Data set enhancement method and device, equipment, storage medium and program product

Abstract

The invention relates to the technical field of data enhancement, and discloses a data set enhancement method and device, equipment, a storage medium and a program product, and the method comprises the steps: obtaining a driving behavior sequence and driving environment data of a vehicle; constructing an original data set based on the driving behavior sequence, and training a preset recurrent neural network by using the original data set and the driving environment data to obtain a trained recurrent neural network; determining a driving behavior initial value of the vehicle and a target driving environment; and based on the driving behavior initial value, the target driving environment and the trained recurrent neural network, generating a driving behavior prediction value of the vehicle in the target driving environment, and obtaining a target data set. According to the invention, the preset recurrent neural network is trained and driving behavior characteristics in different driving environments are learned, so that the trained recurrent neural network is utilized to generate the enhanced target data set, and the diversity of the data set is increased to support more accurate car networking data analysis and application.

Description

Technical Field

[0001] The present invention relates to the field of data enhancement technology, and in particular to a data set enhancement method, device, equipment, storage medium and program product. Background Art

[0002] With the development of Internet of Vehicles technology, the amount of data generated by vehicles has increased dramatically. This data is essential for vehicle status monitoring, fault diagnosis, autonomous driving and other functions. However, the diversity of actual application scenarios makes it difficult for the original data set to cover all working conditions, limiting the generalization ability of the model. Therefore, developing effective data enhancement technology to generate high-quality virtual data has become the key to improving the performance of Internet of Vehicles data analysis models.

[0003] At present, some methods for vehicle network data enhancement have been proposed, such as statistical resampling methods and simple noise addition techniques. However, the resampling method easily introduces a large amount of redundant information and lacks creative simulation of new scenarios, while excessive noise will cover up the real signal characteristics and affect the model learning effect. For another example, vehicle detection data enhancement is based on neural networks. However, related technologies can only perform data enhancement calculations on the original data and cannot add external conditions such as the environment. The generated data cannot cover most driving scenarios, resulting in a small data coverage, low diversity and complexity. Summary of the invention

[0004] In view of this, the present invention provides a data set enhancement method, device, equipment, storage medium and program product to solve the problem that traditional technology cannot add external conditions during the training process, resulting in the generated data failing to cover most driving scenarios and having low diversity and complexity.

[0005] In a first aspect, the present invention provides a data set enhancement method, the method comprising:

[0006] Obtain the vehicle's driving behavior sequence and driving environment data;

[0007] An original data set is constructed based on the driving behavior sequence, and a preset recurrent neural network is trained using the original data set and driving environment data to obtain a trained recurrent neural network;

[0008] Determine the initial value of the vehicle's driving behavior and the target driving environment;

[0009] Based on the initial value of driving behavior, the target driving environment and the trained recurrent neural network, the predicted value of the vehicle's driving behavior in the target driving environment is generated to obtain the target data set.

[0010] Beneficial effects: The original data set is constructed according to the vehicle's driving behavior sequence, and the preset recurrent neural network is trained using the original data set and driving environment data to learn the driving behavior characteristics under different driving environments. The trained recurrent neural network is then used to generate the vehicle's driving behavior prediction value under the target driving environment, generate an enhanced target data set, and increase the diversity of the data set to support more accurate Internet of Vehicles data analysis and application.

[0011] In an optional implementation, the preset recurrent neural network is trained using the original data set and the driving environment data to obtain a trained recurrent neural network, including:

[0012] Encoding the driving environment data to obtain environment encoding data;

[0013] The preset recurrent neural network is iterated multiple times according to the sample sequences of multiple time steps in the original data set, and in each iteration, the state of the recurrent neural network is updated based on the sample sequence of the current time step and the environmental coding data until the recurrent neural network converges to obtain a trained recurrent neural network.

[0014] Beneficial effects: The model is trained by using the environmentally encoded data after the driving environment data is encoded and integrating the driving behavior characteristics of each time step, so that the model can effectively predict future driving behavior data in different driving environments. Through refined data processing, carefully designed model architecture and training strategies, the diversity of driving behavior data and the prediction accuracy of the model are effectively enhanced, providing a powerful tool for Internet of Vehicles data analysis and application.

[0015] In an optional implementation, the recurrent neural network is a long short-term memory network; the state of the recurrent neural network is updated based on the sample sequence and the environment encoding data of the current time step, including:

[0016] Perform weighted fusion or feature concatenation on the sample sequence of the current time step and the environmental coding data to obtain a new sample sequence corresponding to the current time step;

[0017] According to the new sample sequence, the state of the long short-term memory network is updated.

[0018] Beneficial effects: The sample sequence and environmental encoding data are weightedly fused or feature concatenated to form a new sample sequence, and the state of the long short-term memory network is updated, so that the model can learn the changes in driving behavior under different environmental conditions.

[0019] In an optional implementation, the recurrent neural network includes a long short-term memory network, and the input gate, the forget gate, and the output gate of the long short-term memory network are all provided with a weight matrix; the state of the recurrent neural network is updated based on the sample sequence of the current time step and the environment encoding data, including:

[0020] Determine a weight adjustment factor corresponding to the environment encoding data; wherein the weight adjustment factor is used to adjust the weight matrix of the long short-term memory network;

[0021] Update the state of the long short-term memory network according to the weight adjustment factor.

[0022] Beneficial effects: The state of the long short-term memory network is updated according to the weight adjustment factor corresponding to the environmental encoding data, so that the model can learn the changes in driving behavior under different environmental conditions.

[0023] In an optional implementation, the recurrent neural network includes a long short-term memory network, and the long short-term memory network includes an input gate, a forget gate, and an output gate; the state of the recurrent neural network is updated based on the sample sequence of the current time step and the environment encoding data, including:

[0024] Determine the standard gate values ​​corresponding to the input gate, forget gate and output gate in the long short-term memory network, and determine the conditional gate values ​​corresponding to the input gate, forget gate and output gate based on the environment encoding data;

[0025] According to the standard gate value and the conditional gate value, the target gate values ​​corresponding to the input gate, the forget gate and the output gate are calculated respectively;

[0026] According to the target gate value, the state of the long short-term memory network is updated.

[0027] Beneficial effect: According to the conditional gate value corresponding to the environmental encoding data and the standard gate value of each gate in the long short-term memory network, the target gate value is calculated, and the state of the long short-term memory network is updated using the target gate value, so that the model can learn the changes in driving behavior under different environmental conditions.

[0028] In an optional implementation, based on the initial value of the driving behavior, the target driving environment and the trained recurrent neural network, a driving behavior prediction value of the vehicle in the target driving environment is generated to obtain a target data set, including:

[0029] Constructing an initialization sequence based on the initial value of the driving behavior and the target driving environment; wherein the target driving environment includes at least one of the road type, weather conditions and traffic density;

[0030] Input the initialization sequence into the trained recurrent neural network to obtain the predicted value of the vehicle's driving behavior in the target driving environment;

[0031] Based on the predicted value of the vehicle's driving behavior in the target driving environment, a target driving behavior sequence is constructed to obtain a target data set.

