Intelligent networked automobile safety situation assessment method based on multi-source information fusion

Abstract

The invention discloses an intelligent networked automobile safety situation assessment method based on multi-source information fusion. The method comprises the following steps: 1) acquiring training parameters of a speed prediction model; 2) training a neural network by using the speed prediction model training parameters, and constructing a speed prediction model; 3) predicting the speed of the vehicle at the next moment by using the speed prediction model; 4) acquiring driving safety information, and processing the safety information to obtain a safety situation assessment parameter; and 5) processing the safety situation assessment parameter to obtain a current safety situation assessment value so as to determine the driving safety state of the vehicle. According to the method, the self-vehicle speed prediction model and the driving safety situation evaluation model are constructed, and the importance of the safety situation value on the driving safety of the intelligent networked vehicle is clarified by taking the prediction of the self-vehicle speed as a link, so that theoretical and technical support is provided for realizing the safe driving technology of the intelligent networked vehicle.

Description

Technical Field

[0001] This invention relates to the field of vehicle situation recognition technology, specifically a method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion. Background Technology

[0002] The rapid development of the automotive industry has placed higher demands on driving safety, making the improvement of vehicle driving safety an urgent issue. Intelligent connected vehicles, which are more intelligent and connected than traditional vehicles, can be an effective way to improve driving safety.

[0003] Currently, most intelligent connected vehicles do not consider the dynamic time-varying information such as the movement status of the vehicle in front and its own speed in terms of vehicle driving safety. However, in actual driving, safety issues are often determined by both the movement status of the vehicle in front and the speed of the vehicle itself. Existing technologies do not take into account the changes in the vehicle's speed under the influence of the vehicle in front, and therefore cannot effectively meet the safety detection requirements of vehicle driving. Summary of the Invention

[0004] The purpose of this invention is to provide a method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion, comprising the following steps:

[0005] 1) Obtain the training parameters for the velocity prediction model;

[0006] 2) Train the neural network using the training parameters of the speed prediction model to construct the speed prediction model;

[0007] 3) Use a speed prediction model to predict the vehicle speed at the next moment;

[0008] 4) Acquire driving safety information and process it to obtain safety situation assessment parameters:

[0009] 5) Process the security situation assessment parameters to obtain the current security situation assessment value. This allows us to determine the safe driving status of the vehicle.

[0010] Furthermore, the speed prediction model training parameters include driving intention information D. i Historical vehicle speed information V i Workshop motion information M i ;

[0011] Driving Intent Information D i The acquisition steps include: identifying the driving intention based on the information monitored by the pedal sensor, and obtaining the driving intention information D. i .

[0012] Historical vehicle speed information V iThe acquisition steps include: identifying and capturing historical vehicle speed information during the vehicle's driving process based on onboard sensors to obtain historical vehicle speed information V. i ;

[0013] Workshop motion information M i The acquisition steps include: exchanging driving data between the vehicle and the preceding vehicle based on the vehicle-to-vehicle communication protocol, obtaining the relative distance and relative speed between the two vehicles, and obtaining the vehicle-to-vehicle motion information M. i .

[0014] Furthermore, the steps for training a neural network include:

[0015] 2.1) Set the number of nodes in the input layer to n, the number of nodes in the hidden layer to l, and the number of nodes in the output layer to m; initialize the connection weights w between the input layer and the hidden layer. ij The connection weights w between the hidden layer and the output layer jk Threshold a of each neuron in the hidden layer j Threshold b of each neuron in the output layer k ;

[0016] 2.2) Calculate the hidden layer output value H j ,Right now:

[0017]

[0018] In the formula, x is the activation function of the hidden layer; i This serves as input information for training parameters in the velocity prediction model;

[0019] 2.3) Calculate the predicted output value T of the output layer k ,Right now:

[0020]

[0021] 2.4) Calculate the prediction error e k ,Right now:

[0022] e k =Y k -T k , k = 1, 2, ..., m (3)

[0023] In the formula, Y k This refers to the actual output information in the training parameters of the velocity prediction model.

[0024] 2.5) Based on prediction error e k The backpropagation continuously feeds back and corrects the connection weights w ij Connection weight w jk Threshold a j Threshold b kuntil the prediction error e k Less than the preset error threshold;

[0025] Updated connection weight w ij ′、Connection weight w jk ′、threshold a j ′、threshold b k As shown below:

[0026]

[0027]

[0028]

[0029] b k ′=b k +e k , k = 1, 2, ..., m (7)

[0030] Furthermore, when training a neural network, the connection weights w between the input layer and the hidden layer... ij The connection weights w between the hidden layer and the output layer jk Threshold a of each neuron in the hidden layer j Threshold b of each neuron in the output layer k The initial value is determined through the following steps:

[0031] a) Population initialization: The neuron thresholds and connection weights between adjacent neuron layers are integrated into a real number string in the form of chromosome real number encoding, and the boundary range is defined for random initialization;

[0032] b) Calculate fitness J i ,Right now:

[0033]

[0034] In the formula, K is a constant value; m is the number of output layer nodes; and k is the kth node of the output layer.

[0035] c) The selection operation is performed based on the proportional selection method, where the probability p of each individual i being selected is... i As shown below:

[0036]

[0037]

[0038] In the formula, K0 is a constant value; J i ′ represents the fitness value of chromosome i; M represents the population size.

[0039] d) Perform mutual crossover operations based on the real number crossover method. The crossover process is as follows:

[0040]

[0041] In the formula, b is a random number in the interval [0, 1]; a m，l The value of chromosome m at point l; a n，l The value of chromosome n at point l; a m，l ′、a n，l ′ represents the value of chromosomes m and n at point l after the crossover operation;

[0042] e) Select the j-th gene a from the i-th chromosome. ij Performing a mutation operation yields:

[0043]

[0044] In the formula, a max For gene a ij The upper limit of; a min For gene a ij The lower bound value; q is the number of iterations at this time; a ij ′ represents the mutated gene;

[0045] Where f(q) is shown below:

[0046]

[0047] In the formula, r is a random number in the interval [0, 1]; G max This represents the maximum number of evolutions.

[0048] f) Repeat steps b)-e) to obtain the individual with the best fitness, which will be used as the connection weight w between the input layer and the hidden layer. ij The connection weights w between the hidden layer and the output layer jk Threshold a of each neuron in the hidden layer j Threshold b of each neuron in the output layer k The initial value.

[0049] Furthermore, the vehicle's speed at the next moment is shown below:

[0050] V o =f BP-GA (I more ) = V o+1 V o+2 , ..., V o+p (14)

[0051] In the formula, V o For predicting the vehicle speed sequence in the time domain p; f BP-GAThis represents the function mapping relationship within the velocity prediction model; V o+1 V o+2 V o+p Predict the vehicle speed at future time points 1, 2, and p.

[0052] Multi-source information sequence I more As shown below:

[0053]

[0054] In the formula, D i-k D i-k+1 D i Information on the driver's intention; M i-k M i-k+1 M i For workshop motion information; V i-k V i-k+1 V i This is historical vehicle speed information.

[0055] Furthermore, the driving safety information includes the motion information of the vehicle in front, the motion information of the vehicle itself, and the relative information between the vehicles.

[0056] The safety situation assessment parameters include the workshop safety distance d. save and dynamic collision time T DC .

[0057] Furthermore, the workshop safety distance d save As shown below:

[0058]

[0059] In the formula, Δd is the distance between the two vehicles; t1 is the braking reaction time; t2 is the braking gap elimination time; v o v represents the vehicle's speed at the next moment. p The speed of the vehicle in front; a o For the vehicle's acceleration; a p To accelerate the vehicle in front.

