Skid steer loader working stage identification method and device and skid steer loader

Abstract

The invention discloses a method and device for recognizing the working stage of a skid loader and the skid loader, and relates to the technical field of skid loaders. The method comprises the steps that working parameters of a target skid loader are collected in real time; performing data preprocessing on the working parameters acquired at the current moment to obtain preprocessed working parameters at the current moment; performing feature extraction on the preprocessed working parameters at the current moment by using a feature extraction module and a feature extraction method of fractional calculus feature enhancement to obtain feature vectors; the feature extraction module is obtained by optimizing a deep learning model; inputting the feature vector into a working stage recognition model to obtain the working stage of the target skid loader at the current moment; the working stage recognition model is obtained by training a preset model by adopting a data set. According to the invention, the recognition accuracy of the working stage is improved.

Description

Technical Field

[0001] The present application relates to the technical field of skid steer loaders, and particularly to a method and device for identifying the working stage of a skid steer loader and a skid steer loader. Background Art

[0002] As a multi-functional construction machinery, a skid steer loader has the characteristics of flexibility and high efficiency, so it has a wide range of application scenarios in port terminals, landscaping, warehousing handling, agricultural operations and other scenarios. Skid steer loaders widely use engines to drive hydraulic pumps for energy supply, and throttle control is used to distribute and transfer power through a multi-way valve, which will cause serious energy loss. In order to better achieve the efficient operation of skid steer loaders, it is necessary to identify the working stage of each skid steer loader, so as to actively adjust the system operation parameters, improve the matching degree between the power source and the load, and achieve the goal of energy conservation and emission reduction. At the same time, it also has important reference significance in evaluating driver operation habits, production efficiency, etc.

[0003] There are mainly two methods for identifying the working stage of a skid steer loader. The first method is the working stage identification method based on computer vision. This method uses image and video acquisition devices installed at the construction site or on-vehicle cameras to collect images in real time, and uses image processing algorithms such as target clustering and background subtraction to identify the working conditions of the skid steer loader in real time. However, this method requires a large amount of time and economic costs to obtain the image training set, and is greatly affected by environmental illumination, image occlusion, dust particles, etc. The second method is the working stage identification method based on multi-sensor information fusion. This method uses radio technologies such as ultra-wideband, radio frequency identification, and global positioning system installed on the skid steer loader to obtain information such as the acceleration and direction of the device to identify the working stage of the skid steer loader. However, since the installation position, quantity, and transmission distance of the sensors will all affect the identification accuracy, and the working stages that can be identified are relatively simple, it is applicable to devices with relatively simple actions such as trucks and cranes, and the identification accuracy for complex operation devices such as skid steer loaders is relatively low. The third method is to use time-frequency feature vectors such as mean, variance, and standard deviation extracted from signals such as the internal operating parameters of the device itself, such as main pump pressure and operation handle signals, as the judgment basis, and to achieve the identification of the working stage through the constructed classification model. However, since the extraction of feature vectors is greatly affected by human factors and cannot fully reflect the global characteristics of the signals, it is easy to cause problems such as low learning ability and poor accuracy of the classification model. Summary of the Invention

[0004] The purpose of the present application is to provide a method and device for identifying the working stage of a skid steer loader and a skid steer loader, which can improve the accuracy of working stage identification.

[0005] To achieve the above purpose, the present application provides the following solutions: In a first aspect, the present application provides a method for identifying the working stage of a skid steer loader, and the method for identifying the working stage of the skid steer loader includes: Collect the working parameters of the target skid steer loader in real time; Perform data preprocessing on the working parameters collected at the current moment to obtain the preprocessed working parameters at the current moment; Use a feature extraction module and a feature extraction method with fractional-order calculus feature enhancement to extract feature vectors from the preprocessed working parameters at the current moment; the feature extraction module is obtained by optimizing a deep learning model; Input the feature vector into the working stage recognition model to obtain the working stage of the target skid steer loader at the current moment; the working stage recognition model is obtained by training a preset model using a data set.

[0006] In a second aspect, the present application provides a device for identifying the working stage of a skid steer loader, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor executes the computer program to implement the method for identifying the working stage of the skid steer loader according to any one of the above; The device for identifying the working stage of the skid steer loader further includes an in-vehicle display, and the in-vehicle display is used to display the working stage output by the working stage recognition model.