[0032] Beneficial effects: Determine the initial value of driving behavior and the target driving environment, use the trained recurrent neural network to generate the predicted value of driving behavior under the target driving environment, thereby enhancing the data set and obtaining a data-rich target data set.

[0033] In an optional embodiment, before inputting the initialization sequence into the trained recurrent neural network, the method further includes:

[0034] The sample sequence in the original data set is stretched or compressed in time scale to obtain the variation history data, and the initialization sequence is updated based on the variation history data;

[0035] After inputting the initialization sequence into the trained recurrent neural network, the method further includes:

[0036] Interpolate the hidden state of the trained recurrent neural network, or add random variables to the hidden state;

[0037] After obtaining the predicted value of the driving behavior of the vehicle in the target driving environment, the method includes:

[0038] Add random noise to the driving behavior prediction value.

[0039] Beneficial effects: By stretching or compressing the time scale, interpolating, adding random variables and random noise to the data, the data diversity of the target data set is effectively enriched. By generating a large number of driving behavior sequences that reflect various situations in actual driving, and a controllable diversity enhancement strategy, the model output is made richer and more varied, thereby effectively improving the coverage of training data and the generalization ability of the model.

[0040] In an optional implementation, obtaining a driving behavior sequence of a vehicle includes:

[0041] Acquire driving behavior data and perform data cleaning on the driving behavior data; wherein the driving behavior data includes at least one of vehicle speed, engine speed and acceleration, and the data cleaning includes at least one of outlier removal, missing value filling, smoothing, noise reduction and normalization;

[0042] The time series features of the cleaned driving behavior data are extracted to obtain the driving behavior sequence.

[0043] Beneficial effects: The vehicle's driving behavior data is cleaned to optimize data quality, and time series features are extracted from the cleaned driving behavior data to facilitate model training.

[0044] In an optional implementation, constructing an original data set based on a driving behavior sequence includes:

[0045] The driving behavior sequence is divided into multiple sample sequences with continuous time steps through a sliding window;

[0046] Extract the numerical features of each sample sequence, normalize and standardize the numerical features, and obtain the feature vector of each sample sequence;

[0047] For the sample sequence of each time step, the label of the sample sequence of the current time step is determined according to the sample sequence of the next time step; wherein the label is used to characterize the driving behavior data of the sample sequence of the current time step in the future time step;

[0048] Based on the feature vector and label of each sample sequence, the original data set is constructed.

[0049] Beneficial effect: The driving behavior sequence is divided into sample sequences by sliding windows, and the feature vector and label of each sample sequence are extracted to construct an original data set with high data quality.

[0050] In a second aspect, the present invention provides a data set enhancement device, the device comprising:

[0051] An acquisition module, used to acquire the driving behavior sequence and driving environment data of the vehicle;

[0052] The first processing module is used to construct an original data set based on the driving behavior sequence, and train a preset recurrent neural network using the original data set and driving environment data to obtain a trained recurrent neural network;

[0053] A second processing module is used to determine the initial value of the vehicle's driving behavior and the target driving environment;

[0054] The third processing module is used to generate a driving behavior prediction value of the vehicle in the target driving environment based on the initial value of the driving behavior, the target driving environment and the trained recurrent neural network to obtain a target data set.

[0055] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the data set enhancement method of the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0056] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the dataset enhancement method of the first aspect or any corresponding embodiment thereof.

[0057] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions for causing a computer to execute the data set enhancement method of the first aspect or any corresponding embodiment thereof.

[0058] The beneficial effects of the present invention are:

[0059] An original data set is constructed based on the vehicle's driving behavior sequence, and the preset recurrent neural network is trained using the original data set and driving environment data to learn the driving behavior characteristics under different driving environments. The trained recurrent neural network is then used to generate a predicted value of the vehicle's driving behavior in the target driving environment, generating an enhanced target data set, increasing the diversity of the data set to support more accurate Internet of Vehicles data analysis and application. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0061] Figure 1 is a flowchart of a data set enhancement method according to an embodiment of the present invention;

[0062] Figure 2 is a schematic diagram of a data preprocessing process according to an embodiment of the present invention;

[0063] Figure 3 is a flow chart of another data set enhancement method according to an embodiment of the present invention;

[0064] Figure 4 is a schematic diagram of a process of constructing an original data set according to an embodiment of the present invention;

[0065] Figure 5 is a schematic diagram of a data enhancement process according to an embodiment of the present invention;

[0066] Figure 6 is a schematic diagram of a process of model optimization according to an embodiment of the present invention;

[0067] Figure 7 is a flow chart of another data set enhancement method according to an embodiment of the present invention;

[0068] Figure 8 is a structural block diagram of a data set enhancement device according to an embodiment of the present invention;

[0069] Fig. 9 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0070] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings 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 those skilled in the art without creative work are within the scope of protection of the present invention.

[0071] With the development of Internet of Vehicles technology, the amount of data generated by vehicles has increased dramatically. This data is essential for vehicle status monitoring, fault diagnosis, autonomous driving and other functions. However, the diversity of actual application scenarios makes it difficult for the original data set to cover all working conditions, limiting the generalization ability of the model. Therefore, developing effective data enhancement technology to generate high-quality virtual data has become the key to improving the performance of Internet of Vehicles data analysis models.

[0072] In related technologies, some methods for vehicle network data enhancement have been proposed, such as statistical resampling methods and noise addition techniques. Statistical resampling mainly increases the sample size by repeating, rotating or mirroring the original data, and noise addition techniques improve the robustness of the model. However, the resampling method may introduce a large amount of redundant information and lack creative simulation of new scenarios, while excessive noise may mask the true signal characteristics and affect the model learning effect.

[0073] Another example is the use of neural networks to enhance vehicle detection data, but most of them can only perform data enhancement calculations on the original data and cannot add external conditions. However, data enhancement in most scenarios now requires the addition of external conditions and the maintenance of data physical rationality and sequence coherence. The data sets generated by the above methods have a small coverage and low complexity, and cannot ensure the efficiency of model training.

[0074] That is to say, in the process of data enhancement, how to increase the diversity of generated data while ensuring that it is highly similar to real-world data, so as to fully cover various data application scenarios and abnormal conditions, is an urgent problem to be solved. At the same time, for time series data such as Internet of Vehicles data, it is also very important to ensure the temporal logic consistency and accuracy of long-term dependencies within the generated data sequence. In addition, with the continuous enrichment of data types and application scenarios, it is also challenging to ensure the flexibility and scalability of enhancement technology so that it can quickly adapt to new data characteristics and changes in demand.

