A Vehicle State Estimation Method Based on Adaptive Total Variation Filtering

Abstract

The present invention proposes a vehicle state estimation method based on adaptive total variation filtering, and the steps are as follows: Step 1: Acquisition and preprocessing of vehicle original signals; Step 2: Noise level evaluation; Step 3: Teager-Kaiser energy evaluation; Step 4: Construction of an optimization problem; Step 5: Using the filtered signal obtained in Step 4 and applying it to the estimation of vehicle states. The present invention is mainly based on the global noise level characteristics and local intensity change characteristics of vehicle system state data, and realizes parameter adaptive filtering by means of the total difference TVD filtering method, removes the noise existing in the signal to the greatest extent, and retains the spike information of the signal while maintaining the data smoothness, and then uses the signal for vehicle state estimation, working condition recognition, etc.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle signal filtering and state estimation, and relates to a filtering framework for signal restoration. Background Art

[0002] With the continuous improvement of the intelligence and electrification levels of vehicle systems, the noise reduction methods for vehicle data have attracted more and more scholars' attention. The traditional methods for processing vehicle data mainly include low-pass filtering, high-pass filtering, band-pass filtering, etc. The above methods can play a very obvious role in data noise reduction, but while improving the data smoothness, the peak information of the data is often lost.

[0003] Due to vehicle driving characteristics and road features, vehicle data shows a certain degree of sparsity, and the effects of the above filtering methods on sparse data still need to be improved. Total variation difference (TVD) is a filtering method developed for sparse data. It transforms filtering into an optimization problem by establishing an objective function, thereby realizing the filtering of sparse data. The current vehicle data filtering methods mainly have the following problems:

[0004] (1) While traditional filtering methods improve data smoothness and remove noise, it is difficult to retain the peak information of the data, and the peak information of the data often plays a very important role in the determination of vehicle safety thresholds and the identification of vehicle states;

[0005] (2) The total variation difference (TVD) filtering method is very suitable for noise reduction processing of sparse data. However, during the process of data generation, the signal intensity and noise level of the data change in real time. Especially, there are significant differences between the data characteristics under extreme working conditions and general working conditions. How to evaluate the differences in working conditions and achieve adaptive filtering for the local characteristics of vehicle system data is generally lacking in current filtering methods. Summary of the Invention

[0006] The present invention mainly realizes parameter adaptive filtering by means of the total variation difference (TVD) filtering method based on the global noise level characteristics and local intensity change characteristics of vehicle system state data. Specifically, it is embodied in using the Teager-Kaiser energy evaluation method to evaluate the noise level of the signal, and performing noise reduction processing on the signal by setting a noise index threshold ε and arithmetic mean filtering to ensure that the signal has a good noise level when solving the energy adaptive parameters. Then, the energy adaptive parameters of the signal are solved, and combined with the noise adaptive parameters, an optimization problem is constructed. By solving the optimization problem, the filtered signal is obtained, and the noise existing in the signal is removed to the greatest extent, and while maintaining the data smoothness, the peak information of the signal is retained, and then the signal is used for vehicle state estimation, working condition identification, etc.

[0007] A vehicle state estimation method based on adaptive total variation filtering, the steps are as follows:

[0008] Step 1: Collection and preprocessing of vehicle raw signals

[0009] First, for the typical working conditions of the vehicle test, sensors are used to collect data such as vehicle speed, acceleration, wheel speed, yaw rate, motor torque, etc., and the first-order difference f of the signal p is performed. 1 The solution is f 1 (n) = |p(n)-p(n-1)|, providing a data source for subsequent steps 2, 3, and 4.

[0010] Step 2: Noise Level Assessment

[0011] 2.1 Noise level evaluation and determination of noise adaptive parameters

[0012] According to the first-order difference of the signal in step 1, f 1 , perform the second-order difference of the signal f 2 The calculation formula is f 2 (n)=|f 1 (n)-f 1 (n-1)|, based on the second-order difference of the signal f 2 Evaluate the noise level in the signal, the evaluation formula is: Get the noise adaptive parameter K e (n) is the weight coefficient of the smoothness constraint term in the optimization problem in step 4.

[0013] 2.2 Noise Index Threshold Setting

[0014] According to the second-order difference of the signal in step 2.1, f 2 , using I n =AVG(|f 2 |) to obtain the noise index I n . Set the threshold ε, when the signal noise index I n If the signal noise index is greater than the threshold ε, the signal needs to be filtered by arithmetic mean and step 2.1 is executed again. If the signal noise index is less than the threshold ε, step 3 can be directly executed.

