Training method of slip rate prediction model and slip rate prediction method

Abstract

The present application provides a slip rate prediction model training method and a slip rate prediction method, which belongs to the field of slip rate measurement technology. By obtaining sample working parameters of a mobile device; calculating the calibrated slip rate according to the sample working parameters, sample data consisting of the correspondence between the sample working parameters and the calibrated slip rate is obtained; based on the genetic algorithm, the target parameters are determined from the multiple initial parameters included in each deep learning model; after the target parameters are configured in the corresponding deep learning model, the training data in the sample data is used to train each deep learning model; according to the training index, one is selected from multiple trained deep learning models as a slip rate prediction model. Measuring the slip rate based on the slip rate prediction model can improve the accuracy of the slip rate measurement.

Description

Technical Field

[0001] The present application belongs to the technical field of slip rate measurement, and specifically relates to a training method of a slip rate prediction model, a slip rate prediction method, an electronic device and a readable storage medium. Background Art

[0002] During field operations, the driving wheels of mobile devices often slip. If the driving wheels slip excessively, it will reduce field operation efficiency, cause tire wear, energy consumption problems, and also affect crop production. Therefore, it is necessary to accurately measure the slip rate in real time during the operation process so as to control the slip rate within a reasonable range.

[0003] The related technology measures the driving wheel speed through sensors, such as encoder sensors, photoelectric sensors, Hall sensors, magnetoelectric sensors, etc., and measures the body speed of the mobile device through corresponding technologies, such as the Global Navigation Satellite System (GNSS), radar, machine vision, etc. Finally, the slip rate of the mobile device is calculated based on the driving wheel speed and the body speed. The related technology measures the driving wheel speed through a single sensor, and then calculates based on the driving wheel speed and the body speed of the mobile device to obtain the slip rate of the mobile device.

[0004] The related technology uses a single sensor to calculate the rotation speed of the driving wheel. During the measurement process, the measurement signal is prone to noise interference and signal loss, resulting in low accuracy of slip rate measurement. Summary of the invention

[0005] The purpose of the embodiments of the present application is to provide a slip rate prediction model training method, a slip rate prediction method, an electronic device and a readable storage medium, which can solve the problem of low accuracy of slip rate measurement by a single existing sensor.

[0006] In order to solve the above technical problems, this application is implemented as follows:

[0007] In a first aspect, an embodiment of the present application provides a method for training a slip rate prediction model, the method comprising:

[0008] obtaining sample operating parameters of the mobile device;

[0009] Calculating a calibrated slip rate of the mobile device according to the sample operating parameters of the mobile device, and obtaining sample data consisting of a corresponding relationship between the sample operating parameters and the calibrated slip rate;

[0010] Acquire a plurality of deep learning models with different structures, and a plurality of initial parameters included in each of the deep learning models, and determine a target parameter from the plurality of initial parameters included in each of the deep learning models based on a genetic algorithm;

[0011] After configuring the target parameters in the corresponding deep learning models, training each of the deep learning models using the training data in the sample data, and obtaining a training indicator corresponding to each of the deep learning models;

[0012] According to the training index, one of the multiple trained deep learning models is selected as a slip rate prediction model.

[0013] In a second aspect, an embodiment of the present application provides a method for predicting slip rate, the method comprising:

[0014] Get working parameters;

[0015] Inputting the working parameters into a slip rate prediction model to obtain a slip rate; the slip rate prediction model is obtained by training the slip rate prediction model training method described in the first aspect above;

[0016] The working parameters include: one or more of an actual speed of the mobile device obtained by integrating the first speed of the mobile device and the second speed of the mobile device, an acceleration of the mobile device, a pitch angle of the mobile device, a roll angle of the mobile device, a tilling depth of the mobile device, and a resistance of the mobile device.

[0017] In the embodiment of the present application, sample data consisting of the correspondence between the sample working parameters of the mobile device and the calibrated slip rate is obtained, and multiple deep learning models with different structures are obtained, and the target parameters are determined from the multiple initial parameters included in each deep learning model based on the genetic algorithm, and the target parameters are configured in the corresponding deep learning model, and then the training data in the sample data is used to train each deep learning model, and according to the training index, one of the multiple trained deep learning models is selected as a slip rate prediction model. Since the slip rate can be obtained based on the slip rate prediction model, compared with the related art that uses a single sensor to measure the driving wheel speed, and then calculates based on the driving wheel speed and the body speed of the mobile device to obtain the slip rate of the mobile device, it is not affected by the noise interference and signal loss problems of the single sensor measurement signal, and the accuracy of the slip rate measurement can be improved.

[0018] The above description is only an overview of the technical solution of the present application. In order to more clearly understand the technical means of the present application, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present application more obvious and easy to understand, the specific implementation methods of the present application are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0020] Figure 1 A training method for a slip rate prediction model provided in an embodiment of the present application;

[0021] Figure 2 It is a schematic diagram of a deep learning model structure provided in an embodiment of the present application;

[0022] Figure 3 Another method for training a slip rate prediction model provided in an embodiment of the present application;

[0023] Figure 4 is a schematic diagram of a slip rate measurement system for a mobile device provided in an embodiment of the present application;

[0024] Figure 5 is a schematic diagram of a correlation heat map provided in an embodiment of the present application;

[0025] Figure 6 is a flow chart of speed measurement of a photographing device provided in an embodiment of the present application;

[0026] Figure 7 This is a flowchart of selecting target parameters of a deep learning model provided in an embodiment of the present application;

[0027] Figure 8 is a flow chart of obtaining a slip rate prediction model provided by an embodiment of the present application;

[0028] Fig. 9 is a schematic diagram of a slip rate measurement method provided in an embodiment of the present application;

[0029] Fig.10 A slip rate prediction method provided in an embodiment of the present application;

[0030] Fig.11 It is a structural block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0032] The terms "first", "second", etc. in the specification of the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable under appropriate circumstances, so that the embodiments of the present application can be implemented in an order other than those illustrated or described here, and the objects distinguished by "first", "second", etc. are usually a class, and the number of objects is not limited. For example, the first object can be one or more. In addition, "and / or" in the specification represents at least one of the connected objects, and the character " / " generally indicates that the objects associated with each other are in an "or" relationship.

[0033] The following describes in detail the slip rate prediction model training method and the slip rate prediction method provided in the embodiments of the present application through specific embodiments and their application scenarios in conjunction with the accompanying drawings.