[0060] Furthermore, the dynamic collision time T in the workshop is calculated. DC The steps include:

[0061] I) Determine the vehicle state vector J = [d r v o a o D i ] T ;d r The relative distance between the vehicle in front and the vehicle behind;

[0062] II) Establish the vehicle's state matrix J from the previous moment.t-Δp ,Right now:

[0063] J t-Δp =[d r-Δp v o-Δp a o-Δp D i-Δp ] T (17)

[0064] In the formula, d r-Δp v o-Δp a o-Δp D i-Δp The relative distance between the vehicle in front and the vehicle itself at the previous moment, the vehicle's speed, the vehicle's acceleration, and the driver's driving intention;

[0065] III) Construct the vehicle state estimation matrix J at the current time based on the vehicle's state matrix from the previous time step. t ,Right now:

[0066]

[0067] In the formula, Δp is the time difference; v p This refers to the vehicle's speed.

[0068] IV) The mean square error D in predicting the current vehicle state t ,Right now:

[0069]

[0070] In the formula, F is the state transition matrix and Q is the process noise matrix.

[0071] The state transition matrix F is shown below:

[0072]

[0073] In the formula, v o-p The vehicle's speed; a o-p a o-Δp For the vehicle's acceleration;

[0074] V) Calculate the Kalman filter gain K at the current time. t ,Right now:

[0075]

[0076] In the formula, N t For the observation noise matrix, U t The observation matrix;

[0077] Among them, the observation matrix U t As shown below:

[0078]

[0079] VI) Calculate the optimal estimate J at the current time based on the Kalman gain. t ',Right now:

[0080] J t ′=J t +K t (Z t -U t J t ) (twenty three)

[0081] In the formula, Z t The observed value at the current moment;

[0082] VII) Update the state covariance error, i.e.:

[0083] D t ′=(IK t U t )D t (twenty four)

[0084] In the formula, I is the identity matrix with the same dimension as the state vector; D t ′ represents the updated state covariance error.

[0085] VIII) Based on the optimal estimate J t Determine the prediction step size t; calculate the distance between the two vehicles; when the relative distance between the two vehicles is less than or equal to the design minimum safe clearance a, record the step number n up to that moment, and then calculate the dynamic collision time T. DC = n*t.

[0086] Furthermore, the steps for processing the security situation assessment parameters include:

[0087] 5.1) The workshop safety distance d is calculated using the Euclidean distance method. save and dynamic collision time T DC Perform fuzzy clustering to obtain the input parameter set D r ;

[0088] 5.2) Input parameter set D r Substituting the membership function of the triangle to determine the three vertices (a h ,0),(b h ,0),(c h ,0), to estimate the membership function of each fuzzy set;

[0089] Wherein, coordinate parameter a h Coordinate parameter b h Coordinate parameter c h As shown below:

[0090]

[0091] In the formula, N h For parameter set D r Number of elements in D; r_1 D r_2 , D r_i For parameter set D r Elements in;

[0092] 5.3) Establish the input membership number equation, that is:

[0093]

[0094] In the formula, f ch (x) represents the input membership number;

[0095] 5.4) The workshop safety distance d save and dynamic collision time T DC By substituting the parameter set into the membership equation, the safe distance d can be calculated. save and dynamic collision time T DC The membership degree is used as a fuzzy input. According to the "if-then" rule, the fuzzy output of the security situation assessment is obtained. The fuzzy output is then defuzzified and used as the current security situation assessment value.

[0096] Furthermore, the vehicle's driving safety status includes normal driving status, collision warning status, and braking and collision avoidance status;

[0097] Among them, normal driving state means: the distance to the vehicle in front is greater than the preset distance threshold or the vehicle in front does not exist, and there is no need to consider the impact of the changes in the driving conditions of the vehicle in front on the driving conditions of the vehicle itself.

[0098] The collision warning status means that the distance to the vehicle in front is less than the preset distance threshold but there is no risk of collision. It is necessary to consider the impact of the change in the driving conditions of the vehicle in front on the driving conditions of the vehicle itself.

[0099] Braking and collision avoidance status refers to a situation where there is a risk of collision with the vehicle in front under the current driving conditions, and it is necessary to brake immediately to reduce the vehicle's speed to ensure the safety of the driving process.

[0100] The technical effects of this invention are undeniable. This invention constructs a speed prediction model based on a BP neural network by processing the training parameters of the speed prediction model, and optimizes the BP neural network speed prediction model using the GA algorithm, thereby improving the accuracy of predicting the vehicle's speed. By using the BP-GA model to predict the vehicle's speed at the next moment, the safety situation assessment parameters are improved, different driving states of the preceding vehicle are classified, and the safe distance d between them is calculated. saveAnd an extended Kalman filter is introduced to analyze the dynamic collision time T of the workshop. DC Calculations are performed, and a safety situation assessment model is constructed using fuzzy membership functions to obtain a safety situation value map. The intersection points of these maps are used as safety situation distinction boundaries to divide safety situation value intervals, corresponding to different driving safety states. During driving, vehicle speed is predicted in real time, and the current driving safety situation is assessed, quantifying the current safety situation value. Whether the vehicle crosses its original safe driving range determines whether its current safe driving status has changed.

[0101] This invention aims to improve the safety of intelligent connected vehicles in responding to the dynamic and time-varying motion of vehicles ahead during driving. It studies a coupled modeling method for vehicle speed prediction and driving safety in intelligent connected vehicle traffic environments, constructs a vehicle speed prediction model and a driving safety situation assessment model, and uses the prediction of vehicle speed as a link to clarify the importance of safety situation values ​​for the driving safety of intelligent connected vehicles, providing theoretical and technical support for realizing safe driving technology for intelligent connected vehicles. Attached Figure Description

[0102] Figure 1 This is a flowchart of the intelligent connected vehicle safety situation assessment method based on multi-source information fusion provided by the present invention;

[0103] Figure 2 This is a flowchart of processing training parameters for a speed prediction model;

[0104] Figure 3 This is a flowchart of constructing a BP-GA neural network velocity prediction model;

[0105] Figure 4 This is a flowchart of predicting the vehicle speed at the next moment using the BP-GA model;

[0106] Figure 5 It is a flowchart for determining the evaluation parameters of the security situation assessment;

[0107] Figure 6 This is a flowchart for constructing a security situation assessment model;

[0108] Figure 7 This is a schematic diagram of the division of security status value intervals. Detailed Implementation

[0109] The present invention will be further described below with reference to embodiments, but it should not be construed that the scope of the present invention is limited to the following embodiments. Various substitutions and modifications made based on ordinary technical knowledge and common practices in the art without departing from the above-described technical concept of the present invention should be included within the scope of protection of the present invention.

[0110] Example 1:

[0111] See Figures 1 to 7 The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion includes the following steps:

[0112] 1) Obtain the training parameters for the velocity prediction model;

[0113] 2) Train the neural network using the training parameters of the speed prediction model to construct the speed prediction model;

[0114] 3) Use a speed prediction model to predict the vehicle speed at the next moment;

[0115] 4) Acquire driving safety information and process it to obtain safety situation assessment parameters:

[0116] 5) Process the security situation assessment parameters to obtain the current security situation assessment value. This allows us to determine the safe driving status of the vehicle.

[0117] Example 2:

[0118] The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion is the same as in Example 1. Furthermore, the training parameters of the speed prediction model include driving intention information D. i Historical vehicle speed information V i Workshop motion information M i ;

[0119] Driving Intent Information D i The acquisition steps include: identifying the driving intention based on the information monitored by the pedal sensor, and obtaining the driving intention information D. i .

[0120] Historical vehicle speed information V i The acquisition steps include: identifying and capturing historical vehicle speed information during the vehicle's driving process based on onboard sensors to obtain historical vehicle speed information V. i ;

[0121] Workshop motion information M i The acquisition steps include: exchanging driving data between the vehicle and the preceding vehicle based on the vehicle-to-vehicle communication protocol, obtaining the relative distance and relative speed between the two vehicles, and obtaining the vehicle-to-vehicle motion information M. i .

