User driving condition acquisition method, device and equipment and storage medium

Abstract

The invention relates to the technical field of vehicles, and discloses a user driving condition acquisition method and device, equipment and a storage medium, which are used for improving the prediction precision of the user driving condition. The method comprises the following steps: acquiring original vehicle state data, wherein the original vehicle state data comprises original driving data and original vehicle body data; preprocessing the original driving data and the original vehicle body data to obtain target driving data and target vehicle body data; taking the target driving data and the target vehicle body data as two-stage input, and respectively inputting the two-stage input into a pre-trained deep learning model so as to predict the driving working condition of the user; wherein the deep learning model comprises a convolutional neural network, a long-short term memory network and a residual network.

Description

Technical Field

[0001] The present invention relates to the field of vehicle technology, and in particular to a method, device, equipment and storage medium for acquiring a user's driving condition. Background Art

[0002] User driving conditions are required in chassis structural fatigue durability analysis, chassis energy consumption prediction, electric vehicle range calculation, or electric vehicle energy control and other related technologies. However, the actual driving conditions of vehicles are usually difficult to measure directly.

[0003] The inventors found that in order to accurately obtain vehicle operating data, domestic and foreign scholars have mainly proposed two types of analysis technologies: virtual road spectrum technology based on vehicle dynamics analysis and recognition technology based on machine learning. The former first conducts simulation analysis by establishing a vehicle dynamics model including road surface, tires, rigid parts, and elastic parts; the latter first uses a suitable machine learning model to directly predict the operating information of the vehicle's driving process based on easily accessible vehicle operating parameters.

[0004] Among them, the current recognition technology based on machine learning still uses very traditional machine learning models, resulting in poor accuracy in predicting users' driving conditions. Summary of the invention

[0005] The present invention provides a method, device, equipment and storage medium for obtaining a user's driving condition, so as to solve the problem that the machine learning model is very traditional and the traditional solution leads to poor accuracy in condition prediction.

[0006] A method for obtaining a user's driving condition, comprising:

[0007] Acquire original vehicle status data, where the original vehicle status data includes original driving data and original vehicle body data;

[0008] Preprocessing the original driving data and the original vehicle body data to obtain target driving data and target vehicle body data;

[0009] The target driving data and target vehicle body data are used as two-level inputs and are respectively input into a pre-trained deep learning model to predict the user's driving conditions;

[0010] Among them, deep learning models include convolutional neural networks, long short-term memory networks, and residual networks.

[0011] Optionally, the convolutional neural network includes a first convolutional layer, a second convolutional layer, a first pooling layer, a third convolutional layer, a fourth convolutional layer, a second pooling layer, a fifth convolutional layer, a sixth convolutional layer, a seventh convolutional layer and a plurality of eighth convolutional layers connected in sequence, each eighth convolutional layer is connected in sequence, and the target driving data is used to input the first convolutional layer;

[0012] Among them, the output of the fifth convolutional layer and the output of the seventh convolutional layer pass through the first residual network and are input into the first eighth convolutional layer among multiple eighth convolutional layers. The output of the first eighth convolutional layer and the output of the last eighth convolutional layer pass through the second residual network and are input into the long short-term memory network. The target vehicle body data is used to input the long short-term memory network.

[0013] Optionally, the convolutional neural network includes six eighth convolutional layers.

[0014] Optionally, the user's driving condition includes the road surface type and the load borne by the wheel center.

[0015] Optionally, the original driving data includes driving speed, steering wheel angle and vehicle body acceleration; and the original vehicle body data includes suspension type, wheel load and tire height.

[0016] Optionally, preprocessing the original driving data and the original vehicle body data to obtain target driving data and target vehicle body data includes:

[0017] The original driving data and the original body data are respectively trimmed to delete invalid and abnormal data, and the driving trimmed data and the body trimmed data are obtained;

[0018] When it is determined that the first target type data has a data amount less than a preset data amount in the vehicle tailoring data and the vehicle body tailoring data, reconstructing the target type data;

[0019] The reconstructed driving clipping data and body clipping data are normalized respectively to obtain target driving data and target body data.

[0020] Optionally, original vehicle status data is obtained, including:

[0021] Get the current driving scene information;

[0022] When it is determined that the current driving scene meets the preset driving scene based on the driving scene information, the original vehicle state data is obtained.

[0023] A device for acquiring a user's driving condition, comprising:

[0024] An acquisition module is used to acquire original vehicle status data, where the original vehicle status data includes original driving data and original vehicle body data;

[0025] A preprocessing module, used for preprocessing the original driving data and the original vehicle body data to obtain target driving data and target vehicle body data;

[0026] The prediction module is used to input the target driving data and the target vehicle body data as two-level inputs into the pre-trained deep learning model to predict the user's driving condition;

[0027] Among them, deep learning models include convolutional neural networks, long short-term memory networks, and residual networks.

