Method and device for predicting the range of a vehicle

Abstract

The application discloses a kind of vehicle endurance mileage prediction method and prediction device.The method comprises: obtaining multiple groups of time series data of vehicle in running process;Respectively, each group of time series data is input into corresponding convolutional neural network to extract features, obtain the time series features corresponding to each group of time series data;Coupling is carried out to multiple groups of time series features by cross-class attention mechanism, and multiple group features for representing coupling relationship between different physical dimensions are obtained;Feature fusion is carried out to multiple group features, and a feature vector is obtained;The endurance mileage of vehicle is predicted based on the feature vector, and a predicted endurance mileage is obtained.Thereby, the architecture of intra-group feature extraction, inter-group interaction modeling and hierarchical fusion prediction can effectively improve the accuracy of vehicle endurance mileage prediction.

Description

Technical Field

[0001] This application belongs to the field of vehicle control technology, and in particular relates to a method and device for predicting vehicle driving range. Background Technology

[0002] With the increasing popularity of new energy vehicles, the accuracy of vehicle range has become one of the core indicators that users care about most.

[0003] In related technologies, all data from the vehicle during its driving process is typically input into a prediction model, which then outputs a prediction of the vehicle's driving range.

[0004] However, existing prediction models ignore the coupling relationships between data from different physical dimensions, which makes it impossible for them to accurately predict the vehicle's driving range. Summary of the Invention

[0005] This application provides a method and apparatus for predicting vehicle driving range, which can effectively improve the accuracy of vehicle driving range prediction.

[0006] In a first aspect, embodiments of this application provide a method for predicting vehicle range. The method includes: acquiring multiple sets of time-series data of the vehicle during operation; inputting each set of time-series data into a corresponding convolutional neural network for feature extraction to obtain time-series features corresponding to each set of time-series data; coupling multiple sets of time-series features through a cross-class attention mechanism to obtain multiple inter-group features used to characterize the coupling relationship between different physical dimensions; performing feature fusion on multiple inter-group features to obtain a feature vector; and predicting the vehicle range based on the feature vector to obtain the predicted range.

[0007] Secondly, embodiments of this application provide a vehicle range prediction device, which includes: a data acquisition module for acquiring multiple sets of time-series data of the vehicle during operation; a feature extraction module for inputting each set of time-series data into a corresponding convolutional neural network for feature extraction to obtain time-series features corresponding to each set of time-series data; a feature coupling module for coupling multiple sets of time-series features through a cross-class attention mechanism to obtain multiple inter-group features for characterizing the coupling relationship between different physical dimensions; a feature fusion module for fusing multiple inter-group features to obtain a feature vector; and a range prediction module for predicting the vehicle's range based on the feature vector to obtain the predicted range.

[0008] Thirdly, embodiments of this application provide an electronic device, which includes: a processor and a memory storing computer program instructions; the processor executes the computer program instructions to implement the vehicle range prediction method as described in the first aspect.

[0009] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer program instructions that, when executed by a processor, implement the vehicle range prediction method as described in the first aspect.

[0010] Fifthly, embodiments of this application provide a computer program product in which instructions, when executed by a processor of an electronic device, cause the electronic device to perform the vehicle range prediction method as described in the first aspect.

[0011] As described above, in this embodiment, different groups of time-series data correspond to different physical dimensions. By employing independent convolutional neural networks to extract features from different groups of time-series data, the extracted time-series features can not only retain the temporal patterns within each physical dimension but also reduce mutual interference between features of different physical dimensions. By using a cross-class attention mechanism to couple features from multiple groups of time-series features, inter-group features are generated to capture the collaborative or restrictive relationships between different physical dimensions. Then, feature fusion is used to integrate multiple inter-group features into a feature vector, and the driving range is predicted based on this feature vector, which can improve the accuracy of vehicle driving range prediction.

[0012] Therefore, it can be seen that the above-mentioned architecture of intra-group feature extraction, inter-group interaction modeling, and hierarchical fusion prediction can effectively improve the accuracy of vehicle range prediction. Attached Figure Description

[0013] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 This is an architecture diagram of a vehicle range prediction method provided in one embodiment of this application; Figure 2 This is one of the flowcharts illustrating a method for predicting vehicle driving range provided in an embodiment of this application; Figure 3 This is a schematic diagram of the structure of a convolutional neural network provided in one embodiment of this application; Figure 4 This is a schematic diagram of matrix operations of a convolutional neural network provided in one embodiment of this application; Figure 5 This is a framework diagram of an attention mechanism provided in one embodiment of this application; Figure 6 This is a schematic diagram of matrix operations for cross-attention provided in one embodiment of this application; Figure 7 This is a second schematic flowchart of a method for predicting vehicle driving range provided in one embodiment of this application; Figure 8 This is a schematic diagram of the structure of a vehicle range prediction device provided in another embodiment of this application; Figure 9 This is a schematic diagram of the structure of an electronic device provided in another embodiment of this application. Detailed Implementation

[0015] The features and exemplary embodiments of various aspects of this application will be described in detail below. To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only intended to explain this application and not to limit it. For those skilled in the art, this application can be implemented without some of these specific details. The following description of the embodiments is merely to provide a better understanding of this application by illustrating examples.

[0016] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus that includes said element.

[0017] To facilitate understanding, before explaining the solution provided in this application, the background of the solution provided in this application will be explained first.

[0018] With the increasing popularity of new energy vehicles, the accuracy of vehicle range prediction has become one of the core indicators that users care about most. The ability to accurately predict the range is related to users' travel planning and driving experience. If the range prediction is inaccurate, it will cause range anxiety for users, which in turn will affect the driving experience of new energy vehicles.

[0019] In related technologies, the ampere-hour integral method, the operating condition lookup table method, and machine learning methods are commonly used. In the ampere-hour integral method, the battery management system (BMS) monitors the remaining battery charge in real time and combines this with the vehicle's average energy consumption over historical periods to calculate the ratio of remaining charge to average energy consumption. This ratio is then used to predict the vehicle's driving range. In the operating condition lookup table method, the BMS pre-stores energy consumption data for multiple standard test conditions, such as WLTP (Worldwide Harmonized Light Vehicles Test Procedure) and CLTC (China Light-duty Vehicle Test Cycle). When the vehicle is in motion, the BMS matches the closest operating condition to the vehicle's current operating state and determines the driving range corresponding to that condition through a lookup table. In machine learning, data such as vehicle speed, acceleration, SOC (State of Charge), and temperature are collected to train neural networks or models such as XGBoost. This allows the model to learn the mapping relationship between features and driving range from the data, and the battery management system uses the model to predict the vehicle's driving range.

[0020] Compared to the ampere-hour integration method and the operating condition lookup table method, the machine learning method can automatically extract complex nonlinear features related to driving range from multi-dimensional historical data such as vehicle speed, acceleration, and temperature, without relying on fixed operating conditions or average power consumption for simplified calculations, thus improving the accuracy of driving range prediction.

[0021] Currently, there is considerable research on machine learning-based driving range prediction technologies, such as prediction methods based on LSTM (Long Short-Term Memory), CNN (Convolutional Neural Network)-LSTM, and Transformer.