[0122] Example 3:

[0123] The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion has the same technical content as any one of Examples 1-2. Further, the steps for training the neural network include:

[0124] 1) Set the number of nodes in the input layer to n, the number of nodes in the hidden layer to l, and the number of nodes in the output layer to m; initialize the connection weights w between the input layer and the hidden layer. ij The connection weights w between the hidden layer and the output layer jk Threshold a of each neuron in the hidden layer j Threshold b of each neuron in the output layer k ;

[0125] 2) Calculate the hidden layer output value H j ,Right now:

[0126]

[0127] In the formula, x is the activation function of the hidden layer; i This serves as input information for training parameters in the velocity prediction model;

[0128] 3) Calculate the predicted output value T of the output layer k ,Right now:

[0129]

[0130] 4) Calculate the prediction error e k ,Right now:

[0131] e k =Y k -T k , k = 1, 2, ..., m (3)

[0132] In the formula, Y k This refers to the actual output information in the training parameters of the velocity prediction model.

[0133] 5) Based on prediction error e k The backpropagation continuously feeds back and corrects the connection weights w ij Connection weight w jk Threshold a j Threshold b k until the prediction error e k Less than the preset error threshold;

[0134] Example 4:

[0135] The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion has the same technical content as any one of Examples 1-3, and further, the updated connection weight w ij As shown below:

[0136]

[0137] Example 5:

[0138] The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion has the same technical content as any one of embodiments 1-4. Further, the updated connection weight w... jk As shown below:

[0139]

[0140] Example 6:

[0141] The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion has the same technical content as any one of embodiments 1-5, and further, the updated threshold a j As shown below:

[0142]

[0143] Example 7:

[0144] The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion has the same technical content as any one of embodiments 1-6, and further, the updated threshold b k As shown below:

[0145] b k ′=b k +e k , k = 1, 2, ..., m (7)

[0146] Example 4:

[0147] The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion has the same technical content as any one of Examples 1-3. Furthermore, when training the neural network, the connection weights w between the input layer and the hidden layer are... ij The connection weights w between the hidden layer and the output layer jk Threshold a of each neuron in the hidden layer j Threshold b of each neuron in the output layer k The initial value is determined through the following steps:

[0148] 1) Population initialization: The neuron thresholds and connection weights between adjacent neuron layers are integrated into a real number string in the form of chromosome real number encoding, and the boundary range is defined for random initialization;

[0149] 2) Calculate fitness J i ,Right now:

[0150]

[0151] In the formula, K is a constant value; m is the number of output layer nodes; and k is the kth node of the output layer.

[0152] 3) The selection operation is completed based on the proportional selection method, where the probability p of each individual i being selected is... i As shown below:

[0153]

[0154]

[0155] In the formula, K0 is a constant value; J i ′ represents the fitness value of chromosome i; M represents the population size.

[0156] 4) Perform mutual crossover operations based on the real number crossover method. The crossover process is as follows:

[0157]

[0158] In the formula, b is a random number in the interval [0, 1]; a m，l The value of chromosome m at point l; a n，l The value of chromosome n at point l; a m，l ′、a n，l ′ represents the value of chromosomes m and n at point l after the crossover operation;

[0159] 5) Select the j-th gene a from the i-th chromosome. ij Performing a mutation operation yields:

[0160]

[0161] In the formula, a max For gene a ij The upper limit of R; min For gene a ij The lower bound value; q is the number of iterations at this time; a ij ′ represents the mutated gene;

[0162] Where f(q) is shown below:

[0163]

[0164] In the formula, r is a random number in the interval [0, 1]; G max This represents the maximum number of evolutions.

[0165] 6) Repeat steps 2)-5) to obtain the individual with the best fitness, and use it as the connection weight w between the input layer and the hidden layer. ij The connection weights w between the hidden layer and the output layer jk Threshold a of each neuron in the hidden layer j Threshold b of each neuron in the output layer k The initial value.

[0166] Example 5:

[0167] The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion has the same technical content as any one of Examples 1-4. Furthermore, the vehicle speed at the next moment is as follows:

[0168] V o =f BP-GA (I more ) = V o+1 V o+2 , ..., V o+p (14)

[0169] In the formula, V o For predicting the vehicle speed sequence in the time domain p; f BP-GA This represents the function mapping relationship within the velocity prediction model; V o+1 V o+2 V o+p Predict the vehicle speed at future time points 1, 2, and p.

[0170] Multi-source information sequence I more As shown below:

[0171]

[0172] In the formula, D i-k D i-k+1 D i Information on the driver's intentions at different times; M i-k M i-k+1 M i For workshop motion information at different times; V i-k V i-k+1 V i This contains historical vehicle speed information at different times.

[0173] Example 6:

[0174] The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion has the same technical content as any one of embodiments 1-5. Furthermore, the driving safety information includes the motion information of the vehicle in front, the motion information of the vehicle itself, and the relative information between the vehicles.

[0175] The safety situation assessment parameters include the workshop safety distance d. save and dynamic collision time T DC .

[0176] Example 7:

[0177] The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion has the same technical content as any one of Examples 1-6, further specifying the safe distance d between vehicles.save As shown below:

[0178]

[0179] In the formula, Δd is the distance between the two vehicles; t1 is the braking reaction time; t2 is the braking gap elimination time; v o v represents the vehicle's speed at the next moment. p The speed of the vehicle in front; a o For the vehicle's acceleration; a p To accelerate the vehicle in front.

[0180] Example 8:

[0181] The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion has the same technical content as any one of Examples 1-7, but further includes calculating the dynamic collision time T between vehicles. DC The steps include:

[0182] 1) Determine the vehicle state vector J = [d r v o a0, D i ] T ;d r The relative distance between the vehicle in front and the vehicle behind;

[0183] 2) Establish the vehicle's state matrix J from the previous moment. t-Δp ,Right now:

[0184] J t-Δp =[d r-Δp v o-Δp a o-Δp D i-Δp ] T (17)

[0185] In the formula, d r-Δp v o-Δp a o-Δp D i-Δp The relative distance between the vehicle in front and the vehicle itself at the previous moment, the vehicle's speed, the vehicle's acceleration, and the driver's driving intention;

[0186] 3) Based on the vehicle's state matrix from the previous time step, construct the vehicle's state estimation matrix J for the current time step. t ,Right now:

[0187]

[0188] In the formula, Δp is the time difference; v p This refers to the vehicle's speed.

[0189] 4) The mean square error D in predicting the vehicle's current state. t ,Right now:

[0190]

[0191] In the formula, F is the state transition matrix and Q is the process noise matrix.

[0192] The state transition matrix F is shown below:

[0193]

[0194] In the formula, v o-p The vehicle's speed; a o-p a o-Δp For the vehicle's acceleration;

[0195] 5) Calculate the Kalman filter gain K at the current time. t ,Right now:

[0196]

[0197] In the formula, N t For the observation noise matrix, U t The observation matrix;

[0198] Among them, the observation matrix U t As shown below:

[0199]

[0200] 6) Calculate the optimal estimate J at the current time based on the Kalman gain. t ',Right now:

[0201] J t ′=J t +K t (Z t -U t J t ) (twenty three)

[0202] In the formula, Z t The observed value at the current moment;

[0203] 7) Update the state covariance error, i.e.:

[0204] D t ′=(IK t U t )D t (twenty four)

[0205] In the formula, I is the identity matrix with the same dimension as the state vector; D t ′ represents the updated state covariance error.

[0206] 8) Based on the optimal estimate J tDetermine the prediction step size t; calculate the distance between the two vehicles; when the relative distance between the two vehicles is less than or equal to the design minimum safe clearance a, record the step number n up to that moment, and then calculate the dynamic collision time T. DC = n*t.

