Method, device and equipment for automatically identifying vehicle using scene of vehicle

Abstract

The embodiment of the invention provides an automatic identification method, device and equipment for a vehicle using scene of a vehicle. The method comprises the following steps: acquiring vehicle data in each acquisition time window in a plurality of acquisition time windows before the current moment; the adjacent acquisition time windows are overlapped in the time dimension, and the durations of at least two acquisition time windows are different; inputting the vehicle data in the acquisition time window of which the duration is less than a preset threshold value into a first model for identification to obtain a scene feature vector of the acquisition time window; inputting the vehicle data in the acquisition time window of which the duration is greater than or equal to a preset threshold value into a second model for identification to obtain a scene feature vector of the acquisition time window; the scene feature vector represents the features of the vehicle data of the collection time window; and according to the scene feature vector of each time window, determining the vehicle use scene of the vehicle at the current moment. The method is used for achieving the effect of accurately and effectively recognizing the vehicle using scene of the vehicle.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, and in particular to an automatic identification method, device, and equipment for vehicle usage scenarios. Background Technology

[0002] With the development of vehicle intelligence, it is often necessary to identify vehicle usage scenarios in order to control the vehicle or to collect and process data. These usage scenarios include, for example, high-speed driving scenarios and cornering scenarios.

[0003] Therefore, how to accurately and effectively identify the vehicle's usage scenarios is an urgent problem to be solved. Summary of the Invention

[0004] This application provides an automatic identification method, device, and equipment for vehicle usage scenarios, so as to achieve accurate and effective identification of vehicle usage scenarios.

[0005] In a first aspect, embodiments of this application provide a method for automatically identifying vehicle usage scenarios, including:

[0006] Acquire vehicle data within each of multiple acquisition time windows preceding the current time; wherein adjacent acquisition time windows overlap in the time dimension, and at least two acquisition time windows have different durations;

[0007] Vehicle data within a collection time window with a duration less than a preset threshold is input into the first model for identification to obtain the scene feature vector of that collection time window; vehicle data within a collection time window with a duration greater than or equal to the preset threshold is input into the second model for identification to obtain the scene feature vector of that collection time window; wherein, the scene feature vector characterizes the features of the vehicle data within the collection time window.

[0008] Based on the scene feature vectors of each time window, the vehicle's usage scenario at the current moment is determined.

[0009] In one possible implementation, the first model is a lightweight GRU network model; vehicle data within a collection time window with a duration less than a preset threshold is input into the first model for identification to obtain a scene feature vector for that collection time window, including:

[0010] Vehicle data collected within a time window shorter than a preset threshold is input into the lightweight GRU network model for feature extraction to obtain data change features; wherein, the data change features are features that characterize sudden changes in vehicle data.

[0011] Based on the lightweight GRU network model, the data change characteristics are identified to obtain the scene feature vector for that time window.

[0012] In one possible implementation, the second model is an LSTM network model; vehicle data within a collection time window with a duration greater than or equal to a preset threshold is input into the second model for identification to obtain a scene feature vector for that collection time window, including:

[0013] Vehicle data collected within a time window with a duration greater than or equal to a preset threshold is input into the LSTM network model for feature extraction to obtain persistent data features; wherein, the persistent data features are features that characterize the continuous changes in vehicle data, or features that characterize the continuous invariance of vehicle data.

[0014] Based on the LSTM network model, the continuous features of the data are identified to obtain the scene feature vector for the acquisition time window.

[0015] In one possible implementation, the vehicle usage scenario at the current moment is determined based on the scene feature vectors of each of the collection time windows, including:

[0016] Based on the vehicle data within the initial time window, the scene feature vectors of each collection time window are weighted and summed to obtain the fused feature vector of the vehicle; wherein, the initial time window is a time window whose duration before the current time is less than a preset threshold.

[0017] The vehicle's usage scenario at the current moment is determined based on the vehicle's usage scenario in the previous preset time period and the fused feature vector.

[0018] In one possible implementation, based on vehicle data within an initial time window, a weighted summation of scene feature vectors from each of the acquisition time windows is performed to obtain a fused feature vector for the vehicle, including:

[0019] Based on the vehicle data within the initial time window, determine the characteristic rate of change of the initial time window;

[0020] Based on the characteristic change rate of the initial time window, determine the weight coefficient of the scene feature vector for each acquisition time window;

[0021] Based on the weight coefficients of the scene feature vectors of each acquisition time window, the scene feature vectors of each acquisition time window are weighted and summed to obtain the fused feature vector of the vehicle.

[0022] In one possible implementation, determining the characteristic rate of change of the initial time window based on vehicle data within the initial time window includes:

[0023] Based on the specified vehicle data at adjacent moments in the initial time window, determine the data change information at adjacent moments in the initial time window;

[0024] The characteristic rate of change of the initial time window is determined based on the data change information of each adjacent time point in the initial time window.

[0025] In one possible implementation, the specified vehicle data is vehicle speed, then the data change information at adjacent time points is the speed difference at adjacent time points, and the characteristic change rate of the initial time window is the standard deviation of the vehicle speed at the initial time window;

[0026] If the specified vehicle data is the steering angle, then the data change information at adjacent time points is the steering angle difference at adjacent time points, and the characteristic change rate of the initial time window is the steering angle variance of the initial time window.

[0027] In one possible implementation, the weight coefficients of the scene feature vectors for each acquisition time window are determined based on the feature change rate of the initial time window, including:

[0028] If it is determined that the feature change rate of the initial time window is greater than or equal to the preset change rate, then the weight coefficient corresponding to the collection time window with a duration less than the preset threshold is determined, and the weight coefficient corresponding to the collection time window with a duration greater than or equal to the preset threshold is determined.

[0029] If it is determined that the feature change rate of the initial time window is less than the preset change rate, then the weight coefficient corresponding to the collection time window with a duration less than the preset threshold is determined, and the weight coefficient corresponding to the collection time window with a duration less than or equal to the preset threshold is determined.

[0030] In one possible implementation, determining the vehicle's usage scenario at the current moment based on the vehicle's usage scenario during a previous preset time period and the fused feature vector includes:

[0031] The fused feature vector is processed based on the fusion model to obtain the maximum recognition probability of the vehicle at the current moment;

[0032] Based on the vehicle's usage scenario in the previous preset time period, the maximum recognition probability is corrected to obtain the vehicle's usage scenario at the current moment.

[0033] In one possible implementation, the maximum recognition probability is corrected based on the vehicle's usage scenario in a previous preset time period to obtain the vehicle's usage scenario at the current moment, including:

[0034] Based on the vehicle's usage scenario in the previous preset period, retrieve the scenario mapping relationship corresponding to the user scenario of the vehicle in the previous preset period; wherein, the scenario mapping relationship includes correction coefficients for different usage scenarios;

[0035] Based on the vehicle usage scenario corresponding to the highest recognition probability, a correction coefficient corresponding to the vehicle usage scenario corresponding to the highest recognition probability is determined from the scenario mapping relationship;

[0036] Based on the obtained correction coefficient, the maximum recognition probability is corrected to obtain the corrected recognition probability; and based on the corrected recognition probability, the vehicle's usage scenario at the current moment is determined.