[0028] A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the above-mentioned method for acquiring a user's driving condition is implemented.

[0029] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for obtaining a user's driving condition is implemented.

[0030] In the scheme implemented by the above-mentioned method, device, computer equipment and storage medium for obtaining user driving conditions, the deep learning model provided in the embodiment of the present application is used to predict the user's driving conditions. Since the deep learning model is a composite deep learning model including a convolutional neural network, a long short-term memory network and a residual network and has been trained and verified, it is able to quickly and accurately obtain the user's driving conditions during actual use of the user for predicting user driving conditions with long time series, high dynamic characteristics and strong nonlinear characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0032] Figure 1 It is a schematic diagram of an overall framework of a method for obtaining a user's driving condition in one embodiment of the present invention;

[0033] Figure 2 is a schematic diagram of a network structure of a deep learning model in one embodiment of the present invention;

[0034] Figure 3 is a schematic diagram of a training process of a deep learning model in one embodiment of the present invention;

[0035] Figure 4 is a flow chart of a method for obtaining a user's driving condition in one embodiment of the present invention;

[0036] Figure 5 is a structural schematic diagram of a device for acquiring a user's driving condition in one embodiment of the present invention;

[0037] Figure 6 It is a structural diagram of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION

[0038] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] The embodiment of the present invention provides a method for obtaining a user's driving condition, which can be applied to obtaining a user's driving condition of various vehicles. The method for obtaining a user's driving condition can be implemented by a server or a vehicle end, and the server can be implemented by an independent server or a server cluster composed of multiple servers, which is not specifically limited.

[0040] Considering that the current user driving condition recognition technology based on machine learning still has the technical problem that the machine learning model used is very traditional, resulting in poor condition prediction accuracy, the embodiment of the present application provides a method for obtaining user driving conditions based on a new deep learning model. Figure 1 As shown, Figure 1 This is a framework diagram of an embodiment of the present application, including data collection and preprocessing processes. After the data collection and preprocessing processes, two levels of model input are formed, namely input layer 1 and input layer 2, wherein input layer 1 and input layer 2 are different input data, respectively. The input data of input layer 1 is used to input the first layer in the deep learning model, and the input data of input layer 2 is used to input the n+1th layer in the deep learning model. Finally, the output result of the user's driving condition, that is, the prediction result, is output through output layer 1. The input data of input layer 1 is the target driving data obtained after preprocessing, including driving speed, steering wheel angle and vehicle body acceleration, wherein the vehicle body acceleration may include longitudinal acceleration, lateral acceleration and vertical acceleration; the input data of input layer 2 is the target vehicle body data obtained after preprocessing, including suspension type, wheel load and tire height.

[0041] To achieve Figure 1 The prediction scheme shown requires the construction of a suitable deep learning model. To facilitate the understanding of this application, the model structure and input-output relationship of the deep learning model provided in the embodiment of this application are first described. Figure 2 As shown, it is a schematic diagram of the network structure of a deep learning model provided in an embodiment of the present application, from Figure 2It can be seen from the network structure that the deep learning model includes Convolutional Neural Networks (CNN), Long Short-Term Memory Networks (LSTM) and Residual Networks (ResNets). It should be noted that for problems such as user driving condition prediction that have long time series, high dynamic characteristics, and strong nonlinear characteristics, a suitable deep learning model network architecture is very important. Therefore, it is necessary to select the number of network layers, the type of each layer, the number of neurons in each layer, the connection method between each layer, the form of the training error function, the learning rate, the number of batches and other deep learning model structure parameters, and repeatedly train, test and verify each deep learning model to select the optimal prediction model.

[0042] In various deep learning model frameworks, convolutional neural networks (including convolutional layers and pooling layers) can obtain global information of input samples at a high training speed, thereby ensuring that the subsequent predicted wheel center load has a high overall prediction accuracy; the long short-term memory network layer (LSTM layer) can obtain the timing delay of the wheel center load and improve the local peak load prediction accuracy; the residual network (ResNets layer) can improve the convergence efficiency of the multi-layer deep learning model. The inventor combined the prediction results of the user's driving conditions and the characteristics of the input data, and combined these three deep learning network frameworks to build a Figure 2 The composite deep learning model shown can improve the prediction accuracy of problems with long time series, high dynamic characteristics, and strong nonlinear characteristics.

