Driving style based vehicle remaining range fuel consumption prediction method and system

Abstract

The present disclosure belongs to the technical field of vehicle fuel consumption analysis and prediction, and particularly relates to a vehicle residual route fuel consumption prediction method and system based on driving style, comprising: obtaining driving behavior data and driving environment information of a vehicle route that has been traveled, to obtain a driving style representation vector; wherein the driving style representation vector includes vehicle fuel consumption; obtaining route characteristics of the traveled route according to the formed driving style representation vector and vehicle configuration information; obtaining driving environment information of a residual route, combining the driving style representation vector and vehicle configuration information to obtain residual route characteristics; calculating the similarity of the traveled route characteristics and the residual route characteristics, performing weighted summation of the traveled route characteristics and the residual route characteristics based on the obtained similarity, and completing the fuel consumption prediction of the vehicle traveling on the residual route.

Description

Technical Field

[0001] This disclosure belongs to the field of vehicle fuel consumption analysis and prediction technology, specifically involving a method and system for predicting vehicle remaining route fuel consumption based on driving style. Background Technology

[0002] The statements in this section are merely background information relating to this disclosure and do not necessarily constitute prior art.

[0003] With the rapid development of society, the automotive industry is constantly evolving. While automobiles bring convenience to people's lives, they also exacerbate exhaust pollution. Statistics show that over 70% of newly added oil in China is consumed by new vehicles, further highlighting the current energy shortage problem. Predicting vehicle fuel consumption on future roads can provide decision support for drivers' route planning and energy management, which is of great practical significance for energy conservation, environmental protection, cost reduction, and profit improvement.

[0004] Current research primarily employs simple fuel consumption prediction methods to calculate fuel consumption between the start and end points of a vehicle's journey. Physics-based fuel consumption prediction models use a set of mathematical equations corresponding to different vehicle subsystems and components to describe vehicle dynamics at each time step, providing high-precision predictions. However, this comes at the cost of low computational efficiency and requires extensive knowledge of vehicle dynamics and multidimensional maps, which is often unavailable. Traditional machine learning methods analyze factors influencing fuel consumption, determine these factors and the strength of their correlations, and then combine them with traditional machine learning methods such as multiple linear regression to construct fuel consumption prediction models. However, vehicle operating status data contains numerous factors affecting fuel consumption, and these factors are not always linearly correlated with fuel consumption, easily leading to poor model fitting. Fuel consumption prediction models based on deep learning methods possess strong self-learning capabilities and the ability to model complex nonlinear data, and have been widely applied in the field of fuel consumption prediction.

[0005] Research on fuel consumption prediction still has the following problems:

[0006] First, the impact of driving style (such as speed and acceleration habits) on fuel consumption has not been fully studied and utilized;

[0007] Secondly, existing fuel consumption prediction methods mainly focus on predicting fuel consumption along the route between the origin and destination. When a vehicle has been traveling on the current route for a period of time and it is necessary to calculate the fuel consumption required to reach the destination from the current location (remaining route), existing route fuel consumption prediction models use the remaining route and the current time as inputs and provide the fuel consumption that may be required to reach the destination. However, they often ignore the route that has already been traveled from the origin to the driver's parking space (the route already traveled). Making full use of this part of the route may help improve the accuracy of fuel consumption prediction. Summary of the Invention

[0008] To address the aforementioned issues, this disclosure proposes a method and system for predicting fuel consumption for remaining routes based on driving style. The method extracts driving style from historical driving data, utilizes a deep fusion network to calculate driving data and driving environment characteristics of the already driven and remaining routes, and integrates static information such as historical driving style and vehicle configuration to improve the accuracy of fuel consumption prediction.

[0009] According to some embodiments, the first solution of this disclosure provides a method for predicting the remaining fuel consumption of a vehicle based on driving style, employing the following technical solution:

[0010] A method for predicting remaining fuel consumption of a vehicle based on driving style includes:

[0011] The driving behavior data and driving environment information of the vehicle's traveled route are obtained to obtain a driving style representation vector; wherein, the driving style representation vector includes vehicle fuel consumption;

[0012] Based on the generated driving style representation vector and vehicle configuration information, the characteristics of the route already traveled are obtained;

[0013] Obtain the driving environment information of the remaining route, and combine it with the driving style representation vector and vehicle configuration information to obtain the features of the remaining route;

[0014] Calculate the similarity between the features of the traveled route and the features of the remaining route. Based on the obtained similarity, perform a weighted sum of the features of the traveled route and the features of the remaining route to predict the fuel consumption of the vehicle on the remaining route.