[0207] Example 9:

[0208] The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion has the same technical content as any one of Examples 1-8. Further, the steps for processing the safety situation assessment parameters include:

[0209] 1) Use the Euclidean distance method to determine the workshop safety distance d. save and dynamic collision time T DC Perform fuzzy clustering to obtain the input parameter set D r ;

[0210] 2) Input parameter set D r Substituting the membership function of the triangle to determine the three vertices (a h ,0),(b h ,0),(c h ,0), to estimate the membership function of each fuzzy set;

[0211] Wherein, coordinate parameter a h Coordinate parameter b h Coordinate parameter c h As shown below:

[0212]

[0213] In the formula, N h For parameter set D r Number of elements in D; r_1 D r_2 , D r_i For parameter set D r Elements in;

[0214] 3) Establish the input membership number equation, that is:

[0215]

[0216] In the formula, f ch (x) represents the input membership number;

[0217] 4) The workshop safety distance d save and dynamic collision time T DC By substituting the parameter set into the membership equation, the safe distance d can be calculated. save and dynamic collision time T DCThe membership degree is used as a fuzzy input. According to the "if-then" rule, the fuzzy output of the security situation assessment is obtained. The fuzzy output is then defuzzified and used as the current security situation assessment value.

[0218] Example 10:

[0219] The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion has the same technical content as any one of embodiments 1-9. Furthermore, the vehicle's driving safety status includes normal driving status, collision warning status, and braking collision avoidance status.

[0220] Among them, normal driving state means: the distance to the vehicle in front is greater than the preset distance threshold or the vehicle in front does not exist, and there is no need to consider the impact of the changes in the driving conditions of the vehicle in front on the driving conditions of the vehicle itself.

[0221] The collision warning status means that the distance to the vehicle in front is less than the preset distance threshold but there is no risk of collision. It is necessary to consider the impact of the change in the driving conditions of the vehicle in front on the driving conditions of the vehicle itself.

[0222] Braking and collision avoidance status refers to a situation where there is a risk of collision with the vehicle in front under the current driving conditions, and it is necessary to brake immediately to reduce the vehicle's speed to ensure the safety of the driving process.

[0223] Example 11:

[0224] The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion includes the following steps:

[0225] 1. Training parameters for the processing speed prediction model

[0226] Based on the pedal sensor, a fuzzy inference system is used to identify the driving intention, obtaining driving intention information D. i Based on onboard sensors, historical vehicle speed information during the vehicle's driving process is identified and captured to obtain historical vehicle speed information V. i Based on the vehicle-to-vehicle communication protocol, driving data exchange is achieved between the vehicle and the preceding vehicle to obtain the relative distance and relative speed between the two vehicles, thus obtaining the vehicle-to-vehicle motion information M. i .

[0227] 2. Construct a BP neural network speed prediction model

[0228] The steepest descent method, also known as gradient descent, is selected as the BP neural network learning method in this invention. Initial weights and thresholds are determined, and the connection weights w between the input layer and the hidden layer are calculated based on the input vector. jk and the hidden layer threshold a j Calculate the hidden layer output value H j Based on the hidden layer output value H j The connection weights w between the hidden layer and the output layer jkand the output layer threshold b k Calculate the predicted output value T of the output layer. For error calculation, the weights w are continuously corrected based on the backpropagation of the prediction error e. ij w jk and threshold a j b k .

[0229] 3. Optimize the BP neural network velocity prediction model using the GA algorithm.

[0230] The Generative Adversarial (GA) algorithm is used to optimize a backpropagation (BP) neural network. GA utilizes real-number chromosome encoding to represent the initial weights and thresholds of the network, and uses the prediction error of the initialized BP network as the objective function, i.e., the fitness value. GA searches for the optimal fitness individuals through a series of processes including population initialization, determining the fitness function, proportional selection, crossover, and mutation, thereby obtaining the optimal weights and thresholds and ultimately determining the optimal prediction model for the BP neural network.

[0231] 4. Prediction of vehicle speed at the next moment based on the BP-GA model

[0232] Based on various information collection methods for the vehicle, the driving intention information D of the vehicle during the current driving process is analyzed. i Historical vehicle speed information V i Workshop motion information M i Real-time data capture and computation are performed, serving as input to the BP-GA neural network speed prediction model to predict the vehicle speed sequence within the time domain p. Various information is then fed into the BP-GA speed prediction model V. o =f BP-GA (I more ) = V o+1 V o+2 , ..., V o+p The system predicts the vehicle speed at the next time step p and outputs this prediction. The accuracy of the speed prediction is verified using R(k) and R... e The root mean square error in the prediction time domain and the root mean square error for the entire test condition are calculated separately to improve prediction accuracy.

[0233] 5. Determine the parameters for security situation assessment.

[0234] Based on the vehicle's sensors and inter-vehicle communication protocols, driving safety information such as the motion information of the vehicle in front, the vehicle's own motion information, and the relative information with other vehicles is obtained. This safety information is processed to obtain the safety situation assessment parameter: the safe distance between vehicles, d. save and dynamic collision time T DC .

[0235] 6. Calculate the safe distance d between the two sides. save

[0236] Workshop safety distance d save It consists of three parts: the vehicle's braking distance d o The distance d changes due to the movement of the vehicle in front. p And the distance Δd between the two vehicles after the distance between them has stabilized, using the formula d save =d o +Δd-d p The calculation involves considering the relationship between the safe distance between the vehicle and the preceding vehicle, as well as the preceding vehicle's motion state. The braking process is divided into four stages: reaction stage t1, gap elimination stage t2, deceleration stage t3, and braking duration stage t4. The braking distance calculation for each stage is related to time t. Based on the preceding vehicle's driving state, the safe distance is calculated in two scenarios: first, when the current vehicle is traveling at a constant or accelerating speed, the safe distance is determined by calculating the vehicle's braking distance and the preceding vehicle's travel distance; second, when the current vehicle is decelerating or decelerating to a stop, the safe distance is determined by calculating the braking distances of both the current vehicle and the preceding vehicle. In summary, the safe distance d... save The calculation is determined based on the driving status of the vehicle in front and relevant parameters.

[0237] 7. Calculate the dynamic collision time T in the workshop. DC

[0238] Workshop dynamic collision time T DC The dynamic collision time model, established by extended Kalman filtering, is determined to account for the dynamic collision time as the vehicle's motion changes. The vehicle state vector, J = [d], is then determined. r v o a o D i ] T Based on the vehicle's driving state at the previous time step Δp, the vehicle's driving state at the current time is deduced, resulting in the vehicle state estimation matrix J at the current time. t The mean square error D of the vehicle's current state. t Make a prediction and calculate the Kalman filter gain K at the current time. t The optimal estimate for the current moment is calculated based on the Kalman gain, and the state-based collaborative error is updated. By continuously modifying the extrapolation prediction step size t, the distance between the two vehicles is calculated. As time progresses, the two vehicles continuously approach each other. When the relative distance between the two vehicles is less than or equal to the design minimum safe distance a, the step number n up to that moment is recorded, thus obtaining the dynamic collision time T. DC = n*t.

[0239] 8. Construct a security situation assessment model

[0240] The workshop safety distance d is calculated using the Euclidean distance method. save and dynamic collision time T DCFuzzy clustering can ultimately yield a set D of K highly correlated input parameters. r The K parameter sets D after clustering r Substituting the membership function of the triangle to determine the three vertices (a h ,0),(b h ,0),(c h The input membership number f can be obtained from the three vertices of the triangle membership function (0). ch (x) can convert two precise input quantities during the current driving process into fuzzy output quantities, obtain the fuzzy output quantity of the safety situation assessment according to the "if-then" rule, and defuzzify the fuzzy output quantity to obtain the precise current safety situation assessment value. The data is then interpolated to form a security situation value map.

[0241] 9. Divide the security status value range

[0242] The intersection points of the security situation value maps are used as the boundaries for distinguishing different security situations, dividing the map into different security situation intervals. The interval thresholds are as follows: The quantitative value of the security situation is divided into three different security situation ranges.

[0243] The three different safety status intervals correspond to three different driving safety states: normal driving, collision warning, and braking collision avoidance.

