Self-optimizing string-level AEB control method based on multiple extended state observer

Abstract

The application discloses a self-optimization string-level AEB control method based on a multi-expansion state observer, which comprises the following steps: step 1, a double-closed-loop AEB controller based on rolling planning is designed, the double-closed-loop AEB controller comprises an upper controller for dynamically updating a target deceleration a rel and a lower controller for tracking the target deceleration a rel ; the lower controller comprises feedforward control based on a simplified hydraulic system model and feedback control based on an ADRC controller, brake pressure u calculated by the ADRC controller is added to brake pressure P mcp calculated by the feedforward control to obtain P, and P is used as a control quantity and an input of an LSTM network in step 2; step 2, a three-dimensional classifier based on the LSTM network is designed, the LSTM network is used to process the output of a three-axis attitude angle ESO and vehicle data, and classification of road conditions, loads and braking system performance is realized, and the classification result is applied to the double-closed-loop AEB controller in step 1. The application can improve the control precision and stability of the system.

Description

Technical Field

[0001] This invention relates to the field of vehicle active safety technology, and in particular to a self-optimizing cascade AEB (Automatic Emergency Braking) control method based on multiple extended state observers. Background Technology

[0002] With increasing road traffic volume, rear-end collisions are becoming more frequent. Existing Automatic Emergency Braking (AEB) systems rely on sensors to detect hazardous situations and automatically trigger braking. However, traditional AEB systems primarily employ layered collision avoidance strategies and are mostly open-loop control modes. These systems face significant challenges under complex operating conditions and with uncertainties in braking performance, affecting the effectiveness and stability of AEB systems. Summary of the Invention

[0003] The purpose of this invention is to address the technical deficiencies in the existing technology by providing a self-optimizing cascade AEB control method based on multiple extended state observers.

[0004] The technical solution adopted to achieve the purpose of this invention is:

[0005] A self-optimizing cascade AEB control method based on multiple extended state observers includes the following steps:

[0006] Step 1: Design a dual-loop AEB controller based on rolling planning, which includes an upper-level controller and a lower-level controller:

[0007] The upper-level controller is based on a second-order TTC model. It uses the second-order TTC model combined with kinematic principles and the target stopping distance to determine the triggering timing and dynamically update the target deceleration 'a'. rel ;

[0008] The lower-level controller tracks the target deceleration a rel The lower-level controller includes feedforward control based on a simplified hydraulic system model and feedback control based on an ADRC controller, wherein the simplified hydraulic system model derives the target deceleration a. rel Corresponding braking pressure P mcp In the ADRC controller, the target deceleration 'a' output from the upper controller is... rel As input, the braking pressure u serves as the control output;

[0009] The braking pressure u calculated by the ADRC controller and the braking pressure P calculated by the feedforward control. mcp The summation yields P, which serves as both a control variable and an input to the LSTM network in step 2.

[0010] Step 2: Establish the three-axis attitude angle motion equations of the vehicle, and observe the changes in the three-axis attitude angles of the vehicle through the three-axis attitude angle ESO. Design a three-dimensional classifier based on a long short-term memory network (LSTM network), and use the LSTM network to process the output of the three-axis attitude angle ESO and vehicle data to classify the road adhesion coefficient μ, vehicle weight m and braking system performance. The classification results are applied to the dual closed-loop AEB controller in Step 1.

[0011] In the above technical solution, in step 1, the second-order TTC model is:

[0012]

[0013] Where: TTC is the collision time, v rel a rel a lat Let x represent the speed of the vehicle in front relative to this vehicle, the target deceleration, and the lateral acceleration, respectively. A y A The x-coordinate represents the target vehicle's coordinates. B y B This indicates the coordinates of the vehicle.

[0014] In the above technical solution, in step 1, the maximum relative distance s that can trigger AEB is determined based on kinematic principles and the target parking distance. lim for:

[0015]

[0016] Among them, a lim d represents the maximum deceleration that can be provided, and d represents the target stopping distance.

[0017] In the above technical solution, the simplified hydraulic system model in step 1 is as follows:

[0018] P mcp =(F out -F cs -F cf ) / A mc ;

[0019] Among them, P mcp For braking pressure, F out F is the output force of the booster. cs Main cylinder spring preload, F cf For sealing friction, A mc Main cylinder area.