[0015] As a further technical limitation, driving behavior statistics are obtained based on GPS trajectory at different time windows to capture instantaneous vehicle motion characteristics. The captured instantaneous vehicle motion characteristics are represented in matrix form to obtain driving behavior data of the vehicle's traveled route. The obtained driving environment information includes road conditions, geographical area, traffic conditions and weather conditions.

[0016] As a further technical limitation, a statistical matrix is ​​obtained based on the obtained driving behavior data, and an environment matrix is ​​obtained based on the obtained driving environment information. The obtained statistical matrix and environment matrix are then fused and convolutionally learned to obtain a driving style representation vector.

[0017] As a further technical limitation, the driven route features include dynamic data that considers driving environment information and driving behavior during vehicle operation, and static data that considers driving style representation vectors and vehicle configuration information. A recurrent model is used to learn the potential relationship between the driven route features and driving environment information. The recurrent model includes a transformer layer, a concatenate layer, and a GRU layer. The dynamic data and static data are fused under the action of the concatenate layer to obtain the driven route features.

[0018] As a further technical limitation, the remaining route characteristics are related to the remaining route road environment, and the remaining route characteristics represent h. remain for: Where, d remain Represents the dimension of the corresponding vector, and the remaining route feature set E road have Composition, that is Indicates the remaining route segment l i The road environment vector, n l This indicates the number of road segments remaining on the route.

[0019] As a further technical limitation, the road attribute similarity 'a' between the traveled route and the remaining route is: in, This indicates the road environment characteristics of segment i in the route already traveled. Let represent the road environment characteristics of segment j in the remaining route, s represent the total number of segments in the route, and s′ represent the total number of segments in the remaining route.

[0020] As a further technical limitation, the features of the driven route and the remaining route are weighted and summed based on similarity, and the fuel consumption of the vehicle on the remaining route is predicted through a fully connected layer.

[0021] According to some embodiments, the second solution of this disclosure provides a vehicle remaining route fuel consumption prediction system based on driving style, which adopts the following technical solution:

[0022] A vehicle remaining route fuel consumption prediction system based on driving style includes:

[0023] The vector acquisition module is configured to acquire driving behavior data and driving environment information of the vehicle's traveled route to obtain a driving style representation vector; wherein, the driving style representation vector includes vehicle fuel consumption;

[0024] The feature acquisition module is configured to obtain the features of the traveled route based on the formed driving style representation vector and vehicle configuration information; obtain the driving environment information of the remaining route, and combine it with the driving style representation vector and vehicle configuration information to obtain the features of the remaining route.

[0025] The fuel consumption prediction module is configured to calculate the similarity between the features of the traveled route and the features of the remaining route, and then perform a weighted summation of the features of the traveled route and the features of the remaining route based on the obtained similarity to complete the fuel consumption prediction for the vehicle traveling on the remaining route.

[0026] According to some embodiments, a third aspect of this disclosure provides a computer-readable storage medium, employing the following technical solution:

[0027] A computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the steps of the vehicle remaining route fuel consumption prediction method based on driving style as described in the first aspect of this disclosure.

[0028] According to some embodiments, the fourth solution of this disclosure provides an electronic device that adopts the following technical solution:

[0029] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps in the vehicle remaining route fuel consumption prediction method based on driving style as described in the first aspect of this disclosure.

[0030] Compared with the prior art, the beneficial effects of this disclosure are as follows:

[0031] This disclosure fully considers historical driving behavior and the driving environment that generated that behavior. It learns driving style representations from historical driving data based on convolutional networks; it uses deep fusion networks to process different types of fuel consumption influencing factors and extract comprehensive and effective fuel consumption features; it calculates driving behavior statistics and driving environment information for the already driven route, uses Transformer to capture time dependencies, and fuses static information such as historical driving style and vehicle configuration to form features for the already driven route; it calculates driving environment information for the remaining route, uses Transformer to capture time dependencies, and fuses static information such as historical driving style and vehicle configuration with geographic region coding to form features for the remaining route; it uses the similarity between the remaining route and the already driven route to perform attention-weighted prediction on the two parts of the route features, thereby improving the accuracy of fuel consumption prediction. Attached Figure Description

[0032] The accompanying drawings, which form part of this disclosure, are used to provide a further understanding of this disclosure. The illustrative embodiments of this disclosure and their descriptions are used to explain this disclosure and do not constitute an undue limitation of this disclosure.

