Pneumatic brake system pressure estimation method based on data and physical fusion model

Abstract

The application discloses a kind of pressure estimation methods of air pressure brake system based on data and physical fusion model, the method is collected air source pressure, brake chamber pressure and ABS solenoid valve control instruction by test bench, constructs time sequence sliding window sample and normalizes processing;Physical information neural network model containing long short-term memory network and full connection layer is constructed, its output brake chamber pressure estimation value, brake chamber temperature, the gas mass flow of valve port and chamber volume are calculated;According to the differential equation of pressure change rate, the physical law residual is calculated;In training, data fitting loss and physical residual loss jointly constitute total loss function, and network parameters are iteratively updated using Adam optimizer until the model converges;The trained model is deployed in brake system controller, brake chamber pressure can be estimated with high precision, strong robustness and physical consistency online only according to air source pressure and valve instruction.

Description

Technical Field

[0001] This invention relates to the fields of vehicle engineering and artificial intelligence application technology, specifically to a pressure estimation method for air pressure braking systems based on a data and physics fusion model. Background Technology

[0002] Commercial vehicle air pressure braking systems typically consist of components such as an air source, pipelines and connectors, brake valves / relay valves, ABS solenoid valves, and brake chambers. During braking, the brake chamber pressure is the core state variable that generates braking force and achieves pressure increase, holding, and decrease regulation; its dynamic changes directly affect braking response speed, braking stability, and safety. To achieve precise control and health monitoring, the controller usually needs to acquire or estimate the pressure at key nodes and its variation patterns.

[0003] Existing pressure acquisition methods mainly include direct measurement, mechanistic model estimation, and data-driven estimation. Direct measurement requires the placement of pressure sensors at multiple nodes, but due to limitations in cost, installation space, reliability, and maintenance complexity, actual vehicles often only allow sensors to be placed at the air source or a few other locations, resulting in the inability to measure the pressure in intermediate chambers. Mechanistic model estimation typically establishes state equations based on compressible gas flow, orifice flow rate, and valve dynamics. However, the model requires relatively accurate parameters such as the effective valve area, flow coefficient, response time constant, and leakage coefficient. These parameters can drift due to manufacturing differences, wear and aging, temperature changes, and pipeline leaks, leading to increased estimation errors. While pure data-driven methods can fit complex nonlinearities, they are highly dependent on the coverage of training data and are prone to physically inconsistent results in scenarios such as operating condition extrapolation, segmented flow switching, valve dead zone hysteresis, and noise disturbances, resulting in insufficient robustness and interpretability.

[0004] In real-world vehicle applications, common conditions include the ability to measure only the air source pressure and brake chamber pressure, and to obtain control signals such as ABS intake and exhaust valve control commands. However, critical pressure states such as those before and after the valves, and intermediate pipe sections, are difficult to measure directly. Furthermore, differences exist between simulation models and real vehicle systems in terms of valve characteristics, leakage, volume, and pipe damping, making it difficult for models relying on a single simulation or data set to achieve stable adaptation.

[0005] Therefore, there is an urgent need for a method that can embed physical mechanisms such as the ideal gas law, the conservation of control volume mass, orifice flow rate and valve opening and closing action into the learning process in the form of residual constraints under the condition of a small amount of measurable signals, so as to achieve a high-precision, robust and physically consistent estimation method for the pressure of the pneumatic braking system. Summary of the Invention

[0006] The purpose of this invention is to provide a pressure estimation method for pneumatic braking systems based on a data and physics fusion model, in order to solve the problems of low estimation accuracy, poor robustness, physical inconsistency, and over-reliance on sensors that exist in the prior art when relying solely on data-driven or mechanistic models.

[0007] To achieve the above objectives, the present invention provides a pressure estimation method for a pneumatic braking system based on a data and physics fusion model, comprising the following steps:

[0008] S1: Based on bench tests, the air source pressure, brake chamber pressure and ABS solenoid valve control commands under different braking conditions are collected, and the brake chamber pressure is used as the label; the time step and window length are defined to construct a time-series sliding window sample; the brake chamber pressure and discrete variable ABS solenoid valve control commands are normalized.