[0020] In the above technical solution, in the feedforward control of step 1, the target deceleration a rel Through braking torque τ bp Converted into its corresponding braking pressure Pmcp :

[0021]

[0022] In the formula, m represents the vehicle mass, μ represents the road adhesion coefficient, and τ bp For the braking torque, m and μ are dynamically updated by the LSTM network in step 2;

[0023] Braking torque τ bp Represented as:

[0024]

[0025] P wp =P wp (V p );

[0026] V p =x mcp A mc ;

[0027]

[0028] Among them, P wp For the wheel cylinder pressure, P pop As the initial pressure, x mcp For displacement, K bp For braking gain, σ p =sgn(P mcp -P wp ), C qp This is the flow coefficient.

[0029] In the above technical solution, in the feedback control of step 1, the actual braking pressure P act and set pressure P set The relationship between them is:

[0030]

[0031] Where τ is the time constant of the first-order inertial element, and P act P is the actual braking pressure. set The set pressure is the braking pressure u output by the ADRC controller.

[0032] In the above technical solution, the feedback control in step 1 is based on the following model:

[0033]

[0034] Where k is the total proportionality constant, τ is the time constant of the first-order inertial element, u represents the braking pressure, and f represents the total disturbance. k is dynamically corrected by the LSTM network in step 2, as observed by the ADRC controller.

[0035] Let a be the state variable, a = a rel Let u be the control variable, and let x1 = a.

[0036]

[0037] in, k is the overall proportionality constant, τ is the time constant of the first-order inertial element, m represents the vehicle mass, β1 = 2ω0, β2 = ω0 2 ω0 is the observer bandwidth.

[0038] In the above technical solution, the total output of the ADRC controller in step 1 is:

[0039]

[0040] Where, k p These are controller parameters, a res It is the target deceleration input from the upper-level controller, a res =a rel u is the braking pressure output by the ADRC controller; f is the total disturbance observed by the ADRC controller.

[0041] In the above technical solution, in step 2, the three-axis attitude angle motion equation is:

[0042]

[0043] Among them, ψ, θ, These represent yaw angle, pitch angle, and roll angle, respectively. F1 to F8 correspond to the lateral and longitudinal forces on the front and rear axles, respectively. x Let l be the moment of inertia about the x-axis. y Let I be the moment of inertia about the y-axis. z Let l be the moment of inertia about the z-axis. f The distance from the positioning center to the front axle, l r The distance from the positioning center to the rear axle, l wf The lever arm length of the front suspension, l wr This is the lever arm length of the rear suspension.

[0044] In the above technical solution, in step 2, each attitude angle ψ, θ or As state variables, establish the three-axis attitude angle ESO:

[0045]

[0046] In the yaw angle ESO, u=l f F1-l r F2, pitch angle ESO u=l r F4-l r F3, roll angle ESO u=l wf (F5-F6)+l wr (F7-F8), β1=3ω0, β2=3ω0 2 β3=ω0 3 ω0 is the observer bandwidth.

[0047] In the above technical solution, in step 2, the LSTM network includes a ReLU layer and a Dropout layer, mathematically expressed as:

[0048]

[0049] h t =o t *tanh(C t )

[0050]

[0051] y t =W y h t +b y

[0052] x t It is the input matrix, including the three-axis attitude angles ψ, θ, Target deceleration a rel Actual deceleration a, braking pressure u calculated by ADRC controller and braking pressure P calculated by feedforward control mcp The sum of P, the currently used μ, m, k, and the total disturbance f observed by the ADRC controller, and the three disturbances f in the three-axis attitude angle ESO;

[0053] y t It is the output matrix, including the updated μ, m, and k.

[0054] Compared with the prior art, the beneficial effects of the present invention are:

[0055] 1. This algorithm utilizes multiple ESOs to observe the changes in the vehicle's three-axis attitude angles, classifies the vehicle's operating state through an LSTM network, and dynamically adjusts the cascaded dual-loop AEB controller to improve the stability and accuracy of the AEB system.

[0056] 2. Through multi-parameter learning and real-time feedback, online correction of vehicle weight m, braking system performance, and road surface adhesion coefficient μ is achieved, thereby improving the reliability of the braking system under complex working conditions.