[0043] In one embodiment, the convolutional neural network includes a first convolutional layer, a second convolutional layer, a first pooling layer, a third convolutional layer, a fourth convolutional layer, a second pooling layer, a fifth convolutional layer, a sixth convolutional layer, a seventh convolutional layer, and a plurality of eighth convolutional layers connected in sequence, and the positional relationship between the first convolutional layer, the second convolutional layer, the first pooling layer, the third convolutional layer, the fourth convolutional layer, the second pooling layer, the fifth convolutional layer, the sixth convolutional layer, the seventh convolutional layer, and the plurality of eighth convolutional layers is as follows: Figure 2 As shown in the sequence, the target driving data is used to input the first convolutional layer;

[0044] Among them, the output of the fifth convolutional layer and the output of the seventh convolutional layer pass through the first residual network and are input into the first eighth convolutional layer among multiple eighth convolutional layers. The output of the first eighth convolutional layer and the output of the last eighth convolutional layer pass through the second residual network and are input into the long short-term memory network. The target vehicle body data is used to input the long short-term memory network.

[0045] Specifically, for example, Figure 2As shown, the depth of the first convolutional layer and the second convolutional layer is 64, the depth of the third convolutional layer and the fourth convolutional layer is 128, the depth of the fifth convolutional layer, the sixth convolutional layer and the seventh convolutional layer is 256, and the depth of each eighth convolutional layer in the multiple eighth convolutional layers is 512. As an example, the multiple second convolutional layers include six eighth convolutional layers, and the depths are 521 respectively. Among them, 16, 8 and 4 in the input layer represent the converted size of the input data. It should be understood that before the target driving data and the target vehicle body data are respectively input into the convolutional neural network, the data needs to be processed, including conversion into a data form that the convolutional neural network can process. For example, the data is converted into a matrix form, so as to facilitate the convolution and processing operations of each convolutional layer of the convolutional neural network. For example, in Figure 2 In the example, the input data of size 16 is transformed into data of size 8 after being pooled by the pooling layer. Figure 2 structure.

[0046] It should be noted that Figure 2 The specific convolutional layer and pooling layer parameters shown in the example are only for illustrative purposes. In combination with the characteristics of the input data, other convolutional layer and / or pooling layer parameters may also be used, without specific limitation. In addition, the number of the second convolutional layer may also vary according to the situation, without specific limitation.

[0047] In addition, as an example, since the actual collected data usually has a serious distribution imbalance problem, for example, the number of large load samples is very small, and the user's driving condition prediction needs to focus on the prediction accuracy of large load samples, to address this problem, the Focal Loss error evaluation function can be used in the deep model, and the error evaluation of the extreme load samples of the output layer can be weighted to achieve efficient and accurate prediction. Of course, in other examples, other loss functions can also be used, without specific limitation.

[0048] In one embodiment, a method for training a deep learning model is provided, such as Figure 3 As shown, including:

[0049] S101: Acquire training data, where the training data includes original training driving data and original training vehicle body data;

[0050] S102: preprocessing the original training driving data and the original training vehicle body data to obtain target training driving data and target training vehicle body data;

[0051] S103: Using the target training driving data and the target training vehicle body data as two-level inputs, respectively inputting them into the deep learning model for training until the prediction accuracy of the deep learning model reaches a preset accuracy.

[0052] Among them, in order to distinguish the vehicle body data and driving data in subsequent application embodiments, the driving data and vehicle body data during the training process are respectively recorded as original training driving data and original training vehicle body data, and the driving data and vehicle body data after preprocessing are respectively recorded as target training driving data and target training vehicle body data.

[0053] It can be seen that in this application, a framework combining three deep learning networks is provided to build a Figure 2 The composite deep learning model shown, based on the structure of the deep learning model, needs to be trained with training data until a trained deep learning model is obtained.

[0054] In this embodiment, the target interpretation is first required, and the target interpretation mainly includes two points: a. Define the prediction target: For the prediction of user driving conditions based on deep learning, it is first necessary to define the category of user driving conditions that need to be identified. For the road surface type, it is necessary to determine whether the road surface type is divided according to the region or according to the ground roughness. The road surface type usually includes mountain roads, urban roads, general roads and expressways according to the region. According to the road surface roughness, it can be divided into expressways, bump roads, washboard roads, brake roads, pothole roads, cobblestone roads, stone roads, gravel roads, resonance roads, etc.; it should be noted that the above road surface types are only exemplary and do not limit this application. b. Determine the use scenario: For the prediction of user driving conditions, since it will consume huge hardware resources to turn on the prediction function when the user is fully engaged, it is necessary to determine in which scenarios to turn on the function. In one embodiment, the prediction target includes the road surface type and the wheel center load; in other embodiments, the prediction target can be the road surface type or the wheel center load, or even other user driving conditions, which are not specifically limited.