[0075] Therefore, an embodiment of the present invention provides a data set enhancement solution, which performs Internet of Vehicles data enhancement based on a recurrent neural network (RNN). It enriches the vehicle network data set by generating synthetic data, and takes driving behavior analysis scenarios as the target. It effectively enhances the diversity of driving behavior data and the prediction accuracy of the model, and provides a powerful tool for Internet of Vehicles data analysis and application.

[0076] According to an embodiment of the present invention, an embodiment of a data set enhancement method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0077] In this embodiment, a data set enhancement method is provided, which can be used for a computer device or electronic device, such as a computer, a tablet computer, etc., for performing data set enhancement. Figure 1 is a flow chart of a data set enhancement method according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps:

[0078] Step S101, obtaining the driving behavior sequence and driving environment data of the vehicle.

[0079] Specifically, the driving behavior data of the vehicle can be obtained based on the Internet of Vehicles, and the driving behavior data can be cleaned to extract the time series characteristics of the cleaned driving behavior data to obtain the driving behavior sequence. Among them, the driving behavior data mainly includes vehicle speed, engine speed and acceleration, and the data cleaning mainly includes outlier removal, missing value filling, smoothing, noise reduction and normalization.

[0080] It should be noted that the Internet of Vehicles (IoV) refers to a network system that uses technologies such as wireless communications, sensors, and global positioning systems to enable information exchange and sharing between vehicles, vehicles and infrastructure, vehicles and pedestrians, and vehicles and the Internet. It can realize functions such as real-time traffic information, safety warnings, remote diagnosis, and automatic driving assistance, and is an important part of the intelligent transportation system.

[0081] In some optional embodiments, such as Figure 2 As shown, data collection, data cleaning, data format unification, feature construction, feature selection and data segmentation can be performed in sequence to obtain the vehicle's driving behavior sequence and driving environment data.

[0082] For example, first, comprehensive raw data is collected from the Internet of Vehicles big data uploaded by the data acquisition system and the data warehouse. The raw data covers but is not limited to the vehicle's own status such as vehicle speed, engine speed, etc. Driving environment data such as road type, weather conditions, traffic flow, etc. are obtained from the external system to ensure the breadth and integrity of the data source and lay a solid foundation for subsequent analysis. Next, statistical methods are used to identify and eliminate outliers in the data, such as sensor false alarms, extreme or illogical data points, to ensure the authenticity and reliability of the data. This link also includes operations such as filling missing values, smoothing noise, and unifying data formats to optimize data quality. Then, time series features that are significant to driving behavior patterns are extracted from the cleaned data, such as speed change rate, acceleration curve shape, changes in driving direction, etc. These features are converted into a format suitable for RNN model input to prepare for model training.

[0083] In an embodiment of the present application, data cleaning is performed on the driving behavior data of the vehicle to optimize the data quality, and time series features are extracted from the cleaned driving behavior data to facilitate model training.

[0084] Step S102, constructing an original data set based on the driving behavior sequence, and using the original data set and the driving environment data to train a preset recurrent neural network to obtain a trained recurrent neural network.

[0085] Specifically, before training, it is necessary to design the model input layer, including driving behavior sequence input and environmental condition input (weather, road conditions), and implement the model's conditional fusion mechanism. Before each time step or input of the model, through feature splicing, weight adjustment or conditional gating mechanism, ensure that external condition information can affect the model state update.

[0086] Furthermore, the recurrent neural network (RNN) model is trained to comprehensively consider the complexity of the driving behavior sequence and the diversity of external environmental conditions, and to build a highly flexible and efficient prediction architecture. The core of the model revolves around a dual-path input design. On the one hand, it captures the dynamic changes of continuous driving behaviors, such as the timing patterns of speed and acceleration; on the other hand, it ensures that the model can understand and adapt to different driving environments through carefully encoded external environmental condition information flows (such as weather and road conditions). By innovatively incorporating these external conditions into the model, the model's ability to capture long-term dependencies on sequences is enhanced, greatly improving its prediction accuracy and generalization performance in complex driving scenarios. In addition, in the specific RNN model training, different feature paths can be set according to the data enhancement requirements in different scenarios.

[0087] It should be noted that a recurrent neural network is an artificial neural network that has a cyclic structure that allows information to propagate in a sequence. This structure makes RNN particularly suitable for processing problems in the fields of time series data and natural language processing because it can capture the temporal dependencies in sequence data. In the field of Internet of Vehicles, RNN can also be used to analyze vehicle trajectories and predict driving behaviors.

[0088] In addition, model training is the core part of the machine learning process. Through algorithms, the model learns to extract features from data and establish a mapping relationship between input and output. In the Internet of Vehicles data enhancement method, model training aims to enable RNN or other machine learning models to learn to recognize and predict various driving situations of vehicles.

[0089] Step S103, determining the initial value of the vehicle's driving behavior and the target driving environment.

[0090] Specifically, the initial value of driving behavior mainly includes the starting speed, acceleration, etc., among which the starting speed is randomly selected within the legal speed range, such as 30km / h-120km / h; the acceleration is randomly determined within a reasonable range, such as -5m / s 2 To +5m / s 2 , to reflect acceleration, deceleration or constant speed.

[0091] Specifically, the target driving environment mainly includes road type, weather conditions and traffic density, among which the road type can be formed by encoding different roads, such as urban roads, highways, country roads, etc.; weather conditions are also input in coded form, such as sunny days, rainy days, snowy days, etc.; traffic density can be set to three density levels: low, medium and high, which affects the continuity of driving behavior and reaction time.

[0092] Step S104, based on the initial value of the driving behavior, the target driving environment and the trained recurrent neural network, a driving behavior prediction value of the vehicle in the target driving environment is generated to obtain a target data set.

[0093] Specifically, data augmentation is performed using a trained recurrent neural network. Data augmentation is a technique in the field of machine learning that generates additional training samples by transforming the original data, thereby increasing the diversity and scale of the training data set and improving the generalization ability of the model. In the Internet of Vehicles, data augmentation may include slight modifications to vehicle driving data, such as rotating images, translating positions, adding noise, etc., to simulate different driving environments and conditions.

[0094] The present invention uses a recurrent neural network to enhance the data of Internet of Vehicles (such as CAN bus signals), thereby generating new data points that conform to physical laws and logic, and enriching the diversity of data sets. Through deep learning technology, starting from the driving behavior analysis scenario, the real driving behavior sequence is simulated, and the diversity of data sets is increased to support more accurate Internet of Vehicles data analysis and application.

[0095] The data set enhancement method provided in this embodiment constructs an original data set according to the driving behavior sequence of the vehicle, and uses the original data set and driving environment data to train a preset recurrent neural network to learn the driving behavior characteristics under different driving environments, thereby using the trained recurrent neural network to generate a driving behavior prediction value of the vehicle in the target driving environment, generate an enhanced target data set, and increase the diversity of the data set to support more accurate Internet of Vehicles data analysis and application.