[0022] In existing technologies, predicting vehicle range typically involves inputting all vehicle data—such as battery-related, driving-related, and environmental data—into a model. The model then predicts the range based on this input. However, by inputting all data, the internal physical properties of the battery (e.g., the relationship between SOC and voltage, and the impact of temperature on discharge efficiency) are buried under a large number of irrelevant features. The model struggles to learn the intra-group relationships between data with the same physical dimension, resulting in a lack of in-depth understanding of the battery's internal state and reduced accuracy in predicting vehicle range. Furthermore, in practical applications, there are physical coupling relationships between data from different physical dimensions, such as the impact of driving behavior on battery consumption, the impact of environmental factors on battery performance, and the synergistic effect of environment and driving. Mixing data from different physical dimensions erases the physical structure between these data. The model can only learn statistical correlations from the chaotic data, rather than physical laws. This increases the learning difficulty, requires a large amount of data, and may even lead to spurious correlations, reducing the model's generalization ability and impacting the accuracy of vehicle range prediction.

[0023] To address the problems existing in the prior art, this application provides a method and apparatus for predicting vehicle driving range. The prediction method employs intra-group feature learning, inter-group interaction modeling, and hierarchical fusion prediction to predict vehicle driving range.

[0024] For example, Figure 1 An architecture diagram of the method provided in the embodiments of this application is shown, such as... Figure 1As shown, the method architecture consists of four parts: a data acquisition part, a grouped convolution part, a cross-class attention part, and a feature fusion and prediction part. In the data acquisition part, data from different physical dimensions are collected, with data from the same physical dimension grouped together. In this embodiment, physical dimensions can include battery, driving, and environmental dimensions. For example, battery-related data (e.g., battery SOC, voltage, current) can be collected through a BMS (Battery Management System) to obtain battery-dimensional data; driving-related data (e.g., vehicle speed, acceleration) can be obtained through the vehicle's VCU (Vehicle Control Unit); and environmental-related data (e.g., temperature, humidity) can be obtained through an ECU (Electronic Control Unit). In the grouped convolution part, grouped convolution is used to process the data for each physical dimension separately, such as batch normalization and max pooling, to extract the correlation between data within the same physical dimension. In the cross-class attention section, data from different physical dimensions are cross-coupled to obtain multiple feature sets. An attention mechanism is then used to perform average pooling (e.g., temporal average pooling) on ​​the features in each feature set, thereby learning the coupling relationships between data from different physical dimensions. In the feature fusion and prediction section, the features output from the cross-class attention section are fused to obtain a feature vector. This feature vector is then input into a fully connected layer to predict the driving range, thus obtaining the predicted driving range result.

[0025] In this embodiment, data is grouped by physical dimension and input into corresponding convolutional neural networks for feature extraction. This allows the system to learn the features between data with the same physical dimension, reducing the risk of information overload. Cross-category coupling of temporal features from different physical dimensions captures the collaborative or restrictive relationships between them, enabling the system to learn the impact of the coupling between data from different physical dimensions on driving range and improving the accuracy of driving range prediction.

[0026] The method for predicting vehicle driving range provided in this application embodiment will be described below. In this method, a server can act as the execution entity. In this scenario, the server can remotely communicate with the vehicle, predict the vehicle's driving range based on vehicle data, and send the predicted driving range result to the vehicle's controller. The vehicle controller can then control the vehicle based on the predicted driving range result. Alternatively, the vehicle controller can also act as the execution entity. In this scenario, the vehicle controller can transmit data with various control units in the vehicle (e.g., battery management system, ECU, etc.), predict the vehicle's driving range based on the data collected by each control unit, and control the vehicle's operation based on the prediction result. The method described below uses the vehicle controller as the execution entity; the prediction method is similar in scenarios where the server acts as the execution entity.

[0027] Figure 2 A flowchart illustrating a method for predicting vehicle driving range according to an embodiment of this application is shown. Figure 2 As shown, the method includes the following steps S201 to S205: Step S201: Obtain multiple sets of time-series data of the vehicle during operation.

[0028] In step S201, the vehicle can be a new energy vehicle, such as an electric vehicle. In this embodiment, during vehicle operation, data from different control units or sensors is collected in real time via the vehicle bus and organized into multiple sets of time-series data in chronological order. These multiple sets of time-series data correspond to different physical dimensions; that is, data with the same physical dimension form a set of time-series data. The vehicle's physical dimensions may include, but are not limited to, battery dimensions, driving dimensions, and environmental dimensions. In practical applications, the vehicle's physical dimensions can be added or removed as needed.

[0029] For example, regarding the battery dimension, the corresponding time-series data is used to characterize the internal state of the battery, reflecting the vehicle's currently available energy reserves and energy release capabilities. Battery-level time-series data may include, but is not limited to, battery SOC, SOH (State of Health), battery voltage, battery temperature, and battery type code. Regarding the driving dimension, the corresponding time-series data is used to characterize the vehicle's driving behavior, and may include, but is not limited to, vehicle speed, acceleration, and instantaneous power consumption. Regarding the environmental dimension, the corresponding time-series data is used to characterize the impact of external operating conditions on the vehicle's range, and may include, but is not limited to, ambient temperature, tire pressure, altitude, gradient, air conditioning power, and heating power.

[0030] In step S201, data from different physical dimensions can be collected using different sampling frequencies to adapt to the changing characteristics of the corresponding physical dimensions. For example, data from the driving dimension can be collected at a higher sampling frequency (e.g., 10Hz); data from the battery dimension can be collected at a medium sampling frequency (e.g., 1Hz); and data from the environment dimension can be collected at a lower sampling frequency (e.g., 0.1Hz).

[0031] Step S202: Input each set of time series data into the corresponding convolutional neural network for feature extraction to obtain the time series features corresponding to each set of time series data.

[0032] In step S202, time-series data of different physical dimensions are input into corresponding convolutional neural networks to extract features of the time-series data of the corresponding physical dimension. That is, in this embodiment, time-series data of different physical dimensions have dedicated convolutional neural networks. These networks extract time-series features of the corresponding physical dimension by performing convolution, pooling, and other operations on the data. These features reflect the internal evolution of the data within a given time window, reducing mixing and interference with features from other dimensions, thus laying the foundation for accurate prediction of vehicle range.

[0033] Step S203: Couple multiple sets of temporal features through a cross-class attention mechanism to obtain multiple inter-group features used to characterize the coupling relationship between different physical dimensions.

[0034] It's important to note that in practical applications, there's a hierarchical relationship between data from different physical dimensions. Battery state is the energy basis, driving behavior is the manifestation of energy consumption, and environmental factors are external disturbances. Therefore, a vehicle's driving range is not independently influenced by data from a single physical dimension, but rather driven by the interaction of data from multiple physical dimensions.

[0035] In step S203, the vehicle controller calculates the attention weights between the temporal features of different physical dimensions to determine the correlation strength between each physical dimension under the current operating conditions, so as to generate inter-group features that characterize the coupling relationship between physical dimensions. This can improve the accuracy of vehicle range prediction and make the vehicle range prediction results have clear physical interpretability.

[0036] Step S204: Perform feature fusion on multiple group features to obtain feature vectors.