[0055] After the target is interpreted, the data collection work that needs to be carried out mainly includes two parts: a. Use feature selection methods to screen out parameters that have a significant impact on the user's driving conditions and driving style; b. Carry out data collection tests. This process requires reasonable arrangements for the vehicles, personnel, equipment, test sites and environment, data recording tables, data transmission and storage methods, etc.

[0056] Among them, the original driving data and the original body data are obtained as training data, which are recorded as training driving data and original training body data. After obtaining the training driving data and the original training body data, the original training driving data and the original training body data are preprocessed to obtain the target training driving data and the target training body data. In this process, the main thing is to determine the collected user sample data for training and verifying the deep learning model; preprocess the data input to the deep learning model so that it meets the data requirements of the deep learning model and ensures the validity and rationality of the input data, thereby improving the reliability of the deep learning model.

[0057] It should be noted that, as mentioned above, since the user operating condition information predicted by the deep learning model includes two parts: road surface type and wheel center load, therefore, in the process of model training, the following three parts are mainly controlled: a. Control data sampling frequency, segmented sample data length, number of reconstructions of small sample data, division into training set data and validation set data, etc.; b. Model structure parameters such as the network architecture of the deep learning model, the number of network layers, the type of each layer, the number of neurons in each layer, the connection method between each layer, the form of the training error function, the learning rate, and the number of batches; c. Control the road surface type prediction accuracy to reach more than 90%, the wheel center load prediction accuracy to reach more than 80%, and the average pseudo-damage value of the predicted load to reach more than 80% of the real load pseudo-damage. Among them, when the prediction accuracy of the above-trained deep learning model reaches the required preset accuracy, it is determined that the trained deep learning model is used in the actual user driving condition prediction. It should be noted that the above-mentioned specific accuracy is only an exemplary description here and does not limit this application.

[0058] In one embodiment, the trained deep learning model can be mounted on the vehicle for testing and verification, and the deep learning model can be further updated and verified according to any problems that arise, so as to obtain a stable deep learning model for use in predicting user driving conditions.

[0059] During the training process, the trained model is first tested and verified on the offline data that has been collected, and the deep learning model is further updated and verified for any problems that arise. Then, the user's driving condition prediction is carried out on the actual vehicle, and the prediction accuracy and operating efficiency of the model, as well as the software and hardware stability and reliability of the entire prediction method, are judged by comparing the gap between the prediction results and the actual values. In this process, the trained deep learning model is predicted and verified through offline test field data, and predicted and verified through online user data, which can improve the effectiveness of the deep learning model and fully verify it.

[0060] Exemplarily, the road surface type prediction accuracy is controlled to reach more than 90%, the wheel center load prediction accuracy is controlled to reach more than 80%, and the average pseudo-damage value of the predicted load is controlled to reach more than 80% of the actual load pseudo-damage. That is, when the prediction accuracy of the deep learning model trained above reaches the required preset accuracy, the training is completed.

[0061] As an example, the target training driving data may include preprocessed driving speed, steering wheel angle and body acceleration, wherein the body acceleration includes longitudinal acceleration, lateral acceleration and vertical acceleration; the input data of input layer 2 is the target training body data obtained after preprocessing, including suspension type, wheel load and tire height.

[0062] As another example, the target training driving data may include preprocessed driving speed and vehicle body acceleration, wherein the vehicle body acceleration includes longitudinal acceleration, lateral acceleration and vertical acceleration; the input data of input layer 2 is the target training vehicle body data obtained after preprocessing, including suspension type, wheel load and tire height.

[0063] In addition, it may also be other parameter data that is screened out using a feature selection method and has a significant impact on the user's driving conditions and driving style, and the specific embodiments of the present application are not limited thereto.

[0064] In addition, in one embodiment, when collecting training data, the data can be directly obtained on the existing actual vehicle through the sensors installed on the vehicle itself, without the need to install new sensors, which can also reduce costs.

[0065] Among them, the main work that needs to be carried out in the preprocessing process mainly includes the following parts: a. Segment and trim the original training driving data and the original training body data: delete invalid data and abnormal data in the test process, such as invalid transition road data and abnormal burr data. In the trimming process, it is necessary to determine the segment length of each input parameter by analyzing the time lag between the input parameter and the predicted value of the deep learning model; c. The actual collected data usually has uneven distribution of various types of data, and the data values ​​vary greatly. To address these problems, it is necessary to reconstruct a small number of extreme loads, such as replica data with white noise signals; d. Normalize the data and divide the processed sample data into training set, validation set and test set, which is conducive to improving the scope of use scenarios of deep learning models.

