A control method for vehicle starting on a slope

Abstract

The present invention discloses a control method for starting a vehicle on a slope, which relates to the field of vehicle control technology and solves the problems of large motor starting current and uneven motor starting when the motor of a warehouse vehicle is started with load. The control method includes obtaining starting data of the warehouse vehicle on a slope and recording the data acquisition time; calculating the starting parameters of a control mechanism of the warehouse vehicle, and cleaning the starting data and starting parameters; when the warehouse vehicle starts halfway on a slope, controlling the starting current of a kinetic energy device in stages according to the starting parameters to generate a current control data table; calculating a speed control curve and a current control curve of the kinetic energy device; using a monitoring correction model through a monitoring feedback module to monitor the starting data in real time to judge and feedback the starting result of the warehouse vehicle; correcting the speed control curve and the current control curve according to the starting result of the warehouse vehicle; performing speed closed-loop control on the kinetic energy device, and generalizing and correcting the monitoring correction model.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle control, and in particular to a control method for starting a vehicle on a slope. Background Art

[0002] With the acceleration of urbanization and the increasing complexity of road conditions, slopes have become a frequent challenge for drivers. A vehicle parked on a slope must start within the gradient's limits. When starting on a slope, the engine must generate sufficient power to overcome the stress, placing higher demands on the starting control system. The friction coefficient between the tires and the road is crucial when starting on a slope. If the friction coefficient is too low, the vehicle may lose control or slip during starting, making the start more difficult.

[0003] Patent No. CN2020114238082 discloses a control method and device for vehicle hill starting. When controlling the vehicle during hill starting, on the basis of determining the basic required torque of the engine in the vehicle according to the opening position of the clutch in the vehicle, the compensation torque of the engine is added, and the total required torque of the engine is determined based on the basic required torque of the engine and the compensation torque of the engine; then, based on the total required torque of the engine and the torque threshold of the motor in the vehicle, the required torque of the engine and the required torque of the motor are determined, thereby improving the accuracy of the required torque of the engine and the required torque of the motor, so that when the vehicle is controlled to start on a hill according to the required torque of the engine and the required torque of the motor with higher accuracy, precise control of the engine and the motor during the vehicle hill starting process is achieved, thereby avoiding the problems of rolling back and stalling during the vehicle hill starting process.

[0004] Patent No. CN2014103265593 discloses a vehicle and a hill start control method and a hill start control device for the vehicle. The method includes: obtaining vehicle speed, throttle opening, engine speed, clutch shaft speed, road slope and clutch output torque; calculating vehicle acceleration torque and vehicle resistance torque based on the clutch output torque, road slope and clutch shaft speed; calculating target vehicle speed based on the road slope and throttle opening, then calculating vehicle target acceleration torque based on the difference between the calculated target vehicle speed and the vehicle speed, road slope and throttle opening, and calculating engine demand torque and clutch target torque based on the vehicle target acceleration torque, vehicle resistance torque and engine speed; controlling the engine output torque to reach the engine demand torque and the clutch output torque to reach the clutch target torque so that the vehicle speed reaches the target speed and the vehicle acceleration torque reaches the vehicle target acceleration torque, thereby controlling the vehicle speed and acceleration and quickly realizing the driver's intention.

[0005] Although the above patents all mainly use speed closed-loop control for the control method of starting the vehicle on a slope, that is, the controller controls the vehicle's motor according to a given speed command, problems such as large motor starting current and uneven motor starting will occur during load starting, and in severe cases, the controller may even be damaged. Summary of the Invention

[0006] The purpose of the present invention is to provide a control method for vehicle hill starting, which can control the kinetic energy device through a three-stage preset current and calculate the real-time output current control curve and speed control curve of the kinetic energy device; through the monitoring and correction model, the starting data is monitored in real time to judge and feedback the starting results of the warehouse vehicle, and then the speed control curve and the current control curve are corrected; so as to improve the current control capability of the kinetic energy device during startup and effectively reduce the starting current of the kinetic energy device.

