank truck to compile an orthogonal operating condition table; establishing a fine computational fluid dynamics model of the tank body, and inputting the orthogonal operating condition table into the fine computational fluid dynamics model to establish a computational fluid dynamics CFD (Computational Fluid Dynamics) simulation database; establishing a nonlinear high-precision mechanical equivalent sway model; selecting the most concerned degree of freedom to establish a linearized equivalent pendulum model; and calibrating the nonlinear high-precision mechanical equivalent sway model and the linearized equivalent pendulum model using the CFD simulation database to train a neural network to output output characteristics within a predicted time domain step under each degree of freedom and each operating condition of interest.

[0007] Through the above technical solution, the embodiment of the present application can design different excitation levels of the degrees of freedom of interest by selecting the degrees of freedom to compile an orthogonal operating condition table, perform fluid dynamics simulation under orthogonal combination operating conditions, and obtain a CFD data set; to comprehensively consider the six degrees of freedom and transverse and longitudinal coupled flows of the tank-liquid coupling system; and based on the above CFD data set, calibrate the nonlinear high-precision mechanical equivalent sloshing model, and then combine the neural network method to relinearize the nonlinear model in real time to ensure that the linearized model still maintains high precision, which can not only ensure the expression of the fluid sloshing under multi-modal conditions, but also improve the accuracy of long-time domain predictions, thereby providing a solid theoretical basis for dynamic high-precision prediction and observation of the liquid tank truck dynamics system.

[0008] Optionally, in one embodiment of the present application, the refined computational fluid dynamics model is expressed as:

[0009]

[0010] in, It is represented as the time series of the generalized force output by the model; f0(·) represents the refined computational fluid dynamics model; and u(t) represents the time series of the model input excitation vector.

[0011] Through the above technical solution, the embodiment of the present application can establish a CFD simulation database by simulating a fine computational fluid dynamics model. The established CFD data set is used to calibrate the sliding particle model parameters, so that the sliding particle model fully considers the six degrees of freedom of the tank-liquid coupling system and the transverse and longitudinal coupled flows to improve the prediction accuracy.

[0012] Optionally, in one embodiment of the present application, the neural network is represented as:

[0013]

[0014] Where Θ represents the model state vector; w2 is the model parameter vector; Δw2 represents the change in parameter w2; represents a neural network; x1 represents the model state vector.

[0015] Through the above technical solution, the embodiment of the present application can improve the model accuracy by using a neural network reinforcement learning method to fit the optimal parameter changes and state quantity estimates of the dynamic linearization kinetic model of the sliding particle model under different states.

[0016] Optionally, in one embodiment of the present application, the training of the neural network includes: using any algorithm of preset reinforcement learning, wherein the state For x1, action is [Θ, Δw2], reward The orthogonal operating condition table The operating condition evaluation in the given prediction time domain N p Each moment within the step and The sum of the two norms of is negated as a reward.

[0017] Through the above technical solution, the embodiment of the present application can use the sliding particle model combined with the neural network method to relinearize the precise nonlinear model in real time, ensuring that the linearized model still maintains high precision, providing a solid theoretical basis for the dynamic high-precision prediction and observation of the liquid tank truck dynamic system, and reducing the complexity of the calculation.

[0018] Optionally, in one embodiment of the present application, the reward calculation formula is:

[0019]

[0020] in, Indicates reward; Indicates the degree of freedom of concern; N p Represents the prediction time domain time step; N T Indicates the number of working conditions; k indicates the time; represents generalized force, subscript 1 represents STL model, subscript 2 represents Dual-LSP model; w j represents the model parameter vector.

[0021] Through the above technical solution, the embodiment of the present application can make the future N p The cumulative output error of the next N steps will be p The error (such as mean square error, cross entropy) between the predicted output of each step and the true value is accumulated and summed, forcing the model to focus on the coherence of multiple steps in the future, avoiding short-sighted behavior, and improving training stability.

[0022] The second aspect of the present application provides a dynamic linearized dynamic modeling device for tank-liquid coupling of a liquid tank truck, including: a compilation module for selecting at least one degree of freedom of interest from the six spatial degrees of freedom of the tank body of the liquid tank truck to compile an orthogonal operating condition table; a first construction module for establishing a fine computational fluid dynamics model of the tank body, and inputting the orthogonal operating condition table into the fine computational fluid dynamics model to establish a computational fluid dynamics CFD simulation database; a second construction module for establishing a nonlinear high-precision mechanical equivalent sway model; a third construction module for selecting the most concerned degree of freedom to establish a linearized equivalent pendulum model; a modeling module for calibrating the nonlinear high-precision mechanical equivalent sway model and the linearized equivalent pendulum model using the CFD simulation database to train a neural network to output output characteristics within a predicted time domain step under each degree of freedom and each operating condition of interest.

[0023] Through the above technical solution, the embodiment of the present application can design different excitation levels of the degrees of freedom of interest by selecting the degrees of freedom to compile an orthogonal operating condition table, perform fluid dynamics simulation under orthogonal combination operating conditions, and obtain a CFD data set; to comprehensively consider the six degrees of freedom and transverse and longitudinal coupled flows of the tank-liquid coupling system; and based on the above CFD data set, calibrate the nonlinear high-precision mechanical equivalent sway model, and then combine the neural network method to re-linearize the nonlinear model in real time to ensure that the linearized model still maintains high precision, thereby providing a solid theoretical basis for the dynamic high-precision prediction and observation of the liquid tank truck dynamics system.

[0024] Optionally, in one embodiment of the present application, the refined computational fluid dynamics model is expressed as:

[0025]

[0026] in, It is represented as the time series of the generalized force output by the model; f0(·) represents the refined computational fluid dynamics model; and u(t) represents the time series of the model input excitation vector.

[0027] Through the above technical solution, the embodiment of the present application can establish a CFD simulation database by simulating a fine computational fluid dynamics model. The established CFD data set is used to calibrate the sliding particle model parameters, so that the sliding particle model fully considers the six degrees of freedom of the tank-liquid coupling system and the transverse and longitudinal coupled flows to improve the prediction accuracy.

[0028] Optionally, in one embodiment of the present application, the neural network is represented as:

[0029]

[0030] Where Θ represents the model state vector; w2 is the model parameter vector; Δw2 represents the change in parameter w2; represents a neural network; x1 represents the model state vector.