[0009] S2: Construct a physical information neural network model that includes a long short-term memory network layer and a fully connected layer. The input of the physical information neural network model receives time-series sliding window samples, and the output outputs the estimated value of the brake chamber pressure.

[0010] S3: Substitute the estimated value of the brake chamber pressure into the dynamic estimation model of the brake chamber temperature to calculate the brake chamber temperature; substitute the estimated value of the brake chamber pressure into the gas mass flow rate calculation model to calculate the gas mass flow rate; substitute the estimated value of the brake chamber pressure into the dynamic piecewise model of the brake chamber volume to calculate the brake chamber volume.

[0011] S4: Based on the brake chamber temperature, gas mass flow rate, and brake chamber volume, calculate the brake chamber pressure change rate according to the differential equation of the brake chamber pressure change rate; calculate the numerical derivative of the estimated brake chamber pressure with respect to time; calculate the difference between the brake chamber pressure change rate and the numerical derivative to obtain the physical law residual.

[0012] S5: Construct a total loss function that includes physical law residuals, and iteratively update the parameters of the physical information neural network model to minimize the total loss function;

[0013] S6: Deploy the trained physical information neural network model into the pneumatic braking system controller. The pneumatic braking system controller estimates the brake chamber pressure online based on the real-time collected air source pressure and ABS solenoid valve control commands.

[0014] To optimize the above technical solution, the specific measures also include:

[0015] In step S3, the process of substituting the estimated value of the brake chamber pressure into the dynamic estimation model of the brake chamber temperature to calculate the brake chamber temperature is as follows:

[0016] The dynamic estimation model for brake chamber temperature is based on the temperature-pressure coupling relationship established by the thermodynamic polytropic process equation, and the expression is:

[0017] ;

[0018] in, This refers to the temperature of the brake chamber. This is an estimated value for the brake chamber pressure. Atmospheric pressure, For ambient temperature, It is a highly variable index.

[0019] In step S3, the estimated value of the brake chamber pressure is substituted into the gas mass flow rate calculation model to calculate the gas mass flow rate, specifically:

[0020] Furthermore, the gas mass flow rate calculation model is established based on the isentropic flow theory of gas and Bernoulli's equation, and its expression is:

[0021] ;

[0022] in, This is the gas mass flow rate. The effective flow area of ​​the ABS solenoid valve port. and The absolute pressures of upstream and downstream are respectively. For flow coefficient, The gas flow function is determined by the pressure ratio. Decide:

[0023] when hour:

[0024] ;

[0025] Furthermore, when hour:

[0026] ;

[0027] in, It represents the air insulation index.

[0028] In step S3, substituting the estimated value of the brake chamber pressure into the dynamic piecewise model of the brake chamber volume specifically involves:

[0029] The dynamic piecewise model of the brake chamber volume is expressed as follows:

[0030] ;

[0031] in, This refers to the volume of the brake chamber. The initial dead zone volume, This represents the maximum volume after the push rod is fully extended. To overcome the preload pressure of the return spring, The effective area of ​​the air chamber diaphragm. For the return spring stiffness, Preload pressure for the brake chamber diaphragm. This is the pressure value corresponding to the push rod reaching its maximum stroke.

[0032] Step S4, based on the brake chamber temperature, gas mass flow rate, and brake chamber volume, calculates the chamber pressure change rate according to the differential equation of the chamber pressure change rate, specifically as follows:

[0033] Furthermore, based on gas mass flow rate Based on the law of conservation of mass and the ideal gas law, the mass of air in the braking chamber is obtained. :

[0034] ;

[0035] Furthermore, by incorporating the volume derivative and substituting it into the differential equation for the rate of change of pressure in the air chamber, the rate of change of pressure in the air chamber can be calculated. The expression is:

[0036] ;

[0037] When the ABS solenoid valve control command is a boost command When the ABS solenoid valve control command is a pressure reduction command, ;

[0038] in, is the gas constant.