[0007] The present invention utilizes the following technical solutions:

[0008] A method for controlling a vehicle starting on a hill comprises the following steps:

[0009] S1: The data acquisition module obtains the starting data of the storage vehicle on the slope and records the data acquisition time;

[0010] Warehouse vehicles include control mechanisms and kinetic energy devices; starting data includes wheel speed, kinetic energy device torque, kinetic energy device speed, kinetic energy device output current, tilt angle, steering wheel angle, brake pressure, and vehicle mass;

[0011] S2: The data processing module calculates the starting parameters of the control mechanism of the storage vehicle based on the starting data, and cleans the starting data and starting parameters;

[0012] S3: When the warehouse vehicle starts halfway up a slope, the current control module controls the starting current of the kinetic energy device in stages according to the starting parameters to generate a current control data table;

[0013] S4: The curve calculation module calculates the speed control curve and current control curve of the kinetic energy device according to the current control data table;

[0014] S5: During the warehouse vehicle's climbing operation, the monitoring feedback module uses the monitoring correction model to monitor the starting data in real time to determine and provide feedback on the warehouse vehicle's starting results;

[0015] S6: The monitoring feedback module modifies the speed control curve and the current control curve according to the starting result of the storage vehicle;

[0016] S7: The monitoring feedback module performs speed closed-loop control on the kinetic energy equipment according to the corrected speed control curve and current control curve, and generalizes and corrects the monitoring correction model.

[0017] Preferably, in step S1, the data acquisition module uses an inclination sensor to collect the inclination angle of the ramp, uses an inertial sensor to collect the wheel speed, kinetic energy device torque and kinetic energy device speed of the storage vehicle, uses a position sensor to detect the steering wheel angle, uses a pressure sensor to detect the brake pressure of the brake, uses a dynamic weighing instrument to obtain the vehicle mass of the storage vehicle, and uses a current sensor to collect the output current of the kinetic energy device.

[0018] Preferably, in step S2, the control mechanism includes a brake and a steering wheel; the starting parameters include the minimum torque of the kinetic energy device and the starting surge current of the kinetic energy device; the data processing module calculates the starting parameters and cleans the starting parameters and starting data in the following process:

[0019] S21: The data processing module calculates the minimum value of the kinetic energy device torque based on the total mass of the storage vehicle and the inclination angle of the ramp; and calculates the starting surge current of the kinetic energy device based on the kinetic energy device torque, the kinetic energy device speed, the kinetic energy device output current, and the torque constant;

[0020] S22: The data processing module uses a filtering algorithm to remove noise from the starting parameters and starting data;

[0021] S23: The data processing module uses a sorting algorithm to sort the denoised starting parameters and starting data according to the data collection time;

[0022] S24: The data processing module transforms the sorted starting parameters and starting data into the frequency domain using Fourier transform to obtain time series frequency domain features;

[0023] S25: The data processing module uses an association algorithm to associate the starting parameters, starting data, and time series frequency domain features, thereby completing data cleaning.

[0024] Preferably, step S3 includes the following steps:

[0025] S31: When the storage vehicle starts halfway up a slope, a first preset current is started to control the operation of the storage vehicle, and the current control module controls the kinetic energy device to reach the first preset current;

[0026] S32: When the kinetic energy device reaches a first preset current, the current control module obtains the kinetic energy device torque at a first time and the kinetic energy device speed at a first stage;

[0027] S33: During the vehicle climbing operation, the second preset current is started to control the operation of the storage vehicle, and the current control module controls the kinetic energy device to reach the second preset current, thereby obtaining the speed of the kinetic energy device in the second stage in real time;

[0028] S34: After the second preset current has been in operation for a preset time, the third preset current is started to control the operation of the storage vehicle. The current control module controls the kinetic energy device to reach the third preset current, thereby obtaining the speed of the kinetic energy device in the third stage in real time.

[0029] S35: The current control module generates a current control data table according to the first preset current, the second preset current, the third preset current, the first stage kinetic energy device speed, the second stage kinetic energy device speed, and the third stage kinetic energy device speed.

[0030] Preferably, step S4 includes the following steps:

[0031] S41: The curve calculation module uses a clustering algorithm to classify the current control data table according to data type to form a current-time data set and a speed-time data set; the data type includes current and speed;

[0032] S42: The curve calculation module calculates the real-time output current control curve of the kinetic energy device using the current-time data set;

[0033] S43: The curve calculation module calculates the speed control curve of the kinetic energy device using the speed-time data set.