[0034] In field operations, the driving wheels of mobile devices often slip due to soil conditions and other reasons. Excessive slippage of the driving wheels will greatly reduce field operation efficiency, cause tire wear and energy consumption problems, and also cause soil compaction, affecting crop production. Therefore, it is necessary to accurately measure the slip rate in real time during the operation process so as to control the slip rate within a reasonable range.

[0035] Since it is difficult to measure the slip rate directly, an indirect calculation method is usually used to obtain the slip rate. Related technologies measure the speed of the driving wheel through sensors, such as encoder sensors, photoelectric sensors, Hall sensors, magnetoelectric sensors, etc., and measure the speed of the mobile device body through corresponding technologies, such as GNSS, radar, machine vision, etc. Finally, the slip rate of the mobile device is calculated based on the speed of the driving wheel and the speed of the body. However, due to the complex and harsh operating environment of the mobile device, there are large vibrations and uneven ground during the operation process. A single sensor is prone to noise interference and signal loss during the test, resulting in a low accuracy rate of slip rate measurement.

[0036] At present, related technologies such as machine vision and GNSS are becoming more and more mature, and their applications in the field of speed measurement are becoming more and more extensive. At the same time, with the rapid development and application of deep learning, it has gradually been applied to the agricultural field. Based on this, in the embodiment of the present application, by obtaining sample data consisting of the corresponding relationship between the sample working parameters of the mobile device and the calibrated slip rate, and obtaining multiple deep learning models with different structures, the target parameters are determined from the multiple initial parameters included in each deep learning model based on the genetic algorithm, and the target parameters are configured in the corresponding deep learning model, and then the training data in the sample data is used to train each deep learning model, and according to the training index, one of the multiple trained deep learning models is selected as a slip rate prediction model. Since the slip rate can be obtained based on the slip rate prediction model, compared with the related art using a single sensor to measure the driving wheel speed, and then calculating based on the driving wheel speed and the body speed of the mobile device to obtain the slip rate of the mobile device, it is not affected by the noise interference and signal loss problems of the single sensor measurement signal, and the accuracy of the slip rate measurement can be improved.

[0037] Figure 1 This is a training method for a slip rate prediction model provided in an embodiment of the present application, see Figure 1 , the method comprising:

[0038] Step 101: Obtain sample operating parameters of a mobile device.

[0039] In the embodiment of the present application, the mobile device is a device with a moving function, for example, the mobile device may be a tractor. The sample working parameter may be a parameter related to the working state of the mobile device when it is working, and is used to reflect the working state of the mobile device. For example, the sample working parameter may include a first speed of the mobile device , the second speed of the mobile device , the third speed of the mobile device , the acceleration a of the mobile device, the roll angle of the mobile device , the pitch angle of the mobile device , the plowing depth H of the mobile device and the plowing resistance F of the mobile device.

[0040] Step 102: Calculate the calibrated slip rate of the mobile device according to the sample operating parameters of the mobile device, and obtain sample data consisting of the corresponding relationship between the sample operating parameters and the calibrated slip rate.

[0041] In the embodiment of the present application, the calibrated slip rate of the mobile device corresponds to the sample working parameters one by one, that is, one set of sample working parameters corresponds to one calibrated slip rate. The calibrated slip rate is a real value, and the additional information associated with the sample working parameters can be used to guide the learning task of the deep learning model.

[0042] The embodiment of the present application calculates the calibrated slip rate through the sample working parameters of the mobile device, thereby obtaining sample data consisting of the corresponding relationship between the sample working parameters and the calibrated slip rate. After obtaining multiple deep learning models with different structures and the target parameters corresponding to each deep learning model, after configuring the target parameters in the corresponding deep learning model, the training data in the sample data is used to train each deep learning model, which can improve the accuracy of slip rate measurement.

[0043] Step 103: Acquire multiple deep learning models with different structures and multiple initial parameters included in each deep learning model, and determine the target parameter from the multiple initial parameters included in each deep learning model based on a genetic algorithm.

[0044] In an embodiment of the present application, the deep learning model can be an artificial neural network (ANN), which can discover nonlinear and uncertain relationships between input data and output data, and also has good associative ability, high fault tolerance and high storage capacity.

[0045] In the embodiment of the present application, multiple deep learning models with different structures can be understood as having different numbers of hidden layers, numbers of hidden layer nodes, activation functions, and optimization algorithms among the multiple deep learning models. Figure 2 This is a schematic diagram of a deep learning model structure provided in an embodiment of the present application, see Figure 2 , the deep learning model may include an input layer, a hidden layer, and an output layer. Among them, the number of hidden layers is 1, and the number of nodes in the hidden layer is a number (only 8 nodes are shown in the figure). It should be noted that the number of hidden layers is not limited to 1, and can also be multiple layers. Usually, the number of hidden layers is 1 or 2.

[0046] As for the number of nodes in the hidden layer, the number of nodes in the hidden layer can usually be calculated according to the following formula:

[0047]

[0048] in, is the number of hidden layer nodes; is the number of neurons in the input layer; is the number of neurons in the output layer; is a positive integer between 0 and 10;

[0049] For the activation function, the following tansig function and logsig function can be selected.

[0050]

[0051]

[0052] in, are sample working parameters.

[0053] In the regression prediction problem, the activation function of the output layer can directly use a linear function. The loss function can be the mean square error function, which is the following formula:

[0054]

[0055] in, is the calibrated slip rate of the deep learning model corresponding to the sample working parameters; is the predicted slip rate of the deep learning model corresponding to the sample working parameter; N is the number of samples; i is the i-th sample working parameter.

[0056] The optimization algorithm is based on the standard back-propagation algorithm and relies on the gradient information provided by the back-propagation algorithm. The gradient information is used to update the network weights in different ways to minimize the loss, improve the generalization ability or accelerate the convergence speed. For example, the optimization algorithm can be an optimization algorithm for solving nonlinear least squares problems, namely the Levenberg-Marquardt algorithm. The algorithm adjusts the weights based on the gradient of the loss function with respect to the network weights and optimizes the loss function to minimize the loss. For another example, the optimization algorithm can be the Quasi-Newton method, which is used to find the local minimum of unconstrained problems. The algorithm does not require second-order derivatives, namely the Hessian matrix, but updates the solution by iteratively constructing an approximate Hessian matrix. The algorithm uses these gradient information to construct the approximate Hessian matrix and perform weight updates. For another example, the optimization algorithm can be Bayesian Regularization, which adds a regularization term to the loss function. It not only calculates the gradient of the loss function, but also calculates the gradient after adding the Bayesian regularization term, so as to consider the influence of regularization when updating the weights. It avoids overfitting by introducing prior distribution and improves the generalization ability of the model. Among them, the Levenberg-Marquardt algorithm and the Quasi-Newton algorithm are often used for fitting regression problems of small networks, while the Bayesian Regularization algorithm can provide better generalization ability.