[0033] Figure 1 This is a flowchart of the vehicle remaining route fuel consumption prediction method based on driving style in Embodiment 1 of this disclosure;

[0034] Figure 2 This is a framework diagram of the driving style representation vector in Embodiment 1 of this disclosure;

[0035] Figure 3 This is a schematic diagram illustrating the principle of remaining fuel consumption prediction in Embodiment 1 of this disclosure;

[0036] Figure 4 This is a structural block diagram of the vehicle remaining route fuel consumption prediction system based on driving style in Embodiment 2 of this disclosure. Detailed Implementation

[0037] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0038] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of this disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0039] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0040] Where there is no conflict, the embodiments and features described herein can be combined with each other.

[0041] Example 1

[0042] Embodiment 1 of this disclosure introduces a method for predicting the remaining fuel consumption of a vehicle based on driving style.

[0043] like Figure 1 The method shown is a vehicle remaining route fuel consumption prediction method based on driving style, including:

[0044] The driving behavior data and driving environment information of the vehicle's traveled route are obtained to obtain a driving style representation vector; wherein, the driving style representation vector includes vehicle fuel consumption;

[0045] Based on the generated driving style representation vector and vehicle configuration information, the characteristics of the route already traveled are obtained;

[0046] Obtain the driving environment information of the remaining route, and combine it with the driving style representation vector and vehicle configuration information to obtain the features of the remaining route;

[0047] Calculate the similarity between the features of the traveled route and the features of the remaining route. Based on the obtained similarity, perform a weighted sum of the features of the traveled route and the features of the remaining route to predict the fuel consumption of the vehicle on the remaining route.

[0048] This embodiment provides a detailed introduction to the method for predicting the remaining fuel consumption of a vehicle based on driving style, through the driving style representation section and the remaining route fuel consumption prediction section.

[0049] Driving style indication section

[0050] In this embodiment, the driving style representation is calculated from the historical trajectory database based on the driving behavior statistics of each driver, combined with the driving environment in which these behaviors occurred, to extract the driving style representation of each driver. The specific steps are as follows:

[0051] (1) The GPS data conversion module extracts driving behavior statistics for different time windows.

[0052] Instead of feeding raw GPS data into a deep learning model, GPS tracks are converted into more stable statistical features: for each driver, a fixed-length Ls segment of GPS track is taken to learn a representation of driving style.

[0053] In this embodiment, two basic features are used to capture instantaneous vehicle motion characteristics: velocity and acceleration. To reduce the potential impact of outliers, the segment is divided into frames of a fixed size Lf, with an offset. For each frame, seven statistics are calculated for each basic feature, including the mean, minimum, maximum, 25%, 50%, and 75% quartiles, and standard deviation. Each trajectory Tj consists of a series of time-ordered GPS records, and each GPS record gi consists of a timestamp, longitude, latitude, and speed, represented as: gi = <ts, lat, lng, v>; where ts represents the timestamp, lat represents the latitude, lng represents the longitude, and v represents the instantaneous speed.

[0054] Easily calculate speed statistics using driving speed v, and acceleration statistics using timestamp ts and driving speed v; derive a set of statistical feature matrices, each matrix consisting of 2×7=14 rows and columns. The system consists of columns; the statistical feature matrix encodes driving behavior information for the trajectory segment and serves as part of the input to the learning model; if the trajectory segment is shorter than Ls, zeros need to be padded in the matrix to unify the size of all statistical feature matrices. Longer trajectories containing more information about driving behavior are more suitable for model training.