[0015] Step 3: Teager-Kaiser Energy Assessment

[0016] According to the signal obtained in step 2 after noise level evaluation and noise threshold ε, the Teager-Kaiser energy evaluation method is used to comprehensively evaluate the amplitude effect and instantaneous frequency of the signal, and the Teager-Kaiser energy value TKE(f 1Construct energy adaptive parameters It also provides a weight parameter for the signal smoothness constraint term in the optimization problem of step 4.

[0017] Step 4: Construction of the optimization problem

[0018] According to the first-order difference f of the signal in step 1 1 , the noise adaptive parameter Kx in step 2, and the energy adaptive parameter K in step 3 e (n), construct an optimization problem, and obtain the filtered signal by solving the optimization problem:

[0019]

[0020] where y is the filtered signal; K is the regularization parameter; the first term in the formula is the signal fidelity penalty, the second term is the signal smoothness penalty, and the regularization parameter K is used to achieve the trade-off between signal fidelity and signal smoothness.

[0021] The above steps 1-step 4 are the key innovation points of the present invention.

[0022] Step 5, using the filtered signal output in step 4, can be applied to the estimation of vehicle state. This part belongs to the prior art.

[0023] Specifically, taking the estimation of the road ramp angle as an application example to introduce step 5:

[0024] 5.1 Construction of the state vector and parameter initialization

[0025] First, based on the vehicle state signal y and the original vehicle signal required by the vehicle keyboard in step 1, construct a state vector x, and complete the parameter initialization settings of the Kalman model.

[0026] 5.2 Model state update

[0027] Predict the model state according to the state equation of the vehicle system, and the state equation is:

[0028]

[0029] where is the optimal estimate of the system state at time k-1; is the predicted value of the system parameter at time k; A is the process matrix; u k-1 is the control quantity at time k-1, B is the control matrix; w k-1 is the process noise, usually set as Gaussian white noise with variance Q.

[0030] Calculate the covariance matrix of the prediction process The calculation formula is:

[0031]

[0032] where P k-1 is the optimal state estimation covariance at time k - 1.

[0033] 5.3 State Measurement Update

[0034] Calculate the Kalman gain, and the calculation formula is:

[0035]

[0036] where H is the measurement matrix; R is the variance of the measurement white noise v k-1

[0037] Perform state update, and the calculation formula is:

[0038]

[0039] where z k represents the measured quantity.

[0040] Update the covariance matrix P k of the measurement process, and the update formula is:

[0041]

[0042] Output the state of the vehicle, complete the vehicle state estimation process, and at the same time output the state of the vehicle and the covariance matrix P k to step 5.2 for loop execution. Brief Description of the Drawings

[0043] Figure 1 is a schematic diagram of the main process of the present invention;

[0044] Figure 2 is a schematic diagram of the process of step 5 of the present invention. Detailed Embodiment

[0045] The technical solution provided by the present application will be further described below in conjunction with specific embodiments and their accompanying drawings. In combination with the following description, the advantages and features of the present application will be more clear.

[0046] Actual mass-produced vehicles are limited in terms of cost, reliability of sensor performance, etc., and the number and types of sensors are relatively limited. And due to the special structure of the unsprung mass, suspension, and sprung mass of the vehicle, many vehicle states cannot be directly measured. At this time, it is necessary to estimate the state parameters indirectly according to reliable vehicle dynamics parameters through a vehicle dynamics model.

[0047] ​In the present invention, it is assumed that the inherent parameters of the vehicle (such as mass, dimension parameters, etc.) are known, and the input of the sensor to the vehicle dynamics parameters is relatively complete. It is necessary to estimate the ramp angle of the road based on the acceleration of the vehicle. The reliability of the acceleration is crucial. If the noise of the acceleration is too large, it will directly affect the accuracy and reliability of the ramp angle estimation, and further affect the control performance of the actual vehicle.

[0048] The present invention specifically includes the following steps:

[0049] The first part

[0050] Step 1: Acquisition and preprocessing of vehicle raw signals

[0051] 1.1 Measurement of the inherent parameters of the vehicle system

[0052] First, measure and record the gravity G of the vehicle, the mass m, and the signal sampling time interval Δt.