[0057] The number of iterations of the deep learning model can be pre-set, for example, the number of iterations can be 5000 times, that is, after iterating 5000 times, the deep learning model will stop the training process.

[0058] In an embodiment of the present application, the initial parameter may be a threshold weight parameter of a deep learning model.

[0059] In an embodiment of the present application, the multiple threshold weight parameters included in each deep learning model may be threshold weight parameters generated based on different distributions, or may be randomly generated threshold weight parameters. For example, it may be a threshold weight parameter generated based on a uniform distribution, or it may be a threshold weight parameter generated based on a normal distribution. After obtaining the multiple threshold weight parameters included in each deep learning model, the target threshold weight parameter may be determined from the multiple threshold weight parameters included in each deep learning model based on a genetic algorithm, which may improve the performance of the deep learning model, achieve global optimization, and create new conditions for multi-sensor slip measurement.

[0060] In an embodiment of the present application, target parameters are determined from multiple initial parameters included in each deep learning model based on a genetic algorithm, and target parameters can be selected for each deep learning model, so that model training is performed based on the target parameters, which can improve the generalization ability of each deep learning model.

[0061] Step 104: After configuring the target parameters in the corresponding deep learning model, use the training data in the sample data to train each deep learning model, and obtain the training indicators corresponding to each deep learning model.

[0062] In the examples of this application, see Figure 2 , the input of the deep learning model includes the first speed of the mobile device and the second speed of the mobile device The actual speed of the mobile device obtained by fusion , the acceleration a of the mobile device, the roll angle of the mobile device , the pitch angle of the mobile device , the tillage depth H of the mobile device and the plowing resistance F of the mobile device. The slip rate of the mobile device can be measured by using multiple working parameter information to improve the accuracy of slip rate measurement.

[0063] In an embodiment of the present application, after obtaining the target parameters corresponding to each deep learning model, the target parameters are configured in the corresponding deep learning model, and each deep learning model is trained through the training data in the sample data, which can improve the generalization ability of the deep learning model.

[0064] Step 105: Select one from a plurality of trained deep learning models as a slip rate prediction model according to the training index.

[0065] In the embodiment of the present application, the training index may be a determination coefficient ( ), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE). The formulas are as follows:

[0066]

[0067] in, is the calibrated slip rate of the deep learning model corresponding to the sample working parameters; is the predicted slip rate of the deep learning model corresponding to the sample working parameters; is the average value of the calibrated slip rate of the deep learning model corresponding to the sample working parameters; is the average value of the predicted slip rate of the deep learning model corresponding to the sample working parameters; i is the i-th sample working parameter.

[0068]

[0069] in, is the calibrated slip rate of the deep learning model corresponding to the sample working parameters; is the predicted slip rate of the deep learning model corresponding to the sample working parameter; N is the number of samples; i is the i-th sample working parameter.

[0070]

[0071] in, is the calibrated slip rate of the deep learning model corresponding to the sample working parameters; is the predicted slip rate of the deep learning model corresponding to the sample working parameter; N is the number of samples; i is the i-th sample working parameter.

[0072] In the embodiment of the present application, according to , RMSE and MAPE indicators, select one of the multiple trained deep learning models as the slip rate prediction model. For example, in two deep learning models When the values ​​are the same, the RMSE index of the two deep learning models is looked at, and the deep learning model with the lower RMSE index value is used as the slip rate prediction model. By using specific indicators for evaluation, the evaluation process is made more objective and quantifiable, reducing subjectivity and bias.

[0073] In the embodiment of the present application, after the slip rate prediction model is obtained, the slip rate prediction model may be stored, and the stored slip rate prediction model may be used by calling instructions to implement slip rate prediction.

[0074] In the embodiment of the present application, sample data consisting of the correspondence between the sample working parameters of the mobile device and the calibrated slip rate is obtained, and multiple deep learning models with different structures are obtained, and the target parameters are determined from the multiple initial parameters included in each deep learning model based on the genetic algorithm, and the target parameters are configured in the corresponding deep learning model, and then the training data in the sample data is used to train each deep learning model, and according to the training index, one of the multiple trained deep learning models is selected as a slip rate prediction model. Since the slip rate can be obtained based on the slip rate prediction model, compared with the related art that uses a single sensor to measure the driving wheel speed, and then calculates based on the driving wheel speed and the body speed of the mobile device to obtain the slip rate of the mobile device, it is not affected by the noise interference and signal loss problems of the single sensor measurement signal, and the accuracy of the slip rate measurement can be improved.

[0075] Figure 3 is another slip rate prediction model training method provided in the embodiment of the present application, see Figure 3 , methods include:

[0076] Step 201: Obtain sample operating parameters of a mobile device.

[0077] This step may refer to the above step 101 and will not be described again here.

[0078] Step 202: By inputting the first speed of the mobile device and the second speed of the mobile device into the filter, the actual speed of the mobile device output by the filter is obtained; the sample working parameters of the mobile device include: the first speed of the mobile device and the second speed of the mobile device; the first speed of the mobile device is a speed value measured by the shooting equipment, and the second speed of the mobile device is a speed value measured by the global satellite navigation system.

[0079] In the embodiments of the present application, Figure 4 is a schematic diagram of a slip rate measurement system for a mobile device provided in an embodiment of the present application, see Figure 4 , the moving device is a tractor 41, Figure 4 It includes a visual speed measurement unit 42, a GNSS unit 45, an inertial measurement unit (IMU) 43, a tillage depth measurement unit 47, a plowing resistance measurement unit 48, a wheel speed measurement unit 44, a data acquisition unit (not shown in the figure) and a computer (not shown in the figure).

[0080] The visual speed measurement unit 42 installs a monocular camera on the left front of the tractor. The monocular camera faces the ground vertically and is installed at a height of 330 millimeters (mm). The image acquisition frame rate is 60 frames and the image resolution is 1600*1200. The first speed of the tractor is obtained by acquiring images with the monocular camera. .

[0081] See also Figure 4 , the GNSS unit 45 sends the satellite-measured vehicle speed and time information of the tractor 41 to the computer in real time. For example, the mobile station is installed in the center of the cab of the tractor 41, and the base station simultaneously tracks the full system and full frequency points including the Beidou-3 global signal, thereby obtaining the second speed of the tractor .