[0055] In this embodiment, the statistical feature matrix consists of 2 × 7 = 14 rows. A two-dimensional matrix of columns, represented as For example, the position (1,1) in the matrix represents the average velocity of the first frame. Organizing the rows of the matrix in the order of average velocity, minimum velocity, maximum velocity, 25% velocity, 50% velocity, 75% velocity quartile, and standard deviation of velocity, the matrix (3,5) represents the maximum velocity of the fifth frame.

[0056] (2) The driving environment representation module learns the driving environment, such as roads, geographical areas, traffic, and weather.

[0057] Since driving behavior is implicitly influenced by the surrounding driving environment, it is also necessary to consider driving environment information to allow the model to deeply "understand" the driver's behavior, especially in specific driving environments.

[0058] The driving environment information considered in this embodiment includes road conditions, geographical area, traffic conditions, and weather conditions.

[0059] like Figure 2 The driving environment representation module shown fully illustrates how each raw GPS trajectory is processed to generate environmental features, specifically:

[0060] For each GPS trajectory Tj, route matching is performed using map matching technology. Since the GPS data conversion module outputs a statistical feature matrix for each trajectory segment, the driving environment representation module also operates on the corresponding trajectory segment and its associated driving road segment, and generates an environmental feature matrix accordingly.

[0061] Based on road network G and the s-th trajectory segment T js and its trajectory segment R after map matching js Export the representation for each driving environment:

[0062] 1) Road conditions

[0063] Road conditions are characterized by road segment type, length, slope, number of traffic lights in the segment, and number of Points of Interest (POIs). (In Geographic Information Systems, a POI can represent a building, shop, bus stop, etc.; in geospatial prediction, the number of POIs often indicates the "bustling" area around a road segment.) Five vectors, nt, nl, ns, ng, and np (where nt represents the road segment type, nl represents the road segment length, ns represents the road segment slope, ng represents the number of traffic lights in the segment, and np represents the number of POIs), are used to encode the attributes of each road segment. For the road segment type, one-hot encoding is used to generate the attribute vector, given the matched route R. js Derive R js The road type vectors of the covered road segments are sequentially concatenated into a single vector that describes the generated trajectory R of the vehicle after route matching. js The road types in the middle are identified; a linear layer is used to reduce the dimension of the sparse attribute vector, forming a road condition representation vector.

[0064] In this embodiment, the attribute vector is a one-dimensional vector of length 17. There are 17 road types: tertiary, tertiary_link, residential, unclassified, secondary, secondary_link, primary, primary_link, motorway, motorway Jlink, trunk, trunk_link, track, bridleway, livingstreet, path, and service. For example, for the tertiary road type, one-hot encoding is represented by the vector [1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]. For the residential road type, one-hot encoding is represented by the vector [0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0].

[0065] For road segment types, one-hot encoding is used to generate attribute vectors:

[0066] Given the matched route R j =[R j1 ,R j2 ,R j3 R js ], where s represents the number of road segments in the route, R ji Indicates route R j The segment id (1≤i≤s) of the i-th segment in R j1 The endpoint is Rj2 The starting point, R j2 The endpoint is R j3 The starting point, and so on. Each segment R of this route... ji Road types are encoded into one-dimensional vectors using one-hot encoding. Where d nt =17, representing the vector dimension of the road segment type for a single road segment.

[0067] Road type vector nt of road segment j ′ i The road segments are connected in the order they are connected, and a dense layer is used to reduce the dimensionality of the sparse attribute vector to form a road type representation vector:

[0068]

[0069]

[0070] This vector describes the generated trajectory R of the vehicle after route matching. j The type of road in which the vehicle travels.

[0071] The length vector is represented as Where nl ji Indicates route R j The length of the i-th road segment (1≤i≤s).

[0072] Similarly, the vectors representing slope, number of traffic lights, and number of points of interest are respectively:

[0073]

[0074]

[0075]

[0076] The road condition representation vector is given by nt j ,nl j ,ns j ,ng j ,np j Connect the representations and reduce the dimensionality using linear layers:

[0077]

[0078] Where, d context This represents the dimension of the compressed vector.