[0244] The normal driving state is when the distance to the vehicle in front is far or there is no vehicle in front, and there is no need to consider the impact of changes in the driving conditions of the vehicle in front on the driving conditions of the vehicle itself. The vehicle is in a state of free driving.

[0245] The collision warning status indicates that the distance to the vehicle in front is relatively close, and it is necessary to consider the impact of the changes in the driving conditions of the vehicle in front on the driving conditions of the vehicle itself. The vehicle is in a state of cautious driving.

[0246] The braking and collision avoidance state refers to a situation where there is a risk of collision with the vehicle in front under the current driving conditions, and it is necessary to brake immediately to reduce the vehicle speed to ensure the safety of the driving process. The vehicle is in a deceleration driving state.

[0247] 10. Rules for switching security states corresponding to security status values

[0248] During driving, the vehicle will predict its speed in real time and assess the current driving safety situation, quantifying the current safety situation value. Whether the vehicle crosses its original safe driving range determines whether its current safe driving status has changed.

[0249] Example 12:

[0250] The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion includes the following steps:

[0251] I. Training parameters for the processing speed prediction model

[0252] The specific method is as follows:

[0253] 1) Based on the accelerator pedal sensor and brake pedal sensor, information such as the accelerator pedal opening, brake pedal opening, and rate of change of their opening are obtained during vehicle operation to determine the vehicle's acceleration and braking intentions. This driving intention is then identified using a fuzzy inference system to obtain driving intention information D. i .

[0254] 2) Based on the vehicle speed sensor and vehicle acceleration sensor, the historical vehicle speed information during the vehicle's driving process is identified and captured to obtain the historical vehicle speed information V. i .

[0255] 3) Based on vehicle-to-vehicle (V2V) communication protocols, dedicated short-range communication (DSRC) technology, radio frequency identification (RFID) technology, global positioning system (GPS), sensor technology, radar cameras, image processing, and intelligent transportation systems (ITS), the vehicle exchanges driving data with the vehicle in front, obtains the relative distance and relative speed between the two vehicles, and obtains the vehicle's motion information M. i .

[0256] II. Constructing a BP-GA Neural Network Velocity Prediction Model

[0257] The specific method is as follows:

[0258] Constructing a BP neural network speed prediction model:

[0259] 1) The steepest descent method, also known as gradient descent, is used as the BP neural network learning method in this invention. Gradient descent operates along the gradient direction, and typically uses a differentiable sigmoid function as the transfer function. Its function expression is:

[0260]

[0261] 2) Determine initial weights and thresholds. Based on input and output information, determine the number of nodes n, l, and m in each neural layer, and initialize the connection weights w between adjacent neuron layers. ij w jk and the threshold 'a' of each neuron in the hidden layer and output layer j b k .

[0262] 3) Hidden layer output calculation. Based on the input vector, the connection weights w between the input layer and the hidden layer are calculated. jk and the hidden layer threshold a j Calculate the hidden layer output value H j .

[0263]

[0264] In the formula, f(*) is the activation function of the hidden layer, i.e.

[0265] 4) Output layer output calculation. Based on the hidden layer output value H... j The connection weights w between the hidden layer and the output layer jk and the output layer threshold b k Calculate the predicted output value T of the output layer.

[0266]

[0267] 5) Error calculation. Based on the predicted value T and the theoretical output value Y, the prediction error e is obtained.

[0268] e = Y k -T k , k = 1, 2, ..., m (4)

[0269] 6) Weight and threshold update. The weights w are continuously updated based on the backpropagation of the prediction error e. ij w jk and threshold a j b k .

[0270]

[0271] a k =a k +e k , k = 1, 2, ..., m (6)

[0272]

[0273] b k =b k +e k , k = 1, 2, ..., m (8)

[0274] In the formula, η is the learning rate.

[0275] 7) Determine if the gradient error meets the preset requirements. If not, return to step 3.

[0276] Optimizing the BP neural network velocity prediction model using the GA algorithm:

[0277] The GA algorithm for optimizing a BP neural network uses real-number chromosome encoding to represent the initial weights and thresholds of the network. It uses the prediction error of the initialized BP network as the objective function, i.e., the fitness value. GA searches for the best-fit individuals based on selection, crossover, and mutation operations to obtain the optimal weights and thresholds, thereby determining the optimal prediction model for the BP neural network. The specific method is as follows:

[0278] 1) Population initialization. The neuron thresholds and connection weights between adjacent neuron layers are integrated into a real number string in a certain order using chromosome real number encoding, and a boundary range is defined for random initialization. Subsequent optimization operations are all based on chromosomes.

[0279] 2) Determine the fitness function. Assign the initial weights and thresholds obtained through iterative updates to the BP network structure, train the network using sample data, and input test sample data to complete the prediction output. The fitness function value J is... i Select as the predicted output T k and theoretical output Y k The sum of errors is used to determine the objective function:

[0280]

[0281] In the formula, K is a constant value; m is the number of output layer nodes; and k is the kth node of the output layer.

[0282] 3) Selection operation. The selection operation is performed based on the proportional selection method, where the probability p of each individual i being selected is... i for:

[0283]

[0284]

[0285] In the formula, K0 is a constant value; J i is the fitness value of chromosome i; M is the population size.

[0286] 4) Crossover operation. Based on the real number crossover method, for a... m and a n The l-position genes undergo crossover operations, as shown below:

[0287]

[0288] In the formula, b is a random number in the interval [0, 1]; a m，l The value of chromosome m at point l; a n，l This represents the value of chromosome n at point l.

[0289] 5) Mutation operation. Select the j-th gene a from the i-th chromosome. ijThe specific method for performing mutation operations is as follows:

[0290]

[0291] In the formula, a max For gene a ij The upper limit of; a min For gene a ij The lower bound value; q is the number of iterations at this time; where f(q) is as follows:

[0292]

[0293] In the formula, r is a random number in the interval [0, 1]; G max This represents the maximum number of evolutions.

[0294] Optimizing the initial weight threshold of the BP neural network model velocity prediction model using the GA algorithm can improve the prediction accuracy of the BP neural network.

[0295] III. Predicting the vehicle speed at the next moment using the BP-GA model

[0296] The specific method is as follows:

[0297] 1) Real-time vehicle data processing

[0298] Based on information collection methods such as pedal sensors, speed sensors, millimeter-wave radar, and vehicle-to-vehicle communication protocols, the driving intention information of the vehicle during the current driving process is analyzed. i Historical vehicle speed information V i Workshop motion information M i Real-time capture and computation are performed as input to the BP-GA neural network speed prediction model.

[0299] 2) Vehicle speed prediction based on BP-GA multi-source information fusion

[0300] The vehicle driving intention information D obtained in the previous step i Historical vehicle speed information V i Workshop motion information M i Using multi-source information as input variables to a BP-GA neural network, the prediction of the vehicle speed at the next time step p is used as the output. The specific prediction process is as follows:

[0301] Multi-source information sequences as input:

[0302]

[0303] The predicted vehicle speed output at the next moment p is:

[0304] V o =fBP-GA (I more ) = V o+1 V o+2 , ..., V o+p (16)

[0305] In the formula, I more V represents a multi-source information sequence of length k+3; o For predicting the vehicle speed sequence in the time domain p; f BP-GA This represents the function mapping relationships within the BP neural network.

[0306] To verify the accuracy of the vehicle speed prediction results, the root mean square error (RMSE) was selected as the evaluation metric for assessing the quality of the prediction results. The smaller the RMSE, the smaller the error between the expected value and the predicted value, and the higher the prediction accuracy. The evaluation metrics are as follows:

[0307]

[0308]

[0309] In the formula, R(k) is the root mean square error in the prediction time domain; R e The root mean square error represents the total error over the entire test cycle; v(k+i) represents the predicted vehicle speed at that moment; v r (k+i) represents the actual vehicle speed; S represents the total test time.