[0082] Getting the first speed of the tractor and the tractor's second speed After that, set the tractor to the first speed and the tractor's second speed Input into the filter and get the actual speed of the mobile device as the output of the filter By integrating with the filter, it effectively compensates for the large vibration and uneven ground during tractor operation, the reduced speed measurement accuracy of the shooting equipment, and the reduced speed measurement accuracy when the GNSS satellite signal is unstable, which can improve the adaptability and anti-interference ability of the slip rate measurement method.

[0083] Optionally, step 202 may include:

[0084] Sub-step 2021: obtaining a captured video including the screen content of the working process of the mobile device and the time difference between each frame of the captured video, and using a digital stabilization method to de-jitter the video to obtain a de-jittered video.

[0085] Sub-step 2022: extract images frame by frame from the de-jittered video, and perform grayscale processing and distortion correction processing on the images to obtain processed images, so as to eliminate errors caused by uneven lighting and lens distortion.

[0086] Sub-step 2023: extract features from the processed image to obtain feature points, match the feature points between frames of the processed image, and calculate the moving distance of the mobile device.

[0087] Sub-step 2024: Calculate a first speed of the mobile device based on the moving distance of the mobile device and the time difference between frames.

[0088] For sub-steps 2021-2024, after obtaining the captured video including the screen content of the working process of the mobile device, first, extract the captured video frame by frame to extract the image. Traverse all frame images, extract the features in each frame image, and determine the movement trajectory of the feature points between frames. Then, derive the motion trajectory between two consecutive coordinates of the feature point, use a moving average filter to smooth the motion trajectory, and find the difference between the smoothed trajectory and the original trajectory. Finally, compensate the difference to the original trajectory to achieve a smooth transformation and complete the inter-frame motion correction. Loop the frame and apply the above-mentioned smooth transformation, and output the reconstruction frame by frame to achieve video de-jittering. Due to the large vibration during the operation of the mobile device, the above method is used for motion estimation, and the motion trajectory is derived for smooth transformation to complete the inter-frame correction and achieve the de-jittering effect.

[0089] The images are extracted frame by frame from the de-shaking video, and the images are grayed and distorted to obtain the processed images. The Zhang Zhengyou calibration method is used in advance to obtain the camera internal and external parameter matrix and distortion coefficient for image distortion correction to prevent uneven lighting and lens distortion from affecting image quality.

[0090] Extract features from the processed image to obtain feature points, match feature points between frames of the processed image, calculate the actual moving distance between the same feature points, and then calculate the first speed of the mobile device based on the actual moving distance and the time difference between frames. Specifically, the feature points are extracted and matched using the matching method between adjacent reference frames and the Fast Library for Approximate Nearest Neighbors (FLANN) matching method based on the Scale-Invariant Feature Transform (SIFT) algorithm. The vector displacement coordinates ( , )、( , ), and the displacement coordinates in the world coordinate system are obtained through coordinate transformation ( , )、( , ), and based on the displacement coordinate extremes in the world coordinate system, the actual moving distance between the same feature points is obtained, and the first speed of the mobile device is calculated by combining the time difference obtained by the camera frame rate , refer to the following formula for details:

[0091]

[0092] in, is the first speed of the mobile device, in meters per second (m / s); , ) is the coordinate of the feature point of the reference frame; ( , ) is the coordinate of the feature point in the current frame; △t is the time difference between frames.

[0093] Optionally, the mobile device is a tractor, and before step 202, the method further includes:

[0094] Step A1: When the tractor is not connected to a plow, the tractor drives in different gears to obtain a no-load first speed, a no-load second speed and a no-load theoretical speed corresponding to each gear.

[0095] Step A2: taking the no-load first speed and the no-load second speed as the input of the filter, and taking the no-load theoretical speed as the output of the filter, and training the filter to obtain a trained filter.

[0096] In the embodiment of the present application, the no-load first speed is the speed measured by the shooting equipment when the tractor is not attached to the plow, the no-load second speed is the speed value measured by the GNSS unit when the tractor is not attached to the plow, and the no-load theoretical speed is the speed value measured by the multi-tooth code disk and proximity switch installed on the wheel when the tractor is not attached to the plow.

[0097] In the embodiment of the present application, the tractor travels in different gears, which can be understood as the tractor operating at various predetermined gear ratios.

[0098] For step A1-step A2, the tractor is not connected to a plow and is unloaded on a hard road, traveling in different gears. The unloaded first speed, unloaded second speed and unloaded theoretical speed corresponding to each gear are measured by means of a shooting device, a GNSS unit, and a multi-tooth encoder and a proximity switch installed on the wheels. The unloaded first speed and the unloaded second speed are used as inputs of an adaptive Kalman filter, and the unloaded theoretical speed is used as the true value. The adaptive Kalman filter is trained to obtain the adaptive Kalman filter parameters, thereby completing the construction of the Kalman filter.

[0099] In the embodiment of the present application, a tractor is not connected to a plow, is unloaded on a hard road surface, and travels in different gears to obtain an unloaded first speed, an unloaded second speed, and an unloaded theoretical speed corresponding to each gear. The filter is trained by using the unloaded first speed and the unloaded second speed as filter inputs and the unloaded theoretical speed as filter output to obtain a trained filter. Based on the trained filter, the actual speed of the tractor can be obtained by using the first speed of the tractor and the second speed of the tractor. Since the actual speed of the tractor is not affected by a single sensor, the accuracy of slip rate measurement can be improved.

[0100] Step 203, calculate the difference between the actual speed of the mobile device and the third speed of the mobile device, and calculate the ratio of the difference to the third speed of the mobile device to obtain the calibrated slip rate of the mobile device; the sample operating parameters of the mobile device include: the third speed of the mobile device; the third speed of the mobile device is a speed value measured by a multi-tooth code disk and a proximity switch installed on the wheel.

[0101] In the embodiment of the present application, the multi-size code wheel may be a 90-tooth code wheel.

[0102] See also Figure 4 The wheel speed measuring unit 44 includes an iron 90-tooth code disc 441 and a proximity switch 442 mounted on the rim of the driving wheel. The 90-tooth code disc 441 is fixedly connected to the rim by bolts on the rim and rotates with the wheel concentrically. The proximity switch 442 is fixedly connected to the frame by a bracket, and the sensing end is close to the inner teeth of the 90-tooth code disc 441. The third speed of the tractor is obtained according to the number of teeth of the 90-tooth code disc 441 rotated per second obtained by the proximity switch 442, combined with the power radius of the tractor wheel. , the calculation formula is as follows:

[0103]

[0104] in, is the third speed of the tractor, in m / s, is the number of teeth read by the proximity switch per second, in units of pieces. is the dynamic radius of the wheel, in m.