[0037] It should be noted that the inter-group features obtained by coupling temporal features of different physical dimensions describe the coupling relationships under different combinations of physical dimensions, which affect the prediction of vehicle range from different perspectives. Therefore, in step S204, feature fusion is required to obtain a unified feature vector for predicting vehicle range.

[0038] In step S204, the vehicle controller can perform feature fusion on multiple inter-group features using one or more methods such as concatenation, weighted summation, and attention. For example, the vehicle controller can concatenate multiple inter-group features end-to-end to form a high-dimensional feature vector; or, for another example, the vehicle controller can learn a trainable weight for each inter-group feature and then perform weighted summation to obtain a feature vector; or, for yet another example, the vehicle controller can input multiple inter-group features into an attention layer to automatically focus on one or more important coupling relationships under the current operating condition, thereby obtaining a feature vector.

[0039] Step S205: Predict the vehicle's driving range based on the feature vector to obtain the predicted driving range.

[0040] In step S205, after obtaining the feature vector, the vehicle controller can input the feature vector into the prediction network. The fully connected layer of the prediction network performs dimensionality reduction on the feature vector through nonlinear transformation, thereby obtaining the prediction result of the vehicle's driving range.

[0041] Based on the scheme defined in steps S201 to S205 above, it can be understood that in this embodiment, different groups of time-series data correspond to different physical dimensions. By using independent convolutional neural networks to extract features from different groups of time-series data, the extracted time-series features can not only retain the time-series patterns within each physical dimension, but also reduce the mutual interference between features of different physical dimensions. By coupling features of multiple groups of time-series features through a cross-class attention mechanism to generate inter-group features, the collaborative or restrictive relationships between different physical dimensions can be captured. Then, by feature fusion, multiple inter-group features are integrated into a feature vector, and the driving range is predicted based on this feature vector, which can improve the accuracy of vehicle driving range prediction.

[0042] Therefore, it can be seen that the above-mentioned architecture of intra-group feature extraction, inter-group interaction modeling, and hierarchical fusion prediction can effectively improve the accuracy of vehicle range prediction.

[0043] The specific implementation process of the method provided in the embodiments of this application is described below.

[0044] In some embodiments, before predicting the vehicle's driving range, the vehicle controller first needs to acquire multiple sets of time-series data about the vehicle. Specifically, the vehicle controller first acquires the vehicle's driving scenario data, and then, based on the correlation between the driving scenario and dimensional data, determines the target physical dimension required to predict the vehicle's driving range under the driving scenario data, as well as the data indicators corresponding to the target physical dimension; then, it collects the time-series data corresponding to the data indicators to obtain multi-dimensional time-series data.

[0045] In the above embodiments, to improve the accuracy of vehicle range prediction, different physical dimensions and / or time-series data under the physical dimensions are configured for the vehicle in different driving scenarios. For example, tire pressure can be used as time-series data in the driving dimension to reflect its direct correlation with driving behavior, or as time-series data in the environmental dimension to reflect its physical characteristics affected by ambient temperature; similarly, air conditioning power can be used as time-series data in the environmental dimension, or as accessory energy consumption actively controlled by the driver and included in the driving dimension's time-series data.

[0046] In practical applications, the time-series data in each physical dimension can also be adjusted according to actual needs. For example, the battery internal resistance and the number of battery charge-discharge cycles can be extended to the battery dimension; the accelerator pedal opening and brake pedal opening can be extended to the driving dimension; and the weather and air humidity can be extended to the environment dimension.

[0047] For example, the vehicle controller can acquire current driving scenario data through onboard positioning systems, navigation maps, environmental perception sensors (e.g., cameras, LiDAR), and vehicle-to-everything (V2X) communication modules. This driving scenario data can characterize the vehicle's operating environment and driving task type. For instance, driving scenarios can include, but are not limited to, road type scenarios (e.g., highways, urban expressways, urban roads, mountain roads, rural roads), traffic status scenarios (e.g., smooth traffic, slow traffic, congestion, high traffic light density), environmental climate scenarios (e.g., sunny days, rain / snow, high temperatures, low temperatures, strong winds), and driving task scenarios (e.g., long-distance commuting, short-distance travel, commuting to and from get off work, logistics delivery). After determining the vehicle's current driving scenario, the vehicle controller can determine the physical dimensions corresponding to the current driving scenario and the data indicators corresponding to each physical dimension based on the correlation between the driving scenario and dimensional data. For example, in a congested scenario, because its short-term changes are minimal, there is no need to collect battery SOH data frequently; instead, the focus should be on collecting braking frequency, number of start-stop cycles, and air conditioning power in the environment.

[0048] After determining the data metrics required for each target physical dimension, the vehicle controller can collect the sequence values ​​of the data metrics on the time axis through the vehicle bus or the corresponding sensor interface to obtain multi-dimensional time-series data.

[0049] It should be noted that, to improve the accuracy of data processing, after collecting multi-dimensional time-series data, the vehicle controller also performs time alignment processing on the multi-dimensional time-series data to construct a multivariate time series. For example, for the battery dimension, driving dimension, and environment dimension, respectively... , , Where T is the length of the time window. For example, if T is 300, it represents the historical data of the past 300 time steps (about 5 minutes, sampling frequency 1Hz).

[0050] In some embodiments, after acquiring multidimensional time-series data, the vehicle controller performs feature extraction on the multidimensional time-series data. In this embodiment, the vehicle controller inputs the multidimensional time-series data into corresponding convolutional neural networks.

[0051] For example, Figure 3 The structure of a convolutional neural network is shown, consisting of, for example... Figure 3 It can be seen that this convolutional neural network consists of an input layer, a convolutional layer, a downsampling layer, and a fully connected layer stacked together. Figure 3 The figures shown are merely examples of the scales corresponding to each layer. In practical applications, the scale of the corresponding layer can be adjusted according to requirements. By layering, spatial features of time-series data corresponding to the physical dimensions can be extracted, dimensions can be compressed, and information can be integrated, thereby enabling efficient processing of structured input data and completion of classification or prediction tasks.

[0052] In this embodiment, the vehicle controller inputs the first time-series data into convolutional kernels of multiple scales to obtain initial time-series features obtained by the convolutional kernels of multiple scales extracting features of the first time-series data in different time dimensions; then, the initial time-series features of different time dimensions are combined to obtain the time-series features corresponding to the first time-series data.

[0053] In the above embodiments, the first time series data is any one of multiple sets of time series data.

[0054] In the embodiments of this application, the convolution kernel is the core computational unit of the convolutional neural network. For example, in Figure 4In the matrix operation diagram of the convolutional neural network shown, Input Volume is the input matrix, x is the data within the time window of the input matrix, Filter W0 and Filter W1 are the two convolutional kernels, Bias b0 and Bias b1 are the offsets corresponding to the two convolutional kernels, and Output Volume is the output matrix. Figure 4 In this method, the scale of the convolution kernel is 3×3. The convolution kernel slides on the input matrix with a fixed stride and completes feature calculation through dot product operation of element-wise multiplication and summation. It can automatically extract local correlations and feature patterns in the input data.

[0055] In the embodiments of this application, the convolutional neural network has convolutional kernels of multiple scales, wherein convolutional kernels of different sizes can capture short-term, medium-term and long-term change patterns respectively, thereby realizing hierarchical feature representation of data.