[0037] In one possible implementation, determining the vehicle's usage scenario at the current moment based on the corrected recognition probability includes:

[0038] If it is determined that the corrected recognition probability is greater than or equal to the preset probability, then the vehicle usage scenario at the current moment is determined to be the usage scenario corresponding to the maximum recognition probability.

[0039] If it is determined that the corrected recognition probability is less than the preset probability, the vehicle's usage scenario at the current moment is determined to be a transitional scenario.

[0040] In one possible implementation, the method further includes:

[0041] The vehicle data is preprocessed, wherein the preprocessing includes one or more of the following: outlier removal, missing value completion, and feature normalization.

[0042] Secondly, embodiments of this application provide an automatic vehicle usage scenario recognition device, comprising:

[0043] The acquisition module is used to acquire vehicle data within each of multiple acquisition time windows preceding the current time; wherein adjacent acquisition time windows overlap in the time dimension, and at least two acquisition time windows have different durations;

[0044] The identification module is used to input vehicle data within a collection time window with a duration less than a preset threshold into a first model for identification to obtain a scene feature vector for that collection time window; and to input vehicle data within a collection time window with a duration greater than or equal to the preset threshold into a second model for identification to obtain a scene feature vector for that collection time window; wherein, the scene feature vector characterizes the features of the vehicle data within the collection time window.

[0045] The determination module is used to determine the vehicle's usage scenario at the current moment based on the scene feature vectors of each time window.

[0046] In one possible implementation, the first model is a lightweight GRU network model; the identification module includes:

[0047] Vehicle data collected within a time window shorter than a preset threshold is input into the lightweight GRU network model for feature extraction to obtain data change features; wherein, the data change features are features that characterize sudden changes in vehicle data.

[0048] Based on the lightweight GRU network model, the data change characteristics are identified to obtain the scene feature vector for that time window.

[0049] In one possible implementation, the second model is an LSTM network model; the identification module includes:

[0050] Vehicle data collected within a time window with a duration greater than or equal to a preset threshold is input into the LSTM network model for feature extraction to obtain persistent data features; wherein, the persistent data features are features that characterize the continuous changes in vehicle data, or features that characterize the continuous invariance of vehicle data.

[0051] Based on the LSTM network model, the continuous features of the data are identified to obtain the scene feature vector for the acquisition time window.

[0052] In one possible implementation, the determining module includes:

[0053] The processing module is used to perform weighted summation processing on the scene feature vectors of each collection time window based on the vehicle data within the initial time window to obtain the fused feature vector of the vehicle; wherein, the initial time window is a time window whose duration before the current moment is less than a preset threshold.

[0054] The determination submodule is used to determine the vehicle's usage scenario at the current moment based on the vehicle's usage scenario in the previous preset time period and the fused feature vector.

[0055] In one possible implementation, the processing module includes:

[0056] Based on the vehicle data within the initial time window, determine the characteristic rate of change of the initial time window;

[0057] Based on the characteristic change rate of the initial time window, determine the weight coefficient of the scene feature vector for each acquisition time window;

[0058] Based on the weight coefficients of the scene feature vectors of each acquisition time window, the scene feature vectors of each acquisition time window are weighted and summed to obtain the fused feature vector of the vehicle.

[0059] In one possible implementation, determining the characteristic rate of change of the initial time window based on vehicle data within the initial time window includes:

[0060] Based on the specified vehicle data at adjacent moments in the initial time window, determine the data change information at adjacent moments in the initial time window;

[0061] The characteristic rate of change of the initial time window is determined based on the data change information of each adjacent time point in the initial time window.

[0062] In one possible implementation, the specified vehicle data is vehicle speed, then the data change information at adjacent time points is the speed difference at adjacent time points, and the characteristic change rate of the initial time window is the standard deviation of the vehicle speed at the initial time window;

[0063] If the specified vehicle data is the steering angle, then the data change information at adjacent time points is the steering angle difference at adjacent time points, and the characteristic change rate of the initial time window is the steering angle variance of the initial time window.

[0064] In one possible implementation, the weight coefficients of the scene feature vectors for each acquisition time window are determined based on the feature change rate of the initial time window, including:

[0065] If it is determined that the feature change rate of the initial time window is greater than or equal to the preset change rate, then the weight coefficient corresponding to the collection time window with a duration less than the preset threshold is determined, and the weight coefficient corresponding to the collection time window with a duration greater than or equal to the preset threshold is determined.

[0066] If it is determined that the feature change rate of the initial time window is less than the preset change rate, then the weight coefficient corresponding to the collection time window with a duration less than the preset threshold is determined, and the weight coefficient corresponding to the collection time window with a duration less than or equal to the preset threshold is determined.

[0067] In one possible implementation, determining the submodule includes:

[0068] The fused feature vector is processed based on the fusion model to obtain the maximum recognition probability of the vehicle at the current moment;

[0069] Based on the vehicle's usage scenario in the previous preset time period, the maximum recognition probability is corrected to obtain the vehicle's usage scenario at the current moment.

[0070] In one possible implementation, the maximum recognition probability is corrected based on the vehicle's usage scenario in a previous preset time period to obtain the vehicle's usage scenario at the current moment, including:

[0071] Based on the vehicle's usage scenario in the previous preset period, retrieve the scenario mapping relationship corresponding to the user scenario of the vehicle in the previous preset period; wherein, the scenario mapping relationship includes correction coefficients for different usage scenarios;

[0072] Based on the vehicle usage scenario corresponding to the highest recognition probability, a correction coefficient corresponding to the vehicle usage scenario corresponding to the highest recognition probability is determined from the scenario mapping relationship;

[0073] Based on the obtained correction coefficient, the maximum recognition probability is corrected to obtain the corrected recognition probability; and based on the corrected recognition probability, the vehicle's usage scenario at the current moment is determined.

[0074] In one possible implementation, determining the vehicle's usage scenario at the current moment based on the corrected recognition probability includes:

[0075] If it is determined that the corrected recognition probability is greater than or equal to the preset probability, then the vehicle usage scenario at the current moment is determined to be the usage scenario corresponding to the maximum recognition probability.

[0076] If it is determined that the corrected recognition probability is less than the preset probability, the vehicle's usage scenario at the current moment is determined to be a transitional scenario.

[0077] In one possible implementation, the device further includes:

[0078] The vehicle data is preprocessed, wherein the preprocessing includes one or more of the following: outlier removal, missing value completion, and feature normalization.