[0105] In the embodiment of the present application, the calculation of the calibrated slip rate can be seen from the following formula:

[0106]

[0107] Where S is the calibrated slip rate, is the actual speed of the mobile device, is the third speed of the mobile device.

[0108] The sample operating parameters of the mobile device also include the acceleration a of the mobile device, the roll angle of the mobile device , the pitch angle of the mobile device , the tillage depth H of the mobile device and the plowing resistance F of the mobile device, taking the mobile device as a tractor as an example, see Figure 4 , they are introduced one by one below.

[0109] The IMU unit 43 is installed on the chassis of the tractor 41 to be tested and connected to the data acquisition unit. The IMU unit 43 has a built-in three-axis gyroscope and a three-axis accelerometer. When installed, the positive direction of the X axis points to the forward direction of the tractor 41. The X-axis acceleration, the angle of rotation around the X axis, and the angle of rotation around the Y axis measured by the IMU unit 43 correspond to the body acceleration a of the tractor 41, the roll angle And the pitch angle And send the measurement signal to the computer through the data acquisition unit.

[0110] The tillage depth measurement unit 47 installs the inclination sensor on the lifting arm of the tractor three-point suspension 46, measures the angle of the lifting arm, and obtains the relationship between the angle and the tillage depth through field calibration, so as to obtain the tillage depth H of the tractor 41 to be tested in real time when it is in operation. The field calibration can be that when the tractor is connected to the plow, the plow is controlled to perform plowing operations at different heights through the electro-hydraulic suspension control handle, and the angle of the lifting arm and the plowing depth at each height are measured. The angle of the lifting arm and the plowing depth are recorded, and a linear fit is performed to obtain the relationship between the angle of the lifting arm and the tillage depth, so as to obtain the tillage depth H of the tractor 41 when it is in operation through the tillage depth measurement unit 47.

[0111] The plowing resistance measuring unit 48 measures the plowing resistance F of the tractor 41 to be tested when it is operating by means of two force sensors installed at the lower tie rod position of the three-point suspension 46, and sends the data to the computer through the data acquisition unit.

[0112] The tractor 41 performs normal operation in the field at different speeds and tillage depths, and the visual speed measurement unit 42 and the GNSS unit 45 obtain the first speed of the tractor in real time. and tractor second speed And the actual speed of the tractor is obtained by sensor information fusion The IMU unit 43 obtains the vehicle's forward acceleration a and the vehicle body roll angle and pitch angle ; The tillage depth measuring unit 47 is used to monitor the tillage depth H of the tractor; the plowing resistance measuring unit 48 obtains the plowing resistance F of the tractor to be tested; the wheel speed measuring unit 44 obtains the accurate wheel speed to obtain the theoretical speed of the tractor After the above operation information is obtained, the above operation information can be transmitted to the computer through the data acquisition unit.

[0113] Optionally, the method further includes:

[0114] Step B1: draw a scatter plot of each working parameter in the sample data, and remove outliers in the working parameters in the scatter plot.

[0115] Step B2: Analyze the correlation between each operating parameter and the calibrated slip rate corresponding to the operating parameter based on the Spearman correlation, and obtain a first correlation result between each operating parameter and the calibrated slip rate corresponding to the operating parameter.

[0116] Step B3: Analyze the correlation between the various operating parameters based on the Spearman correlation to obtain a second correlation result between the two operating parameters.

[0117] Step B4: if the first correlation result between each working parameter and the calibrated slip rate corresponding to the working parameter does not meet the first preset condition, the working parameter is eliminated; if the first correlation result between each working parameter and the calibrated slip rate corresponding to the working parameter meets the first preset condition, the working parameter is retained.

[0118] Step B5: if the second correlation result between the two working parameters does not satisfy the second preset condition, one of the two working parameters is eliminated; if the second correlation result between the two working parameters satisfies the second preset condition, the two working parameters are retained.

[0119] In the embodiment of the present application, for steps B1 to B5, a scatter plot of each working parameter in the sample data is drawn, the distribution of each working parameter is observed, and the outliers in the working parameters are eliminated. Based on the Spearman correlation analysis, the correlation between each working parameter and the calibrated slip rate corresponding to the working parameter is analyzed to obtain the first correlation result between each working parameter and the calibrated slip rate corresponding to the working parameter. If the absolute value of the first correlation result is greater than or equal to 0.3, the working parameter is retained. If the absolute value of the first correlation result is less than 0.3, the working parameter is eliminated, thereby ensuring the rationality of the selection of the working parameters of the deep learning model.

[0120] The first correlation result may be presented in a correlation heat map to facilitate visualization of the first correlation result. Figure 5 is a schematic diagram of a correlation heat map provided in an embodiment of the present application. Figure 5The actual speed, pitch angle, roll angle, plowing resistance and plowing depth are positively correlated with the slip rate, and the correlation coefficients are 0.60, 0.33, 0.346, 0.67 and 0.43 respectively. The acceleration is negatively correlated with the slip rate, and the correlation coefficient is -0.393. The above six working parameters are significantly correlated at P=0.01.

[0121] Based on the Spearman correlation analysis, the correlation between each working parameter is analyzed to obtain the second correlation result between the two working parameters. If the absolute value of the second correlation result is less than 0.7, the two working parameters are retained. If the absolute value of the second correlation result is greater than or equal to 0.7, the working parameter with a lower correlation result is eliminated based on the correlation result between the working parameter and the calibrated slip rate corresponding to the working parameter. By analyzing the correlation between the working parameters input into the deep learning model, a high correlation between the working parameters can be avoided, thereby reducing the risk of overfitting of the deep learning model due to input redundancy.

[0122] The second correlation results can also be presented as a correlation heat map, which is convenient for visualizing the second correlation results. Figure 5 The absolute values ​​of the correlation coefficients among the actual speed, pitch angle, roll angle, plowing resistance, plowing depth and acceleration are all less than 0.7. Therefore, the above six working parameters as inputs of the deep learning model will not cause the risk of overfitting.

[0123] After performing scatter plot analysis and correlation analysis on the working parameters, in order to prevent the deep learning model from having a preference for features due to the large differences in the numerical ranges of different working parameters, the working parameter values ​​and the calibrated slip rate are normalized respectively to ensure the fairness and accuracy of the deep learning model and improve the convergence speed and stability of the deep learning model. The normalization formula is as follows:

[0124]

[0125] in, is the maximum value of the working parameter value, is the minimum value among the working parameter values, and are the normalized maximum and minimum values, is the working parameter value that needs to be normalized, is the value corresponding to the normalized working parameter value that needs to be normalized.