[0096] In this embodiment, a data set enhancement method is provided, which can be used for a computer device or electronic device, such as a computer, a tablet computer, etc., for performing data set enhancement. Figure 3 is a flow chart of a data set enhancement method according to an embodiment of the present invention. Figure 3 As shown, the process includes the following steps:

[0097] Step S301, obtain the vehicle's driving behavior sequence and driving environment data. For details, please refer to Figure 1 Step S101 of the illustrated embodiment will not be described in detail here.

[0098] Step S302: construct an original data set based on the driving behavior sequence, and train a preset recurrent neural network using the original data set and the driving environment data to obtain a trained recurrent neural network.

[0099] Specifically, the above step S302 includes:

[0100] Step S3021, constructing an original data set based on the driving behavior sequence.

[0101] In some optional implementations, the above step S3021 includes:

[0102] Step a1: Divide the driving behavior sequence into a plurality of sample sequences with continuous time steps through a sliding window.

[0103] Specifically, Figure 4 As shown in the figure, firstly, the sequence features are preprocessed, for example, the driving behavior sequence data (such as vehicle speed and acceleration) is normalized; then the sequence length is defined to determine the length of the model input sequence (also called "window size"); then the sequence is sliced ​​to divide the continuous driving behavior data into multiple sample sequences according to the prediction window length to be simulated.

[0104] For example, the original time series data is segmented by a sliding window. The window slides forward one time step each time (or the step size is adjusted as needed) to generate a series of continuous and partially overlapping subsequences. For example, for the sequence [20, 25, 30, 35, 40, ...], when the window size is 3, the first window is [20, 25, 30], the second window is [25, 30, 35], and so on.

[0105] Step a2: extract the numerical features of each sample sequence, normalize and standardize the numerical features, and obtain the feature vector of each sample sequence.

[0106] Specifically, see again Figure 4 For each window, the numerical features in the window, such as vehicle speed and acceleration, are directly taken. In order to avoid the model learning bias caused by the magnitude difference between different windows, the numerical features in each window are normalized or standardized to form a feature vector. For example, the speed feature vector corresponding to the window [20, 25, 30] is [20, 25, 30].

[0107] In some optional embodiments, see again Figure 4 When performing sequence standardization, the minimum and maximum scaling method can be used to map the values ​​in each window to the [0, 1] interval. For each window, find the maximum and minimum values, and then scale them according to the following formula:

[0108]

[0109] Among them, x represents the current value, min(x) represents the minimum value, and max(x) represents the maximum value.

[0110] In some optional implementations, when performing sequence normalization, normalization can be used to make the mean of the feature in the window 0 and the standard deviation 1, calculate the mean and standard deviation of the feature in each window, and then convert according to the following formula:

[0111]

[0112] Here, μ is the mean and σ is the standard deviation.

[0113] Step a3, for the sample sequence of each time step, determine the label of the sample sequence of the current time step according to the sample sequence of the next time step; wherein the label is used to characterize the driving behavior data of the sample sequence of the current time step in the future time step.

[0114] Specifically, see again Figure 4 , determine the label corresponding to each window, that is, the target variable. For the prediction task, if the window size is n, then the next data point of the last data point in the window is used as the label. For example, if the window is [20, 25, 30], the label may be 35 (that is, the next value in the speed sequence).

[0115] Step a4: construct the original data set based on the feature vector and label of each sample sequence.

[0116] Specifically, see again Figure 4 , generating a sample for each sliding window, including feature vectors (standardized numerical values ​​of driving behavior sequences and possible temporal feature encodings) and corresponding labels, forming training sets, validation sets, and test sets, thereby constructing the original data set (i.e. Figure 4 Build the final dataset).

[0117] In the embodiment of the present invention, the driving behavior sequence is segmented into a sample sequence by sliding window, and the feature vector and label of each sample sequence are extracted to construct an original data set with high data quality.

[0118] Step S3022, encode the driving environment data to obtain environment encoding data.

[0119] Specifically, the environment encoding method can be selected according to the actual training scenario. For details, please refer to the description of related technologies, which will not be repeated here.

[0120] Step S3023, iterate the preset recurrent neural network multiple times according to the sample sequences of multiple time steps in the original data set, and in each iteration, update the state of the recurrent neural network based on the sample sequence of the current time step and the environmental coding data until the recurrent neural network converges to obtain a trained recurrent neural network.

[0121] Specifically, an RNN model based on the Long Short-Term Memory (LSTM) or Gated Recurrent Unit (GRU) can be designed to capture the long-term dependencies of driving behavior sequences. In addition, a dual-path input mechanism is constructed to process driving behavior sequences on the one hand and receive encoded external environmental condition information, i.e., environmental encoding data (such as weather and road conditions), on the other hand, and integrate them before the LSTM layer through a weighted fusion mechanism. The Dropout layer is used to reduce the risk of overfitting, the optimizer is Adam, and the learning rate is adjusted dynamically. Finally, the original data set is divided into a training set, a validation set, and a test set to ensure the generalization ability of the model.

[0122] For example, the RNN model is designed, LSTM or GRU is used as the core unit, and a dual-path input mechanism is constructed. Figure 4 , one input is the driving behavior sequence feature, and the other is the external environmental condition feature, such as weather encoding data. Between the input layer of the RNN model and the first layer of LSTM / GRU units, the driving behavior sequence and external condition information are integrated through a weighted fusion mechanism to enhance the model's adaptability to complex scenarios and realize the conditional fusion mechanism. The Dropout layer (the ratio is set to 0.2 to 0.5) is integrated in the model to reduce the risk of overfitting. The Adam optimizer is selected and a dynamic learning rate strategy is set. The initial learning rate is set to 0.001 and automatically adjusted according to the training process. The processed data set is randomly divided into a training set (70%), a validation set (15%), and a test set (15%) to ensure the generalization ability evaluation and final performance test during the model training process.

[0123] It should be noted that LSTM is a special recurrent neural network architecture. The main goal of LSTM network design is to overcome the limitations of traditional RNN in dealing with long-term dependency issues, that is, in the learning of long-term series data, due to the gradient vanishing and gradient exploding phenomena, it is difficult for the network to capture the long-distance dependencies in the sequence. In addition, GRU is a variant of the recurrent neural network RNN. GRU is designed as a simplified version of the long short-term memory LSTM, aiming to maintain the advantages of LSTM in dealing with long-term dependencies, while reducing the computational complexity and the number of model parameters, making training more efficient.

[0124] In some optional embodiments, the recurrent neural network is a long short-term memory network, which performs weighted fusion or feature concatenation on the sample sequence of the current time step and the environmental coding data to obtain a new sample sequence corresponding to the current time step, and updates the state of the long short-term memory network based on the new sample sequence.

[0125] Specifically, feature concatenation is one of the simplest and most direct methods, which allows external environmental condition information to be directly attached to the current input at each time step, thereby affecting the state update of the LSTM. Assume that there is a conditional vector c (i.e., environmental encoding data) and the input X at the current time step t (i.e., sample sequence), at each time step t, concatenate c to X t On the basis of these, a new input vector [X t ; c], this new input vector is fed into the LSTM unit for state update.