[0057] 3. Experiments have shown that the algorithm of this invention can effectively reduce stopping distance error and improve the control accuracy and stability of the system under different road conditions and vehicle states. Attached Figure Description

[0058] Figure 1 The diagram shown is a schematic of the algorithm framework.

[0059] Figure 2 The diagram shown is a model of the hydraulic system.

[0060] Figure 3 Simplified diagram of master cylinder connection;

[0061] Figure 4 The diagram shows the structure of an LSTM network. Detailed Implementation

[0062] The present invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0063] like Figure 1 As shown, the self-optimizing cascade automatic emergency braking (AEB) control method based on multiple extended state observers includes the following steps:

[0064] Step 1: Design a dual-loop AEB controller based on rolling planning. The dual-loop AEB controller includes an upper-level controller and a lower-level controller. The upper-level controller plans the target deceleration, and the lower-level controller tracks the target deceleration.

[0065] The upper-level controller is based on a second-order collision time model (second-order TTC model). It uses the second-order TTC model in combination with kinematic principles and the target stopping distance to determine the triggering timing and dynamically update the target deceleration.

[0066] The second-order TTC model is:

[0067]

[0068] Where: TTC is the collision time. This paper sets TTC less than 6s as one of the conditions for triggering AEB. rel a rel a lat Let x represent the velocity, acceleration (target deceleration), and lateral acceleration of the vehicle in front relative to this vehicle, respectively. A y A The x-coordinate represents the target vehicle's coordinates. B yB This indicates the coordinates of the vehicle.

[0069] Based on kinematic principles and the target parking distance, the maximum relative distance s that can trigger AEB (Automatic Emergency Braking) lim Defined as:

[0070]

[0071] Among them, a lim d represents the maximum deceleration that can be provided, and d represents the target stopping distance.

[0072] The lower-level controller includes feedforward control based on a simplified hydraulic system model and feedback control based on an ADRC controller.

[0073] Hydraulic system model diagram as follows Figure 2 As shown, the target deceleration a is derived using a simplified hydraulic system model. rel Corresponding braking pressure P mcp .like Figure 3 As shown, the simplified hydraulic system model is a single equivalent brake connected to the master cylinder. The braking pressure P is calculated. mcp as follows:

[0074] P mcp =(F out -F cs -F cf ) / A mc ;

[0075] Among them, F out F is the output force of the booster. cs Main cylinder spring preload, F cf For sealing friction, A mc Main cylinder area.

[0076] Braking torque τ bp Represented as:

[0077]

[0078] P wp =P wp (V p );

[0079] V p =x mcp A mc ;

[0080]

[0081] Among them, P wp For the wheel cylinder pressure, P pop As the initial pressure, x mapFor displacement, K bp For braking gain, σ p =sgn(P mcp -P wp ), C qp Flow coefficient

[0082] The relationship between target deceleration and braking pressure:

[0083]

[0084] In the formula, m represents the vehicle mass, μ represents the road surface adhesion coefficient, and m and μ are dynamically corrected by the LSTM network in step 2.

[0085] In feedback control, the target deceleration 'a' output from the upper controller is... rel As input, the braking pressure u serves as the control output. Specifically, the changes in the vehicle's three-axis attitude angles are observed through an extended state observer (ESO), and combined with the active disturbance rejection control (ADRC) method, precise braking control is achieved.

[0086] The braking system's operation is described by a first-order inertial element, with the actual braking pressure P... act and set pressure P set The relationship between them is:

[0087]

[0088] Where τ is the time constant of the first-order inertial element, and P act P is the actual braking pressure. set The set pressure is the controller's output (braking pressure u);

[0089] Combining the feedforward control model and the braking system operation process model, the model for feedback control is obtained as follows:

[0090]

[0091] Where k is the total proportional constant, uP represents the braking pressure, f represents the total disturbance observed by the ADRC controller, and k is dynamically corrected by the LSTM network in step 2.

[0092] Let a be the state variable, a = a rel Let u be the control variable, and let x1 = a.

[0093]

[0094] in, β1=2ω0,β2=ω0 2 ω0 is the observer bandwidth.

[0095] The total output of the ADRC controller is:

[0096]

[0097] Where, k p These are controller parameters, a res It is the target deceleration input from the upper-level controller, a res =a rel u is the braking pressure output by ADRC.