[0066] In one embodiment, taking acceleration as an example, the judgment process of clipping data in the above data preprocessing process is as follows:

[0067] max(abs(az[t:t+T]))>0.4g or sum(abs(az[t:t+T]))>(0.06*f)g

[0068] Or max(abs(ax[t:t+T]))>0.3g or sum(abs(ax[t:t+T]))>(0.04*f)g

[0069] Or max(abs(ay[t:t+T]))>0.3g or sum(abs(ay[t:t+T]))>(0.06*f)g

[0070] Where T is the duration of each signal segment, f is the sampling frequency, a[t:t+T] represents the acceleration data collected from t to t+T, and this time interval is called the judgment domain, ax, ay, az represent the longitudinal, lateral and vertical accelerations at the center of mass of the vehicle, respectively, and g represents the acceleration of gravity.

[0071] If the above conditions are met, the acceleration data ax[t:t+T], ay[t:t+T], az[t:t+T] and other parameter data during the period from t to t+T / 2 are included in the valid data set; if the above conditions are not met, the data during the period from t to t+T / 2 are discarded; the time interval of each judgment domain is T / 2.

[0072] That is to say, the maximum absolute value of acceleration and the cumulative absolute value of acceleration of the training sample data per unit time will be calculated, and then it will be determined whether the corresponding maximum absolute value of acceleration and the cumulative absolute value of acceleration meet the conditions. If they meet the conditions, it means that these training sample data are valid. The valid training sample data are segmented, and the relatively small number of extreme loads are replicated with white noise signals. Finally, the data is normalized to obtain the target training sample data.

[0073] It should be noted that the data preprocessing process mainly includes cropping, feature selection, segmentation, reconstruction and normalization operations, so that the preprocessed model input data not only meets the data requirements of the deep learning model, but also ensures the validity and rationality of the input data, and improves the reliability of the deep learning model. There is no specific limitation.

[0074] It can be seen that in this embodiment, a deep learning model training method is provided, and the collected data is pre-processed, mainly including cropping, feature selection, segmentation, reconstruction and normalization operations; then a suitable deep learning network architecture is selected, and the deep learning model structure parameters are initialized, and the deep learning model parameters are trained and updated, and the model is trained and iterated using sample data to obtain a deep learning model that can finally be used to predict actual user driving conditions.

[0075] It should be noted that, in the above embodiments, the deep learning model and the training process provided in the embodiments of the present application are described. Next, a method for acquiring user driving conditions based on the deep learning model is described.

[0076] In one embodiment, if Figure 4 As shown, a method for obtaining a user's driving condition is provided, comprising the following steps:

[0077] S10: Acquire original vehicle status data, where the original vehicle status data includes original driving data and original vehicle body data;

[0078] S20: pre-processing the original driving data and the original vehicle body data to obtain target driving data and target vehicle body data;

[0079] S30: taking the target driving data and the target vehicle body data as two-level inputs, respectively inputting the pre-trained deep learning models to predict the user's driving condition; wherein the deep learning model includes a convolutional neural network, a long short-term memory network, and a residual network.

[0080] In this embodiment, after the deep learning model is trained, the newly collected data is used to predict the user's driving condition. First, the original vehicle state data needs to be obtained, and the original vehicle state data includes the original driving data and the original body data. Then, the original driving data and the original body data are preprocessed in a preprocessing manner to obtain the target driving data and the target body data. The content of the deep learning model obtained by training can be seen in the above embodiment, and will not be repeated in this embodiment.

[0081] As an example, the target driving data may include pre-processed driving speed, steering wheel angle and body acceleration, wherein the body acceleration includes longitudinal acceleration, lateral acceleration and vertical acceleration; the target body data includes suspension type, wheel load and tire height.

[0082] As another example, the target driving data may include pre-processed driving speed and vehicle body acceleration, wherein the vehicle body acceleration includes longitudinal acceleration, lateral acceleration and vertical acceleration; and the target vehicle body data includes suspension type, wheel load and tire height.

[0083] In addition, other feature selection methods may be used to filter out parameter data that has a significant impact on the user's driving conditions and driving style, depending on the type of training data used in training the deep learning model, and the specific embodiments of the present application are not limited thereto.

[0084] In one embodiment, the prediction target includes the road surface type and the wheel center load; in other embodiments, the prediction target may be the road surface type or the wheel center load, or even other user driving conditions. For the road surface type, it is necessary to determine whether the road surface type is divided according to region or according to ground roughness. The road surface type generally includes mountain roads, urban roads, general roads and expressways according to region, and can be divided into expressways, bump roads, washboard roads, brake roads, pothole roads, cobblestone roads, stone roads, gravel roads, resonance roads, etc. according to road surface roughness, without specific limitation.

