Self-adaptive telescopic control system applied to driving wheel of electro-tricycle

Abstract

The invention relates to the technical field of telescopic control, in particular to a self-adaptive telescopic control system applied to a driving wheel of an electric tricycle. Comprising the steps that a tricycle attitude extraction module obtains data of a tricycle data acquisition module and processes the data to obtain a turning attitude state matrix, and the turning attitude state matrix comprises a first state vector, a second state vector, a transverse state vector and a third state vector; the suspension parameter generation module comprises an observation state space unit and an RNN-TD3 model unit, and the observation state space unit obtains information of the static data acquisition unit and information of the tricycle attitude extraction module and processes the information to obtain an observation state space; and the RNN-TD3 model unit obtains and processes an observation state space to obtain a suspension parameter adjustment value vector. The suspension parameter adjusting module obtains suspension parameter adjusting value vectors and controls the suspension parameter adjusting value vectors through an incremental PID algorithm, and stable adjustment of self-adaptive telescopic control over the driving wheels of the electro-tricycle is achieved.

Description

Technical Field

[0001] The invention relates to the technical field of telescopic control, and in particular to an adaptive telescopic control system applied to a driving wheel of an electric tricycle. Background Art

[0002] When a tricycle is transporting goods or carrying passengers, the load distribution may be uneven, resulting in changes in the pressure on the drive wheels, affecting the driving force and suspension effects. Especially when making sharp turns or driving on slopes, the center of gravity of the vehicle shifts, which may cause the drive wheels to lose effective contact.

[0003] Due to the three-point support structure, tricycles have poor lateral stability when turning and are prone to rollover risks. If the suspension system of the drive wheel is not adjusted in time, this risk may be exacerbated. When a tricycle turns, by adjusting the suspension height of the drive wheel, lowering the center of gravity of the vehicle and increasing the support force of the outer wheels, and performing real-time monitoring of the vehicle's lateral state combined with suspension control, rollover accidents can be avoided.

[0004] Therefore, an adaptive telescopic control system applied to the driving wheel of an electric tricycle is proposed. Summary of the invention

[0005] The purpose of the present invention is to provide an adaptive telescopic control system applied to the driving wheel of an electric tricycle to improve the stability and maneuverability of the electric tricycle when turning. First, the data of the tricycle data acquisition module is obtained and processed by the tricycle posture extraction module, and a turning posture state matrix is ​​constructed to measure the state of the electric tricycle when turning. The suspension parameter generation module processes the static data and the turning posture state matrix of the tricycle through the RNN-TD3 model to obtain the suspension parameter adjustment value vector. The suspension parameter adjustment module adjusts the parameters smoothly through the incremental PID algorithm, and finally realizes the adaptive telescopic control of the driving wheel of the electric tricycle when turning.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] An adaptive telescopic control system applied to a driving wheel of an electric tricycle, comprising:

[0008] A tricycle data collection module, used to collect tricycle data, including a static data collection unit and a dynamic data collection unit;

[0009] Furthermore, the static data acquisition unit includes:

[0010] A static data sensor is installed on the electric tricycle to collect static data, and the static data includes the length, width and weight of the vehicle body; the static data is obtained and pre-processed and integrated to obtain a static data vector F1.

[0011] Furthermore, the dynamic data acquisition unit comprises:

[0012] A dynamic data sensor is installed on the electric tricycle to collect dynamic data, and the dynamic data includes front wheel collected data, left wheel collected data and right wheel collected data;

[0013] The front wheel collected data includes the front wheel angle, front wheel load, front wheel speed and front wheel friction, wherein the front wheel angle is the rotation angle of the front wheel monitored in real time by the angle sensor, the front wheel load is the load change of the front wheel monitored in real time by the load sensor, and the front wheel friction is the friction data between the front wheel and the ground obtained by the traction sensor;

[0014] The front wheel data is acquired, preprocessed and integrated to obtain the front wheel data vector F2.

[0015] Furthermore, the left wheel collected data and the right wheel collected data include:

[0016] The left wheel collected data includes left wheel speed, left wheel load, left wheel suspension travel, left wheel vibration data, left wheel traction, left wheel braking force and left tire pressure; the left wheel collected data is obtained and pre-processed and integrated to obtain the left wheel data vector F3;

[0017] The right wheel collected data includes right wheel speed, right wheel load, right wheel suspension travel, right wheel vibration data, right wheel traction, right wheel braking force and right tire pressure; the right wheel collected data is obtained and pre-processed and integrated to obtain the right wheel data vector F4;

[0018] The left wheel data vector and the right wheel data vector have the same data format.

