Compensation and fusion method for pressure sensor of intelligent track hinging device

Abstract

According to the intelligent track hinge device pressure sensor compensation and fusion method, modeling and compensation are carried out on temperature response of a pressure sensor by adopting a support vector machine regression method, so that the measurement precision and stability under different temperature conditions are improved; and meanwhile, a multi-sensor data fusion model is constructed by adopting the Belith theory, and measurement data from a plurality of pressure sensors are effectively fused through an optimization algorithm, so that the overall performance and precision of the system are improved. By adopting an advanced nonlinear modeling technology and combining efficient algorithms such as support vector machine regression, the nonlinear response of the pressure sensor under different temperature conditions can be more accurately compensated. The temperature compensation precision is improved, the measurement stability and reliability of the system in a complex environment are remarkably improved, meanwhile, data fusion is carried out on the two pressure sensors through the Belith theory, and the problem that due to noise interference, the left sensor and the right sensor output different numerical values under the same condition is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent track articulated vehicles, and in particular to a pressure sensor compensation and fusion method for an intelligent track articulated device. Background Art

[0002] As a new transportation product, the intelligent tram combines the advantages of modern trams and buses. This new mode of transportation has revolutionized traditional perceptions of urban transportation and provided a new alternative to addressing travel difficulties in large and medium-sized cities. This new mode of transportation utilizes a three-carriage structure with two motors and one trailer, interconnected by articulated joints. These joints, with symmetrically mounted hydraulic dampers on both sides, ensure the intelligent tram's straight-line stability and prevent the front and rear carriages from collapsing during sharp turns. This is crucial for ensuring passenger safety and comfort.

[0003] After searching, CN118565695A discloses a temperature compensation system for a pressure sensor, which mainly focuses on the data receiving module. The system can efficiently receive the sensor data collected by the data acquisition unit, and accurately determine the temperature compensation triggering conditions through the threshold judgment module, thereby improving the accuracy of the temperature compensation triggering. By using the model building module, combined with the least squares method and the interpolation method, the temperature drift problem of the pressure sensor is accurately modeled and processed. Furthermore, through the model training module and the combination optimization module, the system can continuously optimize the model performance to ensure that the efficient and stable compensation effect is always maintained under different working environments, thereby improving the accuracy of the sensor data and the overall performance of the system. It does not involve the comprehensive evaluation of the compensation and fusion of multiple pressure sensors in the articulated device.

[0004] During actual use, the pressure sensor is affected by temperature, and the output value has an error compared to the true value. In addition, noise interferes with the value of the pressure sensor, resulting in inconsistent output results of sensors in different positions. However, in existing patents, only the control system of the articulated device or the structure of the articulated device is described, and the nonlinear influence of temperature and noise on the pressure sensor is not fully considered. Therefore, the accuracy of use is difficult to guarantee.

[0005] In summary, by fully considering the impact of temperature and noise on pressure sensors, a compensation and fusion solution for pressure sensors of intelligent track articulation devices was invented to improve the measurement stability and reliability of the system in complex environments, which has great economic significance and practical value. Summary of the Invention

[0006] This invention addresses the shortcomings of existing technologies by introducing an innovative approach to temperature compensation and data fusion. Using a support vector machine (SVM) model, the invention accurately temperature compensates multiple pressure sensors individually, eliminating the effects of temperature fluctuations on sensor pressure measurements and obtaining accurate compensated pressure values. Then, using advanced data fusion techniques based on Bayesian theory, the compensated pressure values ​​are optimized and integrated to generate a unified, more reliable pressure measurement.

[0007] An innovative data fusion method was introduced, employing a unified scaling method to process the output data from multiple angle sensors, enabling effective comparison based on the same standard. Subsequently, an advanced data fusion algorithm optimized the handling of inconsistent sensor outputs caused by noise and other factors, fusing the multiple angle signals into consistent, reliable angle values. Finally, based on the differential results of the fused angle signals, the two pressure sensors were precisely zero-calibrated, effectively eliminating the effects of temperature drift and other external interference, ensuring sensor measurement accuracy and the long-term stability of the system.