[0126] Step 204: Acquire multiple deep learning models with different structures and multiple initial parameters included in each deep learning model, and determine the target parameter from the multiple initial parameters included in each deep learning model based on a genetic algorithm.

[0127] This step may refer to the above step 103 and will not be described again here.

[0128] Optionally, step 204 may include:

[0129] Sub-step 2041: construct a population according to a plurality of initial parameters; individuals in the population correspond one to one with the initial parameters.

[0130] Sub-step 2042: Calculate the fitness value of each individual in the population using a preset fitness function.

[0131] Sub-step 2043, according to the genetic algorithm and the fitness value of each individual in the population, the parent individuals in the population are subjected to crossover mutation processing through multiple rounds of iterations, and after reaching a preset number of iterations, the target individual is selected from the new population composed of the new individuals generated in the last iteration, and the initial parameters corresponding to the target individual are used as the target parameters.

[0132] In the embodiment of the present application, the multiple initial parameters may be randomly generated initial parameters, or may be initial parameters generated based on distribution, for example, normal distribution, uniform distribution, etc.

[0133] Taking multiple initial parameters as initial parameter A, initial parameter B and initial parameter C as an example, determining the target parameter from the multiple initial parameters included in each deep learning model based on the genetic algorithm is described in detail.

[0134] The initial parameter A can be represented as an array: [2,4,1,2,5,3], the initial parameter B can also be represented as an array: [1,1.2,1.5,2,2.3,3], and the initial parameter C can also be represented as an array: [0.3,0.8,0.1,1,0.6,0.5]. In the genetic algorithm, the initial parameter A, initial parameter B, and initial parameter C can be represented as initial individual A, initial individual B, and initial individual C respectively. The population of the genetic algorithm may include: initial parameter A: [2,4,1,2,5,3], initial parameter B: [1,1.2,1.5,2,2.3,3], initial parameter C: [0.3,0.8,0.1,1,0.6,0.5].

[0135] The preset fitness function can be a mean square error function. For each individual in the population, its fitness is calculated according to the mean square error function. The fitness value represents the initial parameters of the individual representation and the degree of achievement of the deep learning model goal. The smaller the fitness value, the better the individual performance.

[0136] According to the genetic algorithm and the fitness value of each individual in the population, the parent individuals in the population are subjected to crossover mutation through multiple rounds of iterations. For each parent individual, some of the parent individuals are selected for crossover operation according to the preset crossover probability to generate new offspring individuals. The crossover operation refers to splitting two parent individuals in half, splicing the head of one parent individual with the tail of the other parent individual, and obtaining two new individuals. For example, for parent individuals A: [2,4,1,2,5,3] and parent individuals B: [1,1.2,1.5,2,2.3,3]; a crossover operation can divide parent individuals A and B into 4 slices: [2,4,1], [2,5,3], [1,1.2,1.5], [2,2.3,3], and then splice the slices to obtain new individuals D: [1,1.2,1.5,2,4,1] and new individuals E: [2,5,3,2,2.3,3]. For each offspring individual, according to the preset mutation probability, some of the parent individuals are randomly selected for mutation operations to generate new offspring individuals. Mutation operation refers to randomly changing part of the data structure of the parent individuals to generate new individuals. For example, for the parent individual A: [2,4,1,2,5,3] and the parent individual B: [1,1.2,1.5,2,2.3,3]; a mutation operation can randomly change the data structure in the parent individuals A and B to obtain new individuals F: [2.3,4,1.2,2,1.5,3] and new individual G: [1,4,5,2,2.3,3].

[0137] After reaching the preset number of iterations, the fitness value of the new individual generated by the last iteration is calculated according to the mean square error function, the individual with the smallest fitness is selected as the target individual, and the parameters corresponding to the target individual are used as the target parameters.

[0138] The embodiment of the present application sets the population size, number of iterations, crossover probability and mutation probability in the genetic algorithm, and uses the mean square error as the target fitness function. Based on the genetic algorithm, the target parameters are determined from the multiple initial parameters included in each deep learning model, and the target parameters can be selected for each deep learning model, so that the model training is performed based on the target parameters, which can improve the generalization ability of each deep learning model.

[0139] Step 205: After configuring the target parameters in the corresponding deep learning model, use the training data in the sample data to train each deep learning model, and obtain the training indicators corresponding to each deep learning model.

[0140] This step may refer to the above step 104 and will not be described again here.

[0141] Step 206: Select one from a plurality of trained deep learning models as a slip rate prediction model according to the training index.

[0142] This step may refer to the above step 105 and will not be described again here.

[0143] Optionally, the sample data also includes test data, and step 206 includes:

[0144] Sub-step 2061: input the test data into multiple trained deep learning models respectively to obtain the output corresponding to each trained deep learning model.

[0145] Sub-step 2062: Based on the training index, the output corresponding to each trained deep learning model and the calibrated slip rate corresponding to the test data are calculated to obtain the index value corresponding to each trained deep learning model; the index value reflects the degree of deviation between the test data and the calibrated slip rate corresponding to the test data.

[0146] Sub-step 2063: Select one from a plurality of trained deep learning models as a slip rate prediction model according to the indicator value.

[0147] For sub-steps 2061 to 2063, for example, multiple trained deep learning models include deep learning model 1, deep learning model 2, and deep learning model 3. The test data is input into deep learning model 1, deep learning model 2, and deep learning model 3, respectively, to obtain the outputs corresponding to deep learning model 1, deep learning model 2, and deep learning model 3, respectively. Based on the above , RMSE and MAPE training indicators, calculate the calibrated slip rate corresponding to the output of deep learning model 1, deep learning model 2 and deep learning model 3 and the test data, and obtain the indicator values ​​corresponding to deep learning model 1, deep learning model 2 and deep learning model 3. For the R² indicator, the indicator value is usually between 0 and 1. The closer it is to 1, the better the deep learning model effect; for the RMSE and MAPE indicators, the smaller the indicator value, the better the deep learning model effect. Based on , RMSE and MAPE training indicators, and select a deep learning model as the slip rate prediction model.

[0148] The embodiment of the present application verifies the accuracy of the deep learning models by using test data to verify the multiple deep learning models generated by the above training, and evaluates and compares the performance of deep learning models with different structures of configured target parameters according to training indicators, so as to select the best deep learning model for slip rate prediction, which can improve the accuracy of slip rate prediction.