[0126] In this embodiment, the sample sequence and the environmental coding data are weightedly fused or feature concatenated to form a new sample sequence, and the state of the long short-term memory network is updated, so that the model can learn the changes in driving behavior under different environmental conditions.

[0127] In some optional embodiments, the recurrent neural network includes a long short-term memory network, and the input gate, forgetting gate and output gate of the long short-term memory network are all provided with a weight matrix to determine the weight adjustment factor corresponding to the environmental coding data, wherein the weight adjustment factor is used to adjust the weight matrix of the long short-term memory network, and the state of the long short-term memory network is updated according to the weight adjustment factor.

[0128] Specifically, weight adjustment refers to using conditional information in each gate of LSTM (input gate, forget gate, output gate) to adjust the weight matrix. Calculate the weight adjustment factor W corresponding to the conditional vector c c , at each time step t, using W c Adjust the weight matrix W in LSTM and use the adjusted weight matrix to update the state.

[0129] In this embodiment, the state of the long short-term memory network is updated according to the weight adjustment factor corresponding to the environmental coding data, so that the model can learn the changes in driving behavior under different environmental conditions.

[0130] In some optional embodiments, the recurrent neural network includes a long short-term memory network, and the long short-term memory network includes an input gate, a forget gate, and an output gate. The standard gate values ​​corresponding to the input gate, the forget gate, and the output gate in the long short-term memory network are determined, and the conditional gate values ​​corresponding to the input gate, the forget gate, and the output gate are determined based on the environmental coding data. Then, according to the standard gate value and the conditional gate value, the target gate values ​​corresponding to the input gate, the forget gate, and the output gate are calculated. Finally, according to the target gate value, the state of the long short-term memory network is updated.

[0131] Specifically, the conditional gating mechanism introduces conditional information into the gating mechanism of LSTM, for example, considering the influence of the conditional vector when calculating the activation value of the gate. For each gate (input gate, forget gate, output gate), the standard gate value g is calculated, and the conditional vector c is used to calculate an additional gate value g c , the standard threshold value g and the conditional threshold value g c Combined, we get the final gate value g final , use g final Make a status update.

[0132] In this embodiment, the target gate value is calculated according to the conditional gate value corresponding to the environmental coding data and the standard gate value of each gate in the long short-term memory network, and the state of the long short-term memory network is updated using the target gate value, so that the model can learn the changes in driving behavior under different environmental conditions.

[0133] Specifically, when training the model in step S3023, the mean square error (MSE) is used as the loss function, and multiple rounds of training are performed. The performance of the validation set is evaluated after each round. If there is no significant improvement in the performance of the validation set for several consecutive rounds, the training is terminated to avoid overfitting.

[0134] For example, the model is trained multiple times using the training set, and the mean square error (MSE) is used as the loss function to evaluate the deviation between the model prediction and the actual driving behavior. After each round of training, the performance is evaluated using the validation set to monitor the overfitting of the model. An early stopping criterion is set. If there is no significant improvement in the performance of the validation set for five consecutive rounds, the training is stopped to avoid overfitting and optimize the generalization ability of the model.

[0135] Furthermore, the model was evaluated and validated, and the model performance was evaluated using an independent test set, and the prediction accuracy, root mean square error (RMSE), R 2 The prediction accuracy of the model under different driving conditions can be analyzed through visualization tools to compare the prediction results with the actual data and identify potential improvement points.

[0136] Furthermore, the model is optimized and iterated. According to the test results, it may be necessary to adjust the network structure (such as adding hidden layers and number of units), optimization strategy (learning rate adjustment), data preprocessing, etc., to iteratively optimize the model. After each adjustment, the training and verification process is repeated until the model achieves satisfactory performance indicators on the test set to ensure the reliability and practicality of the model.

[0137] In the embodiment of the present application, the model is trained using the environmentally encoded data after the driving environment data is encoded, and the driving behavior characteristics of each time step are integrated, so that the model can effectively predict future driving behavior data in different driving environments. Through refined data processing, carefully designed model architecture and training strategies, the diversity of driving behavior data and the prediction accuracy of the model are effectively enhanced, providing a powerful tool for Internet of Vehicles data analysis and application.

[0138] Step S303, determine the initial value of the vehicle's driving behavior and the target driving environment. For details, please refer to Figure 1 The specific description of step S103 of the illustrated embodiment will not be repeated here.

[0139] Step S304, based on the initial value of the driving behavior, the target driving environment and the trained recurrent neural network, a driving behavior prediction value of the vehicle in the target driving environment is generated to obtain a target data set.

[0140] Specifically, the above step S304 includes:

[0141] S2041: construct an initialization sequence based on the initial value of the driving behavior and the target driving environment.

[0142] The target driving environment includes at least one of road type, weather conditions and traffic density.

[0143] Specifically, Figure 5 As shown, in order to simulate different driving scenarios, it is first necessary to design a series of starting conditions, which may include but are not limited to: starting speed, acceleration, road type, weather conditions, traffic density, etc., and construct an initialization sequence based on these starting conditions.

[0144] In addition, the sample sequence in the original data set can be stretched or compressed in time to obtain the variation history data, and the initialization sequence can be updated based on the variation history data. Take the real data fragments of historical driving behavior and fine-tune them, such as stretching or compressing the real sequence in time, or slightly changing some behavior parameters, and then use the RNN model to "predict" based on these mutated data, further evolve the sequence, and increase data diversity.

[0145] S2042, input the initialization sequence into the trained recurrent neural network to obtain a driving behavior prediction value of the vehicle in the target driving environment.

[0146] Specifically, the initialization sequence is input into the RNN model, and the model predicts the driving behavior data (such as changes in speed and acceleration) at the next time point to obtain the predicted driving behavior value under the target driving environment.

[0147] In addition, the hidden states of the trained recurrent neural network are interpolated, or random variables are added to the hidden states. In order to generate more novel but reasonable driving behaviors, the model can be guided to explore unseen sequence configurations by interpolating or making small perturbations in the hidden state space of the model. This can be achieved by linearly interpolating the hidden states or adding random vectors between consecutive time steps.

[0148] S2043, based on the predicted value of the vehicle's driving behavior in the target driving environment, construct a target driving behavior sequence to obtain a target data set.

[0149] Specifically, see again Figure 5 , concatenate the predicted value with the data at the end of the current initialization sequence to form a new sequence input, repeat this process, gradually generate longer sequences, and obtain the target data set.

[0150] In addition, see again Figure 5 , add random noise to the driving behavior prediction value. To increase the diversity of the data, you can add a small amount of random noise to the generated data points after each prediction. For example, add [-0.5, 0.5] km / h and [-0.2, 0.2] m / s to the speed and acceleration respectively. 2 Random disturbances within the range.