[0007] In a third aspect, the present application provides a skid steer loader including the device for identifying the working stage of the skid steer loader.

[0008] According to the specific embodiments provided by the present application, the following technical effects are disclosed: The present application provides a method, a device, and a skid steer loader for identifying the working stage of a skid steer loader. By using a feature extraction module and a feature extraction method with fractional-order calculus feature enhancement, feature vectors are extracted from the preprocessed working parameters at the current moment. Feature enhancement is performed on the feature extraction of the working parameters through the feature extraction module and fractional-order calculus feature enhancement, which fully reflects the feature details of the extracted working parameters and improves the accuracy of working stage recognition. Description of the Drawings

[0009] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required for use in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0010] Figure 1 It is a schematic flowchart of a method for identifying the working stage of a skid steer loader provided by an embodiment of the present application.

[0011] Figure 2 Schematic diagram of the comparison effect of noise reduction processing on the collected original data provided by an embodiment of the present application.

[0012] Figure 3 Schematic diagram of the operation path provided by an embodiment of the present application.

[0013] Figure 4 Schematic diagram of a skid steer loader working stage identification device provided by an embodiment of the present application.

[0014] Figure 5 Schematic diagram of a skid steer loader structure provided by an embodiment of the present application.

[0015] Reference numerals: 1 - operating handle, 2 - vehicle-mounted processor, 3 - vehicle-mounted memory, 4 - vehicle-mounted display, 5 - oil suction filter, 6 - discrete variable gear pump, 7 - fuel tank, 8 - permanent magnet synchronous motor, 9 - electromagnetic overflow valve, 10 - check valve, 11 - manifold block, 12 - electromagnetic directional valve, 13 - boom cylinder, 14 - bucket cylinder, 15 - first pressure sensor, 16 - second pressure sensor, 17 - third pressure sensor. Detailed implementation manners

[0016] Next, the technical solutions in the embodiments of the present application will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0017] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.

[0018] The present application provides a skid steer loader working stage identification method, as Figure 1 shown, the skid steer loader working stage identification method includes steps 101 to 104.

[0019] Step 101: Real-time collect the working parameters of the target skid steer loader.

[0020] Step 102: Perform data preprocessing on the working parameters collected at the current moment to obtain the preprocessed working parameters at the current moment.

[0021] Step 103: Use a feature extraction module and a feature extraction method enhanced by fractional calculus features to extract feature vectors from the preprocessed working parameters at the current moment; the feature extraction module is obtained by optimizing a deep learning model.

[0022] Step 104: Input the feature vector into the working stage recognition model to obtain the working stage of the target skid steer loader at the current moment; the working stage recognition model is obtained by training a preset model using a data set.

[0023] In an exemplary embodiment, step 101 specifically includes: collecting the working parameters of the target skid steer loader in real time at a set frequency, and executing the processes of steps 102 to 104 for each collected working parameter, so as to continuously obtain the working stage at each sampling moment, realize the perception of the real-time working stage of the whole vehicle of the skid steer loader, and provide a reference for the further control of the skid steer loader.

[0024] When the skid steer loader is a hydraulically powered skid steer loader, the working parameters include the outlet pressure of the working pump, the pressure in the rodless cavity of the boom cylinder, and the pressure in the rodless cavity of the bucket cylinder.

[0025] The working stages include idling travel, digging and loading, full-load travel, boom lifting, tipping unloading, and bucket lowering.

[0026] In an exemplary embodiment, step 102 specifically includes: performing noise reduction and normalization processing on the working parameters collected at the current moment in sequence to obtain the preprocessed working parameters at the current moment.

[0027] The noise reduction method adopted in this application is wavelet filtering noise reduction, specifically using 5-level wavelet decomposition and the unbiased likelihood estimation criterion (Rigorous Sure threshold). Since the pressure signal changes continuously and there is no strong mutation, the soft threshold function is used for processing, and Daubechies 4, which can balance the time and frequency characteristics, is used as the wavelet basis function.

[0028] The wavelet transform is used to convert the preprocessed working parameters at the current moment into different scale spaces, and then the noise is screened out through threshold processing, and then the screened-out noise is discarded. The data after discarding the noise is used for reconstruction to obtain the noise-reduced working parameters.

[0029] The threshold processing mainly includes two parts, one part is the selection of the threshold, and the other part is the selection of the threshold function. The Rigrsure threshold is selected for the threshold, and the soft threshold function is selected for the threshold function.