[0056] In this embodiment, the scale of the convolutional kernel can be flexibly adjusted according to the sampling frequency of the temporal data of the corresponding physical dimension. Multiple convolutional kernels of different scales can be combined according to temporal data; for example, multiple scales of convolutional kernels can be combined as [2,4,6], [4,6,8], or [3,6,9], etc. The number of convolutional kernels at each scale can be selected between 8 and 64 depending on the model size and computational resources, and different numbers can be used for temporal data of different physical dimensions. Adaptive average pooling can be used to accept input sequences of different lengths.

[0057] In the above embodiments, multi-scale convolutional kernels are used to process the same set of time-series data in parallel, thereby extracting features from multiple time dimensions (i.e., different time spans). Then, these initial time-series features from different time dimensions are combined to form the final time-series features of the data set. By using convolutional kernels of different scales, different time scales such as transient, intermediate, and long-term can be taken into account simultaneously, making the extraction of time-series features more comprehensive and robust.

[0058] In some embodiments, taking a convolution kernel of multiple scales, including a first convolution kernel and a second convolution kernel, with the scale of the first convolution kernel being smaller than that of the second convolution kernel, the vehicle controller can input the first time series data into the first convolution kernel and the second convolution kernel respectively to obtain the first initial time series feature output by the first convolution kernel and the second initial time series feature output by the second convolution kernel.

[0059] In the above embodiments, the first initial temporal feature is used to characterize the transient change trend of the first temporal data, and the second initial temporal feature is used to characterize the steady-state change trend of the first temporal data. Vehicle control can simultaneously input the first temporal data into convolutional kernels of multiple scales, so that convolutional kernels of multiple scales can perform convolution processing in parallel to improve the efficiency of feature extraction.

[0060] For example, taking three one-dimensional convolutional kernels with scales of 3, 5, and 7 as examples, 16 convolutional kernels are set for each scale, and 48 channels of features are output. The features are then reduced in dimensionality by batch normalization, ReLU activation, and max pooling layer (pooling size 2).

[0061] Small-scale convolutional kernels with a scale of 3 have a small temporal receptive field and are highly sensitive to rapid, localized changes in the data, enabling them to capture transient changes such as rapid acceleration. For example, when the first time-series data is driving behavior data (such as vehicle speed and acceleration), this small-scale convolutional kernel can effectively capture transient events such as rapid acceleration, rapid deceleration, and rapid release of the pedal; when the first time-series data is battery state data (such as voltage and current), this small-scale convolutional kernel can identify instantaneous voltage drops or current spikes caused by sudden load changes. These initial time-series features can reflect the high-frequency fluctuation components of the data within a short period of time, which is crucial for predicting instantaneous changes in driving range caused by sudden driving behaviors.

[0062] A mesoscale convolutional kernel with a scale of 5 possesses a moderately sized temporal receptive field, capable of capturing mid-term trends or pattern transitions in the data, such as changes in driving conditions. For example, for driving behavior data, this mesoscale convolutional kernel can identify the evolution of a vehicle's driving conditions from constant speed to frequent acceleration and deceleration; for environmental data (such as slope sequences), it can identify the beginning and end of a continuous uphill or downhill section; and for battery state data, it can capture the phased rate of decline in SOC over a timescale of several minutes. These initial temporal features reflect the evolution patterns of the data within a medium time window, which is of great significance for predicting changes in driving range caused by changes in driving conditions.

[0063] Large-scale convolutional kernels with a scale of 7 possess a large temporal receptive field, excelling at capturing long-term trends or slowly evolving features in the data, such as the discharge process. For example, for battery state data (e.g., SOC, battery temperature), this large-scale convolutional kernel can capture the global monotonic decreasing trend of the battery throughout the entire discharge cycle, or the gradual cumulative increase in battery temperature during long-distance driving. For driving behavior data, this large-scale convolutional kernel can capture changes in driving style over a long journey (e.g., a gradual transition from an aggressive style to a more relaxed style). For environmental data, this large-scale convolutional kernel can capture the slow changes in ambient temperature throughout the day. These initial temporal features reflect the long-term dependencies of the data, which is crucial for predicting systematic shifts in driving range caused by factors such as battery aging and continuous environmental changes.

[0064] It should be noted that, in the embodiments of this application, the extraction of temporal features across different physical dimensions can employ the same number and structure of convolutional kernels to ensure the consistency of temporal features across all physical dimensions. For example, the temporal features corresponding to the battery dimension, driving dimension, and environment dimension can be denoted as the battery dimension temporal features, respectively. Driving dimension temporal features Environmental dimension time series characteristics .

[0065] In this embodiment of the application, by setting convolution kernels of different scales to extract features of time series data with different trends, more comprehensive time series features can be extracted to meet the needs of different working conditions.

[0066] In this embodiment, the multiple sets of time-series data include at least: battery state data, driving behavior data, and environmental data. The initial time-series features include initial battery features, initial driving features, and initial environmental features. The vehicle controller inputs the time-series data of the above three physical dimensions into the corresponding convolutional neural networks to extract the corresponding time-series features. The convolutional neural networks corresponding to different physical dimensions adopt a multi-scale parallel convolutional structure.

[0067] The following sections will introduce the feature extraction of time series data in the three physical dimensions mentioned above.

[0068] When the first time series data is battery state data, features are extracted from the battery state data from multiple time dimensions using convolutional kernels of multiple scales to obtain initial battery features that affect the remaining usable energy of the battery.

[0069] In the embodiments of this application, the initial battery characteristics may include, but are not limited to, the synergistic change characteristics of SOC with voltage and temperature, voltage plateau characteristics, and the influence of temperature on the battery's usable capacity. For example, a multi-scale convolutional kernel can identify the energy consumption patterns under different operating conditions such as normal battery discharge and high-current discharge, thereby obtaining the synergistic change characteristics between SOC, voltage, and temperature; a multi-scale convolutional kernel can identify the differences in discharge curves of different battery types (e.g., ternary lithium, lithium iron phosphate, etc.), thereby obtaining the voltage plateau characteristics; a multi-scale convolutional kernel can learn the patterns of battery capacity decay under low-temperature conditions and efficiency decrease under high-temperature conditions, thereby obtaining the influence of temperature on the battery's usable capacity.

[0070] When the first time series data is driving behavior data, features are extracted from the driving behavior data from multiple time dimensions using convolutional kernels of multiple scales to obtain the initial driving features that affect battery energy consumption.

[0071] In the embodiments of this application, the initial driving features may include, but are not limited to, the correlation features between vehicle speed and energy consumption, the correlation features between acceleration and energy consumption, and driving style features. For example, a multi-scale convolutional kernel can obtain the correlation features between vehicle speed and energy consumption by identifying energy consumption features under different operating conditions such as high-speed driving and low-speed congestion; a multi-scale convolutional kernel can obtain the correlation features between acceleration and energy consumption by capturing the impact of transient behaviors such as rapid acceleration and sudden braking on energy consumption; a multi-scale convolutional kernel can obtain driving style features by identifying stable or aggressive driving styles and their impact on range from the vehicle speed change patterns.

[0072] When the first time series data is environmental data, features of the environmental data are extracted from multiple time dimensions using convolutional kernels of multiple scales to obtain the initial environmental features that affect battery life.