[0019] A tricycle posture extraction module is used to obtain and process the data of the dynamic data acquisition unit to obtain a turning posture state matrix; the turning posture state matrix includes a first state vector, a second state vector, a lateral state vector and a third state vector, the first state vector is obtained according to a front wheel data vector and a left wheel data vector, the second state vector is obtained according to a front wheel data vector and a right wheel data vector, the lateral state vector is obtained according to a left wheel data vector and a right wheel data vector, and the third state vector is obtained according to power data of the electric tricycle;

[0020] Furthermore, the specific steps of constructing the turning posture state matrix of the electric tricycle according to the front wheel data vector, the left wheel data vector and the right wheel data vector include:

[0021] Constructing the first state vector f1 according to the front wheel data vector and the left wheel data vector, the first state vector including first state data and second state data, the first state data including data of different properties, and the second state data including data of the same property;

[0022] Constructing the second state vector f2 according to the front wheel data vector and the right wheel data vector, the second state vector comprising third state data and fourth state data, the third state data comprising data of different properties, and the fourth state data comprising data of the same property;

[0023] Constructing the lateral state vector f3 according to the difference between the left wheel data vector and the right wheel data vector;

[0024] Acquiring and processing power data of the electric tricycle to obtain the third state vector f4, wherein the power data includes the output power of the motor;

[0025] The first state vector, the second state vector, the lateral state vector and the third state vector are acquired and integrated to obtain the turning posture state matrix of the electric tricycle.

[0026] A suspension parameter generation module is used to generate a suspension parameter adjustment value vector in real time according to an observed state space, and includes an observed state space unit and an RNN-TD3 model unit; wherein the observed state space unit obtains data from the static data acquisition unit and data from the tricycle posture extraction module, and integrates and processes the data to obtain the observed state space; the RNN-TD3 model unit obtains the observed state space and processes the data through the RNN-TD3 model to obtain the suspension parameter adjustment value vector, and the suspension parameter adjustment value vector includes suspension stiffness and suspension travel;

[0027] Furthermore, the observation state space unit includes:

[0028] Obtain a static data vector and the turning posture state matrix and integrate them to obtain a first observation state space, obtain data of the suspension parameter adjustment value vector of the electric tricycle at the previous moment to obtain a second observation state space, obtain data of the first observation state space and the second observation state space and integrate them to obtain the observation state space.

[0029] Furthermore, the RNN-TD3 model includes:

[0030] A first action network and a first target action network, wherein the first action network and the first target action network have the same action network structure, and the action network structure includes an input layer, an RNN layer, a fully connected layer, an action normalization layer, and an output layer;

[0031] A first evaluation network, a first target evaluation network, a second evaluation network and a second target evaluation network, wherein the first evaluation network, the first target evaluation network, the second evaluation network and the second target evaluation network have the same evaluation network structure, and the evaluation network structure includes an input layer, an RNN layer, a fully connected layer and an output layer.

[0032] Furthermore, the training process of the RNN-TD3 model includes:

[0033] Step S1: Initialize the entire network;

[0034] Step S2: inputting the observed state space and the historical observed state space into the first action network, outputting the suspension parameter adjustment value vector, and obtaining the suspension parameter reward and the next state of the electric tricycle corresponding to the suspension parameter adjustment value vector; constructing a training sample according to the observed state space, the suspension parameter adjustment value vector, the reward and the next state, and storing it in the experience replay pool;

[0035] Step S3: sampling training samples from the experience replay pool, and updating the parameters of the first evaluation network and the second evaluation network according to the training samples;

[0036] Step S4: after the parameters of the first evaluation network and the second evaluation network are updated at intervals of D times, the parameters of the first action network are updated;

[0037] Step S5: updating the parameters of the first target action network, the first target evaluation network and the second target evaluation network according to the soft update rule;

[0038] Step S6: Repeat steps S3 to S5 until a training round is reached.

[0039] Furthermore, the suspension parameter reward includes:

[0040] R=ω1·(-(|θ′|+|θ″|))+ω2·(-|ΔG|)+ω3·(-E)+ω4·(-T);

[0041] Wherein, R represents the suspension parameter reward, ω1 represents the weight coefficient of stability, θ′ represents the lateral inclination angle of the electric tricycle, θ″ represents the longitudinal inclination angle of the electric tricycle, ω2 represents the weight coefficient of the center of gravity offset ΔG, ω3 represents the weight coefficient of the suspension system adjustment energy consumption E, and ω4 represents the weight coefficient of the suspension system response time T.

[0042] The suspension parameter adjustment module is used to obtain the suspension parameter adjustment value vector and control it through an incremental PID algorithm.

[0043] Furthermore, the specific steps of controlling by the incremental PID algorithm include:

[0044] The suspension parameter adjustment value vector ΔE output by the RNN-TD3 model unit is obtained and the parameters are adjusted according to the incremental PID algorithm. The calculation formula is:

[0045]

[0046] in, Indicates the control quantity, K p represents the proportional gain, ΔE(t) represents the suspension parameter adjustment value vector at time t, ΔE(t-1) represents the suspension parameter adjustment value vector at time t-1, ΔE(t-2) represents the suspension parameter adjustment value vector at time t-2, K i Indicates the integral gain, K d Represents the differential gain.

[0047] Compared with the prior art, the present invention has the following beneficial effects:

[0048] 1. The first state vector and the second state vector are combined with the data of the front wheel and the left wheel, and the front wheel and the right wheel respectively, which can not only capture the coordinated relationship between the front wheel and the driving wheel, but also distinguish the dynamic characteristics of the left and right sides; the lateral state vector directly reflects the lateral state and lateral stability characteristics during turning by calculating the data difference between the left and right driving wheels, providing an important basis for optimizing the turning posture; after combining the third state vector of the power data, the turning posture state matrix can comprehensively evaluate the coupling relationship between the drive system and the turning motion, realizing the comprehensive monitoring and quantification of various key characteristics of the tricycle during the turning process.