[0079] Thirdly, embodiments of this application provide an electronic device, including: a memory and a processor;

[0080] The memory stores computer-executed instructions;

[0081] The processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0082] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0083] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0084] This application provides an automatic vehicle usage scenario identification method, apparatus, and device. The method involves acquiring vehicle data within each of multiple acquisition time windows preceding the current moment, with adjacent acquisition time windows overlapping in the time dimension and at least two windows having different durations. Vehicle data within acquisition time windows with durations less than a preset threshold are then input into a first model for identification, yielding a scenario feature vector for that acquisition time window. Simultaneously, vehicle data within acquisition time windows with durations greater than or equal to the preset threshold are input into a second model for identification, also yielding a scenario feature vector for that acquisition time window. The scenario feature vector represents the characteristics of the vehicle data within each acquisition time window. Finally, based on the scenario feature vectors of each time window, the vehicle's usage scenario at the current moment is determined. By employing acquisition time windows of different durations combined with different model identification methods, this approach utilizes both the ability of short-duration windows to capture detailed information and the grasp of overall trends through long-duration windows. This results in more comprehensive and accurate acquisition of scenario feature vectors, thereby achieving a more precise determination of the vehicle's usage scenario at the current moment. This effectively solves the technical problem of traditional methods' difficulty in comprehensively and accurately identifying usage scenarios. Attached Figure Description

[0085] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0086] Figure 1 A flowchart illustrating an automatic vehicle usage scenario recognition method provided in this application embodiment. Figure 1 ;

[0087] Figure 2 A flowchart illustrating an automatic vehicle usage scenario recognition method provided in this application embodiment. Figure 2 ;

[0088] Figure 3 A flowchart illustrating step S204 of an automatic vehicle usage scenario recognition method provided in this application embodiment;

[0089] Figure 4 A schematic diagram of the structure of an automatic vehicle usage scenario recognition device provided in this application embodiment. Figure 1 ;

[0090] Figure 5 A schematic diagram of the structure of an automatic vehicle usage scenario recognition device provided in this application embodiment. Figure 2 ;

[0091] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0092] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0093] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0094] With the continuous advancement of technology, vehicle intelligence has become a significant trend in the automotive industry. Intelligent vehicles aim to achieve a safer, more efficient, and more comfortable driving experience by integrating various advanced technologies, while also providing a foundation for subsequent data analysis and vehicle optimization. Accurately identifying usage scenarios is a crucial step in the process of vehicle intelligence, providing essential data for intelligent vehicle control and the effective collection and processing of data.

[0095] Currently, with the continuous expansion and application of vehicle intelligent functions, higher demands are being placed on the accuracy and effectiveness of vehicle usage scenario recognition. However, existing recognition methods may have certain limitations, such as difficulty in comprehensively and in real-time perceiving the vehicle's surrounding environment and its own state, leading to inaccurate scenario recognition; or high algorithm complexity affecting recognition efficiency and failing to meet the needs of real-time control. These problems make accurately and effectively identifying vehicle usage scenarios a pressing issue that needs to be addressed. Failure to resolve this issue in a timely manner will limit the further development of vehicle intelligent functions, affect user experience, and may even pose a potential threat to driving safety.

[0096] Therefore, the automatic identification method for vehicle usage scenarios provided in this application can solve the above problems.

[0097] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0098] Figure 1 A flowchart illustrating an automatic vehicle usage scenario recognition method provided in this application embodiment. Figure 1 ,like Figure 1 As shown, the method includes:

[0099] S101. Obtain vehicle data within each of multiple acquisition time windows preceding the current time; wherein adjacent acquisition time windows overlap in the time dimension, and at least two acquisition time windows have different durations.

[0100] For example, multiple data collection time windows are a series of time windows that extend backward from the current moment, with each window representing a data collection interval. Vehicle data includes information such as vehicle location, speed, fuel consumption, and sensor readings, which typically come from in-vehicle devices or IoT platforms. Understandably, vehicle data can also include vehicle driving data, environmental and location data, and vehicle status data; vehicle driving data can also include vehicle speed, acceleration, steering angle, braking frequency, engine / motor speed, gear changes, etc.; environmental and location data can also include GPS coordinates, altitude change rate, navigation road type (e.g., highway, urban road), and slope (calculated via acceleration sensors or GPS); vehicle status data can also include suspension compression, tire pressure change rate, door opening / closing status, seat pressure, etc.

[0101] For example, based on the current time, multiple acquisition time windows are defined. The start and end times of these windows are both before the current time, and adjacent windows overlap by adjusting their start and end times. Meanwhile, the window duration is set by parameters to ensure differences.

[0102] In one possible implementation, the number of data collection time windows, the duration of each time window, and the overlap ratio are set; for each time window, vehicle data within that time period is extracted from a database or real-time data stream; and the data from multiple time windows are preprocessed to support subsequent analysis.

[0103] In one possible implementation, assuming the current time is 12:00:00, it is necessary to acquire vehicle data from the previous two acquisition time windows. Window 1 is set from 11:59:55 to 12:00:00 (duration 5 seconds), and window 2 is from 11:59:50 to 12:00:00 (duration 10 seconds, overlapping with window 1 from 11:59:55 to 12:00:00).

[0104] S102. Vehicle data within a collection time window with a duration less than a preset threshold is input into the first model for identification to obtain the scene feature vector of the collection time window; vehicle data within a collection time window with a duration greater than or equal to the preset threshold is input into the second model for identification to obtain the scene feature vector of the collection time window; wherein, the scene feature vector represents the characteristics of the vehicle data in the collection time window.

[0105] For example, when processing vehicle data for each collection time window, the system first calculates the duration of that window (end timestamp - start timestamp) and compares it with a preset threshold. This step, as a routing logic, determines which processing branch the data will flow to.

[0106] Regardless of which branch the data flows to, necessary preprocessing is required before it enters the model. This includes data cleaning (handling missing and outlier values), normalization, and feature extraction. For vehicle data, features may include average speed, speed variance, maximum acceleration, GPS location sequence, steering wheel angle, etc.

[0107] If the duration of the data acquisition window is less than a preset threshold, the preprocessed data will be input into the first model. The first model performs forward propagation calculations and outputs a preliminary scene feature vector for short-term behavior.

[0108] If the duration of the data collection window is greater than or equal to a preset threshold, the data is input into the second model. The second model, leveraging its stronger representational capabilities, performs in-depth analysis on longer behavioral sequences, outputting a more accurate and stable scene feature vector.

[0109] S103. Determine the vehicle usage scenario at the current moment based on the scene feature vectors of each time window.

[0110] For example, the system first collects scene feature vectors for all acquisition time windows within the current preset time period.

[0111] For each vehicle, the system employs one or more aggregation strategies to process its scene feature vector:

[0112] Weighted average method: Assign a weight to each window (e.g., the longer the window duration, the higher the weight; or give all windows equal weight without considering overlap), and then take a weighted average of the probability vectors of all windows to obtain a comprehensive probability vector. Each dimension of this vector represents the average probability of the scene over the entire preset time period.

[0113] Voting method: The most likely scenario (i.e., the scenario with the highest probability) identified in each time window is counted as one vote. Finally, the scenario that receives the most votes throughout the entire preset time period is determined as the final scenario.

[0114] Consider a time-decay weighted approach: assign higher weights to windows closer to the current time, as these better reflect the vehicle's latest state. This can be achieved using an exponentially weighted moving average.

[0115] Based on the aggregation results obtained in the previous step, the final vehicle usage scenario is generated. If a weighted average method is used, the scenario with the highest probability value in the comprehensive probability vector is usually selected as the final decision scenario.

[0116] If all information needs to be retained, this comprehensive probability vector itself can be used as the output, which represents the degree of mixing of vehicle usage scenarios within a preset time period.

