A driver behavior profiling method based on long-term sensor data mobile terminal

Abstract

The present invention proposes a driver behavior profiling method based on a mobile terminal of long-term sensor data, including: a mobile terminal and a cloud server. This method collects data from the mobile terminal and sends it to the cloud platform to construct a data set, performs data cleaning and standardization, and sends the processed data to the mobile terminal to calibrate the coordinates based on the Ground truth information acquisition strategy of the mobile terminal inertial sensor; uses a pre-built driving style classification model to divide driving styles and score safety; introduces the VT‑Micro micro fuel consumption model to draw a long-term fuel consumption trend chart and combines the ISO driving smoothness scoring standard to visualize fuel economy; the cloud finally sends the total score, driving mode, and long-term fuel consumption trend to the mobile terminal; the present invention improves the fuel consumption economy problem, helps car owners develop good economic driving habits, and achieves low-carbon driving. The constructed cloud network database provides a data basis for analyzing economic benefits, emission reduction, and other areas such as bad driving behavior.

Description

Technical Field

[0001] The present invention relates to the field of vehicle assisted driving systems, and in particular to a driver behavior profiling method based on a mobile terminal of long-term sensor data; Background Art

[0002] Currently, energy supply is becoming increasingly scarce, while the proportion of energy consumed by global automobile driving is increasing. At the same time, many drivers' bad habits also lead to huge energy waste. While driving, drivers may trigger acceleration due to driving habits, overtaking, lane changing, and other behaviors. The initial speed, target speed, and aggressiveness of the acceleration process during the acceleration process all affect the vehicle's fuel economy. Therefore, it is necessary to provide a green and safe driving assistance system and method.

[0003] Existing green driving assistance systems mostly focus on changing vehicle operating conditions and temporarily correcting driver behavior. These systems require the installation of additional equipment, increasing application costs and hindering widespread adoption. The proposed method relies solely on mobile devices as sensors, collecting long-term driving data with the primary goal of correcting the driver's driving habits. It evaluates the driving process, aiming to optimize driving strategies and habits, and improve driving safety and economy. Summary of the Invention

[0004] To solve the above problems, the present invention proposes a driver behavior profiling method based on long-term sensor data on a mobile terminal to help drivers improve bad driving habits and improve green driving safety.

[0005] The technical solution of the present invention is a driver behavior profiling method based on a mobile terminal of long-term sensing data;

[0006] The long-period sensor data is characterized by comprising: a mobile terminal and a cloud server;

[0007] The mobile terminal and cloud server transmit data via the 5G network;

[0008] The mobile terminal is used to collect initial data on driving behavior and transmit it to the cloud server for processing and storage in the cloud dataset. It also serves as a visual carrier for driver portraits to improve driver behavior habits.

[0009] The cloud server cleans the initial driving behavior data, identifies states, clusters features, and generates styles, ultimately generating a long-term driver profile and feeding it back to the mobile phone.

[0010] The method for generating a driver behavior profile is characterized by comprising the following steps:

[0011] Step 1: The mobile terminal collects data and sends it to the cloud platform to build a data set;

[0012] Step 2: The cloud platform cleans and standardizes the collected data and sends the processed data to the mobile terminal to calibrate the coordinates;

[0013] Step 3: Calibrate the coordinates based on the Ground Truth information acquisition strategy of the mobile inertial sensor;

[0014] Step 4: Use the pre-built driving style classification model and PAA and SAX to classify the driver's driving style, generate a personalized driving mode profile, and score the safety;

[0015] Step 5: Introduce the VT-Micro microscopic fuel consumption model to draw a long-term fuel consumption trend chart and visualize the fuel economy in combination with the ISO ride comfort rating standard;

[0016] Step 6: The cloud integrates the overall score and sends the personalized driving mode and long-term fuel consumption trend to the mobile app for visualization;

[0017] Preferably, the data collection frequency in step 1 is 50 Hz;

[0018] Preferably, the data set in step 1 is:

[0019] data i ={v i ,a xi ,a yi ,a zi ,β xi ,β yi ,β zi , D}

[0020] GPS = {Lo i , La i}

[0021] i∈[1,I]

[0022] Among them, v i represents the instantaneous speed collected at time i, a xi represents the instantaneous acceleration of the x-axis collected at time i, a yi represents the instantaneous acceleration of the y-axis collected at time i, a zi represents the instantaneous acceleration of the z-axis collected at time i, β xi represents the instantaneous angular acceleration around the x-axis collected at time i, β yi Represents the instantaneous angular acceleration around the y-axis β collected at time i zi represents the instantaneous angular acceleration around the z axis collected at time i, D represents the direction collected by the magnetometer when it is aligned with the coordinate, and Loi Indicates the longitude collected by GPS at time i, La i represents the latitude collected by GPS at time i, and I represents the total number of moments in the dataset;

[0023] As a preference, the specific implementation method of the cloud platform in step 2 for cleaning the collected data is as follows:

[0024] Missing values ​​in vehicle movement: Check whether the time interval of adjacent data is 0.02 seconds, that is:

[0025] t i+1 -t i =0.02

[0026] If the time interval is 0.02 seconds, the data is continuous; if the time is discontinuous, the data segments are separated and each data set is processed separately to ensure the continuity and accuracy of the data;

[0027] Invalid value when the vehicle is moving: If the speed V at time i i is 0, and the speed V at time i+1 i+1 The speed at this moment is also 0, that is:

[0028] V i =V i+1 =0

[0029] Then delete the invalid value V i+1 , making the data concise and accurate;

[0030] As a preference, the specific implementation method of the cloud platform in step 2 for standardizing the collected data is as follows:

[0031] The Z-normalization method is introduced to normalize the driving acceleration data of a certain driver so that its mean is 0 and its standard deviation is 1, making it have the same dimension as other driver data:

[0032]

[0033] Preferably, the coordinates in step 3 are defined as:

[0034] The x-axis points to the head of the phone, the y-axis points to the right of the phone, and the z-axis points to the screen, defining the body coordinate system O. v The x-axis points to the front of the vehicle, the y-axis points to the right of the vehicle, and the z-axis points downward to the chassis. The North-East-Earth navigation coordinate system (NED) is used as the inertial reference system O n : The x-axis points due north, the y-axis points due east, and the z-axis points vertically downward;

[0035] The specific coordinate alignment method of the Ground truth information acquisition strategy based on the mobile phone inertial sensor described in step 3 is:

[0036] First, the relationship between the mobile terminal and the NED reference system is solved in real time based on inertial sensors.

[0037] Second, when the mobile terminal and the vehicle are relatively stationary, the posture of the vehicle body relative to the NED reference system is calculated based on the vehicle's driving direction information. Combined with the mobile terminal's attitude relative to the NED reference system Align the mobile device coordinate system with the vehicle body coordinate system to obtain the ground truth information of the posture between the mobile device and the vehicle, and recalibrate before each trip.

[0038] Preferably, the specific principle of the above method is:

[0039] Rotate the mobile coordinate system to the NED coordinate system. According to the principle of three-dimensional vector rotation transformation, in the NED reference coordinate system, the device coordinate system XYZ axes can be expressed as:

[0040]

[0041] in, Represented by the NED coordinate system O n Convert to device coordinate system O p The transformation matrix of Represented by the device coordinate system O p Convert to NED coordinate system O n The transformation matrix is ​​based on the properties of the unit orthogonal matrix:

[0042]

[0043] Among them, the mobile terminal posture is defined as the transformation matrix Since the coordinate system rotation can be decomposed into three rotations around the ZYX axis, the transformation matrices are:

[0044] Rotation ψ around the Z axis:

[0045] Rotation Φ around X axis:

[0046] Rotate θ around the Y axis:

[0047] Where ψ is the yaw angle around the Z axis, Φ is the roll angle around the X axis, and θ is the pitch angle around the Y axis.

[0048] at this time It can be expressed as:

[0049]

[0050] On the contrary, by The formula for converting to attitude angle is:

[0051] Similarly, the posture of the vehicle body relative to the NED reference frame is defined as

[0052] Preferably, the driving style classification model in step 4 is:

[0053] The driving style classification model includes: feature clustering, situational awareness module, and decision fusion submodule;

[0054] The feature clustering method is as follows:

[0055] Extract features from the raw data using the t-SNE (t-Distributed Stochastic Neighbor Embedding) algorithm, obtain labels based on the features and data, and then send them to the context-aware module for special classification;

[0056] The features are defined as: change in acceleration before maneuver, average value of acceleration before maneuver, P feature before maneuver, change in acceleration during maneuver, average value of acceleration during maneuver, change in acceleration during maneuver, average value of acceleration during maneuver, change in acceleration after maneuver, average value of acceleration after maneuver, and P feature after Manuver;

[0057] The situational awareness module is defined as two parts, including: traffic flow classification and initial style classification;

[0058] The traffic flow classification method is as follows:

[0059] The multi-layer perceptron (MLP) is used to detect the GPS average speed and the y-axis acceleration features measured by the accelerometer to classify the network roads into two categories: normal roads and congested roads. At the same time, the traffic classification module is checked using a window with the highest definition accuracy (10s).

[0060] To overcome the inaccuracy of GPS velocity, the Havenrisine equation is introduced:

[0061]

[0062] d=R·v i

[0063] Where d is the distance between two latitude points and R is the radius of the earth;

[0064] The initial style classification includes: K-NN algorithm, SVM algorithm, MLP algorithm to identify whether the driving action style is normal or dangerous;

[0065] The sigmiod function is introduced to integrate and give different weights to the above three algorithms to evaluate the driving style. The sigmiod function is as follows:

[0066]

[0067] Where x, σ, and μ are the mean, variance, and standard deviation of the output values ​​of the classification algorithm;

[0068] The decision fusion submodule includes: Pre-FIS, Final-FIS;

[0069] The Pre-FIS risk assessment process is as follows:

[0070] Pre-FIS uses the C4.5 decision tree algorithm to extract training set data. After 5-FOLD cross-validation and tree training, three thresholds A, B, and C are defined as 0.619, 0.252, and 0.510, respectively. The Pre-FIS output is saved in the driver file and used in Fianl-FIS;