[0039] In step S5, the total loss function includes a data loss term and a physical loss term, specifically:

[0040] Data loss items The error between the estimated brake chamber pressure output by the physical information neural network and the actual brake chamber pressure label is expressed as:

[0041] ;

[0042] Furthermore, physical loss items The physical law residuals obtained from step S4 are expressed as follows:

[0043] ;

[0044] The total loss function is:

[0045] ;

[0046] in, These are the weighting coefficients.

[0047] In step S5, the iterative update of the physical information neural network model parameters to minimize the total loss function is specifically performed as follows:

[0048] The Adam optimizer is used to iteratively update the parameters of the physical information neural network model. The input is the gradient of the total loss function with respect to the physical information neural network model parameters, and the output is the update amount of the physical information neural network model parameters. The weights and biases of the physical information neural network model are also updated. The iterative update is repeated until the physical information neural network model converges.

[0049] Let the first The parameters of the neural network model at the next iteration are: If the total loss function is Loss, then the gradient is:

[0050] ;

[0051] in, This is the derivative of the total loss function with respect to the current parameters. For neural network model parameters The gradient;

[0052] Furthermore, the Adam optimizer uses exponential moving averages to calculate the first and second moment estimates of the gradient, respectively:

[0053] ;

[0054] in, For first-order moment estimation, For second-order moment estimation, and The attenuation coefficient is... This indicates that the gradient is squared element by element. Indicates time;

[0055] Furthermore, bias correction is introduced, and the first-order moment estimate after bias correction is calculated. With second-order moment estimation :

[0056] ;

[0057] Update the neural network model parameters based on the corrected moment estimates:

[0058] ;

[0059] in, For learning rate, is the numerical stability constant.

[0060] The physical information neural network model includes a data-driven branch and a physical constraint branch, and adopts a serial-parallel hybrid architecture. The data-driven branch adopts a serial structure. The input layer receives sliding window data of a preset length, which is then passed through two LSTM units with 64 and 32 nodes respectively, and connected to two fully connected layers with 64 and 32 nodes respectively. Both layers use the Swish activation function, and the output layer with one neuron outputs the normalized braking chamber pressure.

[0061] Furthermore, the physical constraint branch works in parallel with the data branch during the training phase, constraining and optimizing the weight parameters of the data branch by minimizing the physical law residual.

[0062] The present invention also proposes an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements a pressure estimation method for a pneumatic braking system based on a data and physics fusion model as described above.

[0063] The present invention also proposes a computer-readable storage medium storing a computer program that enables a computer to execute a pressure estimation method for a pneumatic braking system based on a data and physics fusion model as described above.

[0064] Compared with the prior art, the beneficial effects of the present invention are:

[0065] This invention deeply integrates a data-driven approach with a pneumatic braking system mechanism model by constructing a Physical Information Neural Network (PINN). It fully utilizes the statistical regularities of bench test data and ensures that the model output conforms to the basic physical laws of gas flow, thermodynamics, and mass conservation through physical residual constraints. During the training process, this method simultaneously optimizes data fitting error and physical consistency error, significantly improving the accuracy of pressure estimation. It also exhibits stronger robustness and generalization ability in complex scenarios such as working condition extrapolation, noise interference, and system parameter drift.

[0066] This invention only requires easily measurable signals such as air source pressure and ABS solenoid valve control commands to estimate brake chamber pressure in real time online using a trained model, effectively reducing system hardware costs and complexity, and making it suitable for large-scale commercial vehicle applications.

[0067] This invention introduces multi-physics constraints such as the differential equation of the rate of change of air chamber pressure, temperature estimation model, and gas flow rate model into the neural network training, so that the network output has a clear physical meaning, avoiding the physically unreasonable results that may occur in pure data-driven methods under non-training conditions, and enhancing the credibility and interpretability of the model.

[0068] This invention employs the Adam optimizer combined with a dynamically adjusted physical loss term weight strategy. In the early stages of training, it focuses on data fitting, while in the later stages, it strengthens physical constraints to ensure smooth model convergence and avoid getting trapped in local optima. At the same time, it introduces regularization techniques such as Dropout to further improve the model's generalization performance and stability.