[0008] In order to achieve the above objectives, the specific technical solutions of the present invention are as follows:

[0009] A pressure sensor compensation and fusion method for an intelligent track articulation device is disclosed. The support vector machine regression method is used to model and compensate the temperature response of the pressure sensor to improve the measurement accuracy and stability under different temperature conditions. At the same time, the Bayliss theory is used to construct a multi-sensor data fusion model, and the measurement data of multiple pressure sensors are effectively fused through an optimization algorithm, thereby improving the overall performance and accuracy of the system.

[0010] Furthermore, a hinge plate is provided at the center of the articulated device of the intelligent rail vehicle, damping oil cylinders connected to the base are provided on both sides of the hinge plate, and a pressure sensor and a temperature sensor are provided on the damping oil cylinders.

[0011] Furthermore, the process of temperature compensation for pressure sensors using support vector machine regression includes data preprocessing, feature selection, model training, and evaluation steps.

[0012] Furthermore, the data preprocessing forms a data set containing the output of each sensor and the corresponding temperature value. Ideally, the compensated true pressure value of each sensor at different temperatures is collected as training data. The pressure sensors are two pressure sensors S1 and S2, whose output values ​​are z1 and z2, respectively, and the output is affected by temperature T. The outputs of the two pressure sensors are temperature compensated using support vector machine regression.

[0013] Input features: x i =[T1,i ,T 2,i ,z 1,i ,z 2,i ], namely the temperature T and the outputs z1 and z2 of the two sensors;

[0014] Output target: That is the pressure value after compensation.

[0015] Furthermore, in order to deal with the nonlinear relationship between input data, the kernel function is introduced, which is in the form of:

[0016] Among them, σ is the width parameter of the kernel function, which controls the influence range of the kernel.

[0017] Furthermore, after training, the regression function is obtained: Among them, α i is the Lagrange multiplier corresponding to the support vector, which represents the contribution of the model to each training sample; for the new input x new =[T new ,z 1,new ,z 2,new ], the prediction result is:

[0018] Furthermore, the support vector machine regression model will compensate the output of each sensor, that is, the compensated pressure value is obtained by the following formula: and where f(x new ) is the prediction result obtained by the support vector machine regression model, that is, the pressure value after temperature compensation.

[0019] Furthermore, in sensor fusion, the output of each sensor is considered as an observation data z1, z2, ..., z n , the corresponding sensor credibility and noise characteristics are expressed by prior probability P(x) and likelihood function P(z i |x) to describe; in the Bayesian sensor fusion formula, the data obtained by the two pressure sensors correspond to the observation data and Through observation data and Prior distribution P(x), likelihood function and Posterior distribution P(x|z1,z2), fusion results The weighted ratio is dynamically adjusted according to the credibility of different sensors to provide more accurate and reliable pressure value estimation.

[0020] Furthermore, the likelihood function and They respectively indicate that when the real pressure value x is given, the two sensors measure and The probability of , assuming that the measurement noise is Gaussian distributed, the corresponding likelihood function is in the form of:

[0021] in, and are the noise variances of the two sensors, respectively, reflecting the accuracy of each sensor's measurement.

[0022] Furthermore, the fusion results The optimal estimate is obtained by weighted averaging: Among them, z1 and z2 are the measured values ​​of the sensor, and is their noise variance, is the variance of the prior distribution, is the fused pressure estimate.

[0023] Compared with the prior art, the intelligent track articulation device pressure sensor compensation and fusion method of the present invention has the following advantages:

[0024] Compared with traditional sensor solutions, this invention uses advanced nonlinear modeling technology combined with efficient algorithms such as support vector machine regression (SVM) to more accurately compensate for the nonlinear response of pressure sensors under different temperature conditions. This method not only improves the accuracy of temperature compensation, but also significantly enhances the measurement stability and reliability of the system in complex environments, outperforming existing traditional compensation technologies. At the same time, data fusion of the two pressure sensors using Bayesian theory overcomes the problem of noise interference causing the sensors on both sides to output different values ​​under what should be the same conditions. This method not only eliminates the impact of temperature fluctuations and noise on the measurement results, but also improves the overall accuracy and reliability of the system.