[0034] Preferably, the monitoring and correction model first uses the feature extraction layer to extract feature information of the starting parameters, starting data, current control data table, current control curve and speed control curve respectively to obtain the data feature matrix and the curve feature matrix; then uses the iterative training layer to perform several iterative training on the data feature matrix to generate a data feature weight matrix; then, through the identification output layer, the data feature weight matrix is ​​updated and optimized according to the curve feature matrix and the global loss function to obtain the optimal weight matrix; finally, through the judgment and correction layer, the reconstruction decoder is used to decompose and reconstruct the data feature matrix, and the current control curve and the speed control curve are corrected according to the real-time monitoring results of the optimal weight matrix.

[0035] Preferably, the feature extraction layer includes 4 3×3 convolutional layers, 3 3×3 maximum pooling layers and 1 normalization layer; every two 3×3 convolutional layers are connected in parallel and then connected in series with a 3×3 maximum pooling layer, and a 3×3 maximum pooling layer and a normalization layer are connected in parallel in sequence, thereby fusing the feature information of the starting parameters, starting data and current control data table into a data feature matrix; at the same time, the feature information of the current control curve and the speed control curve is fused into a curve feature matrix.

[0036] Preferably, the iterative training layer includes 2 1×1 convolutional layers, 3 3×1 upsampling layers, 3 1×3 downsampling layers, 4 InceptionV4 blocks, 2 InceptionV3 blocks, 6 residual blocks, 4 average pooling layers, 3 inverted residual blocks, 2 Mish activation functions and 2 RELU activation functions; for the data feature matrix, the iterative training layer learning framework is: 2 1×1 convolutional layers are connected in parallel to form two training branches; after the two training branches are spliced ​​using a channel shuffle layer, 2 InceptionV4 blocks, 2 residual blocks, 1 inverted residual block and 1 RELU activation function are connected in series in sequence; the first training branch is connected in series with 2 3×1 upsampling layers, 2 1×3 downsampling layers, 2 average pooling layers and 1 Mish activation function; the second training branch is connected in series with 1 3×1 upsampling layer, 1 1×3 downsampling layer, 2 InceptionV3 blocks, 2 residual blocks, 2 average pooling layers and 1 Mish activation function.

[0037] Preferably, the recognition output layer includes 3 average pooling layers, 2 adaptive pooling layers, 2 fully connected layers, 1 random dropout layer and 2 SoftMax activation functions; each adaptive pooling layer is located between the two average pooling layers, and the random dropout layer and 1 SoftMax activation function are located between the two fully connected layers; the global loss function Loss is:

[0038] Among them, Model 1 Represents the feature extraction layer, Model 2 Represents the iterative training layer, Model 3 represents the recognition output layer, σ represents the number of iterative training layers, G represents the quantization function, μ1 represents the calculation accuracy coefficient, B σ represents the iterative training loss function of each layer, μ2 represents the preset optimization weight coefficient, E σ Indicates the optimization efficiency of each layer's iterative training.

[0039] Preferably, the judgment correction layer first uses a reconstruction decoder to perform dual reconstruction of the structure and attributes of the data feature matrix to generate a reconstructed decoding value;

[0040] S=P(H,A|W)+SM(HH T ) (2)

[0041] Among them, S represents the reconstructed decoding value, P() represents the PRelu activation function, H represents the data feature matrix, SM() represents the SigMoid activation function, and T represents transpose;

[0042] Then, the reconstructed decoded values ​​are arranged in time series according to the real-time monitoring results of the optimal weight matrix to form current time series and speed time series;

[0043] Finally, the judgment correction layer corrects the current control curve and the speed control curve according to the current timing and the speed timing respectively.

[0044] The present invention controls the kinetic energy equipment through a three-stage preset current and calculates the real-time output current control curve and speed control curve of the kinetic energy equipment; through the monitoring and correction model, the starting data is monitored in real time to judge and feedback the starting results of the warehouse vehicle, and then the speed control curve and the current control curve are corrected; the current control capability of the kinetic energy equipment during startup is improved, and the starting current is effectively reduced. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0046] Figure 1 This is a principle block diagram of the control method for starting a vehicle on a hill;

[0047] Figure 2 This is the flow chart of the current control module. DETAILED DESCRIPTION

[0048] The present invention is described in detail below with reference to the accompanying drawings and embodiments:

[0049] like Figure 1-Figure 2 As shown, the control method for starting a vehicle on a hill according to the present invention comprises the following steps:

[0050] S1: The data acquisition module obtains the starting data of the storage vehicle on the slope and records the data acquisition time;

[0051] In the present invention, the storage vehicle includes a control mechanism and a kinetic energy device; the starting data includes wheel speed, kinetic energy device torque, kinetic energy device speed, kinetic energy device output current, tilt angle, steering wheel angle, brake pressure and vehicle mass;

[0052] S2: The data processing module calculates the starting parameters of the control mechanism of the storage vehicle based on the starting data, and cleans the starting data and starting parameters;

[0053] S3: When the warehouse vehicle starts halfway up a slope, the current control module controls the starting current of the kinetic energy device in stages according to the starting parameters to generate a current control data table;

[0054] S4: The curve calculation module calculates the speed control curve and current control curve of the kinetic energy device according to the current control data table;

[0055] S5: During the warehouse vehicle's climbing operation, the monitoring feedback module uses the monitoring correction model to monitor the starting data in real time to determine and provide feedback on the warehouse vehicle's starting results;

[0056] S6: The monitoring feedback module modifies the speed control curve and the current control curve according to the starting result of the storage vehicle;

[0057] S7: The monitoring feedback module performs speed closed-loop control on the kinetic energy equipment according to the corrected speed control curve and current control curve, and generalizes and corrects the monitoring correction model.

[0058] In the present invention, in step S1, the data acquisition module uses an inclination sensor to collect the inclination angle of the ramp, uses an inertial sensor to collect the wheel speed of the storage vehicle, the torque of the kinetic energy device, and the speed of the kinetic energy device, uses a position sensor to detect the steering wheel angle, uses a pressure sensor to detect the brake pressure of the brake, uses a dynamic weighing instrument to obtain the vehicle mass of the storage vehicle, and uses a current sensor to collect the output current of the kinetic energy device;

[0059] In the present invention, in step S2, the control mechanism includes a brake and a steering wheel; the starting parameters include the minimum torque of the kinetic energy device and the starting surge current of the kinetic energy device; the data processing module calculates the starting parameters and cleans the starting parameters and starting data in the following process:

[0060] S21: The data processing module calculates the minimum value of the kinetic energy device torque based on the total mass of the storage vehicle and the inclination angle of the ramp; and calculates the starting surge current of the kinetic energy device based on the kinetic energy device torque, the kinetic energy device speed, the kinetic energy device output current, and the torque constant;

[0061] S22: The data processing module uses a filtering algorithm to remove noise from the starting parameters and starting data;

[0062] S23: The data processing module uses a sorting algorithm to sort the denoised starting parameters and starting data according to the data collection time;

[0063] S24: The data processing module transforms the sorted starting parameters and starting data into the frequency domain using Fourier transform to obtain time series frequency domain features;

[0064] S25: The data processing module uses an association algorithm to associate the starting parameters, starting data, and time series frequency domain features, thereby completing data cleaning;

[0065] In the present invention, step S3 includes the following steps:

[0066] S31: When the storage vehicle starts halfway up a slope, a first preset current is started to control the operation of the storage vehicle, and the current control module controls the kinetic energy device to reach the first preset current;

[0067] S32: When the kinetic energy device reaches a first preset current, the current control module obtains the kinetic energy device torque at a first time and the kinetic energy device speed at a first stage;

[0068] S33: During the vehicle climbing operation, the second preset current is started to control the operation of the storage vehicle, and the current control module controls the kinetic energy device to reach the second preset current, thereby obtaining the speed of the kinetic energy device in the second stage in real time;

[0069] S34: After the second preset current has been in operation for a preset time, the third preset current is started to control the operation of the storage vehicle. The current control module controls the kinetic energy device to reach the third preset current, thereby obtaining the speed of the kinetic energy device in the third stage in real time.