[0085] Finally, the target driving data and target vehicle body data are used as two-level inputs and respectively input into the pre-trained deep learning model to predict the user's driving conditions.

[0086] It can be seen that the embodiment of the present application provides a method for predicting user driving conditions, and adopts the deep learning model provided in the embodiment of the present application to predict user driving conditions. Since the deep learning model is a composite deep learning model including a convolutional neural network, a long short-term memory network, and a residual network and has been trained and verified, it can quickly and accurately obtain the user's driving conditions during actual use for predicting user driving conditions with long time series, high dynamic characteristics, and strong nonlinear characteristics. Among them, the accuracy of the user's driving road type predicted by this method can reach more than 90%, the accuracy of the wheel center load borne by the vehicle during driving is more than 80%, and the accuracy of the wheel center load damage is also more than 80. It can be used to re-formulate the fatigue durability design standards of the automobile chassis and monitor the health status of the chassis wearing parts in real time, and it is highly practical.

[0087] Moreover, the above-mentioned predicted user driving conditions can also provide data support for revising the chassis strength and durability design, which can reduce the redundancy of strength and durability design, and is expected to reduce the weight of chassis structural parts by more than 2%, thereby reducing development costs; the predicted user driving condition information can be used to determine the damage status of chassis wearing parts (such as suspension bushings) in real time, so as to promptly remind customers to check or replace them.

[0088] In one embodiment, when collecting original vehicle status data, the data can be directly obtained on an existing vehicle through sensors installed on the vehicle itself, without the need to install new sensors, thereby reducing costs.

[0089] In one embodiment, the original driving data and the original vehicle body data are preprocessed to obtain target driving data and target vehicle body data, including:

[0090] S31: respectively trimming the original driving data and the original vehicle body data to delete invalid and abnormal data, thereby obtaining trimmed driving data and trimmed vehicle body data;

[0091] S32: when it is determined that the vehicle tailoring data and the body tailoring data contain first target type data whose data amount is less than a preset data amount, reconstruct the target type data;

[0092] S33: respectively normalizing the reconstructed driving clipping data and the body clipping data to obtain target driving data and target body data.

[0093] It should be noted that in practical applications, the data preprocessing and training process types are mainly used to obtain the data required to meet the requirements of the deep learning model.

[0094] During the preprocessing process, the main work that needs to be carried out includes the following parts: a. Segment and trim the original driving data and the original vehicle body data: delete invalid data and abnormal data, such as invalid transition road data and abnormal burr data. During the trimming process, it is necessary to determine the segment length of each input parameter by analyzing the time lag between the input parameters and the predicted values ​​of the deep learning model; c. The actual collected data usually has an uneven distribution of various types of data, and the data values ​​vary greatly. To address these problems, it is necessary to reconstruct a small number of extreme loads, such as replica data with white noise signals; d. Normalize the data to obtain the final input data of the model.

[0095] In one embodiment, taking acceleration as an example, the judgment process of clipping data in the above data preprocessing process is as follows:

[0096] max(abs(az[t:t+T]))>0.4g or sum(abs(az[t:t+T]))>(0.06*f)g

[0097] Or max(abs(ax[t:t+T]))>0.3g or sum(abs(ax[t:t+T]))>(0.04*f)g

[0098] Or max(abs(ay[t:t+T]))>0.3g or sum(abs(ay[t:t+T]))>(0.06*f)g

[0099] Where T is the duration of each signal segment, f is the sampling frequency, a[t:t+T] represents the acceleration data collected from t to t+T, and this time interval is called the judgment domain, ax, ay, az represent the longitudinal, lateral and vertical accelerations at the center of mass of the vehicle, respectively, and g represents the acceleration of gravity.

[0100] If the above conditions are met, the acceleration data ax[t:t+T], ay[t:t+T], az[t:t+T] and other parameter data during the period from t to t+T / 2 are included in the valid data set; if the above conditions are not met, the data during the period from t to t+T / 2 are discarded; the time interval of each judgment domain is T / 2.

[0101] That is to say, the maximum absolute value of acceleration and the cumulative absolute value of acceleration per unit time are calculated, and then it is determined whether the corresponding maximum absolute value of acceleration and the cumulative absolute value of acceleration meet the conditions. If they meet the conditions, it means that these data are valid. The valid data are segmented, and the relatively small number of extreme loads are replicated with white noise signals. Finally, the data are normalized to obtain the target driving data and target vehicle body data.