[0073] In this embodiment, the initial environmental characteristics may include, but are not limited to, the influence of temperature on energy consumption, air conditioning energy consumption characteristics, and tire pressure status characteristics. For example, a multi-scale convolutional kernel can identify the combined effects of decreased battery capacity, increased wind resistance, and increased rolling resistance under low-temperature conditions, thereby obtaining the influence of temperature on energy consumption; a multi-scale convolutional kernel can identify the magnitude and variation of air conditioning power to reflect the energy consumption of accessories on range, thereby obtaining the air conditioning energy consumption characteristics; a multi-scale convolutional kernel can identify the tire pressure level and its variation trend to reflect the influence of rolling resistance on energy consumption, thereby obtaining the tire pressure status characteristics.

[0074] The above embodiments define the initial features of the three types of data and give them clear physical meanings (affecting remaining available energy, affecting energy consumption, and affecting range), making the prediction of vehicle range highly interpretable and physically reasonable, and providing semantically clear input features for subsequent attention coupling.

[0075] In some embodiments, after extracting the intra-group temporal features, the vehicle controller further couples multiple groups of temporal features through a cross-class attention mechanism to obtain multiple inter-group features representing the coupling relationship between different physical dimensions. Specifically, the vehicle controller combines at least two groups of temporal features from the multiple groups of temporal features to obtain multiple feature sets. Then, it cross-couples the temporal features in each feature set through a cross-class attention mechanism to obtain the inter-group features corresponding to each feature set.

[0076] In the above embodiments, the vehicle controller combines at least two sets of temporal features. For example, in a scenario with multiple physical dimensions including battery, driving, and environment dimensions, the vehicle controller combines the temporal features corresponding to the three physical dimensions in pairs to obtain three feature sets: a battery-driving feature set composed of battery and driving temporal features, a battery-environment feature set composed of battery and environment temporal features, and a driving-environment feature set composed of driving and environment temporal features. Then, the vehicle controller inputs these three feature sets into the corresponding attention layers to obtain the inter-group features corresponding to each feature set.

[0077] In the above embodiments, by combining at least two sets of temporal features, it is possible to distinguish the coupling relationship types between different physical dimensions, thereby enabling targeted learning of interaction patterns under different combinations of physical dimensions and avoiding the confusion of coupling relationships caused by simple splicing. Furthermore, by cross-class attention mechanisms to cross-combine temporal features within each feature set, the correlation weights between features of different dimensions can be dynamically calculated. This adaptively highlights important coupling relationships and suppresses irrelevant or noisy interactions under the current operating conditions, giving the generated inter-group features stronger representational capabilities and scene adaptability. In addition, by obtaining multiple inter-group features with clear physical origins, rich and decoupled interaction information is provided for subsequent feature fusion, improving the accuracy and robustness of range prediction.

[0078] In some embodiments, for the cross-coupling of temporal features within a feature set, the vehicle controller first projects the first group of temporal features in the first feature set as query features, projects the second group of temporal features in the first feature set as key features and value features, then calculates the correlation between the query features and the key features to obtain attention weights, and performs a weighted summation of the value features based on the attention weights to obtain the inter-group features.

[0079] In the above embodiments, the first feature set is any one of multiple feature sets, and the first feature set includes a first set of time series features and a second set of time series features.

[0080] For example, Figure 5 A framework diagram of the attention mechanism is shown. Figure 6 A schematic diagram of matrix operations for cross-attention is shown. Figure 5 and Figure 6 In this context, Q represents the query feature, K represents the key feature, and V represents the value feature.

[0081] like Figure 5 As shown, the core architecture of the attention mechanism is based on query ,key ,value Based on three feature branches, the similarity between query feature Q and key feature K is first calculated by dot product. Then, the calculated similarity is scaled, and the normalized Softmax function is applied to the calculated score to convert it into a probability distribution, which is the attention weight. Finally, the attention weight is weighted and summed with the value feature V to obtain the inter-group feature.

[0082] The formula for calculating the attention mechanism can be shown in formula (1): (1) In formula (1), is the scaling factor used to scale the similarity, where Indicates the dimension of the key.

[0083] Figure 6 The computational process of the attention mechanism is illustrated, such as... Figure 6 As shown, the first set of time-series features is projected as query feature Q, and the second set of time-series features is projected as key feature K and value feature V. The similarity between query feature Q and key feature K is calculated via the key dimension. After scaling, attention weights are obtained by SoftMax normalization, and then weighted summation is performed with the value feature V to achieve cross-feature source information interaction and efficient fusion. This can accurately model the inherent coupling relationship between multi-source heterogeneous data, significantly improve the model's ability to jointly model and utilize information from different feature spaces, and is suitable for multi-source feature collaborative modeling scenarios.

[0084] The following sections will explain the cross-coupling of the battery-driving feature set, the battery-environment feature set, and the driving-environment feature set.

[0085] For the battery-driving feature set, the vehicle controller inputs the temporal features in the battery-driving feature set into the battery-driving attention module. This module integrates driving information into the battery temporal features, enabling the battery temporal features to perceive the impact of driving behavior on its state.

[0086] In this embodiment of the application, the battery-driving attention module receives two sets of inputs, one of which is battery timing characteristics. The other group consists of driving sequence characteristics. The battery-driving attention module maps two sets of features to a 64-dimensional space through linear projection to obtain the projected query features. Key features Sum value characteristics Then, calculate the query features. Key features The correlation between the battery time-series features is calculated. Specifically, for each time step's battery time-series features, the battery-driving attention module calculates the similarity between the battery time-series features and the driving time-series features across all time steps, obtaining an attention weight. A higher attention weight indicates a greater impact of driving behavior at that moment on battery consumption. Finally, the battery-driving attention module uses the aforementioned attention weights to evaluate the features. By performing a weighted summation, the fused battery-driving inter-group characteristics can be obtained. As shown in formula (2): (2) For the battery-environment feature set, the vehicle controller inputs the timing features in the battery-environment feature set into the battery-environment attention module. This module integrates environmental information into the battery timing features, enabling the battery timing features to perceive the impact of environmental conditions on its performance.

[0087] In this embodiment, the battery-environment attention module receives two sets of inputs, one of which is battery timing features. The other group consists of environmental temporal characteristics. The battery-environment attention module maps two sets of temporal features to a 64-dimensional space through linear projection, obtaining the projected query features. Key features Sum value characteristics Then, calculate the query features. Key features The correlation between the battery and environment time-series features is calculated. Specifically, for each time-step battery time-series feature, the battery-environment attention module calculates the similarity between the battery time-series feature and the environment time-series features of all time steps, obtaining the attention weight. A larger attention weight indicates a greater impact of environmental conditions on battery performance at that moment. Finally, the battery-environment attention module uses the aforementioned attention weights to evaluate the value features. We perform a weighted summation to obtain the fused battery-environment inter-group characteristics. As shown in formula (3): (3) For the driving-environment feature set, the vehicle controller inputs the temporal features in the driving-environment feature set into the driving-environment attention module. This module integrates environmental information into the driving temporal features, enabling the driving temporal features to perceive the impact of environmental conditions on its energy consumption.