[0117] This application provides an automatic vehicle usage scenario identification method. This method acquires vehicle data from multiple acquisition time windows preceding the current moment, with adjacent acquisition time windows overlapping in the time dimension and at least two windows having different durations. Then, vehicle data from acquisition time windows with durations less than a preset threshold are input into a first model for identification, obtaining a scenario feature vector for that acquisition time window. Simultaneously, vehicle data from acquisition time windows with durations greater than or equal to the preset threshold are input into a second model for identification, also obtaining a scenario feature vector for that acquisition time window. The scenario feature vector represents the characteristics of the vehicle data within each acquisition time window. Finally, based on the scenario feature vectors of each time window, the vehicle's usage scenario at the current moment is determined. By employing acquisition time windows of different durations combined with different model identification methods, this method utilizes both the ability of short-duration windows to capture detailed information and the grasp of overall trends using long-duration windows. This results in more comprehensive and accurate acquisition of scenario feature vectors, leading to a more precise determination of the vehicle's usage scenario at the current moment. This effectively solves the technical problem of traditional methods' difficulty in comprehensively and accurately identifying usage scenarios.

[0118] Figure 2 A flowchart illustrating an automatic vehicle usage scenario recognition method provided in this application embodiment. Figure 2 ,like Figure 2 As shown, the method includes:

[0119] S201. Obtain vehicle data within each of multiple acquisition time windows preceding the current time; wherein adjacent acquisition time windows overlap in the time dimension, and at least two acquisition time windows have different durations.

[0120] For example, this step can refer to step S101 above, and will not be repeated here.

[0121] In one example, vehicle data is preprocessed, which includes one or more of the following: outlier removal, missing value completion, and feature normalization.

[0122] For example, the system first loads the raw vehicle data obtained from the data source into the processing unit. An initial investigation is then conducted to quickly determine the proportion of missing values, the numerical distribution of each feature, and other relevant data to decide on subsequent processing strategies.

[0123] Outlier removal: Anomaly detection rules are defined for each key feature (e.g., velocity, acceleration). Then, the data sequence is scanned, and data points that meet the outlier criteria are marked as invalid. Finally, these invalid points are removed from the dataset or set to null values, awaiting subsequent completion. This process is typically the first step in a data cleaning pipeline to eliminate significant noise.

[0124] Missing value imputation: Identify all missing values ​​that remain after outlier removal. Select an appropriate imputation method based on data characteristics and business logic. Finally, fill in each missing position. This step follows immediately after outlier removal to ensure the integrity of the dataset.

[0125] Feature normalization: Select the numerical features that need to be normalized. Calculate the scaling parameters (such as minimum, maximum, mean, and standard deviation) for these features in the training dataset. Finally, apply these parameters to all data (including subsequent real-time data). This is the final step in the preprocessing workflow, completed before the data enters the model, to ensure that the input features are scaled uniformly.

[0126] S201. The first model is a lightweight GRU network model. Vehicle data within a collection time window with a duration less than a preset threshold is input into the lightweight GRU network model for feature extraction to obtain data change features. Among them, the data change features are features that characterize sudden changes in vehicle data. Based on the lightweight GRU network model, the data change features are identified to obtain the scene feature vector of the time window.

[0127] For example, the Lightweight Gated Recurrent Unit (GRU) network model is a recurrent neural network, particularly adept at processing and identifying long-term dependencies in time-series data. "Lightweight" here refers to the model's deliberate simplification, such as reducing the number of network layers and neurons per layer. This significantly reduces computational complexity and inference time while maintaining the GRU's core ability to capture temporal dynamics, thus meeting the need for real-time or near-real-time processing of vehicle data within acquisition time windows shorter than a preset threshold.

[0128] The preprocessed sequence data is input into a pre-trained lightweight GRU model. This model typically has only one GRU layer, and the number of hidden units may be set to 64 or 128 to balance accuracy and speed.

[0129] The lightweight GRU model processes the sequence step-by-step. Its internal update and reset gates dynamically determine how much past information to retain and how to combine the current input with past states. During this process, the model automatically learns and extracts dynamic patterns in the sequence, i.e., data change characteristics. For sudden changes, the hidden states of the lightweight GRU model will change drastically accordingly.

[0130] The lightweight GRU model uses the last hidden state after processing the entire sequence, or performs average or max pooling on the hidden states across all time steps, as a summary representation of the entire time window. This final hidden state vector is the scene feature vector. It encapsulates the dynamic evolution information of the vehicle data within the entire short time window.

[0131] The feature vector of the scene is output and passed to downstream tasks, such as fusion models.

[0132] S203. The second model is an LSTM network model. Vehicle data within a collection time window with a duration greater than or equal to a preset threshold is input into the LSTM network model for feature extraction to obtain data persistence features. Among them, data persistence features are features that characterize continuous changes in vehicle data or features that remain unchanged in vehicle data. Based on the LSTM network model, the data persistence features are identified to obtain the scene feature vector of the collection time window.

[0133] For example, the Long Short-Term Memory (LSTM) network model is a type of recurrent neural network. Its sophisticated internal gating mechanism (input gate, forget gate, output gate) enables it to effectively learn long-term dependencies in long sequences of data, i.e., the impact of remembering or forgetting information from the distant past on the current state. This makes it well-suited for analyzing persistent patterns in vehicle data within a collection time window with a duration greater than or equal to a preset threshold.

[0134] For example, vehicle data (such as speed, acceleration, GPS coordinates, steering angle, etc.) collected within a time window with a duration greater than or equal to a preset threshold are cleaned, aligned, and standardized. The processed data is then constructed into a complete time series, typically with the shape [sequence length T, feature dimension N]. For instance, a 10-second window sampled at 100 milliseconds will yield a sequence of length 100 and dimension N.

[0135] The long sequence is then fed into a pre-trained LSTM model. The forget gate in the model determines which irrelevant past information (such as momentary noise jitter) needs to be discarded, while the input gate determines which new and important information (such as a slowly starting turning trend) should be stored in the cell state. Through this mechanism, the LSTM progressively updates its cell state across the entire sequence, thereby extracting persistent features of the data throughout the entire time window.

[0136] After processing the entire sequence, the final hidden state of the LSTM (or an aggregation of hidden states from all time steps, such as an attention-weighted average) contains summary information about the long-term pattern of the entire window. This state vector is the scene feature vector for that time window. It accurately encodes macroscopic driving patterns such as continuous acceleration, constant speed driving, or periodic start-stop.

[0137] The feature vector of the scene is output and passed to downstream tasks, such as fusion models.

[0138] S204. Based on the vehicle data within the initial time window, perform weighted summation on the scene feature vectors from each collection time window to obtain the vehicle's fused feature vector; wherein, the initial time window is a time window whose duration before the current moment is less than a preset threshold; based on the vehicle's usage scenario in the previous preset time period and the fused feature vector, determine the vehicle's usage scenario at the current moment.

[0139] For example, based on the vehicle data within the initial time window, the weight coefficients of each scene feature vector obtained in the above steps are determined. The scene feature vectors are then weighted and summed based on these weight coefficients to obtain the vehicle's fused feature vector. The initial time window is a time window where the duration preceding the current time is less than a preset threshold; for example, if the current time is 12:00:00, the initial time window is 11:59:55-12:00:00.