[0151] Specifically, see again Figure 5 , the generated target dataset can be evaluated and feedback looped. The generated sequence needs to pass the rationality check to ensure that it complies with the laws of physics and driving logic. For example, the speed cannot change suddenly and the acceleration should be kept within a reasonable range. According to the results of the rationality check, the parameters such as noise intensity and degree of variation are continuously adjusted to optimize the generation strategy. If it is found that the generation results of the model under certain specific conditions are not good, the model can be retrained or fine-tuned with data of such conditions.

[0152] In the embodiment of the present application, the initial value of driving behavior and the target driving environment are determined, and the predicted value of driving behavior under the target driving environment is generated by using the trained recurrent neural network, so as to enhance the data set and obtain a data-rich target data set. By generating a large number of driving behavior sequences reflecting various situations in actual driving, and a controllable diversity enhancement strategy, the model output is made richer and more varied, thereby effectively improving the coverage of training data and the generalization ability of the model.

[0153] In some optional implementations, the model can be verified and iterated with feedback. The present invention is also equipped with a set of model verification and feedback iteration modules, which designs a set of closed-loop management processes. The closed-loop management process of model verification and feedback iteration is a method for continuously improving model performance, ensuring that every link from data preparation to model deployment can be continuously optimized and adjusted. Figure 6 As shown, this module mainly includes problem identification, strategy adjustment, model retraining, and continuous monitoring and maintenance functions.

[0154] First, perform problem identification. If the model performs poorly in certain scenarios, it is necessary to identify whether it is due to limitations of the model itself or insufficient data. For example, the model may have large errors in predicting driving behavior in extreme weather conditions, perhaps because such data was not adequately considered during enhancement.

[0155] Next, adjust the strategy and adjust the data enhancement strategy based on the results of the performance analysis. More data needs to be generated for areas with poor performance, such as adding driving sequences under extreme weather conditions or increasing the proportion of data for specific driving styles.

[0156] Then, the model is retrained and trained and validated again based on the adjusted dataset and model parameters. This process may require multiple iterations until the model achieves satisfactory performance levels on all key indicators.

[0157] Finally, continuous monitoring and maintenance are required. Even if the model performance meets expectations, it is necessary to continuously monitor its performance after actual deployment, collect new data from actual operation, and periodically fine-tune or retrain the model to adapt to the trend of driving behavior changing over time.

[0158] It should be noted that model verification and feedback iteration is a dynamic and cyclical process, which aims to improve the accuracy and generalization ability of the model in driving behavior analysis tasks through continuous testing, analysis, and adjustment. Through this method, the model can better understand and predict complex and changing driving scenarios, providing solid support for intelligent transportation systems, autonomous driving technology and other fields.

[0159] The present invention uses a recurrent neural network (RNN) for data enhancement, model construction and training for the application of the Internet of Vehicles data enhancement method in driving behavior scenarios. By cleaning and screening the original Internet of Vehicles data, time series features (such as speed changes, acceleration) and external conditions (weather, road conditions) are extracted to prepare a high-quality data set for RNN input. An LSTM / GRU-based RNN model is designed, and a dual-path input mechanism is used to integrate driving behavior sequences and external conditions, and the model's scene understanding is enhanced through a weighted fusion mechanism. Using the trained RNN model, a new driving behavior sequence is generated through random or purposeful design of initial conditions (such as starting speed, acceleration), and strategies such as noise, variant historical data and potential space exploration are introduced to enhance data diversity. The enhanced data set is applied to the model, and the performance is verified and evaluated. The RNN model parameters or data enhancement strategy are adjusted according to the result feedback to achieve continuous optimization of the model, including problem identification, strategy adjustment, retraining and comprehensive testing. An overall closed-loop management process has been formed from demand analysis to model design, data processing, training, verification, feedback iterative optimization and deployment monitoring, ensuring that the model continues to evolve in driving behavior analysis tasks, improving prediction accuracy and generalization capabilities to meet complex and changing driving scenario requirements.

[0160] In the embodiment of the present invention, the dual-path input fusion mechanism and closed-loop feedback iterative optimization can simultaneously process the time series driving behavior data and external environmental condition information, which not only enhances the model's ability to understand complex driving situations, but also optimizes the model's comprehensive evaluation of the impact of different factors through the weighted fusion mechanism, thereby improving the accuracy of prediction and classification. At the same time, a closed-loop management system from model construction to verification, feedback, and retraining is established, which realizes continuous monitoring and self-optimization of model performance, can automatically identify the deficiencies of the model in specific scenarios, and trigger targeted data enhancement or parameter adjustment, ensuring the real-time adaptability and long-term effectiveness of the model.

[0161] In addition, the present invention innovatively applies adaptive data enhancement technology to dynamically adjust the data transformation strategy according to the model training performance. For example, new samples are generated by simulating different driving conditions and behavior patterns, which effectively solves the problems of scarce and insufficient diversity of driving behavior data, greatly enriches the training data set, and improves the generalization performance of the model.

[0162] In summary, in terms of data enhancement, the present invention introduces the RNN model, dynamically adjusts the data processing method according to the model training status, efficiently generates diversified driving scenario samples, effectively solves the problem of data scarcity, greatly improves the model generalization ability, and improves the data enhancement effect in the field of Internet of Vehicles data.

[0163] The data set enhancement method of the present invention is described in detail below with reference to a specific application example. Figure 7 As shown, this application example mainly includes the following steps:

[0164] Step 1: Data preprocessing and feature extraction. Collect raw Internet of Vehicles data, including but not limited to vehicle status, driving environment parameters, etc., clean the data, remove outliers, and ensure the authenticity of the data foundation. Extract time series features, such as speed changes, acceleration patterns, etc., to prepare for RNN input.

[0165] Step 2: Build an RNN model to simulate driving behavior. Design a recurrent neural network structure based on LSTM or GRU to capture the long-term dependencies and sequence characteristics of driving behavior. Introduce a conditional input mechanism to allow the model to consider external conditions such as weather and road conditions during training and generation, and enhance diversity processing capabilities.

[0166] Step 3: Train the RNN model. The preprocessed feature data is used as input, and the real driving behavior sequence (such as the vehicle's driving trajectory, operation behavior, etc.) is used as the target output to perform supervised learning training on the RNN model. In order to improve the generalization ability of the model, strategies such as cross-validation and regularization can be used, and the network architecture parameters (such as the number of hidden layers, the number of units, etc.) and optimization algorithms (such as Adam, RMSprop, etc.) can be adjusted to minimize the prediction error.

[0167] Step 4: Generate augmented data. Using the trained RNN model, generate new driving behavior sequences by random sampling or purposefully designing initial conditions (such as starting speed, acceleration, etc.). These sequences should reflect various situations that may occur in actual driving, including different driving styles, road conditions, traffic density, etc. The generation process can increase diversity by introducing noise, slightly mutated historical data, or exploring the model's potential space.