[0079] 2) Geographical region

[0080] GPS data only reflects instantaneous driving conditions and cannot capture the high-level geographic semantics of a trajectory, such as origin, destination, and driving area. A geographic semantic representation is needed. The geographic area covered by the trajectory is divided into an N×N grid. For each trajectory Rj, a geographic semantic representation matrix M is computed. j If the trajectory Rj intersects the grid [a, b], then M j [a, b] = 1; otherwise M j [a, b] = 0:

[0081] Flattened matrix M j This is then fed into a linear layer to reduce dimensionality, forming a geographic semantic representation vector:

[0082] Where, d context This represents the dimension of the compressed vector.

[0083] 3) Traffic conditions.

[0084] Besides road conditions, another factor that significantly impacts driving activity is real-time traffic conditions. Considering the instantaneous motion of a vehicle and the surrounding traffic conditions allows for a better assessment of driver behavior. Therefore, relative speed (calculated as the ratio between the vehicle's speed and the average speed of the segment in which the vehicle is positioned) is used to represent traffic conditions.

[0085] Real-time traffic conditions are estimated using all available GPS data; for each road segment, its traffic condition can be approximated by the average speed of all vehicles passing through within a time period; all GPS records are classified into road segments based on map matching results. For a given road segment, its average speed is calculated using GPS records falling within that time period. Due to data sparsity, complete traffic conditions for the entire road network across all time periods cannot be obtained. For missing values, spatiotemporal interpolation is used to infer the traffic conditions of uncovered road segments by leveraging the inherent traffic correlations between roads. After obtaining the traffic conditions for all road segments, the number of road segments R covered by the travel route is calculated. js The relative speed is used to form a traffic condition representation based on the relative speed.

[0086] Calculate the road segment R covered by the driving route js The process of relative velocity is as follows:

[0087] After obtaining the traffic conditions for all road segments, calculate the road segment R covered by the travel route. js Relative velocity: Where v js vehicle Indicates that the vehicle is on trajectory R j The driving speed of the s-th road segment, trajectory R jThe s-th road segment has the corresponding road segment ID in the road network as follows: This indicates that at time t, all vehicles are traveling on road segment R. js The average vehicle speed.

[0088] Trajectory R j velocity vector v j relative The relative speed of each road segment is composed of the relative speeds in the order of the road segments, and this is input into a linear layer to reduce the dimensionality, forming a traffic condition representation vector h based on relative speed. j traffic :

[0089] Where s represents the trajectory R j The number of middle road sections, d context This represents the dimension of the compressed vector.

[0090] 4) Weather conditions.

[0091] Weather conditions can also affect driving behavior. We use temperature, humidity, visibility, and precipitation to represent weather conditions, specifically using four vectors: temp, hum, vis, and prec to represent temperature, humidity, visibility, and precipitation respectively. These vectors encode the attributes of each road segment. Where s represents the trajectory R j The number of middle road sections, temp j1 This indicates the weather temperature when the vehicle passes through road segment s. Vectors can be represented similarly.

[0092] The weather condition representation vector is composed of temp, hum, vis, and prec, and is then input into a linear layer to reduce its dimensionality. Where, d context This represents the dimension of the compressed vector.

[0093] When the above four types of information in the driving environment are ready, they are concatenated into a vector, which is then fed into the dense layer to form the driving environment representation vector. To be compatible with the statistical feature matrix, it is reshaped into a driving environment feature matrix of the same size as the statistical feature matrix; specifically, when the above four types of information in the driving environment are ready, they are concatenated into a two-dimensional matrix E. context′ : This is fed into a dense layer to form a driving environment representation vector. To be compatible with the statistical feature matrix, it is reshaped into a driving environment feature matrix of the same size as the statistical feature matrix.

[0094]

[0095] (3) Perform convolutional learning on driving behavior statistics and driving environment information to form a driving style representation vector.

[0096] like Figure 2 As shown in the fusion convolution module, a series of convolutional layers with Leaky ReLU are used to process the input driving behavior statistics matrix and driving environment feature matrix. After a series of convolutions, the driving behavior statistics matrix and driving environment feature matrix are converted into a feature tensor, which is flattened and input into a dense layer to derive the driving style representation vector h. style ,Right now in, This represents the dimension of the driving style vector.