[0030] The normalization method used in this application is the maximum-minimum (Min-Max) normalization method to process the data, and linearly transform it to [-1, 1]. The normalization formula can be specifically expressed as: .

[0031] Where, is the original pressure signal after noise reduction, and z is the data after normalization processing. and are respectively the maximum and minimum values of the original pressure signal data after noise reduction.

[0032] The noise reduction and normalization processing in this application are to avoid the relatively large weight ratio of feature parameters with large values from affecting the true relationship between the pressure signal and the working stage.

[0033] In an exemplary embodiment, step 103 specifically includes: using a feature extraction module to extract features from the preprocessed working parameters at the current moment to obtain a first feature.

[0034] Using a feature extraction method with fractional calculus feature enhancement to extract features from the preprocessed working parameters at the current moment to obtain a second feature.

[0035] Concatenate the first feature and the second feature to obtain a feature vector, so as to avoid manual intervention in the pressure signal feature extraction process. This application can achieve the expansion of the feature domain without additional sensors, and can effectively improve the recognition accuracy of the working stage.

[0036] In an exemplary embodiment, using a feature extraction method with fractional calculus feature enhancement to extract features from the preprocessed working parameters at the current moment to obtain a second feature specifically includes: adopting the Riemann-Liouville theorem method to obtain the continuous forms of the fractional differential and integral of the preprocessed working parameters at the current moment.

[0037] The continuous forms of its fractional differential and integral are as follows respectively: ; ; where n is the order, a is the start time, T is the cut-off time, t is the time slot between the start time and the cut-off time, represents the function at the time point T the fractional differential; represents the function at the time point T the fractional integral, is the gamma function; m represents an integer, here m = 1, represents the function to be integrated and differentiated, is the time slot t the pressure signal collected at time.

[0038] Discretize the continuous forms of the fractional differential and integral respectively, and use the obtained fractional features as the second feature.

[0039] The formula for discretizing the continuous form of fractional differential and integral is as follows: ; ; Among them, k is an integer. Since in this embodiment, it is set that S = 4, the following fractional-order features need to be calculated , where is the collected pressure signal, and the number of types of pressure signals is 3. In summary, the feature vector extracted by fractional calculus is 12-dimensional.

[0040] The fractional calculus feature enhancement technology realizes the expansion of the feature domain through the differentiation and integration of the pressure signal with respect to time, and does not require additional sensors. Any fractional integral and derivative of pressure between -1 and 1 will generate a new feature, which can represent impulse and pressure change rate respectively. The number of fractions S represents the number of fractional steps in the range [-1, 1]. In this application, S = 4, which includes the fractional ranges of [-1, -0.5, 0.5, 1]. The dimension of the feature vector extracted in this part is 12, that is, the vector dimension of the second feature is 12.

[0041] In an exemplary embodiment, the deep learning model is a long short-term memory network integrated with an attention mechanism. The long short-term memory network integrated with an attention mechanism includes a first long short-term memory network, a second long short-term memory network, a dropout layer, and a layer containing a multi-head attention mechanism connected in sequence, which realizes the extraction of features from the input pressure signal from different angles in parallel. The vector dimension of the first feature output by the deep learning model is 6.

[0042] For example, set the number of attention heads to h = 8, and use the following formula to perform a linear transformation on the output data matrix X after the dropout layer: .

[0043] Among them, represents the output of the i-th attention head, , represents the calculation formula of attention, Q represents the query matrix, K represents the key matrix, V represents the value matrix, represents the query transformation matrix of the i-th attention head, represents the key transformation matrix of the i-th attention head, represents the value transformation matrix of the i-th attention head. , is the output dimension of the embedding layer, and the embedding layer is an embedding layer inside the attention mechanism, , is Matrix sum The dimension of the matrix, is the dimension of the matrix during the model training phase and equal, , and both are within the real number range.

[0044] Next, use the following formula to perform scaled dot product calculation. Multiply the generated with by matrix multiplication, divide by , and use the softmax function to convert it into a probability distribution, and then multiply with matrix to perform Attention() calculation to obtain matrix.

[0045] .

[0046] Repeat the above process in parallel h times. Use the Concat function in the following formula to concatenate the matrices and multiply with matrix, , generating a new matrix with the same dimension as the input matrix X to obtain the final output of multi-head attention, which is within the real number range.

[0047] . is the final output of multi-head attention.