[0025] In terms of temperature compensation, this solution utilizes advanced nonlinear modeling techniques, compared to traditional linear regression, quadratic fitting, and polynomial regression methods, to more accurately capture and compensate for the nonlinear response of pressure sensors at different temperatures. By utilizing efficient algorithms such as support vector machine regression (SVM), the accuracy of temperature compensation is significantly improved. This significantly outperforms traditional methods when dealing with complex nonlinear temperature-pressure relationships, ensuring high sensor output accuracy and system stability.

[0026] When it comes to unifying and fusing the output data from two pressure sensors, Baylis's theory, based on statistical methods, more comprehensively considers the actual conditions of each sensor, such as variance and bias, compared to traditional weighted averaging and least squares methods. However, these methods fail to fully account for sensor error characteristics. Baylis's theory rationally adjusts the weights of each sensor's output, fully utilizing these sensor error characteristics. This makes the fused data more accurate, reliable, and closer to the true value. Compared to traditional methods, Baylis's theory can better handle inter-sensor heterogeneity, improving the overall accuracy and stability of the system in complex environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 This is a schematic diagram of the overall system of the pressure sensor compensation and fusion method of the intelligent track articulation device described in the present invention.

[0028] Figure 2 This is an overall flow chart of the pressure sensor compensation and fusion method of the intelligent track articulation device described in the present invention. DETAILED DESCRIPTION

[0029] Below with reference to the embodiments and appended Figure 1-Figure 2 The present invention is further described.

[0030] Example 1

[0031] like Figure 1-Figure 2 As shown, the pressure sensor compensation and fusion method of the intelligent rail articulation device of this embodiment adopts the support vector machine regression method to model and compensate the temperature response of the pressure sensor to improve the measurement accuracy and stability under different temperature conditions; at the same time, the Bayliss theory is used to construct a multi-sensor data fusion model, and the measurement data from multiple pressure sensors are effectively fused through the optimization algorithm, thereby improving the overall performance and accuracy of the system. The intelligent rail vehicle articulation device in this embodiment is provided with an articulation plate 6 at the center, and damping cylinders 5 connected to the base 8 are provided on both sides of the articulation plate 6. A pressure sensor 1 and a temperature sensor 2 are provided on the damping cylinder 5. Specifically, there are two pressure sensors 1, namely pressure sensor 1 and pressure sensor 2; there are two temperature sensors 2, namely temperature sensor 1 and temperature sensor 2.

[0032] SVM regression is a regression method based on support vector machines. It seeks a hyperplane such that the distance (error) between most training data points and the hyperplane is as small as possible, ensuring low model complexity. The core idea of ​​SVM regression is to construct an appropriate hyperplane to best fit the input data and to introduce an ε parameter to tolerate errors, thereby achieving better generalization capabilities.

[0033] The goal of support vector machine regression is to find a function f(x) such that: f(x) = <w,x> + b(1); where: x is the input data (in temperature compensation, it is the temperature value and sensor output); w is the weight vector, which represents the distribution of the data in the feature space; b is the bias term.

[0034] The process of temperature compensation for two pressure sensors using support vector machine regression includes data preparation (data collection and data preprocessing), feature selection (selecting a suitable kernel function), model training and evaluation, etc.

[0035] Specifically, the temperature compensation of the data of the two pressure sensors is performed by SVM regression as follows.

[0036] First, data needs to be collected and organized. The dataset contains the output of each sensor (Z1 and Z2) and the corresponding temperature value (T). Ideally, the actual pressure value of each sensor after compensation at different temperatures is required as training data.

[0037] Assumptions: z1 and z2 are pressure sensor outputs (ie, sensor raw data).

[0038] T1 and T2 are the temperatures corresponding to pressure sensor 1 and pressure sensor 2.

[0039] and is the compensated pressure value (this is the target value that needs to be predicted).