[0049] 2. The RNN-TD3 model can make full use of the historical state sequence information of the electric tricycle in the turning scene provided by the observation state space unit, such as vibration and body posture changes, so that the suspension parameter adjustment is more in line with the actual state; the reward function integrates multiple indicators such as lateral inclination, longitudinal inclination, center of gravity offset, energy consumption and response time, so that the model can take into account energy consumption and real-time performance while improving the performance of the suspension system.

[0050] 3. By controlling the suspension parameter adjustment value vector output by the RNN-TD3 model through the incremental PID algorithm, the real-time, stability, energy efficiency optimization and driving comfort of the electric tricycle suspension system can be comprehensively improved, the robustness and adaptability of the control system can be improved, and the computational complexity and system resource requirements can be reduced. Efficient and stable telescopic control of the drive wheel can be achieved under limited hardware resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1A schematic diagram of the structure of an adaptive telescopic control system applied to a driving wheel of an electric tricycle provided by an embodiment of the present invention;

[0052] Figure 2 A flowchart of calculating a turning posture state matrix provided by an embodiment of the present invention;

[0053] Figure 3 A flow chart for calculating a suspension parameter adjustment value vector provided by an embodiment of the present invention;

[0054] Figure 4 A schematic diagram of the structure of an action network structure provided by an embodiment of the present invention;

[0055] Figure 5 A schematic diagram of the structure of an evaluation network provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] Embodiment 1

[0058] A tricycle manufacturing company has introduced an adaptive telescopic control system for driving wheels of electric tricycles to improve the operability and safety of electric tricycles. The system structure is as follows: Figure 1 As shown, the specific implementation is as follows:

[0059] A tricycle data collection module, used to collect tricycle data, including a static data collection unit and a dynamic data collection unit;

[0060] Furthermore, the static data acquisition unit includes:

[0061] Static data sensors, including distance measuring sensors and load sensors, are installed on the electric tricycle to collect static data, including body length, body width, body weight, etc. The static data is obtained and preprocessed and integrated to obtain a static data vector F1.

[0062] Through the static data acquisition unit, the adaptive telescopic control system of the electric tricycle can obtain the basic structural parameters of the vehicle in real time, provide reliable data and initial conditions for subsequent suspension adjustments and intelligent decision-making, and further significantly improve the accuracy, real-time and environmental adaptability of the suspension system.

[0063] Furthermore, the dynamic data acquisition unit comprises:

[0064] A dynamic data sensor is installed on the electric tricycle to collect dynamic data, and the dynamic data includes front wheel collected data, left wheel collected data and right wheel collected data;

[0065] The front wheel collected data includes the front wheel angle, front wheel load, front wheel speed and front wheel friction, wherein the front wheel angle is the rotation angle of the front wheel monitored in real time by the angle sensor, the front wheel load is the load change of the front wheel monitored in real time by the load sensor, and the front wheel friction is the friction data between the front wheel and the ground obtained by the traction sensor;

[0066] The front wheel data is acquired and preprocessed and integrated to obtain the front wheel data vector F2. Table 1 shows part of the collected front wheel data, where the positive value of the front wheel turning angle is the right deflection angle, and the negative value is the left deflection angle.

[0067] Table 1. Some collected front wheel data

[0068] Number of collections Front wheel angle Front wheel load Front wheel speed No. 129 -13.77° 102.5kg 56.28 rpm 130th -12.12° 105.6kg 59.12 rpm No. 131 -6.56° 101.9kg 61.58 rpm No. 132 -2.45° 108.3kg 65.29 rpm No. 133 -1.29° 105.2kg 76.23 rpm

[0069] The dynamic data acquisition unit collects and processes the front wheel data in real time, providing accurate and real-time input data for the suspension system, ensuring that the suspension system can respond quickly under various road conditions and cornering conditions.

[0070] Furthermore, the left wheel collected data and the right wheel collected data include:

[0071] The left wheel collected data includes left wheel speed, left wheel load, left wheel suspension travel, left wheel vibration data, left wheel traction, left wheel braking force and left tire pressure; the left wheel collected data is obtained and pre-processed and integrated to obtain the left wheel data vector F3;

[0072] The right wheel collected data includes right wheel speed, right wheel load, right wheel suspension travel, right wheel vibration data, right wheel traction, right wheel braking force and right tire pressure; the right wheel collected data is obtained and pre-processed and integrated to obtain the right wheel data vector F4;

[0073] Furthermore, the left wheel data vector and the right wheel data vector have the same data format, as shown in Table 2 which is part of the collected left wheel data.