[0048] In an exemplary embodiment, the deep learning model is optimized, specifically including: using random search to optimize the hyperparameters of the long short-term memory network with a fused attention mechanism; the hyperparameters of the long short-term memory network with a fused attention mechanism include the number of hidden units, the maximum number of iterations, the time step, the initial learning rate, the number of samples in each batch, the dropout rate, and the number of attention heads.

[0049] The deep learning model includes, but is not limited to, the long short-term memory network based on the fused attention mechanism, and can also be the gated recurrent network based on the fused attention mechanism and the convolutional neural network based on the fused attention mechanism, etc.

[0050] In this application, to improve the feature extraction performance of the long short-term memory network with a fused attention mechanism, the hyperparameters of this network structure are randomly optimized 200 times.

[0051] Feature-level data fusion adopts the ordinary splicing method, and the fused feature can be expressed as , where n is the number of feature extractors, represents the feature vector generated by using the nth feature extractor. Since there are only two parts in this application, namely the long short-term memory network feature extractor with a fusion attention mechanism and the fractional-order calculus feature enhancement extractor, and the lengths of the feature vectors are 6 and 12 respectively, an 18-dimensional feature vector is finally constructed.

[0052] In an exemplary embodiment, a preset model is trained using a data set, which specifically includes: based on the data set, using Bayesian optimization to optimize the hyperparameters of the preset model, and finding the best hyperparameter configuration when the hyperparameter range is set relatively large to achieve the highest recognition accuracy in the working stage; the preset model uses LIBSVM or a random forest model, and the hyperparameters of LIBSVM include the penalty coefficient and the gamma value. Based on the constructed 18-dimensional feature vector, the optimized LIBSVM is used to perceive the real-time working stage of the skid steer loader, providing a reference for further system control.

[0053] The following describes the process of optimizing the hyperparameters of LIBSVM using Bayesian optimization.

[0054] Among them, represents the two hyperparameters, the penalty coefficient and the gamma value, represents the hyperparameter a set of hyperparameters in the search space. First, initialize to obtain the data set , to evaluate the objective function.

[0055] .

[0056] Among them, is the pressure data, is the label data, and the label data is the working stage, , and n is the number of samples in the data set.

[0057] The surrogate model uses a Gaussian process, assuming that the objective function obeys a normal distribution, that is: .

[0058] Among them, is the mean function; is the covariance function, represents the hyperparameter another set of hyperparameters in the search space.

[0059] The acquisition function adopted in this embodiment is the expected improvement: .

[0060] Among them, is the current optimal value, represents the hyperparameter The current optimal set of hyperparameters for the search space indicates the likelihood of expected improvement under the new hyperparameters When determining the observation point, the hyperparameter configuration of the observation point is input into the objective function to obtain the function objective value. Then this sample point and the function objective value are used to further update the observation set Continue to repeat the above process until the maximum number of iterations is reached or the performance metric is satisfied. At this point, the optimization is complete, and the model can be used for identification in the actual working stage T In this application, both random optimization and Bayesian optimization evaluate the effect of hyperparameter optimization through four metrics: accuracy, precision, recall, and F1-score

[0061] Accuracy refers to the ratio of the number of samples correctly classified by the model to the total number of samples

[0062] The formula for accuracy is expressed as:

[0063] where .

[0064] Here, is the accuracy represents the number of instances that are actually positive and predicted to be positive represents the number of instances that are actually negative and predicted to be negative represents the number of predictions that are predicted to be positive but are actually negative represents the number of predictions that are predicted to be negative but are actually positive

[0065] Precision refers to the proportion of true positives among all samples predicted to be positive by the model

[0066] The formula for precision is expressed as: , is the precision

[0067] Recall refers to the proportion of true positives correctly predicted as positive by the model among all true positive samples

[0068] The formula for recall is expressed as: , is the recall

[0069] The F1-score is the harmonic mean of precision and recall and is commonly used in class imbalance problems. It takes into account both precision and recall and is a balanced metric between the two

[0070] The formula for the F1-score is expressed as: , is the F1-score

[0071] The method for hyperparameter optimization of the deep learning model or the preset model in this application can also be hyperparameter optimization methods such as grid search and sparrow search.