[0310] IV. Determining Security Situation Assessment Parameters

[0311] The specific method is as follows:

[0312] Based on the vehicle's sensors and inter-vehicle communication protocols, driving safety information such as the motion information of the vehicle in front, the vehicle's own motion information, and the relative information between vehicles is obtained. The safety information is processed to obtain safety situation assessment parameters.

[0313] Workshop safety distance analysis:

[0314] 1) Components of workshop safety distance

[0315] Workshop safety distance d save It can be considered to consist of three parts. When the motion state of the vehicle in front changes, the vehicle takes braking measures to avoid a collision; the vehicle begins braking; and the distance d generated by the braking is... o The distance d caused by the change in the motion state of the vehicle in front p And the distance Δd between the two vehicles after their motion has stabilized and there is no risk of collision.

[0316] d save =do +Δd-d p (19)

[0317] The above formula shows that the safe distance between vehicles is related to the braking distance of the vehicle in front and the motion state of the vehicle in front.

[0318] 2) Braking process analysis

[0319] The calculation of braking distance consists of four stages in the braking process: reaction stage t1; gap elimination stage t2; deceleration stage t3; and braking duration stage t4, where:

[0320] Reaction phase t1: This phase refers to the process when the vehicle detects a change in the motion state of the vehicle in front, analyzes the motion information of the vehicle in front, determines that the vehicle needs to brake to control the distance between itself and the vehicle in front, and then the vehicle begins to perform braking operations. Let the time required for this process be t1.

[0321] Clearance elimination stage t2: When the vehicle begins braking, the braking mechanism cannot immediately generate braking force. Because the vehicle's brakes are designed with a certain braking clearance to prevent continuous contact between the braking components and the brake disc, thus avoiding wear and overheating caused by friction, the vehicle needs to eliminate the braking clearance before generating braking force. Let t2 be the time required for this elimination.

[0322] Deceleration phase t3: After the braking gap is eliminated, the braking mechanism begins to function, generating the braking force and deceleration required for braking. The time required for this deceleration phase—from 0 to the deceleration required to match the current driving conditions—is t3.

[0323] Braking duration phase t4: The process by which the vehicle reaches the desired deceleration and maintains this speed, denoted as t4, is the time required for the deceleration phase.

[0324] 3) Calculation of safe distance in the workshop

[0325] Since the driving status of the vehicle in front has a direct impact on the determination of the safe distance, the invention will analyze the determination of the safe distance in two cases based on the different driving statuses of the vehicle in front.

[0326] ①The current vehicle is moving at a constant speed or accelerating.

[0327] When the vehicle in front is traveling at a constant speed or accelerating, there are three possible speed relationships between the two vehicles: If the speed of the vehicle in front is greater than or equal to the speed of the vehicle in front, a rear-end collision will not occur. If the speed of the vehicle in front is less than the speed of the vehicle in front, then the moment when the speeds of the two vehicles reach equality is the moment when the distance between the two vehicles is smallest. Therefore, by calculating the distance between the two vehicles from the start of deceleration until their speeds reach equality, the safe distance in this case can be determined. The formula is as follows:

[0328] Let the speed of the vehicle in front be v. p The acceleration is a p The predicted speed of the vehicle is v. o The acceleration is a o The braking distance d of the vehicle o It can be represented as:

[0329]

[0330] Simplified to:

[0331]

[0332] Since the reaction time t2 during the gap elimination phase is very short, the squared t2 is calculated as follows. 2 The term can be ignored, and the distance d generated by the vehicle's braking is obtained. o Represented as:

[0333]

[0334] The distance d caused by the change in the motion state of the vehicle in front p Represented as:

[0335]

[0336] Therefore, the safe distance d when the vehicle in front is traveling at a constant speed or accelerating can be obtained. save1 for:

[0337]

[0338] ②The vehicle in front slows down or slows down to a stop.

[0339] When the vehicle in front decelerates or comes to a stop, regardless of whether both vehicles reach 0 speed simultaneously or the vehicle in front reaches 0 speed first, the moment when the distance between the two vehicles is the smallest is when the vehicle in front stops. Therefore, the braking distances of the vehicle in front and the vehicle in front must be calculated separately before determining the safe distance.

[0340] First, the braking distance d of the vehicle o It can be represented as:

[0341]

[0342] For the braking decision of the vehicle itself, corresponding measures are only taken when a change in the driving state of the vehicle in front is detected. Therefore, when solving for the braking distance of the vehicle in front, it is not necessary to consider the reaction process and gap elimination process of the vehicle in front at the beginning of braking; only the deceleration process and the braking duration process need to be considered. The braking distance d of the vehicle in front is then calculated. p It can be represented as:

[0343]

[0344] Therefore, the safe distance d when the vehicle in front decelerates or decelerates to a stop can be obtained. save2 for:

[0345]

[0346] Based on the above analysis, the safe distance d between the two workshops can be obtained. save It can be represented as:

[0347]

[0348] Workshop dynamic collision time analysis:

[0349] The common time to collision (TTC) between two vehicles is the time taken to measure the relative speed of the preceding vehicle from the vehicle in front of it to the relative distance between the two vehicles. Its expression is:

[0350]

[0351] Where d r v represents the relative distance between the preceding vehicle and the vehicle in the workshop. r The relative speed between the preceding vehicle and the vehicle itself.

[0352] This method of calculating collision time does not consider the impact of changes in vehicle motion state on collision time while driving in the workshop. Therefore, this invention introduces extended Kalman filtering to establish a dynamic collision time model to consider the dynamic collision time when the vehicle's motion state changes. The specific method is as follows:

[0353] 1) Determine the vehicle state vector, J = [d r v o a o D i ] T These are the relative distance between the vehicle and the vehicle in front, the vehicle's speed, the vehicle's acceleration, and the driver's driving intention.

[0354] 2) Based on the vehicle's driving state at the previous moment Δp, infer the vehicle's driving state at the current moment. Let the vehicle's state matrix at the previous moment be:

[0355] J t-Δp =[d r-Δp v o-Δp a o-Δp D i-Δp ] T (30)

[0356] Obtain the vehicle state estimation matrix J at the current time. t for:

[0357]

[0358] 3) The mean square error D of the vehicle's state at the current moment t The prediction is expressed as follows:

[0359]

[0360] Where F is the state transition matrix and Q is the process noise matrix. Based on the extended Kalman filter principle, an approximate linear equation is obtained using Taylor series, thus replacing the nonlinear function. The formula for calculating the state transition matrix F is as follows:

[0361]

[0362] 4) Calculate the Kalman filter gain K at the current time. t Its expression is as follows:

[0363]

[0364] Where D t Let N be the mean square error matrix in process 3). t For the observation noise matrix, U t The observation matrix is ​​determined by the observation function. Since the observed values ​​are the relative distance between the vehicle and the vehicle ahead, the vehicle's speed, the vehicle's acceleration, and the driver's driving intention, the observation matrix is:

[0365]

[0366] 5) Calculate the optimal estimate at the current time based on the Kalman gain: The expression is as follows:

[0367] J t =J t +K t (Z t -U t J t (36)

[0368] Z t The value is the observation at the current moment, and J on the right side of the equals sign. t For the vehicle's current driving state predicted in process 2), J on the left side of the equals sign t This is the corrected optimal estimate for the current moment.

[0369] 6) Update the state cooperative defense error, the expression is as follows:

[0370] D t =(IK t U t )D t (37)

[0371] Where I is the identity matrix with the same dimension as the state vector.

[0372] 7) After the above six processes, the extended Kalman filter can be used to predict the driving state of the vehicle, and then the dynamic collision time between the vehicle in front and the vehicle can be obtained.

[0373] The prediction process for dynamic collision time is as follows: Using the current vehicle state parameters as input, an extended Kalman filter is used to predict the vehicle's state parameters for the next moment. The extrapolation prediction step size is t. Simultaneously, the distance between the two vehicles is calculated. As time progresses, the two vehicles continuously approach each other. When the relative distance between the two vehicles is less than or equal to the designed minimum safe clearance a, the step number n up to that moment is recorded, yielding the dynamic collision time T. DC = n*t.