[0140] For example, the fused feature vector obtained in the previous step, along with the vehicle usage scenario from the previous preset time period, is input into a final decision model. The decision model comprehensively analyzes these two pieces of information: the fused vector provides the underlying, continuous feature patterns, while the historical scene sequence provides the top-level, discrete event context. Finally, the model outputs a classification label for the vehicle's usage scenario at the current moment.

[0141] Figure 3 This is a flowchart illustrating step S204 of an automatic vehicle usage scenario recognition method provided in an embodiment of this application. Figure 3 As shown, step S204 includes:

[0142] S2041. Determine the characteristic rate of change of the initial time window based on the vehicle data within the initial time window.

[0143] For example, key features for calculating the rate of change are selected from the raw vehicle data within the initial time window. These features are typically signals with clear physical meaning and sensitive to driving scenarios, such as:

[0144] Vehicle speed: Its rate of change directly reflects longitudinal driving behavior.

[0145] Longitudinal acceleration: Its rate of change (jerk) is a key indicator for measuring driving smoothness and urgency.

[0146] Lateral acceleration / steering angular velocity: its rate of change reflects the abruptness of the steering maneuver.

[0147] For each selected key feature time series, its rate of change over the entire time window is calculated using numerical differentiation. An optional method is first-order differencing.

[0148] For a feature sequence Its rate of change sequence is ,in, It is the time interval between adjacent data points.

[0149] To represent the degree of change of this feature across the entire window using a scalar, the average or root mean square value of the absolute values ​​of the aforementioned rate of change sequence is typically calculated. For example, the average rate of change = mean(abs(rate of change sequence)). This summation value is the rate of change of the feature within this window.

[0150] If the rate of change is calculated for multiple features (such as velocity and lateral acceleration), the rate of change of these individual features can be further combined by weighting or other methods to obtain a comprehensive feature rate of change that represents the overall dynamic level of the time window.

[0151] The calculated feature change rate will be output as a key input for subsequent steps (such as dynamic weight calculation). A window with a high feature change rate will be considered highly informative, thus receiving a higher weight in the weighted summation.

[0152] In one example, based on the specified vehicle data at adjacent times within the initial time window, the data change information of adjacent times within the initial time window is determined; based on the data change information of each adjacent time within the initial time window, the characteristic rate of change of the initial time window is determined.

[0153] For example, the system acquires a sequence of raw vehicle data, such as a speed sequence, within an initial time window. , where n is the total number of sampling points within the window, and the data points are strictly aligned according to the timestamp.

[0154] Iterate through each pair of adjacent data points in the sequence For each pair of data, calculate its first difference: .this It refers to the data change information between the i-th time and the (i+1)-th time.

[0155] To obtain the standardized rate of change, the difference is divided by the time interval between adjacent times. The instantaneous rate of change is obtained as follows: For example, if v is velocity, then It refers to instantaneous acceleration.

[0156] Thus, a sequence consisting of the instantaneous rates of change of all adjacent moments was determined. Then calculate the average absolute rate of change of the sequence, i.e., the characteristic rate of change. Taking the absolute value is to avoid positive and negative changes canceling each other out, thus truly reflecting the severity of the change. Another method is to calculate the root mean square value, which is more sensitive to large changes.

[0157] The calculated feature change rate is output as a key metric for measuring the dynamic characteristics of that time window. Subsequent steps (such as intelligent weighted fusion) will utilize this metric—windows with high feature change rates typically indicate more significant driving events, and therefore their corresponding scene feature vectors should be assigned higher weights.

[0158] In one example, if the specified vehicle data is vehicle speed, then the data change information between adjacent time points is the speed difference between adjacent time points, and the feature change rate of the initial time window is the standard deviation of the vehicle speed in the initial time window; if the specified vehicle data is steering angle, then the data change information between adjacent time points is the steering angle difference between adjacent time points, and the feature change rate of the initial time window is the variance of the steering angle in the initial time window.

[0159] For example, the system receives the raw vehicle data sequence for the initial time window. Based on a preset configuration, it determines whether the current calculation requires a feature change rate based on vehicle speed or steering angle. Understandably, the feature change rate can also be calculated based on other vehicle data, such as the standard deviation of engine / motor speed, altitude change rate, etc.

[0160] Iterate through each pair of adjacent time points (t, t+1) within the window, calculate the absolute value of their speed difference, and form a speed difference sequence; based on the speed difference sequence, determine the standard deviation of the vehicle speed in the current window.

[0161] Iterate through each pair of adjacent time points (t, t+1) within the window, calculate the absolute value of their steering angle difference, and form a steering angle difference sequence; based on the steering angle difference sequence, determine the steering angle variance of the current window.

[0162] The calculated characteristic change rate (vehicle speed standard deviation or steering angle variance) is output and used for subsequent weight coefficient adjustment decisions.

[0163] S2042. Determine the weight coefficients of the scene feature vectors for each acquisition time window based on the feature change rate of the initial time window.

[0164] For example, the weight coefficients of the scene feature vectors for each acquisition time window are determined based on the feature change rate of the initial time window.

[0165] In one example, if it is determined that the feature change rate of the initial time window is greater than or equal to the preset change rate, then the weight coefficient corresponding to the collection time window with a duration less than the preset threshold is determined, and the weight coefficient corresponding to the collection time window with a duration greater than or equal to the preset threshold is determined.

[0166] If the characteristic change rate of the initial time window is determined to be less than the preset change rate, then the weight coefficient corresponding to the collection time window with a duration less than the preset threshold is determined, and the weight coefficient corresponding to the collection time window with a duration less than or equal to the preset threshold is determined.

[0167] For example, if it is determined that the feature change rate of the initial time window is greater than or equal to a preset change rate, then the weight coefficient corresponding to the scene feature vector of the collection time window with a duration less than a preset threshold is determined as the first weight coefficient, and the weight coefficient corresponding to the scene feature vector of the collection time window with a duration greater than the preset threshold is determined as the second weight coefficient; wherein, the first weight coefficient is greater than or equal to the second weight coefficient; it can be understood that the first weight coefficient is 70% and the second weight coefficient is 30%.

[0168] If the feature change rate of the initial time window is determined to be less than the preset change rate, then the weight coefficient corresponding to the scene feature vector of the collection time window with a duration less than the preset threshold is determined to be the third weight coefficient, and the weight coefficient corresponding to the scene feature vector of the collection time window with a duration greater than the preset threshold is determined to be the fourth weight coefficient; where the third weight coefficient is less than the fourth weight coefficient; it can be understood that the third weight coefficient is 30% and the fourth weight coefficient is 70%.

[0169] S2043. Based on the weight coefficients of the scene feature vectors in each acquisition time window, perform weighted summation on the scene feature vectors in each acquisition time window to obtain the vehicle's fused feature vector.

[0170] For example, the system collects scene feature vectors from all acquisition time windows that need to be fused. First, it divides the acquisition time windows into two groups based on their duration: short-time window group (duration < preset threshold) and long-time window group (duration ≥ preset threshold).

[0171] The weight coefficients corresponding to the scene feature vectors of each acquisition time window are determined based on whether the feature change rate of the initial time window is greater than or equal to the preset change rate.