[0072] The construction of the dataset includes: collecting the working parameters under the preset operating conditions of the skid steer loader, that is, the working parameters under typical operating conditions, at the set sampling frequency, and constructing a dataset for the skid steer loader under typical operating conditions. Each sample in the dataset includes working parameters and the working stage corresponding to the working parameters.

[0073] When constructing the dataset, the sampling frequency is preset according to the processing capacity of the in-vehicle processor. For example, when the performance of the in-vehicle processor is weak, the sampling frequency can be set to 10Hz. When the processor has a strong ability to process high-speed information, the sampling frequency can be set to 20Hz or 50Hz. In addition, in order to ensure the accuracy and representativeness of the acquired data, when constructing the dataset, 5 drivers with more than 3 years of operating experience and relevant operation certificates should be hired to complete the data collection. Considering the occurrence of phenomena such as driver errors and data collection interruptions, 40 sets of operation cycles need to be completed for each round of experiments, and finally 200 sets of working cycle experimental sample data are obtained. In this way, a large amount of working stage data, the pressure data corresponding to each sampling time point, and the working stage can be collected within the preset time period.

[0074] According to the basis for dividing the working stage, a large number of collected data samples are divided into working stages. This stage can complete the synchronous collection of the original data by installing displacement sensors and pressure sensors on the boom cylinder and bucket cylinder of the skid steer loader, etc., and use the data that can intuitively reflect the working stage, such as the displacement sensor, as the basis for division determination to avoid the interference of human factors on the stage division. For example, assuming that the sampling frequency of the data is 10Hz and the preset time is 1 hour, then 36,000 sets of operation cycle data corresponding to the pressure signal and the displacement sensor signal of the skid steer loader within the preset time are obtained. Using the displacement sensor signal as the basis for determining the working stage, that is, a sample dataset in which 36,000 pressure signals correspond one-to-one with the working stage is obtained.

[0075] This application performs noise reduction and normalization processing on the data in the dataset. Due to the influence of factors such as environmental interference and vehicle vibration, first, noise reduction processing needs to be performed on the collected original data, such as Figure 2 shown, Figure 2 In part (a), it is the original pressure signal, and in part (b), it is the pressure signal after noise reduction. The noise reduction method for the original data is wavelet filtering noise reduction. Five-level wavelet decomposition and the Rigorous Sure threshold selection criterion are adopted. Since the pressure signal changes continuously and there will be no strong mutations, the soft threshold function is used for processing, and Daubechies 4, which can balance the time and frequency characteristics, is used as the wavelet basis function.

[0076] When constructing the dataset, the driver can collect the original data according to the operation path as Figure 3 shown. It is a typical V-shaped operation condition of a skid steer loader. In stage PⅠ, the driver operates the skid steer loader towards the material pile A; after loading the material, the vehicle drives backward in reverse gear and adjusts the direction in real time, which is stage PⅡ; when the direction adjustment is completed and the front of the skid steer loader faces the dumping pile B, the vehicle adjusts to the forward gear and drives forward, which is stage PⅢ; after the driver completes the unloading, he drives the skid steer loader backward to the original starting point, which is stage PⅣ, and continues the next operation cycle. Stage PⅠ includes two stages: no-load driving and shoveling and loading. Stage PⅡ belongs to the full-load driving stage. The early stage of stage PⅢ belongs to full-load driving. When approaching the dumping pile, it belongs to the boom lifting stage at this time, and the later stage belongs to the tipping and unloading stage. Stage PⅣ includes two stages: bucket retraction and lowering and no-load driving.

[0077] In an exemplary embodiment, the division of the dataset includes, but is not limited to, dividing the training set and the validation set according to 7:3, or it can also be 8:2, etc., or adding a test set, and dividing the dataset into a training set, a test set, and a validation set according to 6:2:2.

[0078] Using the data of the test set, compare the classification performance of the Back-Propagation Neural Network (BPNN), LIBSVM, Long Short-Term Memory (LSTM), Random Forest (RF), the model combining Gated Recurrent Unit and Attention mechanism (GRU-Attention), and the method of this application. The results are shown in Table 1. It can be seen from the data in Table 1 that the method of this application has advantages in terms of accuracy, precision, recall rate, and F1 score.