[0168] Step 5: Model verification and feedback iteration. Apply the enhanced dataset to driving behavior analysis and prediction models to evaluate model performance. Based on the evaluation results, feedback is provided to adjust RNN model parameters or data enhancement strategies to achieve iterative optimization.

[0169] In the present invention, data enhancement is driven by deep learning. Different from the simple data copying or basic transformation of traditional methods, the present invention utilizes recurrent neural networks (RNNs), especially long short-term memory (LSTM) or gated recurrent units (GRUs), to generate new driving behavior sequences. It can learn complex timing patterns based on historical data and create highly relevant and realistic new data instances based on them, thereby enhancing the realism of the data.

[0170] In the present invention, conditional input is added to enhance data diversity. The introduction of the conditional input mechanism enables the RNN model to consider more dimensional external conditions (such as weather, road conditions, etc.) when generating data. This not only enhances the environmental adaptability of the generated data, but also greatly increases the diversity of the data set, so that the data generated by the model can better reflect the various complex scenarios that may be encountered in actual driving.

[0171] The present invention uses a multimodal fusion strategy. In addition to the basic data of vehicle driving, the present invention also advocates the fusion processing of multimodal sensor data such as vision and hearing, which further enriches the dimension and information volume of the data and helps to build a more comprehensive and sophisticated driving behavior model. The multimodal fusion strategy refers to a series of technologies and methods that integrate data information of different modes in processing scenarios containing multiple types of data (such as images, text, audio, video, etc.) to obtain a more comprehensive and accurate feature representation or decision-making basis. It aims to overcome the limitations of single-modal data, utilize the complementarity and redundancy of information between different modalities, and improve the performance, accuracy and robustness of the model.

[0172] The present invention establishes a feedback iterative optimization mechanism, emphasizing the closed-loop process from model verification to feedback adjustment, ensuring that the data enhancement strategy can be continuously optimized based on performance feedback in actual applications, improving the adaptability of the entire system and the effectiveness of the final data enhancement. In addition, the present invention proposes a complete system architecture that covers the entire process from data collection, preprocessing to RNN processing, diversity generation, and final model evaluation and feedback, realizing the practical application of technological innovation at the operational level.

[0173] In summary, the present invention performs Internet of Vehicles data enhancement based on the RNN model, which brings significant improvements to Internet of Vehicles data enhancement, realizes the application of data enhancement in current Internet of Vehicles data scenarios, especially driving behavior analysis scenarios, lays a solid data foundation for autonomous driving technology, and has significant application prospects and economic benefits.

[0174] In this embodiment, a data set enhancement device is also provided, which is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0175] This embodiment provides a data set enhancement device, such as Figure 8 As shown, including:

[0176] The acquisition module 801 is used to acquire the driving behavior sequence and driving environment data of the vehicle;

[0177] The first processing module 802 is used to construct an original data set based on the driving behavior sequence, and train a preset recurrent neural network using the original data set and the driving environment data to obtain a trained recurrent neural network;

[0178] The second processing module 803 is used to determine the initial value of the vehicle's driving behavior and the target driving environment;

[0179] The third processing module 804 is used to generate a driving behavior prediction value of the vehicle in the target driving environment based on the initial value of the driving behavior, the target driving environment and the trained recurrent neural network to obtain a target data set.

[0180] In some optional implementations, the acquisition module 801 is further used to:

[0181] Acquire driving behavior data and perform data cleaning on the driving behavior data; wherein the driving behavior data includes at least one of vehicle speed, engine speed and acceleration, and the data cleaning includes at least one of outlier removal, missing value filling, smoothing, noise reduction and normalization;

[0182] The time series features of the cleaned driving behavior data are extracted to obtain the driving behavior sequence.

[0183] In some optional implementations, the first processing module 802 is further configured to:

[0184] The driving behavior sequence is divided into multiple sample sequences with continuous time steps through a sliding window;

[0185] Extract the numerical features of each sample sequence, normalize and standardize the numerical features, and obtain the feature vector of each sample sequence;

[0186] For the sample sequence of each time step, the label of the sample sequence of the current time step is determined according to the sample sequence of the next time step; wherein the label is used to characterize the driving behavior data of the sample sequence of the current time step in the future time step;

[0187] Based on the feature vector and label of each sample sequence, the original data set is constructed.

[0188] In some optional implementations, the first processing module 802 is further configured to:

[0189] Encoding the driving environment data to obtain environment encoding data;

[0190] The preset recurrent neural network is iterated multiple times according to the sample sequences of multiple time steps in the original data set, and in each iteration, the state of the recurrent neural network is updated based on the sample sequence of the current time step and the environmental coding data until the recurrent neural network converges to obtain a trained recurrent neural network.

[0191] In some optional implementations, the recurrent neural network is a long short-term memory network; the first processing module 802 is further used to:

[0192] Perform weighted fusion or feature concatenation on the sample sequence of the current time step and the environmental coding data to obtain a new sample sequence corresponding to the current time step;

[0193] According to the new sample sequence, the state of the long short-term memory network is updated.

[0194] In some optional implementations, the recurrent neural network includes a long short-term memory network, and the input gate, the forget gate, and the output gate of the long short-term memory network are all provided with a weight matrix; the first processing module 802 is further used for:

[0195] Determine a weight adjustment factor corresponding to the environment encoding data; wherein the weight adjustment factor is used to adjust the weight matrix of the long short-term memory network;

[0196] Update the state of the long short-term memory network according to the weight adjustment factor.

[0197] In some optional implementations, the recurrent neural network includes a long short-term memory network, and the long short-term memory network includes an input gate, a forget gate, and an output gate; the first processing module 802 is further used to:

[0198] Determine the standard gate values ​​corresponding to the input gate, forget gate and output gate in the long short-term memory network, and determine the conditional gate values ​​corresponding to the input gate, forget gate and output gate based on the environment encoding data;

[0199] According to the standard gate value and the conditional gate value, the target gate values ​​corresponding to the input gate, the forget gate and the output gate are calculated respectively;

[0200] According to the target gate value, the state of the long short-term memory network is updated.

[0201] In some optional implementations, the third processing module 804 is further configured to:

[0202] Constructing an initialization sequence based on the initial value of the driving behavior and the target driving environment; wherein the target driving environment includes at least one of the road type, weather conditions and traffic density;

[0203] Input the initialization sequence into the trained recurrent neural network to obtain the predicted value of the vehicle's driving behavior in the target driving environment;

[0204] Based on the predicted value of the vehicle's driving behavior in the target driving environment, a target driving behavior sequence is constructed to obtain a target data set.