[0031] Through the above technical solution, the embodiment of the present application can improve the model accuracy by using a neural network reinforcement learning method to fit the optimal parameter changes and state quantity estimates of the dynamic linearization kinetic model of the sliding particle model under different states.

[0032] Optionally, in one embodiment of the present application, the modeling module is further configured to adopt any algorithm of preset reinforcement learning, wherein the state For x1, action is [Θ, Δw2], reward The orthogonal operating condition table The operating condition evaluation in the given prediction time domain N p Each moment within the step and The sum of the two norms of is negated as a reward.

[0033] Through the above technical solution, the embodiment of the present application can use the sliding particle model combined with the neural network method to relinearize the precise nonlinear model in real time, ensuring that the linearized model still maintains high precision, providing a solid theoretical basis for the dynamic high-precision prediction and observation of the liquid tank truck dynamic system, and reducing the complexity of the calculation.

[0034] Optionally, in one embodiment of the present application, the reward calculation formula is:

[0035]

[0036] in, Indicates reward; Indicates the degree of freedom of concern; N p Represents the prediction time domain time step; N T Indicates the number of working conditions; k indicates the time; represents generalized force, subscript 1 represents STL model, subscript 2 represents Dual-LSP model; w j represents the model parameter vector.

[0037] Through the above technical solution, the embodiment of the present application can make the future N p The cumulative output error of the next N steps will be p The error (such as mean square error, cross entropy) between the predicted output of each step and the true value is accumulated and summed, forcing the model to focus on the coherence of multiple steps in the future, avoiding short-sighted behavior, and improving training stability.

[0038] The third aspect of the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the dynamic linearization dynamic modeling method of the tank-liquid coupling of a tank truck as described in the above embodiment.

[0039] The fourth aspect of the present application provides a computer-readable storage medium, which stores a computer program. When the program is executed by a processor, it implements the above-mentioned dynamic linearization dynamic modeling method of tank-liquid coupling of a tank truck.

[0040] The fifth aspect of the present application provides a computer program product, which stores a computer program that, when executed by a processor, implements the above-mentioned dynamic linearized dynamic modeling method for tank-liquid coupling of a tank truck.

[0041] The embodiment of the present application can establish a detailed computational fluid dynamics model of the tank and liquid in fluid mechanics simulation software, adopt an orthogonal experimental design method to design different excitation levels of the degrees of freedom of interest, perform fluid dynamics simulation under orthogonal combination conditions, and obtain a CFD data set; based on the above CFD data set, a heuristic optimization algorithm is used to calibrate the parameters of the sliding particle model proposed in the present invention, thereby establishing a sliding particle model; finally, a combination of linearized pendulum models is selected according to the degrees of freedom of interest, and a reinforcement learning method is used to fit the optimal parameter changes and state quantity estimates of the dynamic linearized dynamic model of the sliding particle model under different states, thereby establishing a dynamic linearized dynamic model. Compared with the existing equivalent pendulum modeling method, the linearization fitting accuracy of the tank-liquid coupling dynamic system can be greatly improved, and the six degrees of freedom and the transverse and longitudinal coupled flows of the tank-liquid coupling system are fully considered; by combining the sliding particle model with a neural network method, the accurate nonlinear model can be relinearized in real time to ensure that the linearized model still maintains high accuracy, providing a solid theoretical basis for the dynamic high-precision prediction and observation of the liquid tank truck dynamic system. This solves the problems in related technologies such as the inability to express the sloshing of fluids in multimodal conditions and the inability to effectively guarantee the accuracy of long-term predictions.

[0042] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0044] Figure 1This is a flow chart of a dynamic linearization dynamic modeling method for tank-liquid coupling of a tank truck provided according to an embodiment of the present application;

[0045] Figure 2 A schematic diagram of establishing a nonlinear high-precision mechanical equivalent sway model according to a specific embodiment of the present application;

[0046] Figure 3 A schematic diagram of modeling a sliding particle on a two-dimensional curve or three-dimensional surface according to a specific embodiment of the present application;

[0047] Figure 4 A schematic diagram of coordinate definition according to a specific embodiment of the present application;

[0048] Figure 5 is a schematic diagram of a simple pendulum model with a roll input according to a specific embodiment of the present application;

[0049] Figure 6 Schematic diagram of real-time relinearization of a sliding particle model according to a specific embodiment of the present application;

[0050] Figure 7 A schematic diagram of training a soil-filling neural network using a reinforcement learning method according to a specific embodiment of the present application;

[0051] Figure 8 Schematic diagram showing a comparison of the outputs of a sliding mass model and a dynamic linearization model under a working condition according to a specific embodiment of the present application;

[0052] Figure 9 Schematic diagram of a dynamic linearization dynamic modeling device for tank-liquid coupling of a tank truck according to an embodiment of the present application;

[0053] Figure 10 The figure is a schematic diagram of the structure of an electronic device provided according to an embodiment of the present application. DETAILED DESCRIPTION

[0054] The following describes in detail embodiments of the present application, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0055] The following describes the dynamic linearized dynamic modeling method and device of the tank-liquid coupling of the liquid tank truck in the embodiment of the present application with reference to the accompanying drawings. In response to the problems in the related technologies mentioned in the background technology center above that the sloshing of the fluid in multimodal conditions cannot be expressed and the accuracy in long-term prediction cannot be effectively guaranteed, the present application provides a dynamic linearized dynamic modeling method of the tank-liquid coupling of the liquid tank truck. In this method, a detailed computational fluid dynamics model of the tank body and the liquid can be established in the fluid mechanics simulation software, and an orthogonal experimental design method is used to design different excitation levels of the degrees of freedom of interest. Fluid dynamics simulation is performed under orthogonal combination working conditions to obtain a CFD data set; based on the above CFD data set, a heuristic optimization algorithm is used to calibrate the parameters of the sliding particle model proposed in the present invention, thereby establishing a sliding particle model; finally, a combination form of the linearized pendulum model is selected according to the degrees of freedom of interest, and a reinforcement learning method is used to fit the optimal parameter changes and state quantity estimates of the dynamic linearized dynamic model of the sliding particle model in different states, thereby establishing a dynamic linearized dynamic model. Compared to existing equivalent pendulum modeling methods, this method significantly improves the linearization accuracy of the tank-fluid coupled dynamics system, fully accounting for the system's six degrees of freedom and the coupled transverse and longitudinal flows. By combining a sliding particle model with a neural network, it can relinearize precise nonlinear models in real time, ensuring that the linearized model maintains high accuracy. This provides a solid theoretical foundation for high-precision dynamic prediction and observation of tanker truck dynamics systems. This addresses existing issues in related technologies, such as the inability to express multimodal fluid sloshing and the inability to effectively guarantee accuracy in long-term predictions.