[0069] This invention can simultaneously output estimated values ​​of multiple key state quantities such as brake chamber temperature, gas mass flow rate, and chamber volume, providing multi-dimensional information support for the health monitoring, fault diagnosis, and advanced control strategy design of the braking system. Attached Figure Description

[0070] Figure 1 : A schematic diagram of the architecture in an embodiment of the present invention.

[0071] Figure 2 : A flowchart of the workflow in an embodiment of the present invention. Detailed Implementation

[0072] The present invention will be further described in detail below through specific embodiments, but it should not be construed as limiting the scope of the subject matter of the present invention to the following embodiments. All technologies implemented based on the above content of the present invention fall within the scope of the present invention.

[0073] In some implementations, such as Figure 1 As shown, this invention provides a pressure estimation method for a pneumatic braking system based on a data and physics fusion model, comprising the following steps:

[0074] S1: Based on bench tests, air source pressure was collected under different braking conditions. Brake chamber pressure ABS solenoid valve control commands ; using the pressure of the brake chamber As a label;

[0075] Define time step and window length Construct input tensor :

[0076] (1)

[0077] Among them, the feature vector of each time step This timing input enables the network to memorize the pressure delay effect caused by long pipelines and valve body movements in the gas path.

[0078] In some implementations, the Z-score normalization method is used to map the brake chamber pressure to a normal distribution to accelerate network convergence and normalize the discrete variable ABS solenoid valve control commands.

[0079] S2: Construct a physical information neural network model that includes a long short-term memory network layer and a fully connected layer. The input of the physical information neural network model receives time-series sliding window samples, and the output outputs the estimated value of the brake chamber pressure.

[0080] In some implementations, the physical information neural network model includes a data-driven branch and a physical constraint branch, employing a hybrid serial-parallel architecture. The data-driven branch uses a serial structure, with the input layer receiving sliding window data of 20 time steps to cover the aerodynamic hysteresis interval. Temporal features are extracted sequentially through two LSTM layers with 64 and 32 nodes respectively, connected by two fully connected layers with 64 and 32 nodes, both using the Swish activation function. The output layer, containing one neuron, outputs a normalized braking chamber pressure, which is finally denormalized to obtain an estimate of the braking chamber pressure. ;

[0081] The physical constraint branch works in parallel with the data branch during the training phase, and constrains and optimizes the weight parameters of the data branch by minimizing the physical law residual.

[0082] S3: Substitute the estimated value of the brake chamber pressure into the dynamic estimation model (2) of the brake chamber temperature to calculate the brake chamber temperature.

[0083] The dynamic estimation model for brake chamber temperature is based on the temperature-pressure coupling relationship established by the thermodynamic polytropic process equation, and the expression is:

[0084] (2)

[0085] in, This refers to the temperature of the brake chamber. This is an estimated value for the brake chamber pressure. Atmospheric pressure, For ambient temperature, Let be the variability exponent, with a value of 1.3, set at the start of training. ; and It is obtained through iterative updates during training.

[0086] In some implementations, the estimated value of the brake chamber pressure is substituted into the gas mass flow rate calculation model (3) to calculate the gas mass flow rate.

[0087] The gas mass flow rate calculation model is established based on the isentropic flow theory of gas and Bernoulli's equation, and the expression is:

[0088] (3)

[0089] in, This refers to the mass flow rate of the gas passing through the ABS solenoid valve port. The effective flow area of ​​the ABS solenoid valve port is determined by the control command of the ABS solenoid valve. Decision: When When it is a boost command, , This refers to the fully open area of ​​the intake valve; when When it is a decompression command, , The area of ​​the fully open exhaust valve; when When it is a pressure holding command, .

[0090] and The absolute pressures of upstream and downstream are respectively, when When the command is to increase pressure: For gas source pressure , The estimated value of the brake chamber pressure ; .when When it is a decompression command: The estimated value of the brake chamber pressure , atmospheric pressure ; .

[0091] It is the flow coefficient, used to correct for contraction and friction losses during gas flow.

[0092] This is the gas flow function, used to describe the compressibility of a gas, and is determined by the pressure ratio. Decide:

[0093] when hour:

[0094] (4)

[0095] when hour:

[0096] (5)

[0097] in, It represents the air insulation index.