[0088] In this embodiment of the application, the driving-environment attention module receives two sets of inputs, one of which is driving temporal features. The other group consists of environmental temporal characteristics. The driving-environment attention module maps two sets of temporal features to a 64-dimensional space through linear projection, obtaining the projected query features. Key features Sum value characteristics Then, calculate the query features. Key features The correlation between driving time-series features is calculated. Specifically, for each time step's driving time-series features, the driving-environment attention module calculates the similarity between the driving time-series features and the environmental time-series features of all time steps, obtaining an attention weight. A larger attention weight indicates a greater impact of environmental conditions on driving energy consumption at that moment. Finally, the driving-environment attention module uses the aforementioned attention weights to evaluate the value features. We perform a weighted summation to obtain the fused inter-group characteristics of the driving-environment interaction. As shown in formula (4): (4) In this embodiment, the three cross-category attention modules mentioned above achieve hierarchical fusion of multi-source information. Specifically, the battery-driving attention module integrates driving behavior information into battery timing features, enabling the battery to perceive how driving behavior consumes energy; the battery-environment attention module integrates environmental condition information into battery timing features, enabling the battery to perceive how environmental factors affect performance; and the driving-environment attention module integrates environmental condition information into driving timing features, enabling the driver to perceive how environmental factors affect energy consumption. This comprehensively considers the impact of multiple dimensions on range in the process of predicting vehicle range, thereby improving the accuracy of range prediction.

[0089] It should be noted that, in this embodiment, the temporal features output by the grouped convolution are directly input into three cross-class attention modules to establish interaction relationships between different data groups. Each attention module can employ a single-head attention mechanism, focusing on learning a specific type of cross-class physical coupling. Alternatively, each attention module can use multi-head attention (e.g., 2-head, 4-head) instead of single-head attention, enabling different heads to learn different types of interaction patterns. Bidirectional cross-attention can be used to achieve bidirectional information fusion.

[0090] In some embodiments, after obtaining multiple inter-group features through a cross-class attention mechanism, in order to integrate these inter-group features into a fixed-length feature vector as input to the subsequent prediction network, the vehicle controller also performs temporal average pooling on the multiple inter-group features to obtain an inter-group feature vector corresponding to each inter-group feature; then, the multiple inter-group feature vectors are concatenated to obtain a feature vector.

[0091] For example, for each inter-group feature, the vehicle controller performs an average pooling operation along the time axis. Specifically, for each feature dimension of the inter-group feature, its average pooling across all features is calculated. The arithmetic mean at each time step. For battery-driving inter-group characteristics. The corresponding inter-group feature vector can be represented by formula (5): (5) In formula (5), This represents the inter-group feature vector for battery-driving.

[0092] Intergroup characteristics of battery-environment The corresponding inter-group feature vector can be represented by formula (6): (6) In formula (6), This represents the inter-group feature vector of the battery-environment relationship.

[0093] For intergroup characteristics of driving-environment The corresponding inter-group feature vector can be represented by formula (7): (7) In formula (7), This represents the inter-group feature vector of the driving-environment relationship.

[0094] By performing temporal average pooling on the interaction features output by the three cross-class attention modules, a fixed-length feature vector can be obtained. This method can summarize the entire historical information, transforming the time series into static features, which is convenient for subsequent fully connected network processing.

[0095] In this embodiment, after obtaining multiple inter-group feature vectors, the vehicle controller can further concatenate these vectors along the feature dimension to form a high-dimensional feature vector. For example, the vehicle controller can arrange the vector elements corresponding to the multiple inter-group feature vectors in a certain order to obtain the feature vector.

[0096] For example, the vehicle controller can concatenate the three pooled inter-group feature vectors along the channel dimension to obtain a feature vector, as shown in formula (8): (8) In formula (8), The resulting feature vector integrates three types of information: battery-driving interaction features that characterize how driving behavior consumes the battery, battery-environment interaction features that characterize how environmental factors affect battery performance, and driving-environment interaction features that characterize how environmental factors affect driving energy consumption. This brings together multiple dimensions of information that affect driving range.

[0097] After obtaining the feature vector, the vehicle controller inputs the feature vector into a fully connected network to perform dimensionality reduction and nonlinear transformation on the feature vector to obtain the predicted driving range, as shown in formula (9): (9) In formula (9), To predict driving range.

[0098] For example, the specific structure of a fully connected network may include three layers, as follows: First layer: Linear(192, 128) + ReLU activation + Dropout(0.2); Second layer: Linear(128, 64) + ReLU activation + Dropout(0.1); The third layer: Linear(64, 1), outputs the predicted driving range.

[0099] The fully connected network compresses and refines the feature vectors step by step through dimensionality reduction and nonlinear transformation, and outputs a continuous driving range value, completing the end-to-end mapping from multi-source time series data to driving range prediction.

[0100] In some embodiments, Figure 7 A complete flowchart of the method provided in the embodiments of this application is shown, as follows: Figure 7 As shown, the method includes the following steps S701 to S708: In step S701, the vehicle controller first acquires multi-source time-series data of the vehicle, including but not limited to battery data, driving data, and environmental data. Step S702: The vehicle controller groups the multi-source time-series data according to the physical dimension to obtain multiple sets of time-series data; Step S703 involves performing time-series alignment, sliding window sampling, and normalization on each group of time-series data. Step S704: Input the processed time series data into the corresponding convolutional neural network to extract intra-group features and obtain time series features. Step S705: A cross-category attention mechanism is used to cross-fuse the temporal features of each group to obtain inter-group features that characterize the coupling relationship between different physical dimensions. Step S706: Perform time-series averaging on the inter-group features to obtain the inter-group feature vector; Step S707: Concatenate the feature vectors between groups to obtain the feature vector; In step S708, the feature vector is input into the fully connected network to obtain the predicted value of the vehicle's driving range.

[0101] This concludes the introduction of the methods provided in the embodiments of this application.

[0102] As described above, this application's embodiments employ a hierarchical processing architecture that uses grouped convolution to extract intra-group features and cross-category attention to establish inter-group interactions. Through group processing, battery, driving, and environmental data are independently input into their respective convolutional neural networks, allowing each group to specifically learn intra-group features, thus avoiding information overload. Simultaneously, multi-scale convolutional parallel extraction captures features at different time scales, such as rapid acceleration (seconds), operating condition switching (minutes), and discharge processes (hours). The unified output dimensions of the three groups lay the foundation for subsequent cross-category fusion, solving the problem of coarse data fusion methods in existing technologies. Furthermore, by setting up three independent cross-category attention modules to handle three different physical coupling relationships—the battery-driving attention module learning how driving behavior affects battery consumption, the battery-environment attention module learning how environmental factors affect battery performance, and the driving-environment attention module learning how environmental factors affect driving energy consumption—the learning difficulty of the model is reduced, the learning accuracy is improved, and a foundation is laid for accurate prediction of vehicle range.

[0103] This application also provides a vehicle range prediction device, such as... Figure 8 As shown, the device 800 includes: a data acquisition module 801, a feature extraction module 802, a feature coupling module 803, a feature fusion module 804, and a battery life prediction module 805.