[0172] If the feature change rate is greater than or equal to the preset change rate, then all scene feature vectors in the short-term window group share the first weight coefficient (e.g., 70%), and all scene feature vectors in the long-term window group share the second weight coefficient (e.g., 30%).

[0173] If the feature change rate is less than the preset change rate, all scene feature vectors in the short-term window group share the first weight coefficient (e.g., 30%), and all scene feature vectors in the long-term window group share the second weight coefficient (e.g., 70%).

[0174] Since a group may contain multiple feature vectors, the global weights assigned to the group (e.g., 70%) need to be evenly distributed among each vector in the group.

[0175] For example, if there are 2 short-term windows and 3 long-term windows.

[0176] The actual weight of each scene feature vector within the short-time window group is 70% / 2 = 35%.

[0177] The actual weight of each scene feature vector within the long-term window group is 30% / 3 = 10%.

[0178] The weighted scene feature vectors are summed to obtain the final fused feature vector of the vehicle. This fused feature vector is then output to the scene classifier to determine the vehicle's usage scenario at the current moment.

[0179] S2044. Based on the fusion model, the fused feature vector is processed to obtain the maximum recognition probability of the vehicle at the current moment; according to the vehicle's usage scenario in the previous preset period, the maximum recognition probability is corrected to obtain the vehicle's usage scenario at the current moment.

[0180] For example, the fused feature vector is input into a pre-trained fusion model. The model performs forward propagation computation and outputs a probability distribution vector covering all possible scenarios. From the output probability distribution, the probability with the highest value and its corresponding vehicle usage scenario are determined.

[0181] The system retrieves the vehicle's usage scenarios from storage during the previous preset time period.

[0182] The system compares the current candidate scene with the historical scene and applies preset correction rules.

[0183] Based on the current candidate scenario, the vehicle usage scenario of the previous preset time period, and the preset correction rules, the highest recognition probability is corrected to obtain the vehicle usage scenario of the current preset time period.

[0184] Based on the decision of the correction rules, the final, smoothed vehicle usage scenario for the current preset time period is output.

[0185] In one example, based on the vehicle's usage scenario in the previous preset time period, the scene mapping relationship corresponding to the user scenario in the previous preset time period is retrieved; wherein, the scene mapping relationship includes correction coefficients for different usage scenarios; based on the usage scenario corresponding to the highest recognition probability, the correction coefficient corresponding to the usage scenario corresponding to the highest recognition probability is determined from the scene mapping relationship; based on the obtained correction coefficient, the highest recognition probability is corrected to obtain the corrected recognition probability; and based on the corrected recognition probability, the vehicle's usage scenario at the current moment is determined.

[0186] For example, a scene mapping relationship is a predefined set of rules, similar to a lookup table or matrix. It takes historical scenes as input and outputs a list containing correction coefficients. This relationship defines the ease or tendency of switching from a specific historical scene to any other scene (or maintaining the current scene). For instance, switching from highway cruising to city traffic congestion might be difficult (low correction coefficient), while switching to highway driving would be relatively easy (high correction coefficient).

[0187] The correction coefficient is a weighted value (usually a real number greater than 0) used to adjust the current recognition probability. A coefficient of 1 indicates that the historical scene has no bias towards the current candidate scene, and no correction is made. A coefficient greater than 1 encourages switching from the historical scene to the current candidate scene, amplifying its probability. This is typically used for logically connected scenarios (such as likely entering a city for normal driving after parking space search). A coefficient less than 1 discourages or penalizes switching from the historical scene to the current candidate scene, suppressing its probability. This is typically used for scenario jumps that are unlikely to occur in a short time (such as jumping directly from highway cruising to a stationary parking lot).

[0188] For example, the system retrieves the corresponding scenario mapping relationship from a predefined configuration library based on the vehicle's usage scenario in the previous preset time period (denoted as scenario A). This relationship includes correction coefficients from scenario A to all other scenarios (including itself).

[0189] The system finds the vehicle usage scenario (denoted as scenario B) corresponding to the highest recognition probability from the fused feature matrix. Then, it finds the specific correction coefficient from the historical scenario A to the current candidate scenario B from the scenario mapping relationship obtained in step 1.

[0190] The system uses this correction coefficient to weight and adjust the maximum recognition probability. A direct correction formula is: Corrected recognition probability = Original maximum recognition probability × Correction coefficient. Subsequently, all probabilities in the entire fused feature matrix need to be renormalized to ensure that the sum of probabilities for all scenarios is still 1.

[0191] The system compares all probability values ​​in the corrected fused feature matrix. The scenario with the highest probability value is then identified as the vehicle's usage scenario at the current moment. Because the correction coefficients alter the probability distribution, the final result may no longer be the original candidate scenario B.

[0192] In one example, if the corrected recognition probability is greater than or equal to the preset probability, the vehicle's usage scenario at the current moment is determined to be the usage scenario corresponding to the highest recognition probability; if the corrected recognition probability is less than the preset probability, the vehicle's usage scenario at the current moment is determined to be a transitional state scenario.

[0193] For example, the system compares the corrected recognition probability with the preset probability. This constitutes a simple two-branch decision logic.

[0194] If the corrected recognition probability is greater than or equal to the preset probability, the system considers there sufficient evidence to support the candidate scenario. In this case, the system directly confirms that the vehicle's usage scenario at the current moment is the usage scenario corresponding to the highest recognition probability.

[0195] If the corrected recognition probability is less than the preset probability, the system considers the current evidence insufficient and the confidence level too low. Forcibly designating a scenario is too risky. Therefore, the system determines the vehicle's current usage scenario as a transitional state scenario.

[0196] This application provides an automatic vehicle usage scenario identification method. It acquires vehicle data from multiple collection time windows prior to the current moment, where adjacent windows overlap and at least two windows have different durations. A lightweight GRU network model is used as the first model. Vehicle data from collection time windows with durations less than a preset threshold are input, and data change features representing sudden changes in vehicle data are extracted before scene feature vectors are identified. An LSTM network model is used as the second model. Vehicle data from collection time windows with durations greater than or equal to a preset threshold are input, and data continuity features representing continuous changes or invariance in vehicle data are extracted before scene feature vectors are identified. Next, based on vehicle data from the initial time window, the scene feature vectors from each collection time window are weighted and summed to obtain a fused feature vector. Finally, the current usage scenario is determined based on the vehicle's usage scenario in the previous preset time period and the fused feature vector. By combining different time windows with different models, the sudden and continuous characteristics of vehicle data can be accurately captured respectively. The weighted summation further integrates the information, and combined with the previous preset time period scenario, the final determined current vehicle usage scenario is more in line with reality. This effectively solves the problem that traditional methods are difficult to fully and accurately identify vehicle usage scenarios, and achieves the effect of accurately identifying vehicle usage scenarios.

[0197] Figure 4A schematic diagram of the structure of an automatic vehicle usage scenario recognition device provided in this application embodiment. Figure 1 ,like Figure 4 As shown, the automatic vehicle usage scenario recognition device 40 provided in this embodiment includes:

[0198] The acquisition module 401 is used to acquire vehicle data within each of multiple acquisition time windows located before the current time; wherein adjacent acquisition time windows overlap in the time dimension, and at least two acquisition time windows have different durations.