[0070] S35: The current control module generates a current control data table according to the first preset current, the second preset current, the third preset current, the first stage kinetic energy device speed, the second stage kinetic energy device speed, and the third stage kinetic energy device speed;

[0071] In this embodiment, the kinetic energy device is an electric motor; the first preset current is the instantaneous starting current: the current for quick startup and instantaneous operation; the second preset current is the peak current: the current for short-term operation under heavy loads such as climbing; the third preset current is the main current: the current for long-term operation;

[0072] In the present invention, step S4 includes the following steps:

[0073] S41: The curve calculation module uses a clustering algorithm to classify the current control data table according to data type to form a current-time data set and a speed-time data set; the data type includes current and speed;

[0074] S42: The curve calculation module calculates the real-time output current control curve of the kinetic energy device using the current-time data set;

[0075] S43: The curve calculation module calculates the speed control curve of the kinetic energy device using the speed-time data set;

[0076] In the present invention, the monitoring correction model first uses the feature extraction layer to extract feature information from the starting parameters, starting data, current control data table, current control curve and speed control curve to obtain a data feature matrix and a curve feature matrix;

[0077] In the present invention, the feature extraction layer includes four 3×3 convolutional layers, three 3×3 maximum pooling layers, and one normalization layer. Every two 3×3 convolutional layers are connected in parallel and then connected in series with a 3×3 maximum pooling layer. A 3×3 maximum pooling layer and a normalization layer are connected in parallel in sequence. This allows the feature information of the starting parameters, starting data, and current control data table to be fused into a data feature matrix. Simultaneously, the feature information of the current control curve and the speed control curve is fused into a curve feature matrix.

[0078] Then, the data feature matrix is ​​iteratively trained several times using the iterative training layer to generate a data feature weight matrix;

[0079] In the present invention, the iterative training layer includes 2 1×1 convolutional layers, 3 3×1 upsampling layers, 3 1×3 downsampling layers, 4 InceptionV4 blocks, 2 InceptionV3 blocks, 6 residual blocks, 4 average pooling layers, 3 inverted residual blocks, 2 Mish activation functions and 2 RELU activation functions; for the data feature matrix, the iterative training layer learning framework is: 2 1×1 convolutional layers are connected in parallel to form two training branches; after the two training branches are spliced ​​using a channel shuffle layer, 2 InceptionV4 blocks, 2 residual blocks, 1 inverted residual block and 1 RELU activation function are connected in series in sequence; the first training branch is connected in series with 2 3×1 upsampling layers, 2 1×3 downsampling layers, 2 average pooling layers and 1 Mish activation function; the second training branch is connected in series with 1 3×1 upsampling layer, 1 1×3 downsampling layer, 2 InceptionV3 blocks, 2 residual blocks, 2 average pooling layers and 1 Mish activation function;

[0080] Then, the data feature weight matrix is ​​updated and optimized according to the curve feature matrix and the global loss function through the identification output layer to obtain the optimal weight matrix;

[0081] In the present invention, the recognition output layer includes three average pooling layers, two adaptive pooling layers, two fully connected layers, one random dropout layer and two SoftMax activation functions; each adaptive pooling layer is located between the two average pooling layers, and the random dropout layer and one SoftMax activation function are located between the two fully connected layers; the global loss function is:

[0082] Among them, Model 1 Represents the feature extraction layer, Model 2 Represents the iterative training layer, Model 3 represents the recognition output layer, σ represents the number of iterative training layers, G represents the quantization function, μ1 represents the calculation accuracy coefficient, B σ represents the iterative training loss function of each layer, μ2 represents the preset optimization weight coefficient, Eσ Indicates the optimization efficiency of each layer’s iterative training;

[0083] Finally, the judgment correction layer uses the reconstruction decoder to decompose and reconstruct the data feature matrix, and then corrects the current control curve and speed control curve according to the real-time monitoring results of the optimal weight matrix;

[0084] In the present invention, the judgment correction layer first uses the reconstruction decoder to perform dual reconstruction of the structure and attributes of the data feature matrix to generate a reconstructed decoding value;

[0085] S=P(H,A|W)+SM(HH T ) (2)

[0086] Among them, S represents the reconstructed decoding value, P() represents the PRelu activation function, H represents the data feature matrix, SM() represents the SigMoid activation function, and T represents transpose;

[0087] Then, the reconstructed decoded values ​​are arranged in time series according to the real-time monitoring results of the optimal weight matrix to form current time series and speed time series;

[0088] Finally, the judgment correction layer corrects the current control curve and the speed control curve according to the current timing and the speed timing respectively.