[0071] The Pre-FIS risk assessment flow chart is as follows: Figure 2 As shown:

[0072] Where MR is the driving action output result, CT is the traffic flow output result, A, B, and C are 0.619, 0.252, and 0.510 respectively;

[0073] The Final-FIS risk assessment process is as follows:

[0074] Fianl-FIS receives reports from Pre-FIS and makes judgments on the current driving behavior. In order to prevent the driver's emergency driving behavior from being misjudged as dangerous driving behavior during emergency avoidance, this module adopts four relevant fuzzy rules to support these situations, as shown in Table 1;

[0075] Among them, risk is the output of Pre-FIS, Div.KM is the ratio of dangerous driving actions to normal driving actions saved in the driver's profile; Final-FIS finally evaluates the final driving style by considering the driver's profile;

[0076] Preferably, the method for dividing the driver's driving style based on PAA and SAX in step 4 is:

[0077] Among them, the original data is converted into piecewise aggregate approximation (PAA) representation, and then converted into corresponding strings according to the breakpoint position;

[0078] The string conversion method includes: original data dimensionality reduction, data classification and encoding based on Python;

[0079] The original data dimensionality reduction method is:

[0080]

[0081] Among them, C is the original time series, is the w-dimensional space vector after PAA dimensionality reduction, n is the length of the original time series, i is the number of original time series,

[0082] The data classification and coding method based on Python is:

[0083] The SAX algorithm is used to encode and classify the normalized and PAA-processed acceleration data into 6 levels, each level representing a different acceleration level threshold;

[0084] After the coding is completed, Python is used to create a moving index window with a length of 3 and an increment of 1 each time. The frequency of each driving behavior pattern is counted and each driving behavior is output as a TXT file.

[0085] Preferably, the specific method for generating the personalized driving portrait in step 4 is:

[0086] The first 30 driving modes are extracted as typical driving modes, and acceleration and deceleration are used as two independent driving behavior operations to illustrate the driver's driving behavior characteristics;

[0087] Preferably, the driving safety score in step 4 is:

[0088]

[0089] Among them, i is the order of typical driving mode, freq i is the proportion of the typical driving mode to all driving modes, CODE is the code of the typical driving mode, mean is used to calculate the mean of the driving mode code, std is used to calculate the standard deviation of the driving mode code, and abs represents the absolute value;

[0090] Preferably, the VT-Micro microscopic fuel consumption model in step 5 is:

[0091]

[0092] Among them, MOE e is the instantaneous fuel consumption of the vehicle, v i is the instantaneous speed of the vehicle, ax iThe instantaneous acceleration in the vehicle's front direction, When the vehicle accelerates, the speed v i The power is k, the acceleration ax i When the power of h is reached, the fitting coefficient of the instantaneous fuel consumption model is, When the vehicle is decelerating, the speed v i The power is k, the acceleration a xi The fitting coefficient of the instantaneous fuel consumption model when the power of is h;

[0093] The fitting coefficients can be obtained by querying the model coefficient table;

[0094] Preferably, the ISO ride comfort rating standard in step 5 is:

[0095]

[0096] Where T is the period of vibration analysis, a xi is the acceleration in the vehicle head direction, t is the weighted acceleration time history;

[0097] At the same time, the VT-Micro fuel consumption model is used for scoring:

[0098]

[0099] Preferably, the total score in step 6 is defined as:

[0100] SCORE=0.7*score 安全 +0.2*score 平顺 +0.1*score 经济

[0101] The deviation standardization method is used to standardize the scores of the three different magnitude dimensions mentioned above into dimensionless ones, which are defined as follows:

[0102] For the sequence x1, x2, ..., x n Perform the transformation according to the following formula:

[0103]

[0104] Among them, y i is x i The data after standardization, and y i ∈[0,1], n is the number of sequences, i is the number of normalized data, j is the number of original data, i, j∈(0,n);

[0105] As a preference, the specific method of displaying on the mobile terminal in step 6 is as follows:

[0106] After cloud-based data processing, a fuel consumption curve is generated, first generating a fuel consumption graph for the current driving session. The graph and score are then divided by the driving time to obtain the average fuel consumption, which is then used to plot long-term fuel consumption trends. The fuel consumption graph and score are then sent to the mobile device for visualization, and the long-term fuel consumption trend curve is updated. Bad behaviors are visualized on a map route, with the specific time and behavior displayed as coordinate points on the driver's mobile device to inform the driver of corrective actions.

[0107] The present invention provides a system and method for green driving using driving behavior prediction. This system collects and analyzes vehicle driving data in real time to create an economic profile of driving behavior. This improves vehicle fuel economy and helps drivers develop good driving habits, ultimately contributing to low carbon emissions. The cloud-based network database also provides a data foundation and support for analyzing economic benefits, emissions, and other areas, such as unhealthy driving behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0108] Figure 1 : Schematic diagram of system structure.

[0109] Figure 2 : Flowchart of the method.