[0056] Specifically, Figure 1 A flow chart of a dynamic linearization kinetic modeling method for tank-liquid coupling of a liquid tank truck provided in an embodiment of the present application.

[0057] like Figure 1 As shown in FIG, the dynamic linearization dynamic modeling method of the tank-liquid coupling of the tank truck includes the following steps:

[0058] In step S101 , at least one degree of freedom of interest is selected from the six spatial degrees of freedom of the tank body of the liquid tank truck to compile an orthogonal operating condition table.

[0059] Spatial degrees of freedom can be understood as the number of independent parameters that allow an object to independently move or position itself in three-dimensional space. Each degree of freedom represents an independent direction of motion or axis of rotation. For example, in three-dimensional space, a rigid body has a maximum of six degrees of freedom, including three translational degrees of freedom: linear motion along the x, y, and z axes; and three rotational degrees of freedom: rotation about the x, y, and z axes (roll, pitch, and yaw).

[0060] An orthogonal load table can be understood as a tool for experimental design or system analysis. It leverages the principle of orthogonality to efficiently arrange multi-factor, multi-level experimental combinations. Its core purpose is to minimize the number of experiments, covering all factor interactions and analyzing the impact of each factor on the results.

[0061] In the actual implementation process, the embodiment of the present application can select several degrees of freedom of interest from the six spatial degrees of freedom of the tank. where d i and θ i Respectively represent the translational and rotational degrees of freedom in the positive direction of the corresponding coordinate axis, as well as the range of the corresponding degree of freedom excitation, and compile the orthogonal working condition table where N T is the number of required orthogonal working conditions, m i ,n i Respectively represent the number of levels and factors in group i,

[0062] Specifically, this embodiment can select the degree of freedom of interest as That is, translational excitation along the x and y axes and rotational excitation around the x and z axes, because the lateral and longitudinal accelerations as well as roll and yaw excitations are the most common excitations during the driving of the tanker. Then, for the degree excitations in these four directions, an orthogonal working condition table containing n=16 groups of working conditions is formulated. The four excitation sources of the tank that often appear during driving can be considered: longitudinal acceleration a x , lateral acceleration a y , roll angular acceleration Yaw angular acceleration And each is discrete into 4 gears: a x :0,1,3,5m / s 2 , a y :0,1,3,5m / s 2 , 0,0.25,0.5,1rad / s 2 , 0,0.25,0.5,1rad / s 2 , used to design L 16 (4 4 ) Standard orthogonal operating condition table, the orthogonal operating condition table is shown in Table 1:

[0063] Table 1

[0064]

[0065]

[0066] Among them, the four excitation physical quantity units in the above orthogonal working condition table are ax 、a y :m / s 2 ; rad / s 2 .

[0067] Through the above technical solution, the embodiment of the present application can select the degrees of freedom of interest and the corresponding excitation range from all degrees of freedom, convert the liquid sloshing space from the traditional two-dimensional plane to three-dimensional space with multiple degrees of freedom, improve the input of all degrees of freedom of the tank-liquid coupling system, consider the horizontal and vertical coupled flow and swing around the Z axis, so as to realize the motion expression of the liquid in more modes.

[0068] In step S102, a fine computational fluid dynamics model of the tank is established, and the orthogonal operating condition table is input into the fine computational fluid dynamics model to establish a computational fluid dynamics (CFD) simulation database.

[0069] Among them, the computational fluid dynamics simulation database can be understood as a systematic platform for storing, managing and sharing CFD simulation results and related data, which is used to support scientific research, engineering design and decision optimization.

[0070] During the actual implementation process, the embodiment of the present application can establish a computational fluid dynamics model using excitation as the input time series vector, and select the StarCCM+ platform as the computational fluid dynamics modeling platform, wherein the modeling platform and the parameter selection of the computational fluid dynamics model can be adaptively set according to the actual application scenario.

[0071] The embodiment of the present application can perform fluid dynamics simulation under orthogonal combination conditions by establishing a computational fluid dynamics model. The orthogonal table avoids the "dimensionality disaster" of full-factor experiments by evenly covering factor combinations, ensuring that each factor / level is evenly distributed and avoiding local deviations; thereby obtaining a more accurate CFD simulation database, thereby significantly improving the efficiency, accuracy and engineering applicability of fluid dynamics research.

[0072] In step S103 , a nonlinear high-precision mechanical equivalent slosh model is established.

[0073] The nonlinear, high-precision mechanical equivalent sloshing model can be understood as a mathematical model used to simulate complex sloshing behavior. It is commonly applied to the dynamic analysis of systems such as liquid storage tanks (such as aerospace fuel tanks and chemical storage tanks), ship tanks, and vehicle fuel tanks. Its core concept is to equate the nonlinear sloshing effects of liquids to a mechanical system (such as a mass-spring-damper system or a multi-degree-of-freedom oscillator), thereby simplifying calculations and maintaining high accuracy. Those skilled in the art should understand that this model is not fixed and has different definitions and modeling methods depending on different scenarios.

[0074] The embodiments of this application can be used represents a nonlinear high-precision mechanical equivalent sloshing model, where x1 is the model state vector, u is the model input excitation vector, and w1 is the model parameter vector. is the time derivative of the model state vector, Output the generalized forces for the model.

[0075] In some specific embodiments, the nonlinear high-precision mechanical equivalent sway model can be a STL (Sliding Telescopic Libra) model based on sliding particles, which includes two particles sliding on the surface (two "bathtub" models), one particle sliding on the curve ("wire" model) and two "bathtubs" separated from each other. The sliding telescopic balance can use particles sliding on the surface ("bathtub" model) to simulate the transverse and longitudinal coupled flow in the chamber inside the tank, and on this basis use a double bathtub model that can slide on the curve to achieve dynamic fitting of all 6 degrees of freedom of the tank in space. The specific definition of the STL model can be found in Figure 2 .

[0076] The STL model has 12 state variables and 6 degrees of freedom, which are the two degrees of freedom of movement of the balls inside the two bathtubs, x1, y1 and x2, y2, the common sliding freedom of the two bathtubs, x0, and the sliding freedom d relative to each other, so there are 11 state variables in total. The definition of the STL model state variables is shown in Table 2. All state variables are collectively referred to as X STL , the model is a 6-degree-of-freedom model.