[0102] It should be noted that the data preprocessing process mainly includes cropping, feature selection, segmentation, reconstruction and normalization operations, so that the preprocessed model input data not only meets the data requirements of the deep learning model, but also ensures the validity and rationality of the input data, thereby improving the reliability of the deep learning model.

[0103] In one embodiment, obtaining original vehicle status data includes:

[0104] S11: Acquire current driving scene information;

[0105] S12: When it is determined that the current driving scene meets the preset driving scene according to the driving scene information, the original vehicle state data is obtained.

[0106] In this embodiment, the trigger conditions for when data will be collected and the user's driving condition prediction will be made are clearly defined. It should be understood that since turning on the prediction function at all times of the user will consume huge hardware resources, it is necessary to determine in which scenarios to turn on the function. In the embodiment of this application, when the current driving scene is determined to meet the preset driving scene based on the driving scene information, the original vehicle status data is obtained, which can effectively avoid the situation of consuming huge hardware resources and improve the practicality of the solution.

[0107] Among them, the preset driving scenarios can be set according to the actual application scenarios, without specific limitations.

[0108] In one embodiment, the user's driving condition is predicted, and the method further includes: pushing the predicted user's driving condition information to the user terminal. This embodiment can push the predicted user's driving condition information to the user, increase user interaction, and improve the user's driving experience

[0109] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0110] It should be noted that the aforementioned embodiment describes the method for obtaining the user's driving condition provided in the embodiment of the present application. In the present application, a corresponding device for obtaining the user's driving condition is also provided, which will be described separately below.

[0111] In one embodiment, a user driving condition acquisition device is provided, and the user driving condition acquisition device corresponds to the user driving condition acquisition method in the above embodiment. Figure 5 As shown, the user driving condition acquisition device includes an acquisition module 101, a preprocessing module 102 and a prediction module 103. The detailed description of each functional module is as follows:

[0112] The acquisition module 101 is used to acquire original vehicle status data, where the original vehicle status data includes original driving data and original vehicle body data;

[0113] A preprocessing module 102 is used to preprocess the original driving data and the original vehicle body data to obtain target driving data and target vehicle body data;

[0114] The prediction module 103 is used to take the target driving data and the target vehicle body data as two-level inputs, and respectively input them into the pre-trained deep learning model to predict the user's driving condition;

[0115] Among them, deep learning models include convolutional neural networks, long short-term memory networks, and residual networks.

[0116] In one embodiment, the convolutional neural network includes a first convolutional layer, a second convolutional layer, a first pooling layer, a third convolutional layer, a fourth convolutional layer, a second pooling layer, a fifth convolutional layer, a sixth convolutional layer, a seventh convolutional layer, and a plurality of eighth convolutional layers connected in sequence, each eighth convolutional layer is connected in sequence, and the target driving data is used to input the first convolutional layer;

[0117] Among them, the output of the fifth convolutional layer and the output of the seventh convolutional layer pass through the first residual network and are input into the first eighth convolutional layer among multiple eighth convolutional layers. The output of the first eighth convolutional layer and the output of the last eighth convolutional layer pass through the second residual network and are input into the long short-term memory network. The target vehicle body data is used to input the long short-term memory network.

[0118] In one embodiment, the convolutional neural network includes six eighth convolutional layers.

[0119] In one embodiment, the user's driving condition includes the road type and the load borne by the wheel center.

[0120] In one embodiment, the original driving data includes driving speed, steering wheel angle and vehicle body acceleration; and the original vehicle body data includes suspension type, wheel load and tire height.

[0121] In one embodiment, the original driving data and the original vehicle body data are preprocessed to obtain target driving data and target vehicle body data, including:

[0122] The original driving data and the original body data are respectively trimmed to delete invalid and abnormal data, and the driving trimmed data and the body trimmed data are obtained;

[0123] When it is determined that the first target type data has a data amount less than a preset data amount in the vehicle tailoring data and the vehicle body tailoring data, reconstructing the target type data;

[0124] The reconstructed driving clipping data and body clipping data are normalized respectively to obtain target driving data and target body data.

[0125] In one embodiment, obtaining original vehicle status data includes:

[0126] Get the current driving scene information;

[0127] When it is determined that the current driving scene meets the preset driving scene based on the driving scene information, the original vehicle state data is obtained.

[0128] It can be seen that the embodiment of the present application provides a user driving condition prediction device, which uses the deep learning model provided by the embodiment of the present application to predict the user driving condition. Since the deep learning model is a composite deep learning model including a convolutional neural network, a long short-term memory network and a residual network and has been trained and verified, it can quickly and accurately obtain the user's driving condition during the actual use of the user for the prediction of the user's driving condition, which has long time series, high dynamic characteristics and strong nonlinear characteristics. Among them, the accuracy of the user's driving road type predicted by this method can reach more than 90%, the accuracy of the wheel center load borne by the vehicle during driving is more than 80%, and the accuracy of the wheel center load damage is also more than 80. It can be used to re-formulate the fatigue durability design standards of the automobile chassis and monitor the health status of the chassis wearing parts in real time, and it is highly practical.