[0053] 1.2 Acquisition and preprocessing of the vehicle longitudinal acceleration and longitudinal speed signals

[0054] When estimating the road ramp angle, the input vehicle measurement parameters are the vehicle longitudinal acceleration a x , the longitudinal speed v x . Among them, the longitudinal speed v x signal is relatively stable and only requires simple filtering operation; the longitudinal acceleration a x sensor has large noise and is a typical piecewise constant signal, which is very suitable for using the total variation method for filtering, and the accuracy of its value has a great impact on the effectiveness of the vehicle dynamics model. Therefore, select the longitudinal acceleration a x as the filtering object and calculate its first-order difference f 1 to complete the preprocessing process and provide data sources for the filtering processes in steps 2, 3, and 4.

[0055] f 1 (n) = |a x (n) - a x (n - 1)|

[0056] Step 2: Noise level evaluation

[0057] 2.1 Noise level evaluation and solution of noise adaptive parameters

[0058] According to the first-order difference f 1 of the signal in step 1, solve the second-order difference f 2 of the signal.

[0059] f 2 (n) = |f 1 (n) - f1 (n - 1)|

[0060] Evaluate the noise level in the signal based on the second - order difference of the signal to obtain the noise - adaptive parameter K n , which provides a weight coefficient for the smoothness constraint term in the optimization problem of step 4.

[0061]

[0062] 2.2 Noise Index Threshold Setting

[0063] According to the second - order difference f of the signal obtained during the noise level evaluation in step 2.1 2 , solve for the noise index I n .

[0064] I n = AVG(|f 2 |)

[0065] Where AVG represents the mean function.

[0066] To ensure that the noise level of the signal is within an acceptable range for the next operation, set the noise index threshold ε. If the noise index exceeds the threshold range, the signal needs to be arithmetically averaged filtered until the noise index is less than the threshold, and then repeat step 2.1. When the noise index satisfies the threshold range, it indicates that the noise evaluation step has been completed, and step 3 and subsequent operations can be continued.

[0067] Step 3: Teager - Kaiser Energy Evaluation

[0068] 3.1 Teager - Kaiser Energy Evaluation

[0069] Compared with the traditional signal energy evaluation method, Teager - Kaiser energy evaluation can make a comprehensive evaluation of the amplitude effect and frequency characteristics of the signal. According to the first - order difference f of the longitudinal acceleration a obtained in step 1 y , solve for the Teager - Kaiser energy: 1 TKE(f

[0070] (n)) = f 1 (n) 1 - f 2 (n - 1)f 1 (n + 1) 1 (n + 1)

[0071] Where TKE(·) represents the Teager - Kaiser operator.

[0072] 3.2 Construction of Energy - Adaptive Parameter

[0073] In step 3.1, the Teager-Kaiser energy of the signal has been solved, and in this step, an energy adaptive parameter is constructed.

[0074]

[0075] In practical applications, to prevent the situation of no solution in the calculation process, a minimum constant δ can be added to the denominator of the above formula, then

[0076]

[0077] Step 4: Construction of the optimization problem

[0078] The original vehicle longitudinal acceleration signal a is obtained from step 1 x , let x = a x ; The noise adaptive parameter is obtained from step 2 The energy adaptive parameter is obtained from step 3 The optimization problem can be constructed based on the form of the total difference:

[0079]

[0080] where y is the filtered signal; K is the regularization parameter. The first term in the formula is the signal fidelity penalty, the second term is the signal smoothness penalty, and the regularization parameter K is used to achieve the trade-off between signal fidelity and signal smoothness.

[0081] Second part: Application example of vehicle state estimation:

[0082] Taking the filtered signal y in step 4 as the input for "estimating the road ramp angle", and using the Kalman method to estimate the ramp angle, which is regarded as an application example in vehicle state estimation.

[0083] Specifically, it includes the following steps:

[0084] 5.1 Construction of the state vector and parameter initialization

[0085] According to the vehicle longitudinal acceleration a obtained from steps 1 - 4 x the filtered signal y, and selecting the vehicle longitudinal speed v x and the sine value sinα of the road ramp angle, the state vector is constructed as:

[0086] x = [v x y sinα]

[0087] 5.2 Model state update

[0088] According to the vehicle's inherent parameters in step 1, based on the principles of vehicle dynamics, the longitudinal acceleration a measured by the sensor can be obtained x , the road ramp angle value α, and the derivative of the longitudinal speed The relationship between them is

[0089]

[0090] Then, within the signal sampling time interval Δt, the state transition equation of the system can be written as

[0091]