[0040] The data structure is as follows:

[0041] Input features: T1, T2, z1, z2;

[0042] Output target:

[0043] Therefore, assume that there are two pressure sensors S1 and S2, whose outputs are z1 and z2 respectively, and these outputs are affected by temperature T. The outputs of these two pressure sensors are temperature compensated by support vector machine regression (SVR).

[0044] Input features: x i =[T 1,i ,T 2,i ,z 1,i ,z 2,i ], namely the temperature T and the outputs z1 and z2 of the two pressure sensors.

[0045] Output target: That is the pressure value after compensation.

[0046] In regression problems, the goal of support vector machine regression is to fit the data using an ε-insensitive loss function. The goal is to make the prediction error of most data points less than ε while ensuring that the model is as smooth as possible to avoid overfitting.

[0047] The loss function for support vector machine regression is usually defined as:

[0048]

[0049] Where: f(x i ) is the predicted value of the model, y i is the true value.

[0050] ε is a tolerance threshold that allows a certain deviation between the predicted value and the true value.

[0051] Support vector machine regression finds the best fitting function by minimizing this loss function while ensuring that the complexity of the model is not too high.

[0052] By introducing Lagrange multipliers, the optimization problem of support vector machine regression can be transformed into a dual problem. The goal is to maximize the Lagrange dual function to obtain the optimal solution.

[0053] The optimization problem is:

[0054]

[0055] in:

[0056] ξ i is the slack variable, representing the error term.

[0057] C is the regularization parameter that controls the degree of error penalty.

[0058] The constraints are as follows:

[0059]

[0060] In order to deal with the nonlinear relationship between input data, the kernel function is introduced. The commonly used kernel function is the RBF kernel (radial basis function), which has the form:

[0061] Among them, σ is the width parameter of the kernel function, which controls the influence range of the kernel.

[0062] Through the kernel function, the input data can be mapped to a higher-dimensional feature space, thereby realizing linear regression in this space.

[0063] After model training, the regression function is obtained:

[0064]

[0065] Among them, α i is the Lagrange multiplier corresponding to the support vector, which represents the contribution of the model to each training sample. new =[T new ,z 1,new ,z 2,new ], the prediction result is:

[0066]

[0067] Finally, the SVM regression model will compensate the output of each sensor, that is, the compensated pressure value is obtained by the following formula and

[0068]

[0069]

[0070] where f(x new ) is the prediction result obtained by the support vector machine regression model, that is, the pressure value after temperature compensation.

[0071] Support vector machine regression (SVM) can effectively compensate for the effects of temperature on the two pressure sensors. SVM regression finds the nonlinear relationship between temperature and pressure by minimizing error and controlling model complexity. Kernel functions (such as RBF kernels) improve the model's fitting ability, ultimately achieving accurate temperature compensation.

[0072] Bayesian fusion is a statistical method based on Bayes' theorem that combines measurements from multiple sensors to produce an optimal estimate. In sensor fusion, the Bayesian method considers the credibility of each sensor's measurement and dynamically adjusts weights to produce a more accurate estimate.

[0073] The basic idea of ​​the Bayesian fusion method is to obtain the fusion result by calculating the joint posterior probability distribution based on the data of each sensor and its measurement noise characteristics. The basic form of the Bayesian formula is:

[0074]

[0075] in:

[0076] P(x|z) is the posterior probability of the system state x after the observation data z is given, that is, the fusion result that we ultimately want to calculate.

[0077] P(z|x) is the likelihood function, which represents the probability of observing data z given state x.

[0078] P(x) is the prior probability, which represents the estimate of state x when there is no observed data.

[0079] P(z) is the marginal probability of observing the data, which is a constant and can be ignored.

[0080] In sensor fusion, the output of each sensor can be regarded as an observation data z1, z2, ..., z n , the corresponding sensor credibility and noise characteristics are expressed by prior probability P(x) and likelihood function P(z i |x) to describe.

[0081] In the Bayesian sensor fusion formula, the data obtained by pressure sensor 1 and pressure sensor 2 correspond to the observation data and That is, they are the sensor measurements that you want to fuse.