[0129] For the specific definition of the user driving condition prediction device, please refer to the definition of the user driving condition prediction method above, which will not be repeated here. Each module in the above-mentioned user driving condition prediction device can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0130] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 6 As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. The network interface of the computer device is used to obtain required data. When the computer program is executed by the processor, a method for predicting a user's driving condition is implemented.

[0131] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the method for predicting the user driving condition in the above-mentioned embodiment is implemented. Alternatively, when the processor executes the computer program, the functions of each module / unit in this embodiment of the device for predicting the user driving condition are implemented. To avoid repetition, they are not described here.

[0132] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method for predicting the user's driving condition in the above embodiment is implemented. To avoid repetition, it is not described here. Alternatively, when the computer program is executed by a processor, the functions of each module / unit in the embodiment of the user's driving condition prediction device are implemented. To avoid repetition, it is not described here. The computer-readable storage medium may be non-volatile or volatile.

[0133] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and 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-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0134] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0135] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for obtaining a user's driving condition, characterized in that: include: Acquiring original vehicle status data, wherein the original vehicle status data includes original driving data and original vehicle body data; Preprocessing the original driving data and the original vehicle body data to obtain target driving data and target vehicle body data; The target driving data and the target vehicle body data are used as two-level inputs and respectively input into a pre-trained deep learning model to predict the user's driving condition; Among them, the deep learning model includes convolutional neural network, long short-term memory network and residual network.

2. The method for obtaining the user's driving condition according to claim 1, characterized in that: The convolutional neural network includes a first convolutional layer, a second convolutional layer, a first pooling layer, a third convolutional layer, a fourth convolutional layer, a second pooling layer, a fifth convolutional layer, a sixth convolutional layer, a seventh convolutional layer and a plurality of eighth convolutional layers connected in sequence, each of the eighth convolutional layers is connected in sequence, and the target driving data is used to input the first convolutional layer; Among them, the output of the fifth convolutional layer and the output of the seventh convolutional layer are input into the first eighth convolutional layer among the multiple eighth convolutional layers after passing through the first residual network, the output of the first eighth convolutional layer and the output of the last eighth convolutional layer are input into the long short-term memory network after passing through the second residual network, and the target vehicle body data is used to input into the long short-term memory network.

3. The method for obtaining the user's driving condition according to claim 2, characterized in that: The convolutional neural network includes six of the eighth convolutional layers.

4. The method for obtaining a user's driving condition according to any one of claims 1 to 3, characterized in that: The user driving conditions include the road type and the load borne by the wheel center.

5. The method for obtaining a user's driving condition according to any one of claims 1 to 3, characterized in that: The original driving data includes driving speed, steering wheel angle and vehicle body acceleration; and the original vehicle body data includes suspension type, wheel load and tire height.

6. The method for obtaining a user's driving condition according to any one of claims 1 to 3, characterized in that: The preprocessing of the original driving data and the original vehicle body data to obtain target driving data and target vehicle body data includes: The original driving data and the original vehicle body data are respectively trimmed to delete invalid and abnormal data, thereby obtaining driving trimmed data and vehicle body trimmed data; When it is determined that the driving tailored data and the vehicle body tailored data contain first target type data whose data amount is less than a preset data amount, reconstructing the target type data; The reconstructed driving clipping data and the body clipping data are respectively normalized to obtain target driving data and target body data.

7. The method for obtaining a user's driving condition according to any one of claims 1 to 3, characterized in that: The obtaining of original vehicle status data comprises: Get the current driving scene information; When it is determined that the current driving scene meets the preset driving scene according to the driving scene information, the original vehicle state data is acquired.

8. A device for acquiring a user's driving condition, characterized in that: include: An acquisition module, used for acquiring original vehicle status data, wherein the original vehicle status data includes original driving data and original vehicle body data; A preprocessing module, used for preprocessing the original driving data and the original vehicle body data to obtain target driving data and target vehicle body data; A prediction module, used to take the target driving data and the target vehicle body data as two-level inputs, respectively inputting the pre-trained deep learning model to predict the user's driving condition; Among them, the deep learning model includes convolutional neural network, long short-term memory network and residual network.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method for obtaining the user's driving condition as described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for obtaining the user's driving condition as described in any one of claims 1 to 7 is implemented.

Images