[0079] Table 1 Classification accuracy of each model on the skid steer loader

[0080] In an exemplary embodiment, the data collected for the working stage identification of the skid steer loader includes, but is not limited to, the outlet pressure of the working pump, the pressure in the rodless cavity of the boom cylinder, and the pressure in the rodless cavity of the bucket cylinder. It can also be the pilot pressure or electrical signal of the handle corresponding to the boom and bucket movements, or a composite signal composed of the pilot electrical signal of the handle, the outlet pressure of the working pump, the pressure in the rodless cavity of the boom cylinder, and the pressure in the rodless cavity of the bucket cylinder. When the working device of the skid steer loader is driven by a pure electric ball screw actuator, the system parameters collected are the displacements and torques of the boom ball screw actuator and the bucket ball screw actuator.

[0081] In some embodiments, the data acquisition operation conditions include, but are not limited to Figure 3 the V-shaped operation condition described above, and may also be the I-shaped operation condition, etc.

[0082] In an exemplary embodiment, the present application provides a control system for identifying the working stage of a skid steer loader. The control system is a control system that applies a method for identifying the working stage of a skid steer loader. When the working stage result is identified in real time, the control system adjusts the engine operating point based on this to ensure the high efficiency and energy saving of the entire skid steer loader. If the power source of the skid steer loader is an electric motor, the motor speed and the displacement of the connected variable pump or discrete variable gear pump can be adjusted to achieve a certain degree of volume control. In addition, the actual cycle time and the relative cycle time ratio of each working stage of the skid steer loader can be used as feedback for novice operators to specifically improve their driving skills. Furthermore, the working stage identification result of the skid steer loader can also be used for production efficiency analysis, so as to provide a reference for the project manager and dynamically optimize the scheduling of construction site machines to achieve the maximum utilization of resources.

[0083] The present application first obtains the working pump outlet pressure, the pressure of the rodless chamber of the boom cylinder, and the pressure signal data of the rodless chamber of the bucket cylinder of the skid steer loader, and performs noise reduction and normalization processing on the above pressure signal data; then realizes the extraction and construction of feature vectors through a long short-term memory network integrated with an attention mechanism and a fractional order calculus feature enhancement technology; then inputs the feature vectors into a classification model to obtain the working stage identification result of the skid steer loader; finally, uses the identification result as the judgment basis for controlling the variable speed and variable displacement of the hydraulic system of the working device of the skid steer loader to achieve the purpose of energy saving and efficient operation.

[0084] In addition, the working stage identification result can also be used as a criterion for evaluating the operating habits of skid steer loader drivers, and for determining whether a driver needs to receive driving training; the identification result can also be indirectly used to evaluate the time required for a project to be completed, which has reference value for project planning management and productivity evaluation.

[0085] In the present application, there is no need to install external sensors such as cameras, and the working stage of the skid steer loader is identified in real time completely based on the real-time data of the vehicle's own structure, avoiding problems such as interference by human factors and a large number of sensors in traditional feature extraction methods, effectively reducing the number of sensors used while improving the identification accuracy, and reducing the economic cost.

[0086] This application no longer needs to intercept pressure signal segment data through a sliding time window to extract time-frequency features such as mean and variance, avoiding the interference of human factors in the time-frequency feature selection process. Even if features with a relatively large influence in the time-frequency features are extracted through methods such as principal component analysis, the process is relatively cumbersome. The feature vector reflecting the characteristics of pressure signal data can be directly extracted through a long short-term memory network based on a fusion attention mechanism. Through the fractional-order calculus feature enhancement technology, the expansion of the feature domain can be realized without increasing the number of sensors, which is simple, economical, efficient, and has obvious effects.

[0087] Based on the same inventive concept, an embodiment of this application also provides a skid steer loader working stage identification device for implementing the skid steer loader working stage identification method involved above. The solution provided by this device to solve the problem is similar to the solution described in the above method. Therefore, the specific limitations in one or more embodiments of the skid steer loader working stage identification device provided below can refer to the limitations on the skid steer loader working stage identification method in the above text, and will not be repeated here.

[0088] In an exemplary embodiment, this application provides a skid steer loader working stage identification device. The skid steer loader working stage identification device includes: a memory, a processor, and a computer program stored on the memory and executable on the processor. The processor executes the computer program to implement any one of the skid steer loader working stage identification methods.

[0089] Specifically, the memory is an on-vehicle memory 3, and the processor is an on-vehicle processor 2.

[0090] The skid steer loader working stage identification device further includes an on-vehicle display 4, and the on-vehicle display 4 is used to display the working stage output by the working stage identification model.