[0077] Table 2

[0078]

[0079] Among them, the parameters of the STL model include 18, collectively referred to as w1: as shown in Table 3, Table 3 shows the STL parameters and their recognition results when the filling rate is 50%.

[0080] Table 3

[0081]

[0082]

[0083] The input of the STL model contains 6 dimensions, as shown in Table 4. The input variables of the STL model can be defined as:

[0084] Table 4

[0085]

[0086] The STL model can be used for joint simulation with the vehicle dynamics model.

[0087] The mathematical expression of the STL model cannot be fully analyzed, but its calculation process is expressed numerically. This embodiment of the application first defines a three-dimensional surface F as the bathtub, z = F(x, y), where F is required to be continuously differentiable. At the same time, a two-dimensional curve C can be defined as the wire, z = C(x). At the kth moment, this embodiment of the application can describe the simultaneous sliding of two bathtubs along the curve in space as follows:

[0088] Define the tangent vector of the curve at the kth moment (default direction) and the normal vector

[0089]

[0090] Where C1 represents the curve at the kth moment; x represents the state; represents the tangent vector of the curve at the kth moment; Represents the normal vector of the curve at the kth moment.

[0091] The acceleration acting on the common center of mass of the two bathtubs at each moment is the actual x-direction acceleration a x The result of nonlinear mapping of the vertical gravitational acceleration -g.

[0092]

[0093] Among them, a′ k represents the acceleration acting on the common center of mass of the two bathtubs after nonlinear mapping; -g represents the vertical acceleration of gravity.

[0094] Assuming that the centers of mass of the two bathtubs are not constrained by the curve, but can move freely on the two-dimensional plane, the positions of the centers of mass of the two bathtubs during free motion can be simply calculated using the kinematic formula. And the increment of the actual position at this moment relative to the previous moment The specific calculation process diagram is as follows Figure 3 shown.

[0095]

[0096] in, represents the position of the center of mass of the two basins during free motion; represents the sliding center of mass position vector; represents the velocity vector of the particle; It represents the increment of the actual position at the current moment relative to the previous moment; Δt represents the time difference between the current moment and the previous moment.

[0097]

[0098] Among them, θ k represents the characteristic angle at time k; represents the normal vector at time k; Δθ k represents the characteristic angle increment at time k; represents the tangent vector at time k.

[0099]

[0100] z 0,k+1 =C(x 0,k+1 )

[0101]

[0102]

[0103] Among them, c x Indicates the damping coefficient of the bathtub's X-axis movement.

[0104] Then update the separation of each bathtub:

[0105]

[0106] Among them, d k+1 represents the translational degree of freedom in the positive direction of the corresponding coordinate axis at time k+1; s dev Indicates the static offset distance of the bathtub surface; represents the yaw angular velocity of the tank at the kth moment.

[0107] Then update the motion of the particles in each bathtub:

[0108] z 1,k =F(x +,k ,y +,k )

[0109]

[0110] Sliding center of mass position vector In the first component of the two bathtub models, d k For + and -:

[0111]

[0112] Convert the input of the rotation part to:

[0113]

[0114] Among them, a k represents the acceleration vector at the kth moment; a x,k represents the x component of the acceleration at the kth moment of external input; represents the yaw angular acceleration of the tank at the kth moment; represents the unit vector in the y direction; φ k represents the roll angle of the tank at the kth moment; Represents the x-direction unit vector; represents the unit vector in the z direction; a u,k represents the y component of the acceleration at the kth moment of external input; a z,k Represents the z component of the acceleration of the external input at the kth moment.

[0115]

[0116] in, Represents the intermediate operation result vector 1; Represents the intermediate operation result vector 2.

[0117]

[0118] For the definition of the bathtub surface z = F(x, y) and the overall sliding curve z = C(x), refer to the following example:

[0119]

[0120] in Re(·) indicates taking the real part.

[0121]

[0122] Among them, b p represents the minor semi-axis of the sliding plane curve; a p represents the semi-major axis of the sliding plane curve.

[0123] In practice, it is sufficient to ensure that the two functions are continuously differentiable with respect to their inputs; they do not need to follow the given form. However, it is important to note that when the expression structure of the sliding surface and / or curve is adjusted, the parameter definitions of the STL model also need to be modified accordingly. This can be used to simplify or further complicate the STL model.

[0124] The embodiment of the present application can use a sliding particle model to simulate the transverse and longitudinal coupled flow in the tank chamber, and on this basis use a double bathtub model that can slide on a curve to achieve dynamic fitting of all six degrees of freedom of the tank body in space; the sliding particle model used greatly improves the fitting accuracy of the liquid tank dynamic system compared to the existing equivalent pendulum model.

[0125] In step S104 , the most important degree of freedom is selected to establish a linearized equivalent pendulum model.

[0126] The linearized equivalent pendulum model can be understood as an approximation that simplifies the nonlinear sloshing of liquids into a linear pendulum motion. It is suitable for small-amplitude sloshing or linear dynamics analysis. Its core concept is to describe the low-order modal characteristics of liquid sloshing using equivalent mechanical parameters (mass, pendulum length, and damping), thus avoiding complex fluid numerical simulations.

[0127] The linearized equivalent pendulum model can be a linear combination of at least N simple pendulums (N ≥ 1). The double pendulum model used is N = 2, and a longitudinal linearized pendulum can also be added to form a combination, such as a triple pendulum model (double pendulum model + a longitudinal linearized pendulum) and other options.

[0128] The embodiment of the present application can select the degree of freedom that is of most concern And establish a linearized equivalent pendulum model Where Θ is the model state vector, u is the model input excitation vector, and w2 is the model parameter vector. is the time derivative of the model state vector, Output the generalized forces for the model.

[0129] Specifically, the embodiment of the present application can establish a linearized double pendulum model (Dual-linearized-pendulum, DLP), such as Figure 4 As shown, the double pendulum model consists of two single pendulum models with both lateral acceleration and roll angle acceleration input, distributed along the vehicle axis; Figure 4 The coordinate system involved is defined as follows Figure 5 As shown;

[0130] Let's first look at the case of a pendulum, with concentrated masses m0 and m p The position vector is as follows:

[0131]

[0132] in, represents the position vector of the concentrated mass m0; Represents the y-direction position vector; represents the height vector of the fixed mass; x0 represents the overall longitudinal displacement of the bathtub model along the curve; x 1|2 Indicates the front /

[0133] The x position of the rear ball; d represents the relative offset distance between the two bathtub models.