[0199] The identification module 402 is used to input vehicle data within a collection time window with a duration less than a preset threshold into the first model for identification to obtain the scene feature vector of the collection time window; and to input vehicle data within a collection time window with a duration greater than or equal to the preset threshold into the second model for identification to obtain the scene feature vector of the collection time window; wherein, the scene feature vector represents the characteristics of the vehicle data in the collection time window.

[0200] The determination module 403 is used to determine the vehicle usage scenario at the current moment based on the scene feature vectors of each time window.

[0201] This embodiment provides an automatic vehicle usage scenario recognition device that can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0202] Figure 5 A schematic diagram of the structure of an automatic vehicle usage scenario recognition device provided in this application embodiment. Figure 2 ,like Figure 5 As shown, the automatic vehicle usage scenario recognition device 50 provided in this embodiment includes:

[0203] The acquisition module 501 is used to acquire vehicle data within each of multiple acquisition time windows located before the current time; wherein adjacent acquisition time windows overlap in the time dimension, and at least two acquisition time windows have different durations.

[0204] The identification module 502 is used to input vehicle data within a collection time window with a duration less than a preset threshold into the first model for identification to obtain the scene feature vector of the collection time window; and to input vehicle data within a collection time window with a duration greater than or equal to the preset threshold into the second model for identification to obtain the scene feature vector of the collection time window; wherein, the scene feature vector represents the characteristics of the vehicle data in the collection time window.

[0205] The determination module 503 is used to determine the vehicle usage scenario at the current moment based on the scene feature vectors of each time window.

[0206] In one possible implementation, the first model is a lightweight GRU network model; the identification module 502 includes:

[0207] Vehicle data collected within a time window shorter than a preset threshold is input into a lightweight GRU network model for feature extraction to obtain data change features; among them, data change features are features that characterize sudden changes in vehicle data.

[0208] Based on the lightweight GRU network model, the data change characteristics are identified to obtain the scene feature vector for that time window.

[0209] In one possible implementation, the second model is an LSTM network model; the identification module 502 includes:

[0210] Vehicle data collected within a time window with a duration greater than or equal to a preset threshold is input into an LSTM network model for feature extraction to obtain persistent data features. Among these features, persistent data features are those that characterize the continuous changes in vehicle data or those that characterize the continuous invariance of vehicle data.

[0211] Based on the LSTM network model, continuous features of the data are identified to obtain the scene feature vector for the acquisition time window.

[0212] In one possible implementation, the determining module 503 includes:

[0213] The processing module 5031 is used to perform weighted summation processing on the scene feature vectors of each collection time window based on the vehicle data within the initial time window to obtain the vehicle's fused feature vector; wherein, the initial time window is a time window whose duration before the current moment is less than a preset threshold.

[0214] The determination submodule 5032 is used to determine the vehicle's usage scenario at the current moment based on the vehicle's usage scenario in the previous preset time period and the fused feature vector.

[0215] In one possible implementation, the processing module 5031 includes:

[0216] Based on the vehicle data within the initial time window, determine the characteristic rate of change of the initial time window;

[0217] Based on the characteristic change rate of the initial time window, determine the weight coefficients of the scene feature vectors for each acquisition time window;

[0218] Based on the weight coefficients of the scene feature vectors in each acquisition time window, the scene feature vectors in each acquisition time window are weighted and summed to obtain the vehicle's fused feature vector.

[0219] In one possible implementation, determining the characteristic rate of change of the initial time window based on vehicle data within the initial time window includes:

[0220] Based on the specified vehicle data at adjacent moments in the initial time window, determine the data change information at adjacent moments in the initial time window;

[0221] Based on the data change information of each adjacent time point in the initial time window, determine the characteristic rate of change of the initial time window.

[0222] In one possible implementation, if the specified vehicle data is vehicle speed, then the data change information between adjacent time moments is the speed difference between adjacent time moments, and the characteristic change rate of the initial time window is the standard deviation of the vehicle speed of the initial time window.

[0223] If the specified vehicle data is the steering angle, then the data change information between adjacent time points is the steering angle difference between adjacent time points, and the characteristic change rate of the initial time window is the steering angle variance of the initial time window.

[0224] In one possible implementation, the weight coefficients of the scene feature vectors for each acquisition time window are determined based on the feature change rate of the initial time window, including:

[0225] If the characteristic change rate of the initial time window is determined to be greater than or equal to the preset change rate, then the weight coefficient corresponding to the collection time window with a duration less than the preset threshold is determined, and the weight coefficient corresponding to the collection time window with a duration greater than or equal to the preset threshold is determined.

[0226] If the characteristic change rate of the initial time window is determined to be less than the preset change rate, then the weight coefficient corresponding to the collection time window with a duration less than the preset threshold is determined, and the weight coefficient corresponding to the collection time window with a duration less than or equal to the preset threshold is determined.

[0227] In one possible implementation, determining submodule 5032 includes:

[0228] The fusion feature vector is processed based on the fusion model to obtain the maximum recognition probability of the vehicle at the current moment;

[0229] Based on the vehicle's usage scenario in the previous preset time period, the highest recognition probability is corrected to obtain the vehicle's usage scenario at the current moment.

[0230] In one possible implementation, the maximum recognition probability is corrected based on the vehicle's usage scenario in a previous preset time period to obtain the vehicle's usage scenario at the current moment, including:

[0231] Based on the vehicle's usage scenario in the previous preset time period, retrieve the scenario mapping relationship corresponding to the user scenario in the previous preset time period; the scenario mapping relationship includes correction coefficients for different usage scenarios.

[0232] Based on the vehicle usage scenario corresponding to the highest recognition probability, the correction coefficient corresponding to the vehicle usage scenario corresponding to the highest recognition probability is determined from the scenario mapping relationship;

[0233] Based on the obtained correction coefficients, the maximum recognition probability is corrected to obtain the corrected recognition probability; and based on the corrected recognition probability, the vehicle's usage scenario at the current moment is determined.

[0234] In one possible implementation, the vehicle's usage scenario at the current moment is determined based on the corrected recognition probability, including:

[0235] If the corrected recognition probability is determined to be greater than or equal to the preset probability, then the vehicle's usage scenario at the current moment is determined to be the usage scenario corresponding to the highest recognition probability.

[0236] If the corrected recognition probability is determined to be less than the preset probability, the vehicle's current usage scenario is determined to be a transitional scenario.

[0237] In one possible implementation, the device 50 further includes:

[0238] Vehicle data is preprocessed, including one or more of the following: outlier removal, missing value completion, and feature normalization.

[0239] This embodiment provides an automatic vehicle usage scenario recognition device that can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0240] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 6 As shown, the electronic device 60 provided in this embodiment includes at least one processor 601 and a memory 602. Optionally, the device 60 further includes a communication component 603. The processor 601, memory 602, and communication component 603 are connected via a bus 604.

[0241] In a specific implementation, at least one processor 601 executes computer execution instructions stored in memory 602, causing at least one processor 601 to perform the above-described method.

[0242] The specific implementation process of processor 601 can be found in the above method embodiments, and its implementation principle and technical effect are similar. It will not be repeated here.