[0110] Figure 3 : Pre-FIS risk assessment flow chart. Specific implementation plan

[0111] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0112] A driver behavior profiling method based on long-term sensor data on mobile terminals;

[0113] The specific implementation method flow chart is as follows Figure 1 As shown;

[0114] The long-period sensor data is characterized by including: a mobile WeChat applet and a cloud server;

[0115] The mobile terminal system structure is as follows Figure 2 As shown, it includes a three-axis accelerometer, a magnetometer, a GPS, and a gyroscope sensor;

[0116] The mobile terminal and cloud server transmit data via the 5G network;

[0117] The mobile terminal is used to collect initial data on driving behavior and transmit it to the cloud server for processing and storage in the cloud dataset. It also serves as a visual carrier for driver portraits to improve driver behavior habits.

[0118] The cloud server cleans the initial driving behavior data, identifies states, clusters features, and generates styles, ultimately obtaining a long-term driver profile and feeding it back to the mobile terminal.

[0119] The method for generating a driver behavior profile is characterized by comprising the following steps:

[0120] Step 1: The mobile terminal collects data and sends it to the cloud platform to build a data set;

[0121] The data collection frequency in step 1 is 50 Hz;

[0122] The dataset in step 1 is:

[0123] data i ={v i ,a xi , a yi ,a zi , β xi ,β yi , β zi ,D}

[0124] GPS = {Lo i , La i}

[0125] i∈[1,I]

[0126] Among them, v i represents the instantaneous speed collected at time i, a xi represents the instantaneous acceleration of the x-axis collected at time i, a yi represents the instantaneous acceleration of the y-axis collected at time i, a zi represents the instantaneous acceleration of the z-axis collected at time i, β xi represents the instantaneous angular acceleration around the x-axis collected at time i, β yi Represents the instantaneous angular acceleration around the y-axis β collected at time i zi represents the instantaneous angular acceleration around the z axis collected at time i, D represents the direction collected by the magnetometer when it is aligned with the coordinate, and Lo i Indicates the longitude collected by GPS at time i, La i represents the latitude collected by GPS at time i, and I represents the total number of moments in the dataset;

[0127] Step 2: The cloud platform cleans and standardizes the collected data and sends the processed data to the mobile phone to calibrate the coordinates;

[0128] In step 2, the cloud platform performs data cleaning on the collected data. The specific implementation method is as follows:

[0129] Missing values ​​in vehicle movement: Check whether the time interval of adjacent data is 0.02 seconds, that is:

[0130] t i+1 -t i =0.02

[0131] If the time interval is 0.02 seconds, the data is continuous; if the time is discontinuous, the data segments are separated and each data set is processed separately to ensure the continuity and accuracy of the data;

[0132] Invalid value when the vehicle is moving: If the speed V at time i i is 0, and the speed V at time i+1 i+1 The speed at this moment is also 0, that is:

[0133] V i =V i+1 =0

[0134] Then delete the invalid value V i+1 , making the data concise and accurate;

[0135] In step 2, the cloud platform performs data standardization on the collected data. The specific implementation method is as follows:

[0136] The Z-normalization method is introduced to normalize the driving acceleration data of a certain driver so that its mean is 0 and its standard deviation is 1, making it have the same dimension as other driver data:

[0137]

[0138] Step 3: Calibrate the coordinates based on the Ground Truth information acquisition strategy of the mobile inertial sensor;

[0139] The coordinates in step 3 are defined as:

[0140] The x-axis points to the head of the mobile terminal, the y-axis points to the right of the mobile phone, and the z-axis points to the screen, defining the body coordinate system O v The x-axis points to the front of the vehicle, the y-axis points to the right of the vehicle, and the z-axis points downward to the chassis. The North-East-Earth navigation coordinate system (NED) is used as the inertial reference system O n : The x-axis points due north, the y-axis points due east, and the z-axis points vertically downward;

[0141] The specific coordinate alignment method of the Ground truth information acquisition strategy based on the mobile inertial sensor in step 3 is:

[0142] Real-time calculation of the relationship between the mobile phone and the NED reference system based on inertial sensors

[0143] When the mobile terminal and the vehicle are relatively stationary, the posture of the vehicle body relative to the NED reference system is calculated based on the vehicle's driving direction information. Combined with the mobile terminal's attitude relative to the NED reference system Align the mobile device coordinate system with the vehicle body coordinate system to obtain the ground truth information of the posture between the mobile device and the vehicle, and recalibrate before each trip.

[0144] The specific principles of the above method are:

[0145] Rotate the mobile coordinate system to the NED coordinate system. According to the principle of three-dimensional vector rotation transformation, in the NED reference coordinate system, the device coordinate system XYZ axes can be expressed as:

[0146]

[0147] in, Represented by the NED coordinate system O n Convert to device coordinate system O p The transformation matrix of Represented by the device coordinate system O p Convert to NED coordinate system O n The transformation matrix is ​​based on the properties of the unit orthogonal matrix:

[0148]

[0149] Among them, the mobile terminal posture is defined as the transformation matrix Since the coordinate system rotation can be decomposed into three rotations around the ZYX axis, the transformation matrices are:

[0150] Rotation ψ around the Z axis:

[0151] Rotation Φ around X axis:

[0152] Rotate θ around the Y axis:

[0153] Where ψ is the yaw angle around the Z axis, Φ is the roll angle around the X axis, and θ is the pitch angle around the Y axis.