[0098] Substituting the estimated brake chamber pressure into the dynamic piecewise model (6) of the brake chamber volume, the brake chamber volume is calculated, and the expression is:

[0099] (6)

[0100] in, The initial dead zone volume, This represents the maximum volume after the push rod is fully extended. To overcome the preload pressure of the return spring, The effective area of ​​the air chamber diaphragm. For the return spring stiffness, Preload pressure for the brake chamber diaphragm. This is the pressure value corresponding to the push rod reaching its maximum stroke.

[0101] S4: Based on the brake chamber temperature, gas mass flow rate, and brake chamber volume, calculate the brake chamber pressure change rate according to the differential equation of the brake chamber pressure change rate; calculate the numerical derivative of the estimated brake chamber pressure with respect to time; calculate the difference between the brake chamber pressure change rate and the numerical derivative to obtain the physical law residual as the physical loss term in the total loss function.

[0102] Based on the gas mass flow rate through the ABS solenoid valve port According to the law of conservation of mass:

[0103] (7)

[0104] Based on the ideal gas law, the air mass in the brake chamber is obtained as follows:

[0105] (8)

[0106] By combining the volume derivative and substituting it into the differential equation for the rate of change of air chamber pressure, the rate of change of air chamber pressure can be calculated, and the expression is:

[0107] (9)

[0108] When the ABS solenoid valve control command is a boost command, input For positive values, the expression is:

[0109] (10)

[0110] When the ABS solenoid valve control command is a pressure reduction command, input For negative values, the expression is:

[0111] (11)

[0112] S5: Construct the total loss function and iteratively update the parameters of the physical information neural network model to minimize the total loss function.

[0113] The total loss function includes a data loss term and a physical loss term; among which, the data loss term... The error between the estimated brake chamber pressure output by the physical information neural network and the actual brake chamber pressure label is expressed as:

[0114] (12)

[0115] Physical loss item The physical law residuals obtained from step S4 are expressed as follows:

[0116] (13)

[0117] The total loss function is:

[0118] (14)

[0119] in, These are the weighting coefficients.

[0120] The Adam optimizer is used to iteratively update the parameters of the physical information neural network model. The input is the gradient of the total loss function with respect to the physical information neural network model parameters, and the output is the update amount of the physical information neural network model parameters. The weights and biases of the physical information neural network model are also updated. The iterative update is repeated until the physical information neural network model converges.

[0121] Let the first The parameters of the neural network model at the next iteration are: If the total loss function is Loss, then the gradient is:

[0122] (15)

[0123] in, This is the derivative of the total loss function with respect to the current parameters. For neural network model parameters The gradient.

[0124] The Adam optimizer uses exponential moving averages to calculate the first and second moments of the gradient, respectively, and adaptively adjusts the step size for different parameters, thereby improving convergence speed and numerical stability.

[0125] (16)

[0126] in, For first-order moment estimation, For second-order moment estimation, and The attenuation coefficient is... This indicates that the gradient is squared element by element. Indicates time;

[0127] Introducing bias correction, calculating the bias-corrected first-moment estimate. With second-order moment estimation :

[0128] (17)

[0129] Update the neural network model parameters based on the corrected moment estimates:

[0130] (18)

[0131] in, For learning rate, This is a numerical stability constant used to avoid the denominator being zero. Through the above iterative updates, the loss continuously decreases, and the parameters of the trained neural network model are obtained.

[0132] S6: Deploy the trained physical information neural network model into the pneumatic braking system controller. The pneumatic braking system controller estimates the brake chamber pressure online based on the real-time collected air source pressure and ABS solenoid valve control commands.

[0133] In some implementations, such as Figure 2 The diagram shown is a flowchart of the process in an embodiment of the present invention.

[0134] Example 1:

[0135] This embodiment uses the following parameter configuration for model training: the Adam optimizer is selected, and the decay coefficient is set to [value missing]. , The numerical stability constant is set to The initial learning rate was set to 0.001, and an exponential decay strategy was adopted to prevent oscillations near the extreme point in the later stages of training.