[0104] The data acquisition module 801 is used to acquire multiple sets of time-series data of the vehicle during operation; The feature extraction module 802 is used to input each group of time series data into the corresponding convolutional neural network for feature extraction, so as to obtain the time series features corresponding to each group of time series data; The feature coupling module 803 is used to couple multiple sets of temporal features through a cross-class attention mechanism to obtain multiple inter-group features that characterize the coupling relationship between different physical dimensions. The feature fusion module 804 is used to fuse features between multiple groups to obtain feature vectors; The range prediction module 805 is used to predict the vehicle's range based on feature vectors to obtain the predicted range.

[0105] In some embodiments, the convolutional neural network has convolutional kernels of multiple scales, and the feature extraction module includes: The initial feature acquisition module is used to input the first time series data into convolutional kernels of multiple scales respectively, and obtain the initial time series features obtained by the convolutional kernels of multiple scales extracting features of the first time series data in different time dimensions. The first time series data is any one set of time series data from multiple sets of time series data. The feature combination module is used to combine the initial time-series features of different time dimensions to obtain the time-series features corresponding to the first time-series data.

[0106] In some embodiments, the convolutional kernels of multiple scales include a first convolutional kernel and a second convolutional kernel. The scale of the first convolutional kernel is smaller than that of the second convolutional kernel. The initial feature acquisition module is specifically used to input the first time-series data into the first convolutional kernel and the second convolutional kernel respectively to obtain the first initial time-series feature output by the first convolutional kernel and the second initial time-series feature output by the second convolutional kernel. The first initial time-series feature is used to characterize the transient change trend of the first time-series data, and the second initial time-series feature is used to characterize the steady-state change trend of the first time-series data.

[0107] In some embodiments, the multiple sets of time-series data include at least: battery state data, driving behavior data, and environmental data. The initial time-series features include initial battery features, initial driving features, and initial environmental features. Specifically, the initial feature acquisition module is used to extract features from the battery state data from multiple time dimensions using convolutional kernels of multiple scales when the first time-series data is battery state data, to obtain initial battery features that affect the remaining usable energy of the battery; to extract features from the driving behavior data from multiple time dimensions using convolutional kernels of multiple scales when the first time-series data is driving behavior data, to obtain initial driving features that affect battery energy consumption; and to extract features from the environmental data from multiple time dimensions using convolutional kernels of multiple scales when the first time-series data is environmental data, to obtain initial environmental features that affect battery range.

[0108] In some embodiments, the feature coupling module includes a set generation module and a feature cross-coupling module. The set generation module is used to combine at least two sets of temporal features from multiple sets of temporal features to obtain multiple feature sets; the feature cross-coupling module is used to cross-couple the temporal features in each feature set through a cross-class attention mechanism to obtain inter-group features corresponding to each feature set.

[0109] In some embodiments, the feature cross-projection module is specifically used to project a first set of time-series features in the first feature set into query features, and to project a second set of time-series features in the first feature set into key features and value features. The first feature set is any one of multiple feature sets, and the first feature set includes a first set of time-series features and a second set of time-series features. The module calculates the correlation between the query features and the key features to obtain attention weights. Based on the attention weights, the value features are weighted and summed to obtain inter-group features.

[0110] In some embodiments, the feature fusion module includes a vector generation module and a vector concatenation module. The vector generation module is used to perform temporal average pooling on multiple inter-group features to obtain inter-group feature vectors corresponding to each inter-group feature; the vector concatenation module is used to concatenate the multiple inter-group feature vectors to obtain a feature vector.

[0111] In some embodiments, the range prediction module is specifically used to input the feature vector into a fully connected network to perform dimensionality reduction and nonlinear transformation on the feature vector to obtain the predicted range.

[0112] In some embodiments, the data acquisition module is specifically used to acquire vehicle driving scenario data; determine the target physical dimension and the corresponding data indicators for predicting the vehicle's range under the driving scenario data based on the correlation between the driving scenario and the dimensional data; and collect time-series data corresponding to the data indicators to obtain multi-dimensional time-series data.

[0113] The vehicle range prediction device provided in this application embodiment can realize the various processes implemented in the aforementioned method embodiment, and will not be described again here to avoid repetition.

[0114] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is merely an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiments 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. The integrated unit can be implemented in hardware or as a software functional unit. Furthermore, the specific names of the functional units and modules are only for easy differentiation and are not intended to limit the scope of protection of this application. The specific working process of the units and modules in the above system can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0115] Figure 9A schematic diagram of the hardware structure of the electronic device provided in an embodiment of this application is shown.

[0116] The electronic device may include a processor 901 and a memory 902 storing computer program instructions.

[0117] Specifically, the processor 901 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of this application.

[0118] Memory 902 may include mass storage for data or instructions. For example, and not limitingly, memory 902 may include a hard disk drive (HDD), floppy disk drive, flash memory, optical disk, magneto-optical disk, magnetic tape, or Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, memory 902 may include removable or non-removable (or fixed) media. Where appropriate, memory 902 may be internal or external to the integrated gateway disaster recovery device. In a particular embodiment, memory 902 is non-volatile solid-state memory.

[0119] Memory may include read-only memory (ROM), random access memory (RAM), disk storage media devices, optical storage media devices, flash memory devices, and electrical, optical, or other physical / tangible memory storage devices. Therefore, typically, memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software including computer-executable instructions, and when the software is executed (e.g., by one or more processors), it is operable to perform the operations described with reference to the methods according to one aspect of this disclosure.

[0120] The processor 901 reads and executes computer program instructions stored in the memory 902 to implement any of the vehicle range prediction methods in the above embodiments.

[0121] In one example, the electronic device may also include a communication interface 903 and a bus 910. Wherein, as... Figure 9 As shown, the processor 901, memory 902, and communication interface 903 are connected through bus 910 and complete communication with each other.

[0122] The communication interface 903 is mainly used to realize communication between various modules, devices, units and / or equipment in the embodiments of this application.

[0123] Bus 910 includes hardware, software, or both, that couples components of an electronic device together. For example, and not limitingly, the bus may include an Accelerated Graphics Port (AGP) or other graphics bus, an Enhanced Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), HyperTransport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an Infinite Bandwidth Interconnect, a Low Pin Count (LPC) bus, a memory bus, a Microchannel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local (VLB) bus, or other suitable buses, or combinations of two or more of these. Where appropriate, bus 910 may include one or more buses. Although specific buses are described and illustrated in embodiments of this application, this application contemplates any suitable bus or interconnect.

[0124] Furthermore, in conjunction with the vehicle range prediction methods in the above embodiments, this application embodiment can provide a computer-readable storage medium for implementation. This computer-readable storage medium stores computer program instructions; when executed by a processor, these computer program instructions implement any of the vehicle range prediction methods in the above embodiments.

[0125] Furthermore, in conjunction with the vehicle range prediction method in the above embodiments, this application embodiment can provide a computer program product for implementation. When the instructions in this computer program product are executed by the processor of an electronic device, the electronic device performs and implements any of the vehicle range prediction methods in the above embodiments.

[0126] It should be clarified that this application is not limited to the specific configurations and processes described above and shown in the figures. For the sake of brevity, detailed descriptions of known methods are omitted here. In the above embodiments, several specific steps are described and shown as examples. However, the method process of this application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications, and additions, or change the order of steps, after understanding the spirit of this application.