[0097] Remaining route fuel consumption prediction section

[0098] like Figure 3 The remaining route fuel consumption prediction section, as shown, utilizes a deep fusion network to calculate driving data and driving environment characteristics of the already traveled and remaining routes. It integrates static information such as vehicle configuration and historical driving styles learned in the first part to improve the accuracy of fuel consumption prediction. The specific process is as follows:

[0099] (1) Calculate driving behavior statistics and driving environment information of the route already traveled, and integrate static information such as historical driving style and vehicle configuration to form the characteristics of the route already traveled.

[0100] For the route already traveled, the sensors transmit vehicle driving data every Δt time interval. This data can be processed to extract speed, acceleration, RPM, torque, and instantaneous fuel consumption. Therefore, the driving behavior vector of the vehicle at time t can be obtained. It can be represented as:

[0101] In addition to the above, external factors such as road conditions, time, and weather can also affect fuel consumption during vehicle operation. From the road environment perspective, factors such as road type, length, slope, and the number of traffic lights on that road segment were selected. tl ), number of poi (num) poi Five characteristics, represented as follows: The time dimension extracts two features: whether the current day is a weekday and whether the current time period is a peak traffic period, represented as follows: The weather dimension extracts the temperature, humidity, visibility, and cumulative precipitation over 6 hours for the corresponding time, represented as follows: Driving environment characteristics are composed of road environment, time and weather characteristics, and are represented as follows:

[0102] When learning the characteristics of a route that has already been traveled, it is necessary to consider not only dynamic data such as the driving environment and driving behavior of the vehicle when it is traveling on that road, but also static data such as the driver's historical driving style representation vector and vehicle configuration.

[0103] Vehicle configuration reflects the vehicle's basic attributes, including platform type, vehicle type, engine type, clutch type, drive type, transmission type, rear axle type, frame type, wheelbase type, and tire type. Each attribute is processed using one-hot encoding. (The vehicle configuration information is then displayed as h.) vehicle Represented as: h vehicle ={platform,vehicle_type,engine,clutch,drive,transmission,rear_axle,frame,wheelbase,wheel}.

[0104] For the aforementioned different types of features, a recurrent model is used to learn latent representations of driving behavior and the driving environment along the traveled route. The recurrent part consists of a transformer layer, a concatenate layer, and a GRU layer. The transformer layer highlights important periods in the input time-series data through an attention mechanism. The concatenate layer integrates static features (including vehicle configuration features h) into the data. vehicle and historical driving style characteristics h style ) and dynamic time-varying characteristics (including driving behavior) and driving environment This is combined with the GRU layer, which aims to capture temporal patterns in the input data. GRU is an improved version of recurrent neural networks (RNNs), chosen here because it solves the vanishing gradient problem and achieves optimal performance in sequential modeling.

[0105] Specifically, the cyclical model can be expressed as:

[0106] TM(F) = Transformer(F,Θ) TM )

[0107] Rec(F) = GRU(concat([TM(F),h))) style ,h vehicle ]),ΘGRU )

[0108] Where F represents the aforementioned set of driving behavior and driving environment features, Θ TM For the transformer parameter set, Θ GRU This is the GRU parameter set, including the number of hidden layers, the number of encoders, etc. The setting of these parameters may affect the performance of the model.

[0109] Based on the above formula, the cyclic part, which processes the driving behavior and driving environment characteristics of the already traveled route, can be described as follows:

[0110]

[0111]

[0112] Where, d behavior and d context These represent the dimensions of the corresponding vectors, and the driving behavior feature set E of the traveled route. behavior Depend on form:

[0113] Indicates the vehicle at t i The driving behavior vector at time n t This indicates the number of data records transmitted back by sensors along the route the vehicle has traveled.

[0114] Similarly, the driving behavior feature set E of the already traveled route context Depend on form:

[0115] Indicates the vehicle at t i The driving environment vector at any given moment.

[0116] Ultimately, the characteristics of the route already traveled are composed of the driving behavior and driving environment characteristics along the route already traveled, and are described as follows:

[0117]

[0118] (2) Calculate the driving environment information of the remaining route, integrate static information such as historical driving style and vehicle configuration, and form the characteristics of the remaining route.