[0098] The braking pressure u calculated by the ADRC controller and the braking pressure P calculated by the feedforward control. mcp The sum P serves as both a control variable and an input to the LSTM network.

[0099] Step 2: Establish the three-axis attitude angle motion equations of the vehicle, and observe the changes in the three-axis attitude angles of the vehicle through the three-axis attitude angle ESO. Design a three-dimensional classifier based on a long short-term memory network (LSTM network), and use the LSTM network to process the output of the three-axis attitude angle ESO and vehicle data to classify the road adhesion coefficient μ, vehicle weight m and braking system performance. The classification results are applied to the dual closed-loop AEB controller in Step 1.

[0100] The establishment of the three-axis attitude angle ESO includes the following steps:

[0101] Assuming the vehicle is initially symmetrical and subjected to random road excitation, it undergoes minor vibrations around its equilibrium position. The effects of vehicle vibration and air resistance on the vehicle's motion are neglected. The three-axis attitude angle motion equations are:

[0102]

[0103] Among them, ψ, θ, These represent yaw angle, pitch angle, and roll angle, respectively. F1 to F8 correspond to the lateral and longitudinal forces on the front and rear axles, respectively. x Let I be the moment of inertia about the x-axis. y Let I be the moment of inertia about the y-axis. z Let l be the moment of inertia about the z-axis. f The distance from the positioning center to the front axle, l r The distance from the positioning center to the rear axle, l wf The lever arm length of the front suspension, l wr This is the lever arm length of the rear suspension.

[0104] For each attitude angle ψ, θ or Let x1 be the state variable. Establish three-order ESOs respectively and observe the changes in attitude angles:

[0105]

[0106] In the yaw angle ESO, u=l f F1-l r F2, pitch angle ESO u=l r F4-l r F3, roll angle ESO u=l wf (F5-F6)+l wr (F7-F8), β1=3ω0, β2=3ω0 2 β3=ω0 3 ω0 is the observer bandwidth, and f is the perturbation of the three-axis attitude angle ESO.

[0107] like Figure 4 As shown, the LSTM network contains ReLU layers and Dropout layers, which enhance the generalization ability and accuracy of the classifier;

[0108] The mathematical expression for an LSTM network is:

[0109]

[0110] h t =o t *tanh(C t )

[0111]

[0112] y t =W y h t +b y

[0113] x t It is the input matrix, including the three-axis attitude angles ψ, θ, Target deceleration a rel Actual deceleration a, braking pressure u calculated by ADRC controller and braking pressure P calculated by feedforward control mcp The sum of P, the currently used μ, m, k, and the total disturbance f in the ADRC controller, and the three f in the three-axis attitude angle ESO (a total of 13 features);

[0114] y t It is the output matrix, including the updated μ, m, and k.

[0115] The classification performance of the network is improved by using ReLU activation layers and Dropout layers; a training dataset containing 13 features is used, with a sampling time interval of 0.0005s and a sequence duration of 1.5 seconds.

[0116] Step 3, Simulation test under different working conditions: The stopping distance error of the algorithm is verified to be less than 3.5cm through simulation test under different working conditions. Compared with the standard algorithm, the system stability is improved by 95.1%.

[0117] The control accuracy was tested under various road conditions, vehicle loads, and braking system performance. The stopping distance error and control error distribution were compared with and without the integration of the learning module of this invention. The AEB braking distance at each speed was statistically analyzed to ensure that the stopping distance error was less than 3.5cm and the system stability was improved by 95.1%.