[0082] The specific steps are as follows:

[0083] 1. Collect data and observe data and

[0084] These two data are the measurement results of pressure sensor 1 and pressure sensor 2 respectively. Assume that the pressure values ​​measured by pressure sensor 1 and pressure sensor 2 are and

[0085] 2. Define the prior distribution P(x):

[0086] P(x) is the prior distribution of the system state (i.e., the actual pressure value x), an estimate before sensor data is available. A probability distribution (such as a normal distribution) is used to describe the prior distribution of pressure values, P(x), which is determined by historical data or the physical properties of the system.

[0087] The prior distribution can be of the form where x0 is the prior estimate, is the prior variance.

[0088] If there is historical data or previous measurement data, the prior estimate can be determined by statistical analysis of these data (such as calculating the mean). Assuming there are multiple historical measurements, the average of these historical data can be calculated as the prior estimate of pressure:

[0089]

[0090] Among them, x1, x2, ..., x N is the historical data, and N is the number of historical points.

[0091] The formula for calculating the variance of historical data is:

[0092]

[0093] The two pressure sensors of the present invention assume that their historical data are as follows:

[0094] The historical data of pressure sensor 1 is: {p1,p2,...,p N1};

[0095] The historical data of pressure sensor 2 is: {p′1,p′2,...,p' N2};

[0096] Calculate the mean and variance of the historical data of these two sensors as the prior estimate and prior variance:

[0097] The prior estimate x0 can be taken as the weighted average of the historical data of pressure sensor 1 and pressure sensor 2:

[0098]

[0099] in and are the historical data means of pressure sensor 1 and pressure sensor 2, respectively. N1 and N2 are the number of data points of pressure sensor 1 and pressure sensor 2.

[0100] Prior variance It can be calculated by taking the weighted average of the variance of the data from pressure sensor 1 and pressure sensor 2:

[0101] in and are the variances of historical data of pressure sensor 1 and pressure sensor 2 respectively.

[0102] 3. Define the likelihood function and

[0103] These two functions represent the pressure values ​​measured by pressure sensor 1 and pressure sensor 2 respectively when the actual pressure value x is given. and Assuming that the measurement noise is Gaussian distributed, the likelihood function can be written as:

[0104] For sensor one,

[0105] For sensor two,

[0106] in, and are the noise variances of pressure sensor 1 and pressure sensor 2, respectively, reflecting the measurement accuracy of each sensor.

[0107] 4. Posterior distribution P(x|z1,z2):

[0108] The posterior distribution is the probability distribution of the pressure value x calculated based on the sensor measurements z1 and z2 and the prior information. It represents the most likely value of the true pressure value given the sensor data.

[0109] The calculation process uses Bayes' theorem combined with the prior distribution and the likelihood function:

[0110] P(x|z1,z2)∝P(x)P(z1|x)P(z2|x) (15);

[0111] 5. Fusion Results

[0112] After obtaining the posterior distribution, calculate the posterior expected value That is, the estimated value after fusion.

[0113] For the case of Gaussian distribution, the optimal estimate can be obtained by weighted averaging:

[0114]

[0115] z1 and z2 are the measured values ​​of the two pressure sensors, and is their noise variance, is the variance of the prior distribution, is the fused pressure estimate.

[0116] In this way, the Bayesian fusion method can dynamically adjust the weighted ratio according to the credibility of different sensors, providing more accurate and reliable pressure value estimation.

[0117] For temperature compensation, alternatives include linear regression, quadratic fitting, polynomial regression, or empirical compensation methods. These methods compensate by mathematically relating temperature to sensor output.

[0118] In terms of unifying and fusing pressure sensor output data, alternative solutions include weighted averaging, least squares method, or Kalman filtering. These methods improve data stability by weighting or optimizing the outputs of multiple sensors.

[0119] The methods of the present invention are applicable to multi-sensor fusion and can be extended to data fusion tasks involving three, four, or more pressure sensors. These methods can effectively perform temperature compensation or data fusion on the outputs of multiple sensors, ensuring that the system can provide reliable and accurate measurement results in a multi-sensor environment.