[0134]

[0135] in, represents the concentrated mass m p The position vector of represents the position vector of the fixed mass; represents the length vector of the pendulum support; x0 represents the overall longitudinal displacement of the bathtub model along the curve; d represents the relative offset distance between the two bathtub models; l p represents the pendulum length; θ represents the rotational degree of freedom.

[0136] Next, differentiate it to obtain the velocity vectors of the two lumped masses:

[0137]

[0138] in, represents the velocity vector of the concentrated mass m0; represents the roll angular velocity of the tank; φ represents the roll angle of the tank.

[0139]

[0140] in, represents the concentrated mass m p The velocity vector.

[0141] Differentiating again gives the acceleration vector:

[0142]

[0143] in, represents the acceleration vector of the concentrated mass m0; Indicates mass m p The acceleration vector.

[0144] When choosing Plane (see Figure 4 ) is the zero potential energy surface, then the kinetic energy T and potential energy U can be expressed as:

[0145]

[0146] Here, g represents the acceleration due to gravity.

[0147] According to the Euler-Lagrange equation:

[0148]

[0149] The dynamic differential equation of the swinging degree of freedom θ is derived as follows:

[0150]

[0151] The lateral force F acting on the tank y and moment M x for:

[0152]

[0153] Substituting the above position and acceleration vectors into the specific expansion and retaining the linear terms is:

[0154]

[0155] Among them, v x2 Indicates the longitudinal speed of the trailer; represents the sideslip angular velocity of the trailer's center of mass; represents the yaw angular velocity of the tank.

[0156]

[0157] For a general simple pendulum, the lateral acceleration can be expressed as When there are two pendulum models, their swinging degrees of freedom can be named θ1 and θ2 respectively. Both are applicable to the above equation. For the two degrees of freedom θ1 and θ2, the only difference in equation (9) is the lateral acceleration input.

[0158]

[0159] In addition, a separation degree of freedom d needs to be added as shown in formula (7). In the embodiment of the present application, this degree of freedom can be considered by adding an input quantity, or discarding this degree of freedom during relinearization and converting its influence into changes to other parameters. Since it contains nonlinear absolute value terms, it cannot be directly linearized for MPC control, but a new input variable u can be introduced k Instead of Constraints need to be added to ensure equivalence:

[0160]

[0161] At the same time, add the optimization function of MPC By adjusting R, we can control the optimal solution u k Approach However, this method doubles the number of input variables, which has a significant negative impact on solution time, but the negative impact is not significant. On average, doubling the number of input variables increases solution time by about 20% to 30%. However, reducing the prediction horizon by half can reduce solution time by over 50%.

[0162] The lateral acceleration of the front and rear bathtubs is converted into an additional yaw rate term:

[0163]

[0164] in, represents the lateral acceleration; x0 represents the overall longitudinal displacement of the bathtub model along the curve; x1 represents the x-displacement of the ball in bathtub 1; d represents the relative offset distance between the two bathtub models.

[0165] Let x1 = x0 + d and x2 = x0 - d

[0166] Substituting this into the expressions for lateral force and rolling moment:

[0167]

[0168] Where φ represents the roll angle of the tank; θ1 represents the degree of freedom 1; and θ2 represents the degree of freedom 2.

[0169]

[0170] Among them, M x Indicates the moment about the x-axis; M z Indicates the moment about the Z axis.

[0171] The DLP model contains two degrees of freedom, θ1 and θ2, and seven parameters, which are:

[0172] h0h p l p m0x0dc d

[0173] The model contains three outputs, namely F y M y M z , the input is φ.

[0174] The embodiment of the present application can achieve high-precision linear model establishment by introducing degrees of freedom and constraints. Compared with the existing equivalent pendulum modeling method, the dynamic linearization dynamic modeling method can greatly improve the linear fitting accuracy of the tank-liquid coupling dynamic system, and comprehensively consider the six degrees of freedom and transverse and longitudinal coupled flows of the tank-liquid coupling system; through the sliding particle model combined with the neural network method, we can re-linearize the precise nonlinear model in real time to ensure that the linearized model still maintains high precision, providing a solid theoretical basis for the dynamic high-precision prediction and observation of the liquid tank truck dynamic system.

[0175] In step S105, a nonlinear high-precision mechanical equivalent sway model and a linearized equivalent pendulum model are calibrated using a CFD simulation database to train a neural network to output output characteristics within a predicted time domain step under each degree of freedom and each working condition of interest.

[0176] Specifically, the embodiment of the present application can use a heuristic algorithm to calibrate the parameter w1 of the nonlinear high-precision mechanical equivalent sway model f1 and the initial value w of the parameter of the linearized equivalent pendulum model f2. 20 ; Among them, heuristic algorithms include but are not limited to genetic algorithms, particle swarm optimization, simulated annealing, etc.; For example, in the embodiment of the present application, the parameters of the model can be calibrated by a genetic algorithm, and the cost function is the second norm of the distance between the CFD simulation output force / torque sequence and the model output force / torque sequence under 16 weighted working conditions to identify the sliding telescopic balance model parameters of a tank with a 50% filling rate:

[0177]

[0178] The parameter identification results of the sliding telescopic balance when the filling rate is 50% are shown in Table 5:

[0179] Table 5

[0180]

[0181]

[0182] The embodiment of the present application can use the CFD simulation database to calibrate the initial parameter values of the nonlinear high-precision mechanical equivalent sway model and the linearized equivalent pendulum model, fully integrating the CFD simulation data with the model, which can significantly improve the model accuracy, convergence efficiency and engineering applicability.

[0183] Optionally, in one embodiment of the present application, the refined computational fluid dynamics model is expressed as:

[0184]

[0185] Among them, u(t) is the time series of the model input excitation vector, The generalized force time series is output for the model; f0(·) represents the refined computational fluid dynamics model.