[0118] The above description is only a preferred embodiment of the present invention. It should be noted that, for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A self-optimizing cascade AEB control method based on multiple extended state observers, characterized in that, Includes the following steps: Step 1: Design a dual-loop AEB controller based on rolling planning, which includes an upper-level controller and a lower-level controller: The upper-level controller is based on a second-order TTC model. It uses the second-order TTC model combined with kinematic principles and the target stopping distance to determine the triggering timing and dynamically update the target deceleration. ; The lower-level controller tracks the target deceleration. The lower-level controller includes feedforward control based on a simplified hydraulic system model and feedback control based on an ADRC controller, wherein the simplified hydraulic system model is as follows: ; Braking pressure, For the output force of the booster, The preload of the main cylinder spring. To seal friction, The main cylinder area; the target deceleration is derived from a simplified hydraulic system model. Corresponding braking pressure In the ADRC controller, the target deceleration output from the upper controller is... As input, braking pressure As a control output; In feedforward control, target deceleration Through braking torque Converted into its corresponding braking pressure : In the formula, Indicates vehicle mass. Indicates the road surface adhesion coefficient. For braking torque, and The LSTM network from step 2 is used for dynamic updates; Braking torque Represented as: ； ； ； ； in, For the wheel cylinder pressure, As the initial pressure, For displacement, For braking gain, , This is the flow coefficient. Braking pressure calculated by the ADRC controller Braking pressure calculated with feedforward control Sum of P , P It serves both as a control variable and as an input to the LSTM network in step 2; Step 2: Establish the three-axis attitude angle motion equations of the vehicle, and observe the changes in the three-axis attitude angles of the vehicle through the three-axis attitude angle ESO. Design a three-dimensional classifier based on an LSTM network, and use the LSTM network to process the output of the three-axis attitude angle ESO and the vehicle data to realize the road adhesion coefficient. Vehicle weight The classification of braking system performance is applied to the dual closed-loop AEB controller in step 1.

2. The self-optimizing cascade AEB control method as described in claim 1, characterized in that, In step 1, the second-order TTC model is: ； in: For the collision time, , , These represent the speed of the vehicle in front relative to this vehicle, the target deceleration, and the lateral acceleration, respectively. , Indicates the coordinates of the target vehicle. , This indicates the coordinates of the vehicle.

3. The self-optimizing cascade AEB control method as described in claim 1, characterized in that, In step 1, based on kinematic principles and the target parking distance, the maximum relative distance that can trigger AEB (Automatic Emergency Braking) is determined. for: in, Indicates the maximum deceleration that can be provided. Indicates the target parking distance.

4. The self-optimizing cascade AEB control method as described in claim 1, characterized in that, In the feedback control of step 1, the actual braking pressure and setting pressure The relationship between them is: ； in, It is the time constant of a first-order inertial element. This is the actual braking pressure. The set pressure is the braking pressure output by the ADRC controller. .

5. The self-optimizing cascade AEB control method as described in claim 1, characterized in that, The feedback control in step 1 is based on the following model: in, It is the overall proportionality constant. It is the time constant of a first-order inertial element. Indicates braking pressure. This represents the total disturbance, observed by the ADRC controller. Dynamic correction is achieved through the LSTM network in step 2; by For state variables, , To control the quantity, let ,but in, , It is the overall proportionality constant. It is the time constant of a first-order inertial element. Indicates vehicle mass. , , It is the observer bandwidth.

6. The self-optimizing cascade AEB control method as described in claim 1, characterized in that, The total output of the ADRC controller in step 1 is: in, These are controller parameters. It is the target deceleration input from the upper-level controller. , The braking pressure output by the ADRC controller; The total disturbance observed by the ADRC controller.

7. The self-optimizing cascade AEB control method as described in claim 1, characterized in that, In step 2, the three-axis attitude angle motion equations are: in, , , These represent the yaw angle, pitch angle, and roll angle, respectively. These correspond to the lateral and longitudinal forces on the front and rear axles, respectively. Let x be the moment of inertia about the x-axis. The moment of inertia about the y-axis The moment of inertia about the z-axis, The distance from the positioning center to the front axle, The distance from the positioning center to the rear axle, For the lever arm length of the front suspension, This is the lever arm length of the rear suspension.

8. The self-optimizing cascade AEB control method as described in claim 1, characterized in that, In step 2, for each attitude angle , or As state variables, establish the three-axis attitude angle ESO: In the yaw angle ESO, , In pitch angle ESO, , In the roll angle ESO, , , , , , It is the observer bandwidth.

9. The self-optimizing cascade AEB control method as described in claim 1, characterized in that, In step 2, the LSTM network includes a ReLU layer and a Dropout layer, mathematically expressed as: It is the input matrix, including the three-axis attitude angles. , , Target deceleration Actual deceleration 'a', braking pressure calculated by ADRC controller Braking pressure calculated with feedforward control sum P Currently in use , , k and the total disturbance observed by the ADRC controller Three perturbations in the three-axis attitude angle ESO ; It is the output matrix, including the updated one. , , k .

Images