[0092] a x (k) = a x (k - 1) + w 2 (k - 1)

[0093] sinα(k) = sinα(k - 1) + w 3 (k - 1)

[0094] where w 1 (k - 1), w 2 (k - 1), w 3 (k - 1) are the process noises of the three state vectors respectively, and can be set as Gaussian white noises

[0095] (1) Based on the optimal estimation of the system state at time k - 1 Predict the predicted value of the system state at the current time

[0096]

[0097] Among them, the process matrix is set as The process noise matrix is expressed as

[0098] (2) Calculate the covariance matrix of the prediction process The calculation formula is

[0099]

[0100] Among them, Q is the covariance matrix of the process noise

[0101] 5.3 State measurement update

[0102] In this example, mainly for the longitudinal acceleration a of the vehicle x , the longitudinal speed v x Construct the measurement vector z k , so the observation equation can be written as

[0103] z k= H·x k + v k

[0104] where v k is the measurement noise matrix and can be set as Gaussian white noise; is the measurement matrix.

[0105] (1) Calculate the Kalman gain, and the calculation formula is:

[0106]

[0107] where H is the measurement matrix; R is the covariance matrix of the measurement noise v k

[0108] (2) Perform the state update, and the calculation formula is:

[0109]

[0110] where z k represents the measured quantity.

[0111] (3) Update the covariance matrix P k of the measurement process, and the update formula is:

[0112]

[0113] Output the state of the vehicle, which includes the estimated value of the road ramp angle Complete the estimation of the road ramp angle, and at the same time output the state of the vehicle and the covariance matrix P k to step 5.2 for loop execution.

[0114] The above description is only a description of the preferred embodiments of the present application and does not limit the scope of the present application in any way. Any changes or modifications made by any person skilled in the art based on the disclosed technical content shall be regarded as equivalent effective embodiments and fall within the scope of the technical solutions protected by the present application.​

Claims

1. A vehicle state estimation method based on adaptive total variation filtering, characterized in that, the steps are as follows: Step 1: Acquisition and preprocessing of vehicle original signals; Step 2: Noise level evaluation; Step 3: Teager-Kaiser energy evaluation; Step 4: Construction of an optimization problem; Step 5: Using the filtered signal output from Step 4 and applying it to the estimation of vehicle states; The said Step 1: First, for the typical working conditions of vehicle tests, sensors are used to collect data on vehicle speed, acceleration, wheel speed, yaw rate, and motor torque. At the same time, the first-order difference f of signal p is solved, and the solution formula is f 1 The solution formula is f 1 (n) = |p(n) - p(n - 1)|, providing a data source for subsequent steps 2, 3, and 4; The said Step 2 includes: 2.1 Noise level evaluation and solution of noise adaptive parameters According to the first-order difference f of the signal in step 1 1 , perform the calculation of the second-order difference f 2 of the signal. The calculation formula is f 2 (n) = |f 1 (n) - f 1 (n - 1)|. Based on the second-order difference f 2 of the signal, evaluate the noise level in the signal. The evaluation formula is Obtain the noise adaptation parameter K n , and provide it to the weight coefficient of the smoothness constraint term in the optimization problem of step 4; 2.2 Noise index threshold setting According to the second-order difference f of the signal in step 2.1 2 , use I n = AVG(|f 2 |) to solve and obtain the noise index I n ; set the threshold ε. When the noise index I of the signal n is greater than the threshold ε, it is necessary to perform arithmetic mean filtering on the signal and then re-execute step 2.1; if the noise index of the signal is less than the threshold ε, directly proceed to step 3; The said Step 3: For the signal obtained in step 2 after noise level evaluation and noise threshold ε, the Teager-Kaiser energy evaluation method is applied to comprehensively evaluate the amplitude effect and instantaneous frequency of the signal, and an energy adaptive parameter is constructed according to the Teager-Kaiser energy value TKE(f 1 ) also provides a weight parameter for the signal smoothness constraint term in the optimization problem of step 4; The said Step 4: According to the first-order difference f of the signal in step 1 1 , the noise adaptation parameter K in step 2 n and the energy adaptation parameter K in step 3 e (n) construct an optimization problem, and obtain the filtered signal by solving the optimization problem: where y is the filtered signal; K is the regularization parameter; the first term in the formula is the signal fidelity penalty, and the second term is the signal smoothness penalty. The regularization parameter K is used to achieve the trade-off between signal fidelity and signal smoothness.

Images