[0186] It is understood that the computational fluid dynamics model in the embodiment of the present application takes the excitation vector time series as input and the generalized force time series as output;

[0187] In the actual implementation process, the embodiment of the present application can establish a detailed computational fluid dynamics model of the tank and the liquid in the fluid mechanics simulation software, adopt the orthogonal experimental design method, design different excitation levels of the degree of freedom of interest, perform fluid dynamics simulation under the orthogonal combination working condition, and obtain a CFD data set. For example, this embodiment can use the orthogonal working condition table to The content is used as input U(t) to simulate the fine computational fluid dynamics model f0 and establish a CFD simulation database

[0188] The embodiment of the present application can establish a CFD simulation database by simulating the excitation sources of the degrees of freedom in different directions of the tank body with a fine computational fluid dynamics model. The established CFD data set is used to calibrate the sliding particle model parameters. When establishing the sliding particle model, the sliding particle model fully considers the six degrees of freedom of the tank-liquid coupling system and the transverse and longitudinal coupled flows to improve the prediction accuracy.

[0189] Optionally, in one embodiment of the present application, the neural network is represented as:

[0190]

[0191] Where Θ represents the model state vector; w2 is the model parameter vector; Δw2 represents the change in parameter w2; represents a neural network; x1 represents the model state vector.

[0192] It is understandable that neural networks are a type of machine learning model inspired by biological nervous systems and are widely used in tasks such as pattern recognition, prediction, and decision-making. They include but are not limited to graph neural networks, generative adversarial networks, attention mechanisms, etc., and can be adaptively selected based on actual scenarios.

[0193] In the actual implementation process, the embodiment of the present application can obtain Θ and Δw2 by inputting x1 into n, and use them to linearize the equivalent pendulum model f2, that is, The prediction time domain N can be obtained by using the fine computational fluid dynamics model f0 under each degree of freedom and working condition of interest. p In-step output characteristics and A model f2 that is as close as possible.

[0194] The embodiment of the present application can improve the model accuracy by using a neural network reinforcement learning method to fit the optimal parameter changes and state quantity estimates of the dynamic linearization kinetic model of the sliding particle model under different states.

[0195] Optionally, in one embodiment of the present application, training a neural network includes:

[0196] Use any algorithm of preset reinforcement learning, where the state For x1, action is [Θ, Δw2], reward Orthogonal working condition table The working condition evaluation in the given prediction time domain N p Each moment within the step and The sum of the two norms of is negated as a reward.

[0197] Among them, training a neural network can be understood as the process of adjusting model parameters (such as weights and biases) so that it can automatically learn from input data and approximate a specific target function; in the context of machine learning and reinforcement learning, rewards can be understood as feedback signals provided by the environment to the agent, which are used to quantify how well the agent performs an action in a specific state.

[0198] In the actual implementation process, training the neural network The second output in is the change in parameter w2; In the embodiment of the present application, the sliding particle model can be combined with a neural network to relinearize the nonlinear model. The schematic diagram of the relinearization of the model is shown in FIG. Figure 6 As shown in , the parameter change rate and state value of the appropriate DLP model can be obtained according to the state quantity of the STL model. Specifically, Figure 7 As shown, the idea of "model filling" can be used to derive appropriate parameter changes, in order to achieve the purpose of linearizing the STL model more accurately, and embody the dynamics that are difficult for the linear model to reflect in the parameters within a few seconds of interest. The reason why the embodiment of the present application adopts the STL model instead of directly using the CFD calculation results as the true value is two points: first, the speed of CFD calculation is too slow, and the cost of model training is too high; second, it is difficult for CFD to collect a small enough number of state quantities, and it is difficult to describe the current state concisely, so a relatively accurate and quickly calculable model with a finite number (and not too many) of states is needed to characterize the current liquid state. This is the significance of the existence of the STL model. And the STL model can also be used in the joint verification simulation of the control algorithm as a preliminary verification. After verification, the CFD joint simulation verification is carried out again.

[0199] like Figure 8 The figure shows how the DLP model, after neural network relinearization, accurately fits the CFD results within a 3-second prediction time domain, achieving a breakthrough in achieving high precision for linear models under various working conditions.

[0200] The embodiment of the present application can express the STL model in the form of black box input and output:

[0201]

[0202] In the embodiment of the present application, the Dual-LSP model can be expressed in the form of a linear state space equation:

[0203]

[0204] Where X represents the state of the STL model, Θ represents the state quantity of the Dual-LSP model, u represents the input quantity, and the content in the brackets represents the time; represents the generalized force, subscript 1 represents the STL model, and subscript 2 represents the Dual-LSP model; A, B, C, and D represent coefficient matrices; f1 represents the nonlinear high-precision mechanical equivalent shaking model, and f2 represents the linearized equivalent pendulum model.

[0205] The form of the linear state space equation allows it to output Δw2 based on the basic linearization parameter w0 each time the model needs to be relinearized, and finally use w2=w 20 +Δw2 is the actual parameter of the linear model. Currently, there are two methods to train this neural network: supervised learning and reinforcement learning. The supervised learning method requires the optimization results to be prepared in advance. However, the optimization of a state point may take a long time, and it is difficult to generate enough data within a feasible time. data, and depending on the optimizer's results, it is likely that adjacent state points If the parameters are very different, it will easily lead to underfitting in the neural network fitting process. To be as continuous as possible in parameter space, a reinforcement learning scheme is required.

[0206] The embodiment of the present application can use the equivalent force / torque input-output relationship to characterize the influence of the swaying of the liquid in the tank of the liquid tank truck in three-dimensional space on the three directions of the tank body, and then use the sliding particle model combined with the neural network to establish a dynamic linearized dynamic model to fit the above-mentioned equivalent force / torque input-output relationship; through the sliding particle model combined with the neural network method, the precise nonlinear model is relinearized in real time to ensure that the linearized model still maintains high precision, providing a solid theoretical basis for the dynamic high-precision prediction and observation of the liquid tank truck dynamic system, and reducing the complexity of the calculation.

[0207] Optionally, in one embodiment of the present application, the reward calculation formula is:

[0208]

[0209] in, Indicates reward; Indicates the degree of freedom of concern; N p Represents the prediction time domain time step; N T Indicates the number of working conditions; k indicates the time; represents generalized force, subscript 1 represents STL model, subscript 2 represents Dual-LSP model; w j represents the model parameter vector.

[0210] It is understandable that the embodiment of the present application can pThe cumulative output error of each step is used as a reward signal to instantly evaluate the environment's behavior of the agent in reinforcement learning.

[0211] In some embodiments, in a reinforcement learning scheme, the state can be set to be a [x1(k)Θ(k)] vector, the action can be set to be a Δw2 vector, and the reward can be set to be the future N p The input quantity in the simulation process can be a set of typical working condition inputs (assuming there are N T Group working condition), the reward is:

[0212]

[0213] In each group of working conditions, the coefficient w n Each output component is also weighted by a coefficient of w j weighted.