[0119] The remaining route represents the planned route the vehicle will enter, for which driving behavior data has not yet been generated. Furthermore, the actual time and weather conditions at the time of entry are uncertain. Therefore, only the road environment is processed for the remaining route. The processing of the road environment is similar to that of the already traveled route, and the characteristics of the remaining route are represented as follows:

[0120]

[0121] Where, d remain Represents the dimension of the corresponding vector, and the remaining route feature set E road Depend on form:

[0122] Indicates the remaining route segment l i The road environment vector, n l This indicates the number of road segments remaining on the route.

[0123] (3) The similarity between the remaining route and the already traveled route is used to apply attention weights to the two route features, and a multilayer perceptron is used to predict the vehicle's fuel consumption on the remaining route.

[0124] For the already traveled route feature h traveled and remaining route features h remain It is necessary to calculate the road attribute similarity 'a' between the route already traveled and the remaining route:

[0125]

[0126] in, This indicates the road environment characteristics of segment i in the route already traveled. Let represent the road environment characteristics of segment j in the remaining route, s represent the total number of segments in the route, and s′ represent the total number of segments in the remaining route.

[0127] If the number of road segments traveled by the vehicle while it is still in motion is less than s, fill the vector with 0 values.

[0128] The calculation method is as follows:

[0129] For length, slope, num tl ,num poi After normalization,

[0130] in

[0131] After obtaining the similarity, the two feature sets are weighted and summed using the similarity α. Finally, the fuel consumption of the commercial vehicle on the remaining route is predicted through a fully connected layer. Specifically, this can be expressed as:

[0132]

[0133] This embodiment fully considers historical driving behavior and the driving environment in which that behavior occurred. It learns driving style representations from historical driving data using a convolutional network; it utilizes a deep fusion network to process different types of fuel consumption influencing factors, extracting comprehensive and effective fuel consumption features; it calculates driving behavior statistics and driving environment information for the already driven route, uses a Transformer to capture time dependencies, and fuses static information such as historical driving style and vehicle configuration to form features for the already driven route; it calculates driving environment information for the remaining route, uses a Transformer to capture time dependencies, and fuses static information such as historical driving style, vehicle configuration, and geographic region encoding to form features for the remaining route; and it uses the similarity between the remaining route and the already driven route to perform attention-weighted prediction on the two parts of the route features, improving the accuracy of fuel consumption prediction.

[0134] Example 2

[0135] Embodiment 2 of this disclosure introduces a vehicle remaining route fuel consumption prediction system based on driving style.

[0136] like Figure 4 The system shown is a vehicle remaining route fuel consumption prediction system based on driving style, comprising:

[0137] The vector acquisition module is configured to acquire driving behavior data and driving environment information of the vehicle's traveled route to obtain a driving style representation vector; wherein, the driving style representation vector includes vehicle fuel consumption;

[0138] The feature acquisition module is configured to obtain the features of the traveled route based on the formed driving style representation vector and vehicle configuration information; obtain the driving environment information of the remaining route, and combine it with the driving style representation vector and vehicle configuration information to obtain the features of the remaining route.

[0139] The fuel consumption prediction module is configured to calculate the similarity between the features of the traveled route and the features of the remaining route, and then perform a weighted summation of the features of the traveled route and the features of the remaining route based on the obtained similarity to complete the fuel consumption prediction for the vehicle traveling on the remaining route.

[0140] The detailed steps are the same as those of the vehicle remaining route fuel consumption prediction method based on driving style provided in Example 1, and will not be repeated here.

[0141] Example 3

[0142] Embodiment 3 of this disclosure provides a computer-readable storage medium.

[0143] A computer-readable storage medium having a program stored thereon that, when executed by a processor, implements the steps of the vehicle remaining route fuel consumption prediction method based on driving style as described in Embodiment 1 of this disclosure.

[0144] The detailed steps are the same as those of the vehicle remaining route fuel consumption prediction method based on driving style provided in Example 1, and will not be repeated here.

[0145] Example 4

[0146] Embodiment 4 of this disclosure provides an electronic device.

[0147] An electronic device includes a memory, a processor, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps in the vehicle remaining route fuel consumption prediction method based on driving style as described in Embodiment 1 of this disclosure.

[0148] The detailed steps are the same as those of the vehicle remaining route fuel consumption prediction method based on driving style provided in Example 1, and will not be repeated here.