[0154] at this time It can be expressed as:

[0155]

[0156] On the contrary, by The formula for converting to attitude angle is:

[0157] Similarly, the posture of the vehicle body relative to the NED reference frame is defined as

[0158] Step 4: Use the pre-built driving style classification model and PAA and SAX to classify the driver's driving style, generate a personalized driving mode profile, and score the safety;

[0159] The driving style classification model in step 4 is:

[0160] The driving style classification model includes: feature clustering, situational awareness module, and decision fusion submodule;

[0161] The feature clustering method is as follows:

[0162] Extract features from the raw data using the t-SNE (t-Distributed Stochastic Neighbor Embedding) algorithm, obtain labels based on the features and data, and then send them to the context-aware module for special classification;

[0163] The features are defined as: change in acceleration before maneuver, average value of acceleration before maneuver, P feature before maneuver, change in acceleration during maneuver, average value of acceleration during maneuver, change in acceleration during maneuver, average value of acceleration during maneuver, change in acceleration after maneuver, average value of acceleration after maneuver, and P feature after Manuver;

[0164] The situational awareness module is defined as two parts, including: traffic flow classification and initial style classification;

[0165] The traffic flow classification method is as follows:

[0166] The multi-layer perceptron (MLP) is used to detect the GPS average speed and the y-axis acceleration features measured by the accelerometer to classify the network roads into two categories: normal roads and congested roads. At the same time, the traffic classification module is checked using a window with the highest definition accuracy (10s).

[0167] To overcome the inaccuracy of GPS velocity, the Havenrisine equation is introduced:

[0168]

[0169] d=R·v i

[0170] Where d is the distance between two latitude points and R is the radius of the earth;

[0171] The initial style classification includes: K-NN algorithm, SVM algorithm, MLP algorithm to identify whether the driving action style is normal or dangerous;

[0172] The sigmiod function is introduced to integrate and give different weights to the above three algorithms to evaluate the driving style. The sigmiod function is as follows:

[0173]

[0174] Where x, σ, and μ are the mean, variance, and standard deviation of the output values ​​of the classification algorithm;

[0175] The decision fusion submodule includes: Pre-FIS, Final-FIS;

[0176] The Pre-FIS risk assessment process is as follows:

[0177] Pre-FIS uses the C4.5 decision tree algorithm to extract training set data. After 5-FOLD cross-validation and tree training, three thresholds A, B, and C are defined as 0.619, 0.252, and 0.510, respectively. The Pre-FIS output is saved in the driver file and used in Fianl-FIS;

[0178] The Pre-FIS risk assessment flow chart is as follows: Figure 3 As shown:

[0179] Where MR is the driving action output result, CT is the traffic flow output result, A, B, and C are 0.619, 0.252, and 0.510 respectively;

[0180] The Final-FIS risk assessment process is shown in Table 1, specifically:

[0181] Table 1 Final judging rules of Fianl-FIS

[0182]

[0183] Fianl-FIS receives reports from Pre-FIS and makes judgments on the current driving behavior. In order to prevent the driver's emergency driving behavior from being misjudged as dangerous driving behavior during emergency avoidance, this module adopts four relevant fuzzy rules to support these situations, as shown in Table 1;

[0184] Among them, risk is the output of Pre-FIS, Div.KM is the ratio of dangerous driving actions to normal driving actions saved in the driver's profile; Final-FIS finally evaluates the final driving style by considering the driver's profile;

[0185] The method for classifying the driver's driving style based on PAA and SAX in step 4 is as follows:

[0186] Among them, the original data is converted into the piecewise aggregate approximation PAA representation, and then converted into the corresponding string according to the breakpoint position;

[0187] The string conversion method includes: original data dimensionality reduction, data classification and encoding based on Python;

[0188] The original data dimensionality reduction method is:

[0189]

[0190] Among them, C is the original time series, is the w-dimensional space vector after PAA dimensionality reduction, n is the length of the original time series, i is the number of original time series,

[0191] The data classification and coding method based on Python is:

[0192] The SAX algorithm is used to encode and classify the normalized and PAA-processed acceleration data into 6 levels, each level representing a different acceleration level threshold;

[0193] After the coding is completed, Python is used to create a moving index window with a length of 3 and an increment of 1 each time. The frequency of each driving behavior pattern is counted and each driving behavior is output as a TXT file.