[0149] In the embodiment of the present application, sample data consisting of the correspondence between the sample working parameters of the mobile device and the calibrated slip rate is obtained, and multiple deep learning models with different structures are obtained, and the target parameters are determined from the multiple initial parameters included in each deep learning model based on the genetic algorithm, and the target parameters are configured in the corresponding deep learning model, and then the training data in the sample data is used to train each deep learning model, and according to the training index, one of the multiple trained deep learning models is selected as a slip rate prediction model. Since the slip rate can be obtained based on the slip rate prediction model, compared with the related art that uses a single sensor to measure the driving wheel speed, and then calculates based on the driving wheel speed and the body speed of the mobile device to obtain the slip rate of the mobile device, it is not affected by the noise interference and signal loss problems of the single sensor measurement signal, and the accuracy of the slip rate measurement can be improved.

[0150] Figure 6 is a flow chart of speed measurement of a photographing device provided in an embodiment of the present application. Figure 6 , the speed measurement process includes:

[0151] Step T1: Capture video by setting the height, focal length and acquisition frame rate of the shooting device.

[0152] Step T2: De-jitter the captured video to obtain a de-jittered video.

[0153] Step T3: extract images from the de-shaking video frame by frame, and perform grayscale processing and distortion correction on the images to obtain processed images.

[0154] Step T4: extract features from the processed image, calculate the actual moving distance between the same feature points, and thereby calculate the first speed of the mobile device based on the actual moving distance and the time difference between frames.

[0155] Figure 7 This is a flowchart of selecting target parameters of a deep learning model provided in an embodiment of the present application, see Figure 7 , the method comprising:

[0156] Step M1: Initialize the parameters of multiple deep learning models to obtain multiple initial parameters corresponding to each deep learning model.

[0157] Step M2: Set the population size, number of iterations, crossover probability, and mutation probability of the genetic algorithm, and use the mean square error as the fitness to select the target individual in the population.

[0158] Step M3: The individuals in the population undergo crossover and mutation to form a population consisting of a new generation of individuals.

[0159] Step M4: Calculate the fitness of the new generation of individuals.

[0160] Step M5, determine whether the iteration is completed. If the iteration is completed, jump to step M6. If the iteration is not completed, jump to step M3.

[0161] Step M6: Select a target individual based on the fitness of the new generation of individuals, and use the parameters corresponding to the target individual as the target parameters.

[0162] Step M7: configure the target parameters in the corresponding deep learning model, and use the training data to train each deep learning model.

[0163] Step M8, determine whether the iteration is completed. If the iteration is completed, jump to step M9; if the iteration is not completed, jump to step M7.

[0164] Step M9: obtain multiple trained deep learning models.

[0165] Figure 8 is a flow chart of obtaining a slip rate prediction model provided in an embodiment of the present application, see Figure 8 , the method comprising:

[0166] Step N1, preprocessing the sample working parameters, including: removing abnormal values ​​in the sample working parameter values, performing correlation analysis on the sample working parameters, normalizing the sample data, and dividing the sample data into training data and test data.

[0167] Step N2: Determine the number of hidden layers, the number of hidden layer nodes, the activation function, and the optimization algorithm of the deep learning model to obtain multiple deep learning models with different structures.

[0168] Step N3: Initialize multiple deep learning models with different structures to obtain multiple initial parameters included in each deep learning model.

[0169] Step N4, set the population size, number of iterations, crossover probability, mutation probability and fitness function of the genetic algorithm, perform crossover mutation on the individuals in the population, and after reaching the number of iterations, select the target individual from the new population composed of the new individuals generated by the last iteration, and use the initial parameters corresponding to the target individual as the target parameters.

[0170] Step N5: configure the target parameters in the corresponding deep learning model, and train each deep learning model using the training data.

[0171] Step N6: After the number of iterations is reached, one is selected from multiple deep learning models based on the training index as a slip rate prediction model.

[0172] Step N7: storing the slip rate prediction model.

[0173] Fig. 9is a schematic diagram of a slip rate measurement method provided in an embodiment of the present application, see Fig. 9 , the method comprising:

[0174] Step P1, obtaining a first speed of the mobile device, a second speed of the mobile device, a third speed of the mobile device, an acceleration of the mobile device, a roll angle of the mobile device, a pitch angle of the mobile device, a plowing depth of the mobile device, and a plowing resistance of the mobile device.

[0175] Step P2: input the first speed of the mobile device and the second speed of the mobile device into an adaptive Kalman filter to obtain an actual speed of the mobile device.

[0176] Step P3: Build multiple deep learning models with different structures.

[0177] Step P4: Select one from the multiple deep learning models as a slip rate prediction model based on the training index, and store the slip rate prediction model.

[0178] Step P5: calling the stored slip rate prediction model through a calling instruction to predict the slip rate.

[0179] In the embodiment of the present application, sample data consisting of the correspondence between the sample working parameters of the mobile device and the calibrated slip rate is obtained, and multiple deep learning models with different structures are obtained, and the target parameters are determined from the multiple initial parameters included in each deep learning model based on the genetic algorithm, and the target parameters are configured in the corresponding deep learning model, and then the training data in the sample data is used to train each deep learning model, and according to the training index, one of the multiple trained deep learning models is selected as a slip rate prediction model. Since the slip rate can be obtained based on the slip rate prediction model, compared with the related art that uses a single sensor to measure the driving wheel speed, and then calculates based on the driving wheel speed and the body speed of the mobile device to obtain the slip rate of the mobile device, it is not affected by the noise interference and signal loss problems of the single sensor measurement signal, and the accuracy of the slip rate measurement can be improved.

[0180] Fig.10 This is a slip rate prediction method provided in the embodiment of the present application, see Fig.10 , the method comprising:

[0181] Step 301: Obtain working parameters.

[0182] Step 302: Inputting the working parameters into the slip rate prediction model to obtain the slip rate; the slip rate prediction model is obtained by training using a slip rate prediction model training method.

[0183] In an embodiment of the present application, the working parameters include: one or more of the actual speed of the mobile device obtained by merging the first speed of the mobile device and the second speed of the mobile device, the acceleration of the mobile device, the pitch angle of the mobile device, the roll angle of the mobile device, the tilling depth of the mobile device, and the resistance of the mobile device.

[0184] In the embodiment of the present application, sample data consisting of the correspondence between the sample working parameters of the mobile device and the calibrated slip rate is obtained, and multiple deep learning models with different structures are obtained, and the target parameters are determined from the multiple initial parameters included in each deep learning model based on the genetic algorithm, and the target parameters are configured in the corresponding deep learning model, and then the training data in the sample data is used to train each deep learning model, and according to the training index, one of the multiple trained deep learning models is selected as a slip rate prediction model. Since the slip rate can be obtained based on the slip rate prediction model, compared with the related art that uses a single sensor to measure the driving wheel speed, and then calculates based on the driving wheel speed and the body speed of the mobile device to obtain the slip rate of the mobile device, it is not affected by the noise interference and signal loss problems of the single sensor measurement signal, and the accuracy of the slip rate measurement can be improved.