[0074] Table 2. Some left wheel collection data

[0075] Number of collections Left wheel speed Left wheel load Left wheel suspension travel Left wheel traction Left tire pressure No. 129 55.12 rpm 112.3kg 12.23mm 229.2N 301.5kPa 130th 58.34 rpm 115.2kg 10.35mm 230.5N 298.3kPa No. 131 60.38 rpm 108.6kg 8.12mm 215.4N 290.2kPa No. 132 64.19 rpm 111.9kg 5.93mm 212.9N 286.8kPa No. 133 76.02 rpm 108.4kg 3.11mm 210.3N 282.6kPa

[0076] Through high-precision sensors and standardized data processing procedures, and ensuring that the data formats of the left and right wheels remain consistent, detailed and reliable dynamic data support is provided for the suspension control system of the electric tricycle.

[0077] A tricycle posture extraction module is used to obtain and process the data of the dynamic data acquisition unit to obtain a turning posture state matrix; the turning posture state matrix includes a first state vector, a second state vector, a lateral state vector and a third state vector, the first state vector is obtained according to a front wheel data vector and a left wheel data vector, the second state vector is obtained according to a front wheel data vector and a right wheel data vector, the lateral state vector is obtained according to a left wheel data vector and a right wheel data vector, and the third state vector is obtained according to power data of the electric tricycle;

[0078] Further, the specific steps of constructing the turning posture state matrix of the electric tricycle according to the front wheel data vector, the left wheel data vector and the right wheel data vector are as follows: Figure 2 As shown, including:

[0079] Constructing the first state vector f1 according to the front wheel data vector and the left wheel data vector, the first state vector including first state data and second state data, the first state data including data of different properties, and the second state data including data of the same property;

[0080] Furthermore, the first state data includes data of different nature, for example, the front wheel steering angle in the front wheel data vector does not exist in the left wheel data vector, so it is data of different nature and is included in the first state data; the front wheel load data in the front wheel data vector and the left wheel load in the left wheel data vector are data of the same nature, so they can be further processed, such as subtracting or averaging, and included in the second state data.

[0081] Constructing the second state vector f2 according to the front wheel data vector and the right wheel data vector, the second state vector comprising third state data and fourth state data, the third state data comprising data of different properties, and the fourth state data comprising data of the same property;

[0082] Further, the third state data is similar to the first state data, and the fourth state data is similar to the second state data.

[0083] Constructing the lateral state vector f3 according to the difference between the left wheel data vector and the right wheel data vector;

[0084] Furthermore, the left wheel data vector and the right wheel data vector have the same data structure, so the properties of their data are also the same, and further processing can be performed, such as weighted averaging, difference or analysis of time series data to extract features to obtain the lateral state vector.

[0085] Acquiring and processing power data of the electric tricycle to obtain the third state vector f4, wherein the power data includes the output power of the motor;

[0086] The first state vector, the second state vector, the lateral state vector and the third state vector are acquired and integrated to obtain the turning posture state matrix of the electric tricycle.

[0087] The first state vector, the second state vector, the lateral state vector and the third state vector are integrated to form the turning posture state matrix of the electric tricycle, which comprehensively reflects the dynamic state of the tricycle when turning and provides efficient and accurate data support for the suspension control system.

[0088] A suspension parameter generation module is used to generate a suspension parameter adjustment value vector in real time according to an observed state space, and includes an observed state space unit and an RNN-TD3 model unit; wherein the observed state space unit obtains data from the static data acquisition unit and data from the tricycle posture extraction module, and integrates and processes the data to obtain the observed state space; the RNN-TD3 model unit obtains the observed state space and processes the data through the RNN-TD3 model to obtain the suspension parameter adjustment value vector, and the suspension parameter adjustment value vector includes several suspension-related parameters such as suspension stiffness and suspension travel;

[0089] Furthermore, the calculation process of the suspension parameter adjustment value vector is as follows: Figure 3 shown.

[0090] Furthermore, the observation state space unit includes:

[0091] Obtain a static data vector and the turning posture state matrix and integrate them to obtain a first observation state space, obtain data of the suspension parameter adjustment value vector of the electric tricycle at the previous moment to obtain a second observation state space, obtain data of the first observation state space and the second observation state space and integrate them to obtain the observation state space.

[0092] The first observation state space combines the static characteristics and dynamic behaviors of the vehicle, providing comprehensive input data for the adaptive telescopic control of the drive wheels; the second observation state space reflects the historical adjustment effect of the suspension system, providing a basis for the next step of control optimization; through the fusion of the first observation state space and the second observation state space, a comprehensive, accurate and real-time observation state space is constructed.

[0093] Furthermore, the RNN-TD3 model includes:

[0094] A first action network and a first target action network, wherein the first action network and the first target action network have the same action network structure, and the action network structure is as follows: Figure 4 As shown, it includes input layer, RNN layer, fully connected layer, action normalization layer and output layer;

[0095] A first evaluation network, a first target evaluation network, a second evaluation network, and a second target evaluation network, wherein the first evaluation network, the first target evaluation network, the second evaluation network, and the second target evaluation network have the same evaluation network structure, and the evaluation network structure is as follows: Figure 5 As shown, it includes input layer, RNN layer, fully connected layer and output layer.