[0089] Example:

[0090] The data acquisition module uses an inclination sensor to collect the inclination angle of the ramp, an inertial sensor to collect the wheel speed, kinetic energy device torque, and kinetic energy device speed of the warehouse vehicle, a position sensor to detect the steering wheel angle, a pressure sensor to detect the brake pressure, a dynamic weighing instrument to obtain the total vehicle mass, and a current sensor to collect the output current of the kinetic energy device. The starting data includes wheel speed, kinetic energy device torque, kinetic energy device speed, kinetic energy device output current, inclination angle, steering wheel angle, brake pressure, and total vehicle mass.

[0091] The data processing module calculates the minimum torque of the kinetic energy device based on the vehicle mass and the slope angle. It calculates the starting inrush current of the kinetic energy device based on the kinetic energy device torque, kinetic energy device speed, kinetic energy device output current, and torque constant. The data processing module uses a filtering algorithm to denoise the starting parameters and starting data. The data processing module uses a sorting algorithm to sort the denoised starting parameters and starting data based on data acquisition time. The data processing module uses Fourier transform to transform the sorted starting parameters and starting data into the frequency domain to obtain time-series frequency domain features. The data processing module uses a correlation algorithm to correlate the starting parameters, starting data, and time-series frequency domain features to complete data cleaning.

[0092] When the storage vehicle starts halfway up a slope, the current control module starts the first preset current to control the operation of the storage vehicle, and controls the real-time output current of the kinetic energy device to reach the first preset current; when the kinetic energy device reaches the first preset current, the current control module obtains the kinetic energy device torque at the first time and the first-stage kinetic energy device speed; during the vehicle climbing operation, the current control module starts the second preset current to control the operation of the storage vehicle, and thereby obtains the second-stage kinetic energy device speed in real time; after the running time of the second preset current reaches the preset time, the current control module starts the third preset current to control the operation of the storage vehicle, and thereby obtains the third-stage kinetic energy device speed in real time; the current control module generates a current control data table according to the first preset current, the second preset current, the third preset current, the first-stage kinetic energy device speed, the second-stage kinetic energy device speed, and the third-stage kinetic energy device speed;

[0093] The curve calculation module uses a clustering algorithm to classify the current control data table according to data type to form a current-time data set and a speed-time data set; the data type includes current and speed; the curve calculation module uses the current-time data set to calculate the real-time output current control curve of the kinetic energy device; and uses the speed-time data set to calculate the speed control curve of the kinetic energy device;

[0094] After a period of starting time, when the warehouse vehicle transitions to the normal driving state, the monitoring feedback module uses the feature extraction layer of the monitoring correction model, using 4 3×3 convolutional layers, 3 3×3 maximum pooling layers and 1 normalization layer; every 2 3×3 convolutional layers are connected in parallel and then connected in series with 1 3×3 maximum pooling layer, and 1 3×3 maximum pooling layer and 1 normalization layer are connected in parallel in sequence, so as to fuse the feature information of the starting parameters, starting data and current control data table into a data feature matrix; at the same time, the feature information of the current control curve and the speed control curve is fused into a curve feature matrix; then, through the iterative training layer, 2 1×1 convolutional layers, 3 3×1 upsampling layers, 3 1×3 downsampling layers, 4 InceptionV4 blocks, 2 InceptionV3 blocks, 6 residual blocks, 4 average pooling blocks, layer, 3 inverted residual blocks, 2 Mish activation functions, and 2 RELU activation functions; for the data feature matrix, the iterative training layer learning framework is: 2 1×1 convolutional layers are connected in parallel to form two training branches; the two training branches are spliced ​​using a channel shuffle layer, and then 2 InceptionV4 blocks, 2 residual blocks, 1 inverted residual block, and 1 RELU activation function are connected in series in sequence; the first training branch sequentially connects 2 3×1 upsampling layers, 2 1×3 downsampling layers, 2 average pooling layers, and 1 Mish activation function; the second training branch sequentially connects 1 3×1 upsampling layer, 1 1×3 downsampling layer, 2 InceptionV3 blocks, 2 residual blocks, 2 average pooling layers, and 1 Mish activation function, and performs several iterative training on the data feature matrix to generate the data feature weight matrix;

[0095] Then, the recognition output layer uses three average pooling layers, two adaptive pooling layers, two fully connected layers, one random dropout layer, and two SoftMax activation functions; each adaptive pooling layer is located between the two average pooling layers, and the random dropout layer and one SoftMax activation function are located between the two fully connected layers. According to the curve feature matrix and the global loss function, the data feature weight matrix is ​​updated and optimized to obtain the optimal weight matrix;

[0096] Finally, the judgment correction layer uses the reconstruction decoder to perform dual reconstruction of the data feature matrix in terms of structure and attributes to generate reconstructed decoding values; then, the reconstructed decoding values ​​are arranged in time series according to the real-time monitoring results of the optimal weight matrix to form current time series and speed time series; finally, the judgment correction layer corrects the current control curve and speed control curve according to the current time series and speed time series respectively.