[0136] Every 100 training iterations, the learning rate decays to 0.9 times its original value. Considering the continuity of time-series data and memory usage, the batch size is set to 128. A smaller batch size helps introduce noise, helping the model escape local optima. The maximum number of training iterations is set to 500 to 1000. A patience value of 50 is set; if the validation set loss does not decrease within 50 consecutive iterations, training is terminated early, and the model weights with the smallest validation set error are saved. For LSTM layers and fully connected layers, an initialization method is used to maintain the consistency of signal variance during forward propagation due to the use of the Swish activation function. Dropout mechanisms are introduced between two LSTM layers and between fully connected layers, with a dropout rate of 0.2 to prevent overfitting and enhance the model's generalization ability.

[0137] In this embodiment, Instead of a fixed constant, a dynamic adjustment strategy is adopted: in the early stages of training, Set it to a smaller value to allow the network to preferentially fit the data trend; gradually increase it as the number of rounds increases. This forces the network output to strictly conform to the constraints of the physical equations.

[0138] In another embodiment of the present invention, an electronic device is proposed, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements a pressure estimation method for a pneumatic braking system based on a data and physics fusion model as described above.

[0139] In another embodiment of the present invention, a computer-readable storage medium is provided storing a computer program that causes a computer to execute a pressure estimation method for a pneumatic braking system based on a data and physics fusion model as described above.

[0140] In the embodiments disclosed in this application, the computer storage medium may be a tangible medium that may contain or store programs for use by or in conjunction with an instruction execution system, apparatus, or device. The computer storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of computer storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing. The above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention in any way. Any simple modifications, equivalent substitutions, and improvements made by those skilled in the art to the above embodiments without departing from the scope of the present invention and based on the technical essence of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A pressure estimation method for a pneumatic braking system based on a data and physics fusion model, characterized in that, Includes the following steps: S1: Based on bench tests, the air source pressure, brake chamber pressure and ABS solenoid valve control commands under different braking conditions are collected, and the brake chamber pressure is used as the label; the time step and window length are defined to construct a time-series sliding window sample; the brake chamber pressure and discrete variable ABS solenoid valve control commands are normalized. S2: Construct a physical information neural network model that includes a long short-term memory network layer and a fully connected layer. The input of the physical information neural network model receives time-series sliding window samples, and the output outputs the estimated value of the brake chamber pressure. S3: Substitute the estimated value of the brake chamber pressure into the dynamic estimation model of the brake chamber temperature to calculate the brake chamber temperature; substitute the estimated value of the brake chamber pressure into the gas mass flow rate calculation model to calculate the gas mass flow rate; substitute the estimated value of the brake chamber pressure into the dynamic piecewise model of the brake chamber volume to calculate the brake chamber volume. S4: Based on the brake chamber temperature, gas mass flow rate, and brake chamber volume, calculate the brake chamber pressure change rate according to the differential equation of the brake chamber pressure change rate; calculate the numerical derivative of the estimated brake chamber pressure with respect to time; calculate the difference between the brake chamber pressure change rate and the numerical derivative to obtain the physical law residual. S5: Construct a total loss function that includes physical law residuals, and iteratively update the parameters of the physical information neural network model to minimize the total loss function; S6: Deploy the trained physical information neural network model into the pneumatic braking system controller. The pneumatic braking system controller estimates the brake chamber pressure online based on the real-time collected air source pressure and ABS solenoid valve control commands.

2. The pressure estimation method for a pneumatic braking system based on a data and physics fusion model according to claim 1, characterized in that: In step S3, the process of substituting the estimated value of the brake chamber pressure into the dynamic estimation model of the brake chamber temperature to calculate the brake chamber temperature is as follows: The dynamic estimation model for brake chamber temperature is based on the temperature-pressure coupling relationship established by the thermodynamic polytropic process equation, and the expression is: ； in, This refers to the temperature of the brake chamber. This is an estimated value for the brake chamber pressure. Atmospheric pressure, For ambient temperature, It is a highly variable index.