[0096] The RNN-TD3 model introduces a dual structure of action network and evaluation network, combined with the time series feature extraction capability of recurrent neural network, to construct an efficient, stable and accurate suspension control mechanism. This design significantly improves the adaptive telescopic control capability of the driving wheels of electric tricycles under turning conditions.

[0097] Furthermore, the training process of the RNN-TD3 model includes:

[0098] Step S1: Initialize the entire network;

[0099] Step S2: inputting the observed state space and the historical observed state space into the first action network, outputting the suspension parameter adjustment value vector, and obtaining the suspension parameter reward and the next state of the electric tricycle corresponding to the suspension parameter adjustment value vector; constructing a training sample according to the observed state space, the suspension parameter adjustment value vector, the reward and the next state, and storing it in the experience replay pool;

[0100] Step S3: sampling training samples from the experience replay pool, and updating the parameters of the first evaluation network and the second evaluation network according to the training samples;

[0101] Step S4: after the parameters of the first evaluation network and the second evaluation network are updated at intervals of D times, the parameters of the first action network are updated;

[0102] Step S5: updating the parameters of the first target action network, the first target evaluation network and the second target evaluation network according to the soft update rule;

[0103] Step S6: Repeat steps S3 to S5 until a training round is reached.

[0104] The RNN layer realizes the efficient extraction of time series features, improves training efficiency, and enhances the generalization ability of the model; through the reinforcement learning framework, the model can generate dynamically optimized suspension parameter adjustment strategies according to different driving conditions, effectively reducing the vehicle's lateral inclination, longitudinal inclination and center of gravity offset, and providing precise adaptive telescopic control for the electric tricycle's drive wheels.

[0105] Furthermore, the suspension parameter reward includes:

[0106] R=ω1·(-(|θ′|+|θ″|))+ω2·(-|ΔG|)+ω3·(-E)+ω4·(-T);

[0107] Wherein, R represents the suspension parameter reward, ω1 represents the weight coefficient of stability, θ′ represents the lateral inclination angle of the electric tricycle, θ″ represents the longitudinal inclination angle of the electric tricycle, ω2 represents the weight coefficient of the center of gravity offset ΔG, ω3 represents the weight coefficient of the suspension system adjustment energy consumption E, and ω4 represents the weight coefficient of the suspension system response time T.

[0108] The design of the suspension parameter reward not only focuses on the dynamic stability of the vehicle, but also comprehensively considers factors such as energy consumption and response speed, optimizes the stability of the tricycle, improves the handling of the tricycle, and achieves efficient and energy-saving suspension adjustment.

[0109] The suspension parameter adjustment module is used to obtain the suspension parameter adjustment value vector and control it through an incremental PID algorithm.

[0110] Furthermore, the specific steps of controlling by the incremental PID algorithm include:

[0111] The suspension parameter adjustment value vector ΔE output by the RNN-TD3 model unit is obtained and the parameters are adjusted according to the incremental PID algorithm. The calculation formula is:

[0112]

[0113] in, Indicates the control quantity, K p represents the proportional gain, ΔE(t) represents the suspension parameter adjustment value vector at time t, ΔE(t-1) represents the suspension parameter adjustment value vector at time t-1, ΔE(t-2) represents the suspension parameter adjustment value vector at time t-2, K iIndicates the integral gain, K d Represents the differential gain.

[0114] By controlling the suspension parameter adjustment value vector output by the RNN-TD3 model through the incremental PID algorithm, comprehensive improvements in the real-time, stability, energy efficiency optimization, and driving comfort of the electric tricycle suspension system can be achieved, and the robustness and adaptability of the control system can be enhanced. At the same time, the computational complexity and system resource requirements can be reduced, and efficient and stable suspension control can be achieved under limited hardware resources.

[0115] Table 3 shows partial data of the electric tricycle in the 2352nd round of the simulation phase, which reflects that the control system provided by the present invention brings good stability and controllability to the electric tricycle.

[0116] Table 3. Partial system response data

[0117] Simulation steps Lateral inclination Longitudinal inclination Center of gravity offset Response time Step 21 2.21° 1.34° 20.3cm 12.1ms Step 22 1.98° 1.25° 16.4cm 11.8ms Step 23 1.63° 0.99° 14.2cm 12.0ms Step 24 1.39° 0.82° 9.1cm 11.5ms Step 25 0.91° 0.71° 4.5cm 11.3ms

[0118] An adaptive telescopic control system for driving wheels of an electric tricycle provided by the present invention first collects all-round data of the tricycle through a static data collection unit and a dynamic data collection unit of a tricycle data collection module, thereby providing data support for control; then a turning posture state matrix of the tricycle is extracted through a tricycle posture extraction module, thereby comprehensively reflecting the state of the electric tricycle when turning; a suspension parameter generation module and a suspension parameter adjustment module, as a whole, jointly execute suspension control, thereby finally realizing efficient and stable adaptive telescopic control of the driving wheel under limited hardware resources.