[0243] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0244] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0245] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0246] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0247] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0248] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0249] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0250] The division of units is merely a logical functional division; in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0251] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0252] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0253] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0254] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0255] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. An automatic identification method for vehicle usage scenarios, characterized in that, include: Acquire vehicle data within each of multiple acquisition time windows preceding the current time; wherein adjacent acquisition time windows overlap in the time dimension, and at least two acquisition time windows have different durations; Vehicle data within a collection time window with a duration less than a preset threshold is input into the first model for identification to obtain the scene feature vector of that collection time window; vehicle data within a collection time window with a duration greater than or equal to the preset threshold is input into the second model for identification to obtain the scene feature vector of that collection time window; wherein, the scene feature vector characterizes the features of the vehicle data within the collection time window. Based on the scene feature vectors of each time window, the vehicle's usage scenario at the current moment is determined.

2. The method according to claim 1, characterized in that, The first model is a lightweight GRU network model; vehicle data within a collection time window with a duration less than a preset threshold is input into the first model for identification to obtain the scene feature vector for that collection time window, including: Vehicle data collected within a time window shorter than a preset threshold is input into the lightweight GRU network model for feature extraction to obtain data change features; wherein, the data change features are features that characterize sudden changes in vehicle data. Based on the lightweight GRU network model, the data change characteristics are identified to obtain the scene feature vector for that time window.

3. The method according to claim 1, characterized in that, The second model is an LSTM network model; vehicle data within a collection time window with a duration greater than or equal to a preset threshold are input into the second model for recognition, obtaining the scene feature vector for that collection time window, including: Vehicle data collected within a time window with a duration greater than or equal to a preset threshold is input into the LSTM network model for feature extraction to obtain persistent data features; wherein, the persistent data features are features that characterize the continuous changes in vehicle data, or features that characterize the continuous invariance of vehicle data. Based on the LSTM network model, the continuous features of the data are identified to obtain the scene feature vector for the acquisition time window.

4. The method according to claim 1, characterized in that, Based on the scene feature vectors of each of the aforementioned collection time windows, determine the vehicle's usage scenario at the current moment, including: Based on the vehicle data within the initial time window, the scene feature vectors of each collection time window are weighted and summed to obtain the fused feature vector of the vehicle; wherein, the initial time window is a time window whose duration before the current time is less than a preset threshold. The vehicle's usage scenario at the current moment is determined based on the vehicle's usage scenario in the previous preset time period and the fused feature vector.

5. The method according to claim 4, characterized in that, Based on the vehicle data within the initial time window, the scene feature vectors from each of the acquisition time windows are weighted and summed to obtain the fused feature vector of the vehicle, including: Based on the vehicle data within the initial time window, determine the characteristic rate of change of the initial time window; Based on the characteristic change rate of the initial time window, determine the weight coefficient of the scene feature vector for each acquisition time window; Based on the weight coefficients of the scene feature vectors of each acquisition time window, the scene feature vectors of each acquisition time window are weighted and summed to obtain the fused feature vector of the vehicle.

6. The method according to claim 5, characterized in that, Based on the vehicle data within the initial time window, determine the characteristic rate of change of the initial time window, including: Based on the specified vehicle data at adjacent moments in the initial time window, determine the data change information at adjacent moments in the initial time window; The characteristic rate of change of the initial time window is determined based on the data change information of each adjacent time point in the initial time window.

7. The method according to claim 6, characterized in that, If the specified vehicle data is vehicle speed, then the data change information at adjacent time points is the speed difference at adjacent time points, and the characteristic change rate of the initial time window is the standard deviation of the vehicle speed at the initial time window; If the specified vehicle data is the steering angle, then the data change information at adjacent time points is the steering angle difference at adjacent time points, and the characteristic change rate of the initial time window is the steering angle variance of the initial time window.

8. The method according to claim 5, characterized in that, Based on the characteristic change rate of the initial time window, the weight coefficients of the scene feature vectors for each acquisition time window are determined, including: If it is determined that the feature change rate of the initial time window is greater than or equal to the preset change rate, then the weight coefficient corresponding to the collection time window with a duration less than the preset threshold is determined, and the weight coefficient corresponding to the collection time window with a duration greater than or equal to the preset threshold is determined. If it is determined that the feature change rate of the initial time window is less than the preset change rate, then the weight coefficient corresponding to the collection time window with a duration less than the preset threshold is determined, and the weight coefficient corresponding to the collection time window with a duration less than or equal to the preset threshold is determined.

9. The method according to claim 4, characterized in that, Based on the vehicle's usage scenario in the previous preset time period and the fused feature vector, the vehicle's usage scenario at the current moment is determined, including: The fused feature vector is processed based on the fusion model to obtain the maximum recognition probability of the vehicle at the current moment; Based on the vehicle's usage scenario in the previous preset time period, the maximum recognition probability is corrected to obtain the vehicle's usage scenario at the current moment.

10. The method according to claim 9, characterized in that, Based on the vehicle's usage scenario in the previous preset time period, the maximum recognition probability is corrected to obtain the vehicle's usage scenario at the current moment, including: Based on the vehicle's usage scenario in the previous preset period, retrieve the scenario mapping relationship corresponding to the user scenario of the vehicle in the previous preset period; wherein, the scenario mapping relationship includes correction coefficients for different usage scenarios; Based on the vehicle usage scenario corresponding to the highest recognition probability, a correction coefficient corresponding to the vehicle usage scenario corresponding to the highest recognition probability is determined from the scenario mapping relationship; Based on the obtained correction coefficient, the maximum recognition probability is corrected to obtain the corrected recognition probability; and based on the corrected recognition probability, the vehicle's usage scenario at the current moment is determined.

11. The method according to claim 10, characterized in that, Based on the corrected recognition probability, the vehicle's usage scenario at the current moment is determined, including: If it is determined that the corrected recognition probability is greater than or equal to the preset probability, then the vehicle usage scenario at the current moment is determined to be the usage scenario corresponding to the maximum recognition probability. If it is determined that the corrected recognition probability is less than the preset probability, the vehicle's usage scenario at the current moment is determined to be a transitional scenario.

12. The method according to any one of claims 1-11, characterized in that, The method further includes: The vehicle data is preprocessed, wherein the preprocessing includes one or more of the following: outlier removal, missing value completion, and feature normalization.

13. An automatic vehicle usage scenario recognition device, characterized in that, include: The acquisition module is used to acquire vehicle data within each of multiple acquisition time windows preceding the current time; wherein adjacent acquisition time windows overlap in the time dimension, and at least two acquisition time windows have different durations; The identification module is used to input vehicle data within a collection time window with a duration less than a preset threshold into a first model for identification to obtain a scene feature vector for that collection time window; and to input vehicle data within a collection time window with a duration greater than or equal to the preset threshold into a second model for identification to obtain a scene feature vector for that collection time window; wherein, the scene feature vector characterizes the features of the vehicle data within the collection time window. The determination module is used to determine the vehicle's usage scenario at the current moment based on the scene feature vectors of each time window.

14. An electronic device, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-12.

15. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-12.

16. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method described in any one of claims 1-12.

Images