[0194] The specific method for generating the personalized driving profile described in step 4 is as follows:

[0195] The first 30 driving modes are extracted as typical driving modes, and acceleration and deceleration are used as two independent driving behavior operations to illustrate the driver's driving behavior characteristics;

[0196] The driving safety score in step 4 is:

[0197]

[0198] Among them, i is the order of typical driving mode, freq i is the proportion of the typical driving mode to all driving modes, CODE is the code of the typical driving mode, mean is used to calculate the mean of the driving mode code, std is used to calculate the standard deviation of the driving mode code, and abs represents the absolute value;

[0199] Preferably, the VT-Micro microscopic fuel consumption model in step 5 is:

[0200]

[0201] Among them, MOE e is the instantaneous fuel consumption of the vehicle, v i is the instantaneous speed of the vehicle, ax i The instantaneous acceleration in the vehicle's front direction, When the vehicle accelerates, the speed v i The power is k, the acceleration ax i When the power of h is reached, the fitting coefficient of the instantaneous fuel consumption model is, When the vehicle is decelerating, the speed v i The power is k, the acceleration a xi The fitting coefficient of the instantaneous fuel consumption model when the power of is h;

[0202] The fitting coefficients can be obtained by querying the model coefficient table;

[0203] Step 5: Introduce the VT-Micro microscopic fuel consumption model to draw a long-term fuel consumption trend chart and visualize the fuel economy in combination with the ISO ride comfort rating standard;

[0204] The ISO ride comfort rating criteria described in step 5 are:

[0205]

[0206] Where T is the period of vibration analysis, a xi is the acceleration in the vehicle head direction, t is the weighted acceleration time history;

[0207] At the same time, the VT-Micro fuel consumption model is used for scoring:

[0208]

[0209] Step 6: The cloud integrates the total score and sends the personalized driving mode and long-term fuel consumption trend to the mobile app for visualization;

[0210] The total score in step 6 is defined as:

[0211] SCORE=0.7*score 安全 +0.2*score 平顺 +0.1*score 经济

[0212] The deviation standardization method is used to standardize the scores of the three different magnitude dimensions mentioned above into dimensionless ones, which are defined as follows:

[0213] For the sequence x1, x2, ..., x n Perform the transformation according to the following formula:

[0214]

[0215] Among them, y i is x i The data after standardization, and y i ∈[0,1], n is the number of sequences, i is the number of normalized data, j is the number of original data, i, j∈(0,n);

[0216] The specific method of displaying on the mobile phone in step 6 is as follows:

[0217] After cloud-based data processing, a fuel consumption curve is generated, first generating a fuel consumption graph for the current driving session. The graph and score are then divided by the driving time to obtain the average fuel consumption, which is then used to plot long-term fuel consumption trends. The fuel consumption graph and score are then sent to the mobile app for visualization and update of the long-term fuel consumption trend curve. Bad behaviors are visualized on a map route, with the specific time and behavior displayed as coordinate points on the driver's phone to inform the driver of corrective actions.

[0218] The present invention provides a system and method for green driving using driving behavior prediction. This system collects and analyzes vehicle driving data in real time to create an economic profile of driving behavior. This improves vehicle fuel economy and helps drivers develop good driving habits, ultimately contributing to low carbon emissions. The cloud-based network database also provides a data foundation and support for analyzing economic benefits, emissions, and other areas, such as unhealthy driving behavior.

[0219] It should be understood that parts not elaborated in detail in this specification belong to the prior art.

[0220] Although this document frequently uses the terms "mobile terminal" and "cloud server," the use of other terms is not excluded. These terms are used solely to more conveniently describe the essence of the present invention, and interpreting them as any additional limitations is contrary to the spirit of the present invention.

[0221] It should be understood that the above description of the preferred embodiment is relatively detailed and cannot be regarded as limiting the scope of protection of the patent of the present invention. Under the guidance of the present invention, ordinary technicians in this field can also make substitutions or modifications without departing from the scope of protection of the claims of the present invention, which all fall within the scope of protection of the present invention. The scope of protection requested by the present invention shall be based on the attached claims.

Claims

1. A driver behavior profiling method based on long-term sensor data mobile terminal, characterized in that: The long-period sensor data mobile terminal includes: a mobile terminal and a cloud server; The mobile terminal and cloud server transmit data via 5G network; The mobile terminal is used to collect initial data on driving behavior and transmit it to the cloud server for processing and storage in the cloud dataset. It also serves as a visual carrier for driver portraits to improve driver behavior habits. The cloud server cleans the initial driving behavior data, identifies states, clusters features, and generates styles, ultimately generating a long-term driver profile and feeding it back to the mobile phone. The method for generating a driver behavior profile comprises the following steps: Step 1: The mobile terminal collects data and sends it to the cloud platform to build a data set; Step 2: The cloud platform cleans and standardizes the collected data and sends the processed data to the mobile terminal to calibrate the coordinates; Step 3: Calibrate the coordinates based on the Ground Truth information acquisition strategy of the mobile inertial sensor; Step 4: Use the pre-built driving style classification model and PAA and SAX to classify the driver's driving style, generate a personalized driving mode profile, and score the safety; The driving style classification model includes: feature clustering, situational awareness module, and decision fusion submodule; The method for classifying the driver's driving style based on PAA and SAX in step 4 is as follows: Among them, the original data is converted into the piecewise aggregate approximation PAA representation, and then converted into the corresponding string according to the breakpoint position; The string conversion method includes: original data dimensionality reduction, data classification and encoding based on Python; The original data dimensionality reduction method is: Among them, C is the original time series, is the w-dimensional space vector after PAA dimensionality reduction, n is the length of the original time series, i is the number of original time series, The data classification and coding method based on Python is: The SAX algorithm is used to encode and classify the normalized and PAA-processed acceleration data into 6 levels, each level representing a different acceleration level threshold; After the coding is completed, Python is used to create a moving index window with a length of 3 and an increment of 1 each time. The frequency of each driving behavior pattern is counted and each driving behavior is output as a TXT file. The specific method for generating the personalized driving profile described in step 4 is as follows: The first 30 driving modes are extracted as typical driving modes, and acceleration and deceleration are used as two independent driving behavior operations to illustrate the driver's driving behavior characteristics; The driving safety score in step 4 is: Among them, i is the order of typical driving mode, freq i is the proportion of the typical driving mode to all driving modes, CODE is the code of the typical driving mode, mean is used to calculate the mean of the driving mode code, std is used to calculate the standard deviation of the driving mode code, and abs represents the absolute value; Step 5: Introduce the VT-Micro microscopic fuel consumption model to draw a long-term fuel consumption trend chart and visualize the fuel economy in combination with the ISO ride comfort rating standard; Step 6: The total score is integrated in the cloud, and the personalized driving mode and long-term fuel consumption trend are sent to the mobile application for visualization.