[0119] Embodiment 2

[0120] When a company was tuning the suspension of a tricycle, it introduced an adaptive telescopic control system for the driving wheel of an electric tricycle provided by the present invention to achieve adaptive telescopic control of the driving wheel of the electric tricycle when turning. The specific implementation method is as follows:

[0121] A tricycle data collection module, used to collect tricycle data, including a static data collection unit and a dynamic data collection unit;

[0122] Furthermore, a static data sensor is installed on the electric tricycle to collect static data, including body length, body width and body weight; the static data is obtained and preprocessed and integrated to obtain a static data vector F1. Table 4 shows the data of some models of tricycles.

[0123] Table 4. Static data

[0124] Tricycle No. Vehicle length Body width Vehicle weight A001 2.4m 1.4m 290kg A002 2.4m 1.4m 300kg A003 2.4m 1.4m 310kg A004 2.4m 1.4m 320kg A005 2.4m 1.4m 330kg

[0125] Furthermore, a dynamic data sensor is installed on the electric tricycle to collect dynamic data, and the dynamic data includes front wheel collected data, left wheel collected data and right wheel collected data. Table 5 shows part of the right wheel collected data;

[0126] Table 5. Part of the right wheel collection data

[0127] Number of collections Right wheel speed Right wheel load Right wheel suspension travel Right wheel traction Right tire pressure No. 39 77.31 rpm 125.4kg 11.52mm 230.4N 298.3kPa 40th 76.82 rpm 127.8kg 10.48mm 229.8N 300.1kPa 41st 79.38 rpm 126.3kg 9.24mm 229.4N 299.4kPa 42nd 76.40 rpm 127.2kg 8.29mm 230.3N 300.2kPa 43rd 77.49 rpm 129.5kg 8.78mm 230.6N 299.5kPa

[0128] A tricycle posture extraction module is used to obtain and process the data of the dynamic data acquisition unit to obtain a turning posture state matrix; the turning posture state matrix includes a first state vector, a second state vector, a lateral state vector and a third state vector, the first state vector is obtained according to a front wheel data vector and a left wheel data vector, the second state vector is obtained according to a front wheel data vector and a right wheel data vector, the lateral state vector is obtained according to a left wheel data vector and a right wheel data vector, and the third state vector is obtained according to power data of the electric tricycle;

[0129] Furthermore, the specific steps of constructing the turning posture state matrix of the electric tricycle according to the front wheel data vector, the left wheel data vector and the right wheel data vector include:

[0130] Constructing the first state vector f1 according to the front wheel data vector and the left wheel data vector, the first state vector including first state data and second state data, the first state data including data of different properties, and the second state data including data of the same property;

[0131] Constructing the second state vector f2 according to the front wheel data vector and the right wheel data vector, the second state vector comprising third state data and fourth state data, the third state data comprising data of different properties, and the fourth state data comprising data of the same property;

[0132] Constructing the lateral state vector f3 according to the difference between the left wheel data vector and the right wheel data vector;

[0133] Acquiring and processing power data of the electric tricycle to obtain the third state vector f4, wherein the power data includes the output power of the motor;

[0134] The first state vector, the second state vector, the lateral state vector and the third state vector are acquired and integrated to obtain the turning posture state matrix of the electric tricycle.

[0135] A suspension parameter generation module is used to generate a suspension parameter adjustment value vector in real time according to an observed state space, and includes an observed state space unit and an RNN-TD3 model unit; wherein the observed state space unit obtains data from the static data acquisition unit and data from the tricycle posture extraction module, and integrates and processes the data to obtain the observed state space; the RNN-TD3 model unit obtains the observed state space and processes the data through the RNN-TD3 model to obtain the suspension parameter adjustment value vector, and the suspension parameter adjustment value vector includes suspension stiffness and suspension travel;

[0136] Furthermore, the observation state space unit includes:

[0137] Obtain a static data vector and the turning posture state matrix and integrate them to obtain a first observation state space, obtain data of the suspension parameter adjustment value vector of the electric tricycle at the previous moment to obtain a second observation state space, obtain data of the first observation state space and the second observation state space and integrate them to obtain the observation state space.

[0138] Furthermore, the RNN-TD3 model includes:

[0139] A first action network and a first target action network, wherein the first action network and the first target action network have the same action network structure, and the action network structure includes an input layer, an RNN layer, a fully connected layer, an action normalization layer, and an output layer;

[0140] A first evaluation network, a first target evaluation network, a second evaluation network and a second target evaluation network, wherein the first evaluation network, the first target evaluation network, the second evaluation network and the second target evaluation network have the same evaluation network structure, and the evaluation network structure includes an input layer, an RNN layer, a fully connected layer and an output layer.

[0141] Furthermore, the training process of the RNN-TD3 model includes:

[0142] Step S1: Initialize the entire network;

[0143] Step S2: inputting the observed state space and the historical observed state space into the first action network, outputting the suspension parameter adjustment value vector, and obtaining the suspension parameter reward and the next state of the electric tricycle corresponding to the suspension parameter adjustment value vector; constructing a training sample according to the observed state space, the suspension parameter adjustment value vector, the reward and the next state, and storing it in the experience replay pool;

[0144] Step S3: sampling training samples from the experience replay pool, and updating the parameters of the first evaluation network and the second evaluation network according to the training samples;

[0145] Step S4: after the parameters of the first evaluation network and the second evaluation network are updated at intervals of D times, the parameters of the first action network are updated;

[0146] Step S5: updating the parameters of the first target action network, the first target evaluation network and the second target evaluation network according to the soft update rule;

[0147] Step S6: Repeat steps S3 to S5 until a training round is reached.