Claims

1. A method for controlling a vehicle starting on a hill, characterized by: The following steps are involved: S1: The data acquisition module obtains the starting data of the storage vehicle on the slope and records the data acquisition time; Warehouse vehicles include control mechanisms and kinetic energy devices; starting data includes wheel speed, kinetic energy device torque, kinetic energy device speed, kinetic energy device output current, tilt angle, steering wheel angle, brake pressure, and vehicle mass; S2: The data processing module calculates the starting parameters of the control mechanism of the storage vehicle based on the starting data, and cleans the starting data and starting parameters; In step S2, the control mechanism includes a brake and a steering wheel; the starting parameters include the minimum torque of the kinetic energy device and the starting surge current of the kinetic energy device; the data processing module calculates the starting parameters and cleans the starting parameters and starting data in the following process: S21: The data processing module calculates the minimum value of the kinetic energy device torque based on the total mass of the storage vehicle and the inclination angle of the ramp; and calculates the starting surge current of the kinetic energy device based on the kinetic energy device torque, the kinetic energy device speed, the kinetic energy device output current, and the torque constant; S22: The data processing module uses a filtering algorithm to remove noise from the starting parameters and starting data; S23: The data processing module uses a sorting algorithm to sort the denoised starting parameters and starting data according to the data collection time; S24: The data processing module transforms the sorted starting parameters and starting data into the frequency domain using Fourier transform to obtain time series frequency domain features; S25: The data processing module uses an association algorithm to associate the starting parameters, starting data, and time series frequency domain features, thereby completing data cleaning; S3: When the warehouse vehicle starts halfway up a slope, the current control module controls the starting current of the kinetic energy device in stages according to the starting parameters to generate a current control data table; The step S3 includes the following steps: S31: When the storage vehicle starts halfway up a slope, a first preset current is started to control the operation of the storage vehicle, and the current control module controls the kinetic energy device to reach the first preset current; S32: When the kinetic energy device reaches a first preset current, the current control module obtains the kinetic energy device torque at a first time and the kinetic energy device speed at a first stage; S33: During the vehicle climbing operation, the second preset current is started to control the operation of the storage vehicle, and the current control module controls the kinetic energy device to reach the second preset current, thereby obtaining the speed of the kinetic energy device in the second stage in real time; S34: After the second preset current has been in operation for a preset time, the third preset current is started to control the operation of the storage vehicle. The current control module controls the kinetic energy device to reach the third preset current, thereby obtaining the speed of the kinetic energy device in the third stage in real time. S35: The current control module generates a current control data table according to the first preset current, the second preset current, the third preset current, the first stage kinetic energy device speed, the second stage kinetic energy device speed, and the third stage kinetic energy device speed; The first preset current is the instantaneous starting current: the current for quick startup and instantaneous operation; the second preset current is the peak current: the current for short-term operation under heavy loads such as climbing; the third preset current is the main current: the current for long-term operation; S4: The curve calculation module calculates the speed control curve and current control curve of the kinetic energy device according to the current control data table; S5: During the warehouse vehicle's climbing operation, the monitoring feedback module uses the monitoring correction model to monitor the starting data in real time to determine and provide feedback on the warehouse vehicle's starting results; S6: The monitoring feedback module modifies the speed control curve and the current control curve according to the starting result of the storage vehicle; S7: The monitoring feedback module performs speed closed-loop control on the kinetic energy equipment according to the corrected speed control curve and current control curve, and generalizes and corrects the monitoring correction model.

2. The method for controlling a vehicle hill start according to claim 1, wherein: In step S1, the data acquisition module uses an inclination sensor to collect the inclination angle of the ramp, uses an inertial sensor to collect the wheel speed, kinetic energy device torque and kinetic energy device speed of the storage vehicle, uses a position sensor to detect the steering wheel angle, uses a pressure sensor to detect the brake pressure of the brake, uses a dynamic weighing instrument to obtain the vehicle mass of the storage vehicle, and uses a current sensor to collect the output current of the kinetic energy device.