[0214] During training, the embodiment of the present application can sample the possible state space in a way (tentatively uniformly) to perform N p During the training process, for example, the DDPG algorithm can be used to adapt to the continuous action space (Δw2 is a continuous vector).

[0215] The embodiment of the present application can p The cumulative output error of the next N steps will be p The error (such as mean square error, cross entropy) between the predicted output of each step and the true value is accumulated and summed, forcing the model to focus on the coherence of multiple steps in the future, avoiding short-sighted behavior, and improving training stability.

[0216] According to the dynamic linearized dynamic modeling method of the tank-liquid coupling of a liquid tank truck proposed in the embodiment of the present application, a detailed computational fluid dynamics model of the tank body and the liquid can be established in the fluid mechanics simulation software. The orthogonal experimental design method is used to design different excitation levels of the degrees of freedom of interest. The fluid dynamics simulation is performed under the orthogonal combination working condition to obtain a CFD data set. Based on the above CFD data set, the parameters of the sliding particle model proposed in the present invention are calibrated using a heuristic optimization algorithm to establish the sliding particle model. Finally, the combination of linearized pendulum models is selected according to the degrees of freedom of interest. The optimal parameter changes and state quantity estimates of the dynamic linearized dynamic model of the sliding particle model under different states are fitted using a reinforcement learning method to establish a dynamic linearized dynamic model. Compared with the existing equivalent pendulum modeling method, the linearized fitting accuracy of the tank-liquid coupling dynamic system can be greatly improved, and the six degrees of freedom and the lateral and longitudinal coupled flows of the tank-liquid coupling system are fully considered. By combining the sliding particle model with a neural network method, the accurate nonlinear model can be relinearized in real time to ensure that the linearized model still maintains high accuracy, providing a solid theoretical foundation for the dynamic high-precision prediction and observation of the liquid tank truck dynamic system. This solves the problems in related technologies such as the inability to express the sloshing of fluids in multimodal conditions and the inability to effectively guarantee the accuracy of long-term predictions.

[0217] Next, refer to the attached Figure 9 The present invention describes a dynamic linearization kinetic modeling device for tank-liquid coupling of a tank truck proposed in an embodiment of the present application.

[0218] Figure 9 It is a block diagram of a dynamic linearized kinetic modeling device for tank-liquid coupling of a tank truck according to an embodiment of the present application.

[0219] like Figure 9 As shown, the dynamic linearization dynamics modeling device 10 for tank-liquid coupling of a liquid tank truck includes: a compilation module 100 , a first construction module 200 , a second construction module 300 , a third construction module 400 , and a modeling module 500 .

[0220] The compilation module 100 is used to select at least one degree of freedom of interest from the six spatial degrees of freedom of the tank body of the liquid tank truck to compile an orthogonal operating condition table.

[0221] The first building module 200 is used to establish a fine computational fluid dynamics model of the tank body and input the orthogonal operating condition table into the fine computational fluid dynamics model to establish a computational fluid dynamics CFD simulation database.

[0222] The second building module 300 is used to establish a nonlinear high-precision mechanical equivalent sway model.

[0223] The third building block 400 is used to select the most important degree of freedom to establish a linearized equivalent pendulum model.

[0224] The modeling module 500 is used to calibrate the nonlinear high-precision mechanical equivalent sway model and the linearized equivalent pendulum model using the CFD simulation database to train the neural network to output the output characteristics within the predicted time domain step under each degree of freedom and each working condition of interest.

[0225] Optionally, in one embodiment of the present application, the refined computational fluid dynamics model is expressed as:

[0226]

[0227] in, It is represented as the time series of the generalized force output by the model; f0(·) represents the refined computational fluid dynamics model; and u(t) represents the time series of the model input excitation vector.

[0228] Optionally, in one embodiment of the present application, the neural network is represented as:

[0229]

[0230] Where Θ represents the model state vector; w2 is the model parameter vector; Δw2 represents the change in parameter w2; represents a neural network; x1 represents the model state vector.

[0231] Optionally, in one embodiment of the present application, the modeling module 500 is further configured to adopt any algorithm of preset reinforcement learning, wherein the state For x1, action is [Θ, Δw2], reward Orthogonal working condition table The working condition evaluation in the given prediction time domain N p Each moment within the step and The sum of the two norms of is negated as a reward.

[0232] Optionally, in one embodiment of the present application, the reward calculation formula is:

[0233]

[0234] in, Indicates reward; Indicates the degree of freedom of concern; N p Represents the prediction time domain time step; N T Indicates the number of working conditions; k indicates the time; represents generalized force, subscript 1 represents STL model, subscript 2 represents Dual-LSP model; w jrepresents the model parameter vector.

[0235] It should be noted that the above explanation of the embodiment of the dynamic linearization dynamic modeling method for the tank-liquid coupling of the liquid tank truck is also applicable to the dynamic linearization dynamic modeling device for the tank-liquid coupling of the liquid tank truck of this embodiment, and will not be repeated here.

[0236] According to the dynamic linearized dynamic modeling device for the tank-liquid coupling of a liquid tank truck proposed in the embodiment of the present application, a detailed computational fluid dynamics model of the tank and liquid can be established in fluid mechanics simulation software. The orthogonal experimental design method is used to design different excitation levels for the degrees of freedom of interest. Fluid dynamics simulation is performed under orthogonal combination conditions to obtain a CFD data set. Based on the above CFD data set, the parameters of the sliding mass model proposed in the present invention are calibrated using a heuristic optimization algorithm to establish the sliding mass model. Finally, a combination of linearized pendulum models is selected according to the degrees of freedom of interest. The optimal parameter changes and state quantity estimates of the dynamic linearized dynamic model of the sliding mass model under different states are fitted using a reinforcement learning method to establish a dynamic linearized dynamic model. Compared with the existing equivalent pendulum modeling method, the linearized fitting accuracy of the tank-liquid coupling dynamic system can be greatly improved, and the six degrees of freedom and the lateral and longitudinal coupled flows of the tank-liquid coupling system are fully considered. By combining the sliding mass model with a neural network method, the accurate nonlinear model can be relinearized in real time to ensure that the linearized model still maintains high accuracy, providing a solid theoretical foundation for the dynamic high-precision prediction and observation of the liquid tank truck dynamic system. This solves the problems in related technologies such as the inability to express the sloshing of fluids in multimodal conditions and the inability to effectively guarantee the accuracy of long-term predictions.