[0374] V. Constructing a Security Situation Assessment Model

[0375] The specific method is as follows:

[0376] 1) Determine the degree of membership of the vehicle's driving safety status.

[0377] The workshop safety distance d obtained from the previous step save and dynamic collision time T DC As input parameters for the safety situation assessment model, the workshop safety distance d save and dynamic collision time T DC Perform fuzzy clustering and calculate the number of input members for its security situation assessment model.

[0378] The similarity relationship between the input parameters is established using the Euclidean distance method. For any two parameters D... i and D j Similarity relationship between them:

[0379]

[0380] Where δ is a constant to ensure R(D)i D j )∈[0,1], the formula for calculating δ is as follows:

[0381]

[0382] Among them, D max This represents the maximum value among the input parameters. This allows us to obtain the similarity matrix R with respect to the parameter set D. D And R is calculated according to the Washall algorithm. D Fuzzy equivalence matrix R e The process of change is as follows:

[0383]

[0384] when At that time, it is considered that the fuzzy equivalent matrix

[0385] R is obtained through the threshold λ. e The λ-cut matrix R eλ Using R eλ Classify the parameter set D, if R eλ (D i D j If ) = 1, then D i With D j They belong to the same class. Where λ∈[0,1], and as λ decreases, the fuzzy equivalence matrix R... e The identified elements within a class will be gradually merged, eventually forming a single class, resulting in a set D of K highly relevant input parameters. r .

[0386] The K parameter sets D after clustering r Substituting the membership function of the triangle to determine the three vertices (a h ,0),(b h ,0),(c h ,0), to approximately estimate the membership function of each fuzzy set:

[0387]

[0388] Each parameter set D r Contains N h Each element.

[0389] After determining the three vertices of the triangle membership function, the input membership number can be obtained as follows:

[0390]

[0391] 2) Construct a security situation assessment model

[0392] The workshop safety distance d save and dynamic collision time T DC Substituting the parameter set into the membership degree calculation formula above, we obtain the information about the workshop safety distance d. save and dynamic collision time T DC Membership degree of the security situation model input:

[0393] safe distance d save As a precise input, it is fuzzily divided into three membership degrees—high (H), medium (M), and low (L)—using the membership number formula, thus obtaining the safe distance d. save The membership degree set is {H, M, L}.

[0394] Dynamic collision time T DC As a precise input, it is fuzzily divided into three membership degrees—high (H), medium (M), and low (L)—using the membership number formula to obtain the dynamic collision time T. DC The membership degree set is {H, M, L}.

[0395] The two precise input quantities during the current driving process are converted into fuzzy output quantities. The fuzzy output quantity of the safety situation assessment is obtained according to the "if-then" rule. The fuzzy output quantity of the safety situation assessment is represented by S, and its membership set is defined as {H, M, L}, corresponding to five fuzzy membership degree sets: high risk (H), medium-high risk (MH), medium risk (M), medium-low risk (ML), and low risk (L).

[0396] The "if-then" rule is defined as follows:

[0397] If(d save is H)and(T DC is H)then(Sis H)

[0398] If(d save is H)and(T DC is M)then(Sis MH)

[0399] If(d save is H)and(T DC is L)then(Sis ML)

[0400] If(d save is M)and(T DC is H)then(Sis MH)

[0401] If(d save is M)and(T DC is M)then(Sis M)

[0402] If(d save is M)and(T DC is L)then(Sis ML)

[0403] If(d save is L)and(T DC is H)then(Sis H)

[0404] If(d save is L)and(T DC is M)then(Sis ML)

[0405] If(d save is L)and(T DC is L)then(Sis L)

[0406] To deblur the output fuzzy value, the region centroid method is used, and the formula is as follows:

[0407]

[0408] Where S represents the fuzzy output quantity, f ch (S) indicates its membership degree.

[0409] To obtain an accurate assessment of the current security situation. The data is then interpolated to form a security situation value map.

[0410] 3) Divide the security status value range

[0411] Based on the quantification of safety status under different driving conditions, a safety status quantification value map is obtained. The intersection point of the safety status quantification values ​​is used as the critical point for distinguishing safety statuses. The critical point is mapped to the x-axis, dividing the safety status quantification value map into three different safety status intervals. The critical points of the intervals are s1 and s2, where s1 is the critical point between the normal driving interval and the collision warning interval, and s2 is the critical point between the collision warning interval and the braking collision avoidance interval. Figure 7 The diagram shows the division of security status value intervals.

[0412] The three different safety status intervals correspond to three different driving safety states: normal driving, collision warning, and braking collision avoidance.

[0413] The normal driving state is when the distance to the vehicle in front is far or there is no vehicle in front, and there is no need to consider the impact of changes in the driving conditions of the vehicle in front on the driving conditions of the vehicle itself. The vehicle is in a state of free driving.

[0414] The collision warning status indicates that the distance to the vehicle in front is relatively close, and it is necessary to consider the impact of the changes in the driving conditions of the vehicle in front on the driving conditions of the vehicle itself. The vehicle is in a state of cautious driving.

[0415] The braking and collision avoidance state refers to a situation where there is a risk of collision with the vehicle in front under the current driving conditions, and it is necessary to brake immediately to reduce the vehicle speed to ensure the safety of the driving process. The vehicle is in a deceleration driving state.

[0416] 4) Rules for switching security statuses corresponding to security status values

[0417] During driving, the vehicle will predict its speed in real time and assess the current driving safety situation, quantifying the current safety situation value. Whether the vehicle crosses the original safe situation range, it determines whether the vehicle's safety status has changed in the current driving scenario.

[0418] When the driving mode of the current vehicle changes, but the safety status value remains within the safety status range, the safety status of the vehicle remains unchanged.

[0419] When the vehicle's driving mode changes and the safety situation quantification value also changes within the safety situation range, the vehicle's safety situation value under the current driving conditions is... Switch to a safety status suitable for the current driving conditions. When the safety status value... When the vehicle's driving mode switches to normal driving mode, the safety situation value When the vehicle's driving mode is switched to collision warning mode, and the safety situation value... The vehicle's driving mode has been switched to collision avoidance mode.

[0420] By using a method that maps safety situation values ​​to safety states, the calculation of safety situation values ​​and the corresponding or switching of safety states can be completed simultaneously with the prediction of vehicle speed, thereby improving the driving safety of the vehicle during operation.

Claims

1. A method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion, characterized in that, Includes the following steps: 1) Obtain the training parameters of the speed prediction model. 2) Train the neural network using the training parameters of the speed prediction model to construct the speed prediction model. 3) Use a speed prediction model to predict the vehicle speed at the next moment; 4) Acquire driving safety information and process it to obtain safety situation assessment parameters: 5) Process the security situation assessment parameters to obtain the current security situation assessment value. This allows us to determine the safe driving status of the vehicle.

2. The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion according to claim 1, characterized in that, The speed prediction model training parameters include driving intention information D. i Historical vehicle speed information V i Workshop motion information M i ; Driving Intent Information D i The acquisition steps include: identifying the driving intention based on the information monitored by the pedal sensor, and obtaining the driving intention information D. i . Historical vehicle speed information V i The acquisition steps include: identifying and capturing historical vehicle speed information during the vehicle's driving process based on onboard sensors to obtain historical vehicle speed information V. i ; Workshop motion information M i The acquisition steps include: exchanging driving data between the vehicle and the preceding vehicle based on the vehicle-to-vehicle communication protocol, obtaining the relative distance and relative speed between the two vehicles, and obtaining the vehicle-to-vehicle motion information M. i .