2. The driver behavior profiling method based on long-term sensor data mobile terminal according to claim 1 is characterized in that: The data collection frequency in step 1 is 50 Hz; The dataset in step 1 is: data i ={v i ,a xi ,a yi ,a zi ,b xi ,b yi ,b zi ,D} GPS={The i ,The i } i∈[1,I] Among them, v i represents the instantaneous speed collected at time i, a xi represents the instantaneous acceleration of the x-axis collected at time i, a yi represents the instantaneous acceleration of the y-axis collected at time i, a zi represents the instantaneous acceleration of the z-axis collected at time i, β xi represents the instantaneous angular acceleration around the x-axis collected at time i, β yi Represents the instantaneous angular acceleration around the y-axis β collected at time i zi represents the instantaneous angular acceleration around the z axis collected at time i, D represents the direction collected by the magnetometer when it is aligned with the coordinate, and Lo i Indicates the longitude collected by GPS at time i, La i represents the latitude collected by GPS at time i, and I represents the total number of moments in the dataset.

3. The driver behavior profiling method based on long-term sensor data mobile terminal according to claim 1 is characterized in that: In step 2, the cloud platform performs data cleaning on the collected data. The specific implementation method is as follows: Missing values ​​in vehicle movement: Check whether the time interval of adjacent data is 0.02 seconds, that is: t i+1 -t i =0.02 If the time interval is 0.02 seconds, the data is continuous; if the time is discontinuous, the data segments are separated and each data set is processed separately to ensure the continuity and accuracy of the data; Invalid value when the vehicle is moving: If the speed V at time i i is 0, and the speed V at time i+1 i+1 The speed at this moment is also 0, that is: V i =V i+1 =0 Then delete the invalid value V i+1 , making the data concise and accurate; In step 2, the cloud platform performs data standardization on the collected data. The specific implementation method is as follows: The Z-normalization method is introduced to normalize the driving acceleration data of a certain driver so that its mean is 0 and its standard deviation is 1, making it have the same dimension as other driver data:

4. The driver behavior profiling method based on long-term sensor data mobile terminal according to claim 1 is characterized in that: The coordinates in step 3 are defined as: The x-axis points to the head of the phone, the y-axis points to the right of the phone, and the z-axis points to the screen, defining the body coordinate system O. v The x-axis points to the front of the vehicle, the y-axis points to the right of the vehicle, and the z-axis points downward to the chassis. The North-East-Earth navigation coordinate system NED is used as the inertial reference system O. n : The x-axis points due north, the y-axis points due east, and the z-axis points vertically downward; The specific coordinate alignment method of the Ground truth information acquisition strategy based on the mobile phone inertial sensor described in step 3 is: First, the relationship between the mobile terminal and the NED reference system is solved in real time based on inertial sensors. Second, when the mobile terminal and the vehicle are relatively stationary, the posture of the vehicle body relative to the NED reference system is calculated based on the vehicle's driving direction information. Combined with the mobile terminal's attitude relative to the NED reference system Align the mobile device coordinate system with the vehicle body coordinate system to obtain the ground truth information of the posture between the mobile device and the vehicle, and recalibrate before each trip; The specific principles of the above method are: Rotate the mobile coordinate system to the NED coordinate system. According to the principle of three-dimensional vector rotation transformation, in the NED reference coordinate system, the device coordinate system XYZ axes can be expressed as: in, Represented by the NED coordinate system O n Convert to device coordinate system O p The transformation matrix of Represented by the device coordinate system O p Convert to NED coordinate system O n The transformation matrix is ​​based on the properties of the unit orthogonal matrix: Among them, the mobile terminal posture is defined as the transformation matrix Since the coordinate system rotation can be decomposed into three rotations around the ZYX axis, the transformation matrices are: Rotation ψ around the Z axis: Rotation Φ around X axis: Rotate θ around the Y axis: Where ψ is the yaw angle around the Z axis, Φ is the roll angle around the X axis, and θ is the pitch angle around the Y axis. at this time It can be expressed as: On the contrary, by The formula for converting to attitude angle is: Similarly, the posture of the vehicle body relative to the NED reference frame is defined as 5. The driver behavior profiling method based on long-term sensor data mobile terminal according to claim 1 is characterized in that: The driving style classification model described in step 4 is: The feature clustering method is as follows: Based on t-SNE, t-Distributed Stochastic Neighbor Embedding, the algorithm extracts features from the raw data, obtains labels based on the features and the data, and then sends them to the context-aware module for special classification; The features are defined as: change in acceleration before maneuver, average value of acceleration before maneuver, P feature before maneuver, change in acceleration during maneuver, average value of acceleration during maneuver, change in acceleration during maneuver, average value of acceleration during maneuver, change in acceleration after maneuver, average value of acceleration after maneuver, and P feature after Manuver; The situational awareness module is defined as two parts, including: traffic flow classification and initial style classification; The traffic flow classification method is as follows: The multi-layer perceptron (MLP) is used to detect the GPS average speed and the Y-axis acceleration features measured by the accelerometer, and the network roads are classified into two categories: normal roads and congested roads. At the same time, the traffic classification module is checked using a window of 10 seconds with the highest accuracy. To overcome the inaccuracy of GPS velocity, the Havenrisine equation is introduced: d=R·v i Where d is the distance between two latitude points and R is the radius of the earth; The initial style classification includes: K-NN algorithm, SVM algorithm, MLP algorithm to identify whether the driving action style is normal or dangerous; The sigmiod function is introduced to integrate and give different weights to the above three algorithms to evaluate the driving style. The sigmiod function is as follows: Where x, σ, and μ are the mean, variance, and standard deviation of the output values ​​of the classification algorithm; The decision fusion submodule includes: Pre-FIS, Final-FIS; The Pre-FIS risk assessment process is as follows: Pre-FIS uses the C4.5 decision tree algorithm to extract training set data. After 5-FOLD cross-validation and tree training, three thresholds A, B, and C are defined as 0.619, 0.252, and 0.510, respectively. The Pre-FIS output is saved in the driver file and used in Fianl-FIS; Among them, MR is the driving action output result, CT is the traffic flow output result, A, B, and C are 0.619, 0.252, and 0.510 respectively; The Final-FIS risk assessment process is as follows: Fianl-FIS receives reports from Pre-FIS and makes judgments on current driving behavior. To prevent the driver's emergency driving behavior from being misjudged as dangerous driving behavior during emergency avoidance, this module adopts four related fuzzy rules to support these situations. Among them, risk is the output of Pre-FIS, Div.KM is the ratio of dangerous driving actions to normal driving actions saved in the driver's file; Final-FIS finally evaluates the final driving style by considering the driver's file.

6. The driver behavior profiling method based on long-term sensor data mobile terminal according to claim 1 is characterized in that: The VT-Micro microscopic fuel consumption model described in step 5 is: Among them, MOE e is the instantaneous fuel consumption of the vehicle, v i is the instantaneous speed of the vehicle, a xi The instantaneous acceleration in the vehicle's front direction, When the vehicle accelerates, the speed v i The power is k, the acceleration a xi When the power of h is the fitting coefficient of the instantaneous fuel consumption model, When the vehicle is decelerating, the speed v i The power is k, the acceleration a xi The fitting coefficient of the instantaneous fuel consumption model when the power of is h; The fitting coefficients can be obtained by querying the model coefficient table; The ISO ride comfort rating criteria described in step 5 are: Where T is the period of vibration analysis, a xi is the acceleration in the vehicle head direction, t is the weighted acceleration time history; At the same time, the VT-Micro fuel consumption model is used for scoring: The total score in step 6 is defined as: SCORE=0.7*score 安全 +0.2*score 平顺 +0.1*score 经济 The deviation standardization method is used to standardize the scores of the three different magnitude dimensions mentioned above into dimensionless ones, which are defined as follows: For the sequence x1,x2,…,x n Perform the transformation according to the following formula: Among them, y i is x i The data after standardization, and y i ∈[0,1], n is the number of sequences, i is the number of normalized data, j is the number of original data, ij∈(0,n).

7. The driver behavior profiling method based on long-term sensor data mobile terminal according to claim 1 is characterized in that: The specific method of mobile terminal display in step 6 is as follows: After cloud data processing, a fuel consumption curve is formed. The fuel consumption graph of the current driving is first generated, and the average fuel consumption is obtained by integrating and dividing by the driving time as data for drawing long-term fuel consumption trends; the fuel consumption graph and score are sent to the mobile terminal for visualization and the long-term fuel consumption trend curve is updated; the marked bad behaviors are visualized through the map route, and the specific time and behavior are reflected in the owner's mobile terminal as coordinate points to inform the owner to correct the direction.

Images