[0091] The skid steer loader working stage identification device further includes a pressure sensor, and the pressure sensor is used to collect the outlet pressure of the working pump, the pressure in the rodless cavity of the boom cylinder, and the pressure in the rodless cavity of the bucket cylinder.

[0092] As Figure 4 shown, the skid steer loader working stage identification device further includes the operation handle 1 of the skid steer loader, the oil suction filter 5 of the working device hydraulic system, the discrete variable gear pump 6, the fuel tank 7, the permanent magnet synchronous motor 8 driving the gear pump, the electromagnetic overflow valve 9, the one-way valve 10, the manifold block 11, the electromagnetic directional valve 12, the boom cylinder 13, and the bucket cylinder 14. Since the control system of the bucket cylinder 14 adopts the same hydraulic system as that of the boom cylinder 13, Figure 4 it is not shown repeatedly here, and only the bucket cylinder 14 is schematically shown. Figure 4It can be understood that when the operating handle 1 moves, the hydraulic system needs to provide corresponding flow rate at this time to drive the boom cylinder 13 to complete corresponding actions. However, it can be seen that the permanent magnet synchronous motor 8 synchronously drives two discrete variable gear pumps 6 with different displacements. For a certain value of flow rate demand, at this time, due to the permutations and combinations of displacement and motor speed, there can be different solutions to achieve the target flow rate. The purpose of this embodiment is to use the real-time recognition result of the working stage of the skid steer loader as an additional input to reasonably switch the motor speed and the displacement of the gear pump. The motor refers to the permanent magnet synchronous motor 8, and the gear pump is the discrete variable gear pump 6, so as to keep the motor and the gear pump in the high-efficiency area as much as possible on the premise of minimizing the switching impact, and finally realize the efficient utilization of the vehicle energy and improve the endurance.

[0093] The skid steer loader working stage recognition device further includes a first pressure sensor 15, a second pressure sensor 16 and a third pressure sensor 17. As Figure 4 shown, the first pressure sensor 15 is used to detect the pressure signal of the rodless cavity of the boom cylinder to obtain the pressure of the rodless cavity of the boom cylinder, the second pressure sensor 16 is used to detect the pressure signal of the rodless cavity of the bucket cylinder to obtain the pressure of the rodless cavity of the bucket cylinder, and the third pressure sensor 17 is used to detect the pressure signal at the outlet of the working pump to obtain the pressure at the outlet of the working pump.

[0094] In an exemplary embodiment, the skid steer loader working condition recognition result is applied to the control system. For example, when the system recognizes that it is in the shoveling and loading stage at this time, since in the typical working condition, the subsequent stages are full-load driving and boom lifting stages, the full-load driving stage mainly affects the driving part of the vehicle and is not within the scope of consideration in this embodiment. In the subsequent boom lifting stage, since there are certain requirements for the lifting time, the hydraulic system needs to provide the maximum flow rate, and the flow rate has a certain proportional relationship with the handle tilt angle (0-20°). During shoveling and loading, the boom is in a micro-motion state. Therefore, when the handle continuously increases the tilt angle and does not exceed 5°, the motor first increases the speed. When it is subsequently detected that the angle of the boom handle continues to increase beyond 5° and is in the boom lifting stage, the control system first continuously changes the displacement. After all the gear pumps are connected to the system, the motor then continues to increase the speed. This embodiment is only a simple application of the skid steer loader working stage recognition result on the control system, and the application of the present application includes but is not limited to this.

[0095] In an exemplary embodiment, the identification result of the working stage of the skid steer loader can be used to evaluate the driving habits of the driver and make targeted improvements. For example, by identifying the working stage of the skid steer loader, the actual cycle time and relative cycle time of the six stages of the skid steer loader, namely, empty-load driving, shoveling and loading, full-load driving, boom lifting, and tipping bucket, can be obtained. The training company can use the identification result information of each stage as feedback to provide targeted training for inexperienced drivers and improve their operating skills.

[0096] In an exemplary embodiment, the identification result of the working stage of the skid steer loader can be used to evaluate the production efficiency of each machine. When there is a strict construction deadline for an earthwork construction project, the site manager, contractor, and construction company can track and monitor the operating efficiency of each machine during the operation, so as to optimize the project plan, reasonably dispatch the machines inside and outside the site, ensure the effective utilization of resources, maximize the equipment utilization rate, and accurately budget the project construction process.