[0149] The above description is merely a preferred embodiment of this disclosure and is not intended to limit this disclosure. Various modifications and variations can be made to this disclosure by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

[0150] While the specific embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of this disclosure. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of this disclosure are still within the scope of protection of this disclosure.

Claims

1. A method for predicting remaining fuel consumption of a vehicle based on driving style, characterized in that, include: The driving behavior data and driving environment information of the vehicle's traveled route are obtained to obtain a driving style representation vector; wherein, the driving style representation vector includes vehicle fuel consumption; Based on the generated driving style representation vector and vehicle configuration information, the characteristics of the route already traveled are obtained; Obtain the driving environment information of the remaining route, and combine it with the driving style representation vector and vehicle configuration information to obtain the features of the remaining route; Calculate the similarity between the features of the traveled route and the features of the remaining route. Based on the obtained similarity, perform a weighted sum of the features of the traveled route and the features of the remaining route to complete the fuel consumption prediction of the vehicle on the remaining route. The similarity of road attributes between the already traveled route and the remaining route for: ;in, Indicates the section of the route already traveled. The characteristics of the road environment, Indicates the remaining route segments The characteristics of the road environment, This indicates the number of all road segments in the route. Indicates the number of remaining route segments; The features of the already traveled route and the remaining route are weighted and summed based on similarity, and the fuel consumption of the vehicle on the remaining route is predicted through a fully connected layer. : .

2. The method for predicting vehicle remaining route fuel consumption based on driving style as described in claim 1, characterized in that, based on GPS The trajectory acquires driving behavior statistics at different time windows, captures instantaneous vehicle motion features, and represents the captured instantaneous vehicle motion features in matrix form to obtain driving behavior data of the vehicle's traveled route. The acquired driving environment information includes road conditions, geographical area, traffic conditions, and weather conditions.

3. The method for predicting vehicle remaining route fuel consumption based on driving style as described in claim 1, characterized in that, A statistical matrix is ​​obtained based on the obtained driving behavior data, and an environment matrix is ​​obtained based on the obtained driving environment information. The obtained statistical matrix and environment matrix are fused and convolutionally learned to obtain a driving style representation vector.

4. The method for predicting vehicle remaining route fuel consumption based on driving style as described in claim 1, characterized in that, The features of the traveled route include dynamic data that considers driving environment information and driving behavior during vehicle travel, and static data that considers driving style representation vectors and vehicle configuration information. A recurrent model is employed to learn the potential relationship between features of the traveled route and driving environment information. The recurrent model includes... transformer layer, concatenate Layers and GRU Layer, in concatenate The layer enables the fusion of dynamic and static data to obtain the characteristics of the traveled route.

5. The method for predicting vehicle remaining route fuel consumption based on driving style as described in claim 1, characterized in that, The remaining route characteristics are related to the road environment of the remaining route, and the remaining route characteristics represent... for: ;in, Represents the dimension of the corresponding vector, and the remaining route feature set. have Composition, that is , Indicates the remaining route segments The road environment vector, This indicates the number of road segments remaining on the route.

6. A vehicle remaining route fuel consumption prediction system based on driving style, employing the vehicle remaining route fuel consumption prediction method based on driving style as described in any one of claims 1-5, characterized in that, include: The vector acquisition module is configured to acquire driving behavior data and driving environment information of the vehicle's traveled route to obtain a driving style representation vector; wherein, the driving style representation vector includes vehicle fuel consumption; The feature acquisition module is configured to obtain the features of the traveled route based on the formed driving style representation vector and vehicle configuration information; obtain the driving environment information of the remaining route, and combine it with the driving style representation vector and vehicle configuration information to obtain the features of the remaining route. The fuel consumption prediction module is configured to calculate the similarity between the features of the traveled route and the features of the remaining route, and then perform a weighted summation of the features of the traveled route and the features of the remaining route based on the obtained similarity to complete the fuel consumption prediction for the vehicle traveling on the remaining route.

7. A computer-readable storage medium having a program stored thereon, characterized in that, When the program is executed by the processor, it implements the steps in the vehicle remaining route fuel consumption prediction method based on driving style as described in any one of claims 1-5.

8. An electronic device comprising a memory, a processor, and a program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps in the vehicle remaining route fuel consumption prediction method based on driving style as described in any one of claims 1-5.

Images