[0237] Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device may include:

[0238] A memory 1001 , a processor 1002 , and a computer program stored in the memory 1001 and executable on the processor 1002 .

[0239] When the processor 1002 executes the program, the dynamic linearization dynamics modeling method of the tank-liquid coupling of the tank truck provided in the above embodiment is implemented.

[0240] Furthermore, the electronic device further includes:

[0241] The communication interface 1003 is used for communication between the memory 1001 and the processor 1002 .

[0242] The memory 1001 is used to store computer programs that can be run on the processor 1002 .

[0243] The memory 1001 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0244] If the memory 1001, the processor 1002, and the communication interface 1003 are implemented independently, the communication interface 1003, the memory 1001, and the processor 1002 can be connected to each other via a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 10 Only one thick line is used in the diagram, but this does not mean that there is only one bus or one type of bus.

[0245] Optionally, in a specific implementation, if the memory 1001, the processor 1002 and the communication interface 1003 are integrated on a chip, the memory 1001, the processor 1002 and the communication interface 1003 can communicate with each other through an internal interface.

[0246] The processor 1002 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0247] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned dynamic linearization dynamic modeling method of tank-liquid coupling of a liquid tank truck.

[0248] An embodiment of the present application also provides a computer program product having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned dynamic linearization dynamic modeling method of tank-liquid coupling of a liquid tank truck.

[0249] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or N embodiments or examples in a suitable manner. In addition, those skilled in the art can combine and combine different embodiments or examples described in this specification and features of different embodiments or examples without contradiction.

[0250] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be understood to indicate or imply relative importance or implicitly specify the number of technical features indicated. Thus, a feature specified as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of this application, "N" means at least two, for example, two, three, etc., unless otherwise specifically defined.

[0251] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or N executable instructions for implementing a custom logical function or process step, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may be performed in a different order than shown or discussed, including performing functions in a substantially simultaneous manner or in a reverse order depending on the functions involved, which should be understood by those skilled in the art to which the embodiments of the present application pertain.

[0252] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or N wires (electronic devices), a portable computer disk cartridge (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and programmable read-only memory (EPROM or flash memory), fiber optic devices, and a portable compact disc read-only memory (CDROM). Additionally, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically by optically scanning the paper or other medium and then editing, interpreting, or otherwise processing in a suitable manner as necessary, and then storing it in a computer memory.

[0253] It should be understood that various parts of the present application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiment, the N steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented using hardware, as in another embodiment, it can be implemented using any one or a combination of the following technologies known in the art: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0254] Those skilled in the art will appreciate that all or part of the steps in the method for implementing the above-mentioned embodiment can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable storage medium. When the program is executed, it includes one or a combination of the steps of the method embodiment.

[0255] In addition, the functional units in the various embodiments of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into a module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0256] The storage medium mentioned above may be a read-only memory, a magnetic disk, or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present application. Persons skilled in the art may make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present application.

Claims

1. A dynamic linearization dynamic modeling method for tank-liquid coupling of a tank truck, characterized by: The following steps are involved: Select at least one degree of freedom of interest from the six spatial degrees of freedom of the tank body of the tank truck to compile an orthogonal working condition table; Establishing a fine computational fluid dynamics model of the tank body, and inputting the orthogonal operating condition table into the fine computational fluid dynamics model to establish a computational fluid dynamics (CFD) simulation database; Establish a nonlinear high-precision mechanical equivalent sloshing model; Select the most interesting degrees of freedom to build a linearized equivalent pendulum model; The nonlinear high-precision mechanical equivalent sway model and the linearized equivalent pendulum model are calibrated using the CFD simulation database to train a neural network to output output characteristics within a predicted time domain step under each degree of freedom and each working condition of interest.

2. The method according to claim 1, characterized in that The detailed computational fluid dynamics model is expressed as: in, It is represented as the time series of the generalized force output by the model; f0(·) represents the refined computational fluid dynamics model; and u(t) represents the time series of the model input excitation vector.

3. The method according to claim 1, characterized in that The neural network is represented as: Where Θ represents the model state vector; w2 is the model parameter vector; Δw2 represents the change in parameter w2; represents a neural network; x1 represents the model state vector.

4. The method according to claim 3, characterized in that The training of the neural network comprises: Use any algorithm of preset reinforcement learning, where the state For x1, action is [Θ, Δw2], reward The orthogonal operating condition table The operating condition evaluation in the given prediction time domain N p Each moment within the step and The sum of the two norms of is negated as a reward.

5. The method according to claim 4, characterized in that The reward calculation formula is: in, Indicates reward; Indicates the degree of freedom of concern; N p Represents the prediction time domain time step; N T Indicates the number of working conditions; k indicates the time; represents generalized force, subscript 1 represents STL model, subscript 2 represents Dual-LSP model; w j represents the model parameter vector.

6. A dynamic linearization dynamic modeling device for tank-liquid coupling of a tank truck, comprising: A compilation module, used for selecting at least one degree of freedom of interest from the six spatial degrees of freedom of the tank body of the liquid tank truck to compile an orthogonal working condition table; A first construction module is used to establish a fine computational fluid dynamics model of the tank body and input the orthogonal operating condition table into the fine computational fluid dynamics model to establish a computational fluid dynamics (CFD) simulation database; The second building block is used to establish a nonlinear high-precision mechanical equivalent sloshing model; The third building block is used to select the most interesting degrees of freedom to establish a linearized equivalent pendulum model; A modeling module is used to calibrate the nonlinear high-precision mechanical equivalent sway model and the linearized equivalent pendulum model using the CFD simulation database to train a neural network to output output characteristics within a predicted time domain step under each degree of freedom and each working condition of interest.

7. The device according to claim 6, characterized in that The detailed computational fluid dynamics model is expressed as: in, It is represented as the time series of the generalized force output by the model; f0(·) represents the refined computational fluid dynamics model; and u(t) represents the time series of the model input excitation vector.

8. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the dynamic linearization dynamic modeling method for tank-liquid coupling of a tank truck as described in any one of claims 1 to 5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the dynamic linearization dynamic modeling method of tank-liquid coupling of a tank truck as described in any one of claims 1 to 5.

10. A computer program product comprising a computer program, characterized in that The computer program is executed to implement the dynamic linearization dynamics modeling method of tank-liquid coupling of a tank truck according to any one of claims 1 to 5.

Images