[0148] Furthermore, the suspension parameter reward includes:

[0149] R=ω1·(-(|θ′|+|θ″|))+ω2·(-|ΔG|)+ω3·(-E)+ω4·(-T);

[0150] Wherein, R represents the suspension parameter reward, ω1 represents the weight coefficient of stability, θ′ represents the lateral inclination angle of the electric tricycle, θ″ represents the longitudinal inclination angle of the electric tricycle, ω2 represents the weight coefficient of the center of gravity offset ΔG, ω3 represents the weight coefficient of the suspension system adjustment energy consumption E, and ω4 represents the weight coefficient of the suspension system response time T.

[0151] Furthermore, the formula of the soft update rule is:

[0152] θ t ' ar =τ·θ cur +(1-τ)·θ tar ;

[0153] Among them, θ t ' ar represents the parameters of the updated target network, θ cur Represents the parameters of the current network, namely the evaluation network and the action network, θ tar represents the parameters of the target network before updating, τ represents the soft update coefficient, and its value range is 0 to 1.

[0154] The suspension parameter adjustment module is used to obtain the suspension parameter adjustment value vector and control it through an incremental PID algorithm.

[0155] Table 6. Partial system response data

[0156] Simulation steps Lateral inclination Longitudinal inclination Center of gravity offset Response time Step 19 2.01° 1.23° 10.32cm 11.2ms Step 20 1.92° 1.21° 10.98cm 10.9ms Step 21 2.02° 1.19° 9.38cm 10.5ms Step 22 1.97° 1.15° 9.57cm 10.3ms Step 23 1.98° 1.18° 9.19cm 10.6ms

[0157] Furthermore, the specific steps of controlling by the incremental PID algorithm include:

[0158] The suspension parameter adjustment value vector ΔE output by the RNN-TD3 model unit is obtained and the parameters are adjusted according to the incremental PID algorithm. The calculation formula is:

[0159]

[0160] in, Indicates the control quantity, K p represents the proportional gain, ΔE(t) represents the suspension parameter adjustment value vector at time t, ΔE(t-1) represents the suspension parameter adjustment value vector at time t-1, ΔE(t-2) represents the suspension parameter adjustment value vector at time t-2, K i Indicates the integral gain, K d Represents the differential gain.

[0161] Table 6 shows some system response data in the 4828th round of the simulation stage. It can be seen that the adaptive telescopic control system applied to the driving wheel of the electric tricycle provided by the present invention can realize the efficient operation and telescopic of the driving wheel, shorten the response time, and improve the driving stability of the electric tricycle when turning.

[0162] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An adaptive telescopic control system for driving wheels of an electric tricycle, characterized in that: include: A tricycle data collection module, used to collect tricycle data, including a static data collection unit and a dynamic data collection unit; A tricycle posture extraction module is used to obtain and process the data of the dynamic data acquisition unit to obtain a turning posture state matrix; the turning posture state matrix includes a first state vector, a second state vector, a lateral state vector and a third state vector, the first state vector is obtained according to a front wheel data vector and a left wheel data vector, the second state vector is obtained according to a front wheel data vector and a right wheel data vector, the lateral state vector is obtained according to a left wheel data vector and a right wheel data vector, and the third state vector is obtained according to power data of the electric tricycle; A suspension parameter generation module is used to generate a suspension parameter adjustment value vector in real time according to an observed state space, and includes an observed state space unit and an RNN-TD3 model unit; wherein the observed state space unit obtains data from the static data acquisition unit and data from the tricycle posture extraction module, and integrates and processes the data to obtain the observed state space; the RNN-TD3 model unit obtains the observed state space and processes the data through the RNN-TD3 model to obtain the suspension parameter adjustment value vector, and the suspension parameter adjustment value vector includes suspension stiffness and suspension travel; The suspension parameter adjustment module is used to obtain the suspension parameter adjustment value vector and control it through an incremental PID algorithm.

2. The adaptive telescopic control system for driving wheels of an electric tricycle according to claim 1, characterized in that: The static data acquisition unit comprises: A static data sensor is installed on the electric tricycle to collect static data, and the static data includes the length, width and weight of the vehicle body; the static data is obtained and pre-processed and integrated to obtain a static data vector F1.

3. The adaptive telescopic control system for driving wheels of an electric tricycle according to claim 1, characterized in that: The dynamic data acquisition unit comprises: A dynamic data sensor is installed on the electric tricycle to collect dynamic data, and the dynamic data includes front wheel collected data, left wheel collected data and right wheel collected data; The front wheel collected data includes the front wheel angle, front wheel load, front wheel speed and front wheel friction, wherein the front wheel angle is the rotation angle of the front wheel monitored in real time by the angle sensor, the front wheel load is the load change of the front wheel monitored in real time by the load sensor, and the front wheel friction is the friction data between the front wheel and the ground obtained by the traction sensor; The front wheel data is acquired, preprocessed and integrated to obtain the front wheel data vector F2.