[0097] In an exemplary embodiment, as Figure 5 shown, the present application provides a skid steer loader, and the skid steer loader includes a skid steer loader working stage identification device.

[0098] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0099] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the steps in the above method embodiments are implemented.

[0100] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0101] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the various embodiments provided in this application can include at least one of non-volatile and volatile memories. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0102] The databases involved in the various embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the various embodiments provided in this application can be general-purpose processors, central processors, graphics processors, digital signal processors, data processing logics of programmable logics, etc., without limitation.

[0103] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as these combinations of technical features do not conflict, they should all be considered as within the scope described in this specification.

[0104] In this article, specific examples are used to elaborate on the principles and implementation manners of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, according to the idea of this application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. A method for identifying the working stage of a skid steer loader, characterized in that, The method for identifying the working stage of the skid steer loader includes: Collecting the working parameters of the target skid steer loader in real time; Performing data preprocessing on the working parameters collected at the current moment to obtain the preprocessed working parameters at the current moment; Using a feature extraction module and a feature extraction method enhanced by fractional calculus features to extract feature vectors from the preprocessed working parameters at the current moment; the feature extraction module is obtained by optimizing a deep learning model; Inputting the feature vectors into a working stage identification model to obtain the working stage of the target skid steer loader at the current moment; the working stage identification model is obtained by training a preset model using a data set.

2. The method for identifying the working stage of a skid steer loader according to claim 1, wherein When the skid steer loader is a hydraulically powered skid steer loader, the working parameters include the outlet pressure of the working pump, the pressure in the rodless chamber of the boom cylinder, and the pressure in the rodless chamber of the bucket cylinder; The working stages include no-load travel, digging and loading, full-load travel, boom lifting, tipping and discharging, and bucket retracting and lowering.

3. The method for identifying the working stage of a skid steer loader according to claim 1, characterized in that, Performing data preprocessing on the working parameters collected at the current moment to obtain the preprocessed working parameters at the current moment, specifically including: Successively performing noise reduction and normalization processing on the working parameters collected at the current moment to obtain the preprocessed working parameters at the current moment.

4. The method for identifying the working stage of a skid steer loader according to claim 1, wherein Using a feature extraction module and a feature extraction method enhanced by fractional calculus features to extract feature vectors from the preprocessed working parameters at the current moment, specifically including: Using the feature extraction module to extract the first feature from the preprocessed working parameters at the current moment; Using the feature extraction method enhanced by fractional calculus features to extract the second feature from the preprocessed working parameters at the current moment; Concatenating the first feature and the second feature to obtain a feature vector.

5. The method for identifying the working stage of a skid steer loader according to claim 4, wherein Using the feature extraction method enhanced by fractional calculus features to extract the second feature from the preprocessed working parameters at the current moment, specifically including: Adopting the Riemann-Liouville theorem method to obtain the continuous form of the fractional differential and integral of the preprocessed working parameters at the current moment; Respectively performing discretization processing on the continuous form of the fractional differential and integral, and taking the obtained fractional features as the second feature.

6. The method for identifying the working stage of a skid steer loader according to claim 1, wherein The deep learning model is a long short-term memory network integrated with an attention mechanism. The long short-term memory network integrated with an attention mechanism includes a first long short-term memory network, a second long short-term memory network, a dropout layer, and a layer containing a multi-head attention mechanism connected in sequence.

7. The method for identifying the working stage of a skid steer loader according to claim 6, wherein Optimizing the deep learning model, specifically including: Using random search to optimize the hyperparameters of the long short-term memory network integrated with an attention mechanism; the hyperparameters of the long short-term memory network integrated with an attention mechanism include the number of hidden units, the maximum number of iterations, the time step, the initial learning rate, the number of samples in each batch, the dropout rate, and the number of attention heads.

8. The method for identifying the working stage of a skid steer loader according to claim 1, wherein Training a preset model using a data set, specifically including: Based on the data set, using Bayesian optimization to optimize the hyperparameters of the preset model; the preset model uses LIBSVM or a random forest model.

9. A working stage recognition device for a skid steer loader, comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that the processor executes the computer program to implement the skid steer loader working stage identification method according to any one of claims 1-8; The skid steer loader working stage identification device further includes an in-vehicle display, and the in-vehicle display is used to display the working stage output by the working stage identification model.

10. A skid steer loader, characterized in that, It includes the skid steer loader working stage identification device according to claim 9.

Images