[0205] In some optional embodiments, before inputting the initialization sequence into the trained recurrent neural network, the apparatus is further configured to:

[0206] The sample sequence in the original data set is stretched or compressed in time scale to obtain the variation history data, and the initialization sequence is updated based on the variation history data;

[0207] After the initialization sequence is fed into the trained recurrent neural network, the device is also used to:

[0208] Interpolate the hidden state of the trained recurrent neural network, or add random variables to the hidden state;

[0209] After obtaining the predicted value of the driving behavior of the vehicle in the target driving environment, the device is also used to:

[0210] Add random noise to the driving behavior prediction value.

[0211] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0212] The data set enhancement device in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0213] The embodiment of the present invention also provides a computer device having the above Figure 8 The dataset augmentation device shown.

[0214] See also Fig. 9 , Fig. 9 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Fig. 9As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Fig. 9 A processor 10 is taken as an example.

[0215] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0216] The memory 20 stores instructions executable by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0217] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0218] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0219] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means. Fig. 9 The example of connecting through bus is taken in the following.

[0220] The input device 30 can receive input digital or character information, and generate key signal input related to the user settings and function control of the computer device, such as a touch screen, a keypad, a mouse, a track pad, a touch pad, an indicator bar, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 may include a display device, an auxiliary lighting device (e.g., an LED) and a tactile feedback device (e.g., a vibration motor), etc. The above-mentioned display device includes but is not limited to a liquid crystal display, a light emitting diode, a display and a plasma display. In some optional embodiments, the display device can be a touch screen.

[0221] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0222] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0223] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A data set enhancement method, characterized in that: The method comprises: Obtain the vehicle's driving behavior sequence and driving environment data; Constructing an original data set based on the driving behavior sequence, and training a preset recurrent neural network using the original data set and the driving environment data to obtain a trained recurrent neural network; Determine the initial value of the vehicle's driving behavior and the target driving environment; Based on the initial value of the driving behavior, the target driving environment and the trained recurrent neural network, a driving behavior prediction value of the vehicle in the target driving environment is generated to obtain a target data set.

2. The method according to claim 1, characterized in that The method of training a preset recurrent neural network using the original data set and the driving environment data to obtain a trained recurrent neural network includes: Encoding the driving environment data to obtain environment encoding data; The preset recurrent neural network is iterated multiple times according to the sample sequences of multiple time steps in the original data set, and in each iteration, the state of the recurrent neural network is updated based on the sample sequence of the current time step and the environmental coding data until the recurrent neural network converges to obtain a trained recurrent neural network.

3. The method according to claim 2, characterized in that The recurrent neural network is a long short-term memory network; the state of the recurrent neural network is updated based on the sample sequence of the current time step and the environment encoding data, including: Perform weighted fusion or feature concatenation on the sample sequence of the current time step and the environmental coding data to obtain a new sample sequence corresponding to the current time step; According to the new sample sequence, the state of the long short-term memory network is updated.

4. The method according to claim 2, characterized in that: The recurrent neural network includes a long short-term memory network, and the input gate, the forget gate and the output gate of the long short-term memory network are all provided with a weight matrix; the state of the recurrent neural network is updated based on the sample sequence of the current time step and the environment coding data, including: Determining a weight adjustment factor corresponding to the environmental coding data; wherein the weight adjustment factor is used to adjust the weight matrix of the long short-term memory network; The state of the long short-term memory network is updated according to the weight adjustment factor.

5. The method according to claim 2, characterized in that: The recurrent neural network includes a long short-term memory network, and the long short-term memory network includes an input gate, a forget gate, and an output gate; the state of the recurrent neural network is updated based on the sample sequence of the current time step and the environment encoding data, including: Determine the standard gate values ​​corresponding to the input gate, the forget gate and the output gate in the long short-term memory network, and determine the conditional gate values ​​corresponding to the input gate, the forget gate and the output gate based on the environmental encoding data; According to the standard gate value and the conditional gate value, the target gate values ​​corresponding to the input gate, the forget gate and the output gate are calculated respectively; The state of the long short-term memory network is updated according to the target gate value.

6. The method according to any one of claims 1 to 5, characterized in that The method generates a driving behavior prediction value of the vehicle in the target driving environment based on the driving behavior initial value, the target driving environment and the trained recurrent neural network, and obtains a target data set, including: Based on the initial value of the driving behavior and the target driving environment, constructing an initialization sequence; wherein the target driving environment includes at least one of road type, weather conditions and traffic density; Inputting the initialization sequence into a trained recurrent neural network to obtain a driving behavior prediction value of the vehicle in the target driving environment; Based on the predicted driving behavior value of the vehicle in the target driving environment, a target driving behavior sequence is constructed to obtain a target data set.

7. The method according to claim 6, characterized in that Before inputting the initialization sequence into the trained recurrent neural network, the method further includes: Performing time scale stretching or time scale compression on the sample sequence in the original data set to obtain variation history data, and updating the initialization sequence based on the variation history data; After inputting the initialization sequence into the trained recurrent neural network, the method further includes: Interpolating the hidden state of the trained recurrent neural network, or adding a random variable to the hidden state; After obtaining the predicted value of the driving behavior of the vehicle in the target driving environment, the method includes: Random noise is added to the driving behavior prediction value.

8. The method according to any one of claims 1 to 5, characterized in that The step of obtaining the driving behavior sequence of the vehicle includes: Acquire driving behavior data, and perform data cleaning on the driving behavior data; wherein the driving behavior data includes at least one of vehicle speed, engine speed, and acceleration, and the data cleaning includes at least one of outlier removal, missing value filling, smoothing, noise reduction, and normalization; The time series features of the cleaned driving behavior data are extracted to obtain the driving behavior sequence.

9. The method according to claim 8, characterized in that The constructing of the original data set based on the driving behavior sequence comprises: The driving behavior sequence is divided into multiple sample sequences with continuous time steps through a sliding window; Extracting numerical features of each sample sequence, normalizing and standardizing the numerical features, and obtaining feature vectors of each sample sequence; For the sample sequence of each time step, determine the label of the sample sequence of the current time step according to the sample sequence of the next time step; wherein the label is used to characterize the driving behavior data of the sample sequence of the current time step in the future time step; Based on the feature vector and label of each sample sequence, an original data set is constructed.

10. A data set enhancement device, characterized in that: The device comprises: An acquisition module, used to acquire the driving behavior sequence and driving environment data of the vehicle; A first processing module, configured to construct an original data set based on the driving behavior sequence, and train a preset recurrent neural network using the original data set and the driving environment data to obtain a trained recurrent neural network; A second processing module is used to determine the initial value of the vehicle's driving behavior and the target driving environment; The third processing module is used to generate a driving behavior prediction value of the vehicle in the target driving environment based on the initial value of the driving behavior, the target driving environment and the trained recurrent neural network, so as to obtain a target data set.

11. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the data set enhancement method according to any one of claims 1 to 9 by executing the computer instructions.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the data set enhancement method according to any one of claims 1 to 9.

13. A computer program product, characterized in that The method comprises computer instructions for causing a computer to execute the data set enhancement method according to any one of claims 1 to 9.

Images