3. The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion according to claim 1, characterized in that, The steps for training a neural network include: 1) Set the number of nodes in the input layer to n, the number of nodes in the hidden layer to l, and the number of nodes in the output layer to m; initialize the connection weights w between the input layer and the hidden layer. ij The connection weights w between the hidden layer and the output layer jk Threshold a of each neuron in the hidden layer j Threshold b of each neuron in the output layer k ; 2) Calculate the hidden layer output value H j ,Right now: In the formula, x is the activation function of the hidden layer; i This serves as input information for training parameters in the velocity prediction model; 3) Calculate the predicted output value T of the output layer k ,Right now: 4) Calculate the prediction error e k ,Right now: e k =Y k -T k ,k=1,2,...,m (3) In the formula, Y k This refers to the actual output information in the training parameters of the velocity prediction model. 5) Based on prediction error e k The backpropagation continuously feeds back and corrects the connection weights w ij Connection weight w jk Threshold a j Threshold b k until the prediction error e k Less than the preset error threshold; Updated connection weight w ij ′、Connection weight w jk ′、threshold a j ′、threshold b k As shown below: b k ′=b k +e k ,k=1,2,...,m (7) In the formula, η is the learning rate.

4. The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion according to claim 1, characterized in that, When training a neural network, the connection weights w between the input layer and the hidden layer ij The connection weights w between the hidden layer and the output layer jk Threshold a of each neuron in the hidden layer j Threshold b of each neuron in the output layer k The initial value is determined through the following steps: 1) Population initialization: The neuron thresholds and connection weights between adjacent neuron layers are integrated into a real number string in the form of chromosome real number encoding, and the boundary range is defined for random initialization; 2) Calculate fitness J i ,Right now: In the formula, K is a constant value; m is the number of output layer nodes; and k is the kth node of the output layer. 3) The selection operation is completed based on the proportional selection method, where the probability p of each individual i being selected is... i As shown below: In the formula, K0 is a constant value; J i ′ represents the fitness value of chromosome i; M represents the population size. 4) Perform mutual crossover operations based on the real number crossover method. The crossover process is as follows: In the formula, b is a random number in the interval [0, 1]; a m，l The value of chromosome m at point l; a n，l The value of chromosome n at point l; a m，l ′、a n，l ′ represents the value of chromosomes m and n at point l after the crossover operation; 5) Select the j-th gene a from the i-th chromosome. ij Performing a mutation operation yields: In the formula, a max For gene a ij The upper limit of; a min For gene a ij The lower bound value; q is the number of iterations at this time; a ij ′ represents the mutated gene; The function f(q) is shown below: In the formula, r is a random number in the interval [0, 1]; G max This represents the maximum number of evolutions. 6) Repeat steps 2)-5) to obtain the individual with the best fitness, and use it as the connection weight w between the input layer and the hidden layer. ij The connection weights w between the hidden layer and the output layer jk Threshold a of each neuron in the hidden layer j Threshold b of each neuron in the output layer k The initial value.

5. The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion according to claim 1, characterized in that, The vehicle's speed at the next moment is shown below: V o =f BP-GA (I more )=V o+1 ,V o+2 ,...,V o+p (14) In the formula, V o For predicting the vehicle speed sequence in the time domain p; f BP-GA This represents the function mapping relationship within the velocity prediction model; V o+1 V o+2 V o+p Predict the vehicle speed at future time points 1, 2, and p. Multi-source information sequence I more As shown below: In the formula, D i-k D i-k+1 D i Information on the driver's intention; M i-k M i-k+1 M i For workshop motion information; V i-k V i-k+1 V i This is historical vehicle speed information.

6. The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion according to claim 1, characterized in that, The driving safety information includes the movement information of the vehicle in front, the movement information of the vehicle itself, and the relative information between the vehicles. The safety situation assessment parameters include the workshop safety distance d. save and dynamic collision time T DC。 7. The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion according to claim 1, characterized in that, Workshop safety distance d save As shown below: In the formula, Δd is the distance between the two vehicles; t1 is the braking reaction time; t2 is the braking gap elimination time; v o v represents the vehicle's speed at the next moment. p The speed of the vehicle in front; a o For the vehicle's acceleration; a p Acceleration for the vehicle in front.

8. The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion according to claim 1, characterized in that, Calculate the dynamic collision time T in the workshop DC The steps include: 1) Determine the vehicle state vector J = [d r v o a o D i ] T ;d r The relative distance between the vehicle in front and the vehicle itself; 2) Establish the vehicle's state matrix J from the previous time step. t-Δp ,Right now: J t-Δp =[d r-Δp ,v o-Δp ,a o-Δp ,D i-Δp ] T (17) In the formula, d r-Δp v o-Δp a o-Δp D i-Δp The relative distance between the vehicle in front and the vehicle itself at the previous moment, the vehicle's speed, the vehicle's acceleration, and the driver's driving intention; 3) Based on the vehicle's state matrix from the previous time step, construct the vehicle's state estimation matrix J for the current time step. t ,Right now: In the formula, Δp is the time difference; v p This refers to the vehicle's speed. 4) The mean square error D in predicting the vehicle's current state. t ,Right now: In the formula, F t-1 Let Q be the state transition matrix from the previous time step. t-1 Let D be the noise matrix of the previous time step. t-1 This represents the mean square error of the vehicle's state at the previous moment; The state transition matrix F is shown below: In the formula, v o-p The vehicle's speed; a o-p a o-Δp For the vehicle's acceleration; 5) Calculate the Kalman filter gain K at the current time. t ,Right now: In the formula, N t For the observation noise matrix, U t The observation matrix; Among them, the observation matrix U t As shown below: 6) Calculate the optimal estimate J at the current time based on the Kalman gain. t ',Right now: J t ′=J t +K t (Z t -U t J t ) (23) In the formula, Z t The observed value at the current moment; 7) Update the state covariance error, i.e.: D t ′=(I-K t U t )D t (24) In the formula, I is the identity matrix with the same dimension as the state vector; D t 'This represents the updated state covariance error. 8) Based on the optimal estimate J t Determine the prediction step size t; calculate the distance between the two vehicles; when the relative distance between the two vehicles is less than or equal to the design minimum safe clearance a, record the step number n up to that moment, and then calculate the dynamic collision time T. DC = n*t.

9. The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion according to claim 1, characterized in that, The steps for processing security situation assessment parameters include: 1) Use the Euclidean distance method to determine the workshop safety distance d. save and dynamic collision time T DC Perform fuzzy clustering to obtain the input parameter set D r ; 2) Input parameter set D r Substituting the membership function of the triangle to determine the three vertices (a h ,0),(b h ,0),(c h ,0), to estimate the membership function of each fuzzy set; Wherein, coordinate parameter a h Coordinate parameter b h Coordinate parameter c h As shown below: In the formula, N h For parameter set D r Number of elements in D; r_1 D r_2 , D r_i For parameter set D r Elements in; 3) Establish the input membership number equation, that is: In the formula, f ch (x) represents the input membership number; 4) The workshop safety distance d save and dynamic collision time T DC By substituting the parameter set into the membership equation, the safe distance d can be calculated. save and dynamic collision time T DC The membership degree is used as a fuzzy input. According to the "if-then" rule, the fuzzy output of the security situation assessment is obtained. The fuzzy output is then defuzzified and used as the current security situation assessment value.

10. The method for assessing the safety situation of intelligent connected vehicles based on multi-source information fusion according to claim 1, characterized in that: The vehicle's driving safety status includes normal driving status, collision warning status, and braking and collision avoidance status. Among them, normal driving state means: the distance to the vehicle in front is greater than the preset distance threshold or the vehicle in front does not exist, and there is no need to consider the impact of the changes in the driving conditions of the vehicle in front on the driving conditions of the vehicle itself. The collision warning status means that the distance to the vehicle in front is less than the preset distance threshold but there is no risk of collision. It is necessary to consider the impact of the change in the driving conditions of the vehicle in front on the driving conditions of the vehicle itself. Braking and collision avoidance status refers to a situation where there is a risk of collision with the vehicle in front under the current driving conditions, and it is necessary to brake immediately to reduce the vehicle's speed to ensure the safety of the driving process.

Images