[0120] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A pressure sensor compensation and fusion method for an intelligent track articulation device, characterized in that: The support vector machine regression method is used to model and compensate the temperature response of the pressure sensor to improve the measurement accuracy and stability under different temperature conditions. At the same time, the Bayliss theory is used to construct a multi-sensor data fusion model, and the measurement data of multiple pressure sensors are effectively fused through the optimization algorithm, thereby improving the overall performance and accuracy of the system.

2. The intelligent track articulation device pressure sensor compensation and fusion method according to claim 1 is characterized in that: The intelligent rail vehicle articulation device is provided with an articulation plate (6), damping oil cylinders (5) connected to a base (8) are provided on both sides of the articulation plate (6), and a pressure sensor (1) and a temperature sensor (2) are provided on the damping oil cylinders (5).

3. The intelligent track articulation device pressure sensor compensation and fusion method according to claim 2 is characterized in that: The process of temperature compensation of pressure sensor (1) using support vector machine regression includes data preprocessing, feature selection, model training and evaluation steps.

4. The intelligent track articulation device pressure sensor compensation and fusion method according to claim 3 is characterized in that: The data preprocessing forms a data set including the output of each sensor and the corresponding temperature value; the compensated real pressure value of each sensor at different temperatures is collected as training data; the pressure sensor (1) is two pressure sensors S1 and S2, and the output is affected by the temperature T; the output of the two pressure sensors is temperature compensated by support vector machine regression; Input features: x i =[T 1,i ,T 2,i ,z 1,i ,z 2,i ], namely the temperature T and the outputs z1 and z2 of the two pressure sensors; Output target: That is, the pressure value after compensation.

5. The intelligent track articulation device pressure sensor compensation and fusion method according to claim 4 is characterized in that: In order to deal with the nonlinear relationship between input data, the kernel function is introduced, which is in the form of: Among them, σ is the width parameter of the kernel function.

6. The intelligent track articulation device pressure sensor compensation and fusion method according to claim 5 is characterized in that: After model training, the regression function is obtained: Among them, α i is the Lagrange multiplier corresponding to the support vector, which represents the contribution of the model to each training sample; for the new input x new =[T new ,z 1,new ,z 2,new ], the prediction result is:

7. The intelligent track articulation device pressure sensor compensation and fusion method according to claim 6 is characterized in that: The support vector machine regression model will compensate the output of each sensor, that is, the compensated pressure value is obtained through the following formula and for sensor 1; for sensor 2; where f(x new ) is the prediction result obtained by the support vector machine regression model, that is, the pressure value after temperature compensation.

8. The intelligent track articulation device pressure sensor compensation and fusion method according to claim 7 is characterized in that: In sensor fusion, the output of each sensor is considered as an observation data z1, z2, ..., z n , the corresponding sensor credibility and noise characteristics are expressed by prior probability P(x) and likelihood function P(z i |x) to describe; in the Bayesian sensor fusion formula, the data obtained by the two pressure sensors correspond to the observation data and Through observation data and Prior distribution P(x), likelihood function and Posterior distribution P(x|z1,z2), fusion results The weighted ratio is dynamically adjusted according to the credibility of different sensors to provide more accurate and reliable pressure value estimation.

9. The intelligent track articulation device pressure sensor compensation and fusion method according to claim 8, characterized in that: Likelihood function and They respectively indicate that when the real pressure value x is given, the two sensors measure and The probability of , assuming that the measurement noise is Gaussian distributed, the corresponding likelihood function is in the form of: in, and are the noise variances of the two sensors, respectively, reflecting the accuracy of each sensor's measurement.

10. The intelligent track articulation device pressure sensor compensation and fusion method according to claim 9, characterized in that: Fusion results The optimal estimate is obtained by weighted averaging: Among them, z1 and z2 are the measured values ​​of the sensor, and is their noise variance, is the variance of the prior distribution, is the fused pressure estimate.

Images