3. The pressure estimation method for a pneumatic braking system based on a data and physics fusion model according to claim 1, characterized in that: In step S3, the estimated value of the brake chamber pressure is substituted into the gas mass flow rate calculation model to calculate the gas mass flow rate, specifically: The gas mass flow rate calculation model is established based on the isentropic flow theory of gas and Bernoulli's equation, and the expression is: ； in, This is the gas mass flow rate. The effective flow area of ​​the ABS solenoid valve port. and The absolute pressures of upstream and downstream are respectively. For flow coefficient, The gas flow function is determined by the pressure ratio. Decide: when hour: ； when hour: ； in, It represents the air insulation index.

4. The pressure estimation method for a pneumatic braking system based on a data and physics fusion model according to claim 1, characterized in that: In step S3, substituting the estimated value of the brake chamber pressure into the dynamic piecewise model of the brake chamber volume specifically involves: The dynamic piecewise model of the brake chamber volume is expressed as follows: ； in, This refers to the volume of the brake chamber. The initial dead zone volume, This represents the maximum volume after the push rod is fully extended. To overcome the preload pressure of the return spring, The effective area of ​​the air chamber diaphragm. For the return spring stiffness, Preload pressure for the brake chamber diaphragm. This is the pressure value corresponding to the push rod reaching its maximum stroke.

5. The pressure estimation method for a pneumatic braking system based on a data and physics fusion model according to claim 1, characterized in that: Step S4, based on the brake chamber temperature, gas mass flow rate, and brake chamber volume, calculates the chamber pressure change rate according to the differential equation of the chamber pressure change rate, specifically as follows: Gas mass flow rate Based on the law of conservation of mass and the ideal gas law, the mass of air in the braking chamber is obtained. : ； By combining the volume derivative and substituting it into the differential equation for the rate of change of pressure in the air chamber, the rate of change of pressure in the air chamber can be calculated. The expression is: ； When the ABS solenoid valve control command is a boost command When the ABS solenoid valve control command is a pressure reduction command, ; in, This is an estimated value for the brake chamber pressure. This is the actual brake chamber pressure. This refers to the volume of the brake chamber. The air insulation index. It is the gas constant; For ambient temperature, This refers to the temperature of the brake chamber.

6. The pressure estimation method for a pneumatic braking system based on a data and physics fusion model according to claim 1, characterized in that: In step S5, the total loss function includes a data loss term and a physical loss term, specifically: Data loss items The error between the estimated brake chamber pressure output by the physical information neural network and the actual brake chamber pressure label is expressed as: ； Physical loss item The physical law residuals obtained from step S4 are expressed as follows: ； The total loss function is: ； in, These are the weighting coefficients. This is an estimated value for the brake chamber pressure. This is the actual brake chamber pressure. This represents the rate of change of air chamber pressure.

7. The pressure estimation method for a pneumatic braking system based on a data and physics fusion model according to claim 1, characterized in that: In step S5, the iterative update of the physical information neural network model parameters to minimize the total loss function specifically involves: The Adam optimizer is used to iteratively update the parameters of the physical information neural network model. The input is the gradient of the total loss function with respect to the physical information neural network model parameters, and the output is the update amount of the physical information neural network model parameters. The weights and biases of the physical information neural network model are also updated. The iterative update is repeated until the physical information neural network model converges.

8. The pressure estimation method for a pneumatic braking system based on a data and physics fusion model according to claim 7, characterized in that: Let the first The parameters of the physical information neural network model at the next iteration are: If the total loss function is Loss, then the gradient is: ； in, This is the derivative of the total loss function with respect to the current parameters. For neural network model parameters The gradient; The Adam optimizer uses exponential moving averages to calculate the first and second moment estimates of the gradient, respectively. ； in, For first-order moment estimation, For second-order moment estimation, and The attenuation coefficient is... This indicates that the gradient is squared element by element. Indicates time; Introducing bias correction, calculating the bias-corrected first-moment estimate. With second-order moment estimation : ； Update the neural network model parameters based on the corrected moment estimates: ； in, For learning rate, is the numerical stability constant.

9. An electronic device, characterized in that, include: The device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements a pressure estimation method for a pneumatic braking system based on a data and physics fusion model as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: The computer program causes the computer to execute a pressure estimation method for a pneumatic braking system based on a data and physics fusion model as described in any one of claims 1 to 8.

Images