4. The adaptive telescopic control system for driving wheels of an electric tricycle according to claim 3, characterized in that: The left wheel data collection and the right wheel data collection include: The left wheel collected data includes left wheel speed, left wheel load, left wheel suspension travel, left wheel vibration data, left wheel traction, left wheel braking force and left tire pressure; the left wheel collected data is obtained and pre-processed and integrated to obtain the left wheel data vector F3; The right wheel collected data includes right wheel speed, right wheel load, right wheel suspension travel, right wheel vibration data, right wheel traction, right wheel braking force and right tire pressure; the right wheel collected data is obtained and pre-processed and integrated to obtain the right wheel data vector F4; The left wheel data vector and the right wheel data vector have the same data format.

5. The adaptive telescopic control system for driving wheels of an electric tricycle according to claim 1, characterized in that: The specific steps of constructing the turning posture state matrix of the electric tricycle according to the front wheel data vector, the left wheel data vector and the right wheel data vector include: Constructing the first state vector f1 according to the front wheel data vector and the left wheel data vector, wherein the first state vector includes first state data and second state data; Constructing the second state vector f2 according to the front wheel data vector and the right wheel data vector, wherein the second state vector includes the third state data and the fourth state data; Constructing the lateral state vector f3 according to the difference between the left wheel data vector and the right wheel data vector; Acquiring and processing power data of the electric tricycle to obtain the third state vector f4, wherein the power data includes the output power of the motor; The first state vector, the second state vector, the lateral state vector and the third state vector are acquired and integrated to obtain the turning posture state matrix E of the electric tricycle.

6. The adaptive telescopic control system for driving wheels of an electric tricycle according to claim 1, characterized in that: The observation state space unit comprises: Obtain a static data vector and the turning posture state matrix and integrate them to obtain a first observation state space, obtain data of the suspension parameter adjustment value vector of the electric tricycle at the previous moment to obtain a second observation state space, obtain data of the first observation state space and the second observation state space and integrate them to obtain the observation state space.

7. The adaptive telescopic control system for driving wheels of an electric tricycle according to claim 1, characterized in that: The RNN-TD3 model includes: A first action network and a first target action network, wherein the first action network and the first target action network have the same action network structure, and the action network structure includes an input layer, an RNN layer, a fully connected layer, an action normalization layer, and an output layer; A first evaluation network, a first target evaluation network, a second evaluation network and a second target evaluation network, wherein the first evaluation network, the first target evaluation network, the second evaluation network and the second target evaluation network have the same evaluation network structure, and the evaluation network structure includes an input layer, an RNN layer, a fully connected layer and an output layer.

8. The adaptive telescopic control system for driving wheels of an electric tricycle according to claim 7, characterized in that: The training process of the RNN-TD3 model includes: Step S1: Initialize the entire network; Step S2: inputting the observed state space and the historical observed state space into the first action network, outputting the suspension parameter adjustment value vector, and obtaining the suspension parameter reward and the next state of the electric tricycle corresponding to the suspension parameter adjustment value vector; constructing a training sample according to the observed state space, the suspension parameter adjustment value vector, the reward and the next state, and storing it in the experience replay pool; Step S3: sampling training samples from the experience replay pool, and updating the parameters of the first evaluation network and the second evaluation network according to the training samples; Step S4: after the parameters of the first evaluation network and the second evaluation network are updated at intervals of D times, the parameters of the first action network are updated; Step S5: updating the parameters of the first target action network, the first target evaluation network and the second target evaluation network according to the soft update rule; Step S6: Repeat steps S3 to S5 until a training round is reached.

9. The adaptive telescopic control system for driving wheels of an electric tricycle according to claim 8, characterized in that: The suspension parameter rewards include: R=ω1·(-(|θ′|+|θ″|))+ω2·(-|ΔG|)+ω3·(-E)+ω4·(-T); Wherein, R represents the suspension parameter reward, ω1 represents the weight coefficient of stability, θ′ represents the lateral inclination angle of the electric tricycle, θ″ represents the longitudinal inclination angle of the electric tricycle, ω2 represents the weight coefficient of the center of gravity offset ΔG, ω3 represents the weight coefficient of the suspension system adjustment energy consumption E, and ω4 represents the weight coefficient of the suspension system response time T.

10. The adaptive telescopic control system for driving wheels of an electric tricycle according to claim 1, characterized in that: The specific steps of control through the incremental PID algorithm include: The suspension parameter adjustment value vector ΔE output by the RNN-TD3 model unit is obtained and the parameters are adjusted according to the incremental PID algorithm. The calculation formula is: in, Indicates the control quantity, K p represents the proportional gain, ΔE(t) represents the suspension parameter adjustment value vector at time t, ΔE(t-1) represents the suspension parameter adjustment value vector at time t-1, ΔE(t-2) represents the suspension parameter adjustment value vector at time t-2, K i Indicates the integral gain, K d Represents the differential gain.

Images