[0185] The present application embodiment discloses an electronic device, such as Fig.11 As shown, it includes: a processor 401 and a memory 402, wherein the memory 402 stores programs or instructions that can be run on the processor 401, and when the program or instruction is executed by the processor 401, any of the methods described above is implemented.

[0186] An embodiment of the present application discloses a computer-readable storage medium, wherein a computer program stored in the storage medium enables a processor to execute any of the methods described above.

[0187] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0188] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present application, all of which are within the protection of the present application.

Claims

1. A method for training a slip rate prediction model, characterized in that: include: obtaining sample operating parameters of the mobile device; The mobile device includes a tractor; sample working parameters of the mobile device include: a first speed of the mobile device, a second speed of the mobile device, a third speed of the mobile device, an acceleration of the mobile device, a roll angle of the mobile device, a pitch angle of the mobile device, a tilling depth of the mobile device, and a plowing resistance of the mobile device; the first speed of the mobile device is a speed value measured by a photographing device; the second speed of the mobile device is a speed value measured by a global satellite navigation system; the third speed of the mobile device is a speed value measured by a multi-tooth code disk and a proximity switch mounted on a wheel; By inputting the first speed of the mobile device and the second speed of the mobile device into a filter, an actual speed of the mobile device output by the filter is obtained; Calculating a difference between an actual speed of the mobile device and a third speed of the mobile device, and calculating a ratio of the difference to the third speed of the mobile device to obtain a calibrated slip rate of the mobile device, and obtaining sample data consisting of a correspondence between the sample operating parameters and the calibrated slip rate; Acquire a plurality of deep learning models with different structures, and a plurality of initial parameters included in each of the deep learning models, and determine a target parameter from the plurality of initial parameters included in each of the deep learning models based on a genetic algorithm; After configuring the target parameters in the corresponding deep learning models, training each of the deep learning models using the training data in the sample data, and obtaining a training indicator corresponding to each of the deep learning models; According to the training index, one of the multiple trained deep learning models is selected as a slip rate prediction model.

2. The training method according to claim 1, characterized in that: The obtaining of sample operating parameters of the mobile device includes: Acquire a captured video including the screen content of the working process of the mobile device and the time difference between each frame of the captured video, and remove jitter from the video using a digital stabilization method to obtain a jitter-removed video; Extracting the image frame by frame from the de-shaking video, and performing grayscale processing and distortion correction processing on the image to obtain a processed image, so as to eliminate errors caused by uneven illumination and lens distortion; Extracting features from the processed image to obtain feature points, matching the feature points between frames of the processed image, and calculating the moving distance of the mobile device; A first speed of the mobile device is calculated based on the moving distance of the mobile device and the inter-frame time difference.

3. The training method according to claim 1, characterized in that: The mobile device is a tractor; before obtaining the actual speed of the mobile device by inputting the first speed of the mobile device and the second speed of the mobile device into the filter, the method further includes: When the tractor is not connected to a plow, the tractor drives at different gears to obtain a no-load first speed, a no-load second speed and a no-load theoretical speed corresponding to each gear; The no-load first speed and the no-load second speed are used as filter inputs, and the no-load theoretical speed is used as the filter output. The filter is trained to obtain a trained filter.

4. The training method according to claim 1, characterized in that: The method further comprises: Draw a scatter plot of each of the operating parameters in the sample data, and remove abnormal values ​​in the operating parameters in the scatter plot; Analyzing the correlation between each of the operating parameters and the calibrated slip rate corresponding to the operating parameter based on the Spearman correlation, to obtain a first correlation result between each of the operating parameters and the calibrated slip rate corresponding to the operating parameter; Analyze the correlation between the operating parameters based on the Spearman correlation to obtain a second correlation result between two operating parameters; If the first correlation result between each of the working parameters and the calibrated slip rate corresponding to the working parameter does not meet the first preset condition, the working parameter is eliminated; if the first correlation result between each of the working parameters and the calibrated slip rate corresponding to the working parameter meets the first preset condition, the working parameter is retained; If the second correlation result between the two working parameters does not meet the second preset condition, one of the two working parameters is eliminated; if the second correlation result between the two working parameters meets the second preset condition, the two working parameters are retained.

5. The training method according to claim 1, characterized in that: The determining the target parameter from the multiple initial parameters included in each of the deep learning models based on the genetic algorithm includes: Constructing a population according to the multiple initial parameters; the individuals in the population correspond one to one to the initial parameters; Using a preset fitness function, respectively calculating the fitness value of each individual in the population; According to the genetic algorithm and the fitness value of each individual in the population, crossover mutation processing is performed on the parent individuals in the population through multiple rounds of iterations, and after reaching a preset number of iterations, a target individual is selected from a new population composed of new individuals generated in the last iteration, and the initial parameters corresponding to the target individual are used as target parameters.

6. The training method according to claim 1, characterized in that: The sample data also includes test data; The selecting, according to the training indicator, one of the multiple trained deep learning models as a slip rate prediction model comprises: Inputting the test data into the plurality of trained deep learning models respectively to obtain an output corresponding to each of the trained deep learning models; Based on the training index, the output corresponding to each of the trained deep learning models and the calibrated slip rate corresponding to the test data are calculated to obtain an index value corresponding to each of the trained deep learning models; the index value reflects the degree of deviation between the test data and the calibrated slip rate corresponding to the test data; According to the index value, one is selected from the plurality of trained deep learning models as a slip rate prediction model.

7. A method for predicting slip rate, characterized in that: include: Get working parameters; Inputting the working parameters into a slip rate prediction model to obtain a slip rate; The slip rate prediction model is obtained by training the slip rate prediction model training method according to any one of claims 1 to 6; The working parameters include: one or more of an actual speed of the mobile device obtained by integrating the first speed of the mobile device and the second speed of the mobile device, an acceleration of the mobile device, a pitch angle of the mobile device, a roll angle of the mobile device, a tilling depth of the mobile device, and a resistance of the mobile device.

8. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a program or instruction that can be run on the processor, and when the program or instruction is executed by the processor, the method according to any one of claims 1 to 7 is implemented.

9. A readable storage medium, characterized in that: When the instructions in the readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the method as described in any one of method claims 1-7.

Images