[0127] The functional modules shown in the above-described block diagram can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application-specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, etc. When implemented in software, the elements of this application are programs or code segments used to perform the required tasks. Programs or code segments can be stored on a machine-readable medium or transmitted over a transmission medium or communication link via data signals carried on a carrier wave. "Machine-readable medium" can include any medium capable of storing or transmitting information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, etc. Code segments can be downloaded via computer networks such as the Internet, intranets, etc.

[0128] It should also be noted that the exemplary embodiments mentioned in this application describe methods or systems based on a series of steps or apparatus. However, this application is not limited to the order of the above steps; that is, the steps can be performed in the order mentioned in the embodiments, or in a different order, or several steps can be performed simultaneously.

[0129] The foregoing flowcharts and / or block diagrams of the vehicle range prediction method and apparatus according to embodiments of the present disclosure have described various aspects of the present disclosure. It should be understood that each block in the flowcharts and / or block diagrams, and combinations of blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to create a machine such that these instructions, executable via the processor of the computer or other programmable data processing apparatus, enable the implementation of the functions / actions specified in one or more blocks of the flowcharts and / or block diagrams. Such a processor can be, but is not limited to, a general-purpose processor, a special-purpose processor, a special application processor, or a field-programmable logic circuit. It is also understood that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can also be implemented by dedicated hardware performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0130] The above description is merely a specific implementation of this application. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, modules, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here. It should be understood that the protection scope of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the protection scope of this application.

Claims

1. A method for predicting vehicle driving range, characterized in that, include: Acquire multiple sets of time-series data of the vehicle during operation; Each set of time-series data is input into the corresponding convolutional neural network for feature extraction to obtain the time-series features corresponding to each set of time-series data; By coupling multiple sets of temporal features through a cross-class attention mechanism, multiple inter-group features are obtained to characterize the coupling relationship between different physical dimensions. Feature fusion is performed on multiple inter-group features. Obtain the feature vector; The predicted driving range of the vehicle is obtained by predicting the driving range based on the feature vector. The convolutional neural network has convolutional kernels of multiple scales. Each set of time-series data is input into the corresponding convolutional neural network for feature extraction to obtain the time-series features corresponding to each set of time-series data, including: The first time series data is input into the convolutional kernels of the multiple scales respectively, and the initial time series features obtained by the convolutional kernels of the multiple scales extracting features of the first time series data in different time dimensions are obtained. The first time series data is any one of the multiple sets of time series data. The initial time-series features of different time dimensions are combined to obtain the time-series features corresponding to the first time-series data.

2. The method according to claim 1, characterized in that, The multiple scale convolutional kernels include a first convolutional kernel and a second convolutional kernel, wherein the scale of the first convolutional kernel is smaller than the scale of the second convolutional kernel. The step of inputting the first temporal data into the multiple scale convolutional kernels respectively, and obtaining the initial temporal features obtained by the multiple scale convolutional kernels performing feature extraction on the first temporal data at different time dimensions, includes: The first time series data is input into the first convolution kernel and the second convolution kernel respectively to obtain the first initial time series feature output by the first convolution kernel and the second initial time series feature output by the second convolution kernel. The first initial time series feature is used to characterize the transient change trend of the first time series data, and the second initial time series feature is used to characterize the steady-state change trend of the first time series data.

3. The method according to claim 1, characterized in that, The multiple sets of time-series data include at least: battery state data, driving behavior data, and environmental data. The initial time-series features include initial battery features, initial driving features, and initial environmental features. The initial time-series features obtained by inputting the first time-series data into the multiple scale convolutional kernels and extracting features from the first time-series data using the multiple scale convolutional kernels at different time dimensions include: When the first time series data is the battery state data, the battery state data is used to extract features from multiple time dimensions through the convolutional kernels of the multiple scales to obtain the initial battery features that affect the remaining usable energy of the battery. When the first time series data is the driving behavior data, the driving behavior data is feature extracted from multiple time dimensions using the convolutional kernels of multiple scales to obtain the initial driving features that affect battery energy consumption. When the first time-series data is the environmental data, features are extracted from the environmental data from multiple time dimensions using convolutional kernels of multiple scales to obtain initial environmental features that affect battery life.

4. The method according to any one of claims 1 to 3, characterized in that, The method involves coupling multiple sets of temporal features through a cross-class attention mechanism to obtain multiple inter-group features used to characterize the coupling relationship between different physical dimensions, including: At least two sets of time-series features from the multiple sets of time-series features are combined to obtain multiple feature sets; By cross-category attention mechanism, the temporal features in each feature set are cross-coupled to obtain the inter-group features corresponding to each feature set.

5. The method according to claim 4, characterized in that, The step of cross-combining the temporal features in each feature set to obtain the inter-group features corresponding to each feature set includes: The first set of time-series features in the first feature set is projected as query features, and the second set of time-series features in the first feature set is projected as key features and value features. The first feature set is any one of the plurality of feature sets, and the first feature set includes the first set of time-series features and the second set of time-series features. Calculate the correlation between the query feature and the key feature to obtain the attention weight; The value features are weighted and summed based on the attention weights to obtain the inter-group features.

6. The method according to any one of claims 1 to 3, characterized in that, The feature fusion is performed on multiple inter-group features. The feature vector is obtained, including: Temporal average pooling is performed on each of the multiple inter-group features to obtain an inter-group feature vector corresponding to each inter-group feature. The feature vector is obtained by concatenating the feature vectors of the multiple groups.

7. The method according to claim 6, characterized in that, The step of predicting the vehicle's range based on the feature vector to obtain the predicted range includes: The feature vector is input into a fully connected network to perform dimensionality reduction and nonlinear transformation on the feature vector to obtain the predicted driving range.

8. The method according to any one of claims 1 to 3, characterized in that, The acquisition of multiple sets of time-series data of the vehicle during operation includes: Obtain the driving scenario data of the vehicle; Based on the correlation between driving scenarios and dimensional data, the target physical dimension required to predict the vehicle's range under the driving scenario data and the corresponding data indicators of the target physical dimension are determined. Collect time-series data corresponding to the data indicators to obtain multidimensional time-series data.

9. A device for predicting vehicle driving range, characterized in that, include: The data acquisition module is used to acquire multiple sets of time-series data of the vehicle during operation. The feature extraction module is used to input each group of time series data into the corresponding convolutional neural network for feature extraction, so as to obtain the time series features corresponding to each group of time series data; The feature coupling module is used to couple multiple sets of temporal features through a cross-class attention mechanism to obtain multiple inter-group features that characterize the coupling relationship between different physical dimensions. The feature fusion module is used to perform feature fusion on multiple inter-group features. Obtain the feature vector; The range prediction module is used to predict the range of the vehicle based on the feature vector to obtain the predicted range. The convolutional neural network has convolutional kernels of multiple scales. The feature extraction module is used to input each group of time-series data into the corresponding convolutional neural network for feature extraction, obtaining the time-series features corresponding to each group of time-series data, including: The first time series data is input into the convolutional kernels of the multiple scales respectively, and the initial time series features obtained by the convolutional kernels of the multiple scales extracting features of the first time series data in different time dimensions are obtained. The first time series data is any one of the multiple sets of time series data. The initial time-series features of different time dimensions are combined to obtain the time-series features corresponding to the first time-series data.

Images