3. The method for controlling a vehicle hill start according to claim 1, wherein: The step S4 includes the following steps: S41: The curve calculation module uses a clustering algorithm to classify the current control data table according to data type to form a current-time data set and a speed-time data set; the data type includes current and speed; S42: The curve calculation module calculates the real-time output current control curve of the kinetic energy device using the current-time data set; S43: The curve calculation module calculates the speed control curve of the kinetic energy device using the speed-time data set.

4. The method for controlling a vehicle hill start according to claim 1, wherein: The monitoring and correction model first uses a feature extraction layer to extract feature information from starting parameters, starting data, current control data table, current control curve, and speed control curve to obtain a data feature matrix and a curve feature matrix. Then, an iterative training layer is used to iterate the data feature matrix several times to generate a data feature weight matrix. Then, the data feature weight matrix is ​​updated and optimized according to the curve feature matrix and the global loss function through the identification output layer to obtain the optimal weight matrix; Finally, the data feature matrix is ​​decomposed and reconstructed by the reconstruction decoder through the judgment correction layer, and the current control curve and speed control curve are corrected according to the real-time monitoring results of the optimal weight matrix.

5. The method for controlling a vehicle hill start according to claim 4, characterized in that: The feature extraction layer includes Four 3×3 convolutional layers, three 3×3 maximum pooling layers, and one normalization layer. Every two 3×3 convolutional layers are connected in parallel and then connected in series with a 3×3 maximum pooling layer. A 3×3 maximum pooling layer and a normalization layer are connected in parallel in sequence. This integrates the feature information of the starting parameters, starting data, and current control data table into a data feature matrix. At the same time, the feature information of the current control curve and speed control curve is integrated into a curve feature matrix.

6. The method for controlling a vehicle hill start according to claim 4, characterized in that: The iterative training layer includes 2 1×1 convolutional layers, 3 3×1 upsampling layers, 3 1×3 downsampling layers, 4 InceptionV4 blocks, 2 InceptionV3 blocks, 6 residual blocks, 4 average pooling layers, 3 inverted residual blocks, 2 Mish activation functions, and 2 RELU activation functions; for the data feature matrix, the iterative training layer learning framework is: 2 1×1 convolutional layers in parallel to form two training branches; After the two training branches are spliced ​​using the channel shuffle layer, two InceptionV4 blocks, two residual blocks, one inverted residual block and 1 RELU activation function; the first training branch sequentially connects 2 3×1 upsampling layers, 2 1×3 downsampling layers, 2 average pooling layers and 1 Mish activation function; the second training branch sequentially connects 1 3×1 upsampling layer, 1 1×3 downsampling layer, 2 InceptionV3 blocks, 2 residual blocks, 2 average pooling layers and 1 Mish activation function.

7. The method for controlling a vehicle hill start according to claim 4, characterized in that: The recognition output layer includes 3 average pooling layers, 2 adaptive pooling layers, 2 fully connected layers, 1 random dropout layer and 2 SoftMax activation functions; Each adaptive pooling layer is located between two average pooling layers, and the random dropout layer and a SoftMax activation function are located between two fully connected layers; the global loss function is: in, represents the feature extraction layer, represents the iterative training layer, Represents the recognition output layer, Loss represents the global loss function, σ represents the number of iterative training layers, G represents the quantization function, represents the calculation accuracy coefficient, Represents the loss function of each layer iterative training, Indicates the preset optimization weight coefficient, Indicates the optimization efficiency of each layer's iterative training.

8. The method for controlling a vehicle hill start according to claim 4, wherein: The judgment correction layer first uses a reconstruction decoder to perform dual reconstruction of the structure and attributes of the data feature matrix to generate a reconstructed decoding value; Among them, S represents the reconstructed decoding value, P() represents the PRelu activation function, H represents the data feature matrix, SM() represents the SigMoid activation function, and T represents transpose; Then, the reconstructed decoded values ​​are arranged in time series according to the real-time monitoring results of the optimal weight matrix to form current time series and speed time series; Finally, the judgment correction layer corrects the current control curve and the speed control curve according to the current timing and the speed timing respectively.

Images