CN116882505A - Speed prediction method and system in self-vehicle expressway scene - Google Patents

Speed prediction method and system in self-vehicle expressway scene Download PDF

Info

Publication number
CN116882505A
CN116882505A CN202311139549.4A CN202311139549A CN116882505A CN 116882505 A CN116882505 A CN 116882505A CN 202311139549 A CN202311139549 A CN 202311139549A CN 116882505 A CN116882505 A CN 116882505A
Authority
CN
China
Prior art keywords
vehicle
self
bayesian network
variable
vehicle speed
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202311139549.4A
Other languages
Chinese (zh)
Other versions
CN116882505B (en
Inventor
于会龙
陈研博
席军强
王海睿
闫国富
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Institute of Technology BIT
Original Assignee
Beijing Institute of Technology BIT
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Institute of Technology BIT filed Critical Beijing Institute of Technology BIT
Priority to CN202311139549.4A priority Critical patent/CN116882505B/en
Publication of CN116882505A publication Critical patent/CN116882505A/en
Application granted granted Critical
Publication of CN116882505B publication Critical patent/CN116882505B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00—Computing arrangements based on specific mathematical models
    • G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/10—Services
    • G06Q50/26—Government or public services
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
    • Y02T10/00—Road transport of goods or passengers
    • Y02T10/10—Internal combustion engine [ICE] based vehicles
    • Y02T10/40—Engine management systems

Abstract

The invention discloses a vehicle speed prediction method and a system under a self-vehicle expressway scene, which relate to the technical field of vehicle speed prediction and comprise the following steps: acquiring an environment data set corresponding to a self-vehicle in a historical period based on a dynamic environment model in a highway scene; calculating a vehicle speed limit value to be predicted according to a preset prediction time domain and the environment data set by adopting an adaptive window search algorithm; training a preset Bayesian network according to a training sample set based on a expectation maximization algorithm to obtain a trained Bayesian network; based on the environment data set, a joint tree algorithm is adopted, and vehicle speed prediction output is determined according to the trained Bayesian network and the vehicle speed limit value to be predicted; and determining the final predicted speed of the self-vehicle according to the predicted output of the vehicle speeds corresponding to the preset prediction time domains by adopting an adaptive weight algorithm. The invention improves the real-time performance and accuracy of vehicle speed prediction.

Description

Speed prediction method and system in self-vehicle expressway scene
Technical Field
The invention relates to the technical field of vehicle speed prediction, in particular to a vehicle speed prediction method and system in a self-vehicle expressway scene.
Background
The speed prediction has important theoretical value and wide application value for intelligent automobile development. Firstly, the future speed is predicted, so that the man-machine coordination consistency can be improved, the lane changing operation of a driver is assisted, and the driving burden of the driver is reduced. For example, in an overtaking scenario, an electronic control unit (Electronic Control Unit, ECU) predicts the speed of the own vehicle in advance for a period of time in the future, thereby inferring the need for a driver to quickly overtake, assisting the driver in achieving a quick overtake by delaying upshifts to improve the vehicle's power performance. Second, the predicted vehicle speed profile may be used in a safety warning system. If the predicted speed differs from the actual speed by more than a certain amount, it may give the driver an appropriate warning to improve the safety of driving. Finally, the introduction of uncertain future speed information may be used in a control system to improve vehicle dynamics, comfort, and fuel economy. Therefore, the method has important research significance for accurately and rapidly predicting the speed of the man-machine co-driving vehicle.
Achieving accurate speed prediction is a critical and complex task, however, because it is highly dependent on the driving manoeuvre characteristics of the human driver, the dynamically time-varying vehicle conditions and the constraints of the dynamic environment. Furthermore, most existing studies fail to achieve satisfactory prediction accuracy in a dynamic environment in which the number of surrounding vehicles varies with time. In consideration of processing highly uncertain driving behaviors, existing prediction methods based on probability models, such as Gaussian mixture hidden Markov (Gaussian Mixture Model-Hidden Markov Model, GMM-HMM), dynamic Bayesian networks (Dynamic Bayesian Network, DBN) and the like, are widely applied to speed prediction, and GMM-HMM is a DBN with a special structure, but because the GMM-HMM is simple and fixed in structure and limited in performance in complex vehicle speed prediction tasks, the prediction process using the DBN model is very complex, the traditional traversal search method is high in complexity, actual driving characteristics are not considered, and the real-time requirements of prediction cannot be met.
Disclosure of Invention
The invention aims to provide a vehicle speed prediction method and a vehicle speed prediction system under a self-vehicle expressway scene, and the real-time performance and accuracy of vehicle speed prediction are improved.
In order to achieve the above object, the present invention provides the following solutions:
a speed prediction method in self-vehicle expressway scene includes:
constructing a dynamic environment model under a highway scene; the dynamic environment model comprises a vehicle of a lane where a self vehicle is currently located, an adjacent lane of the lane where the self vehicle is currently located and a vehicle of the adjacent lane;
acquiring an environment data set corresponding to the self-vehicle in a history period based on the dynamic environment model; the environment data set comprises a self-vehicle speed acceleration data set, a self-vehicle adjacent lane data set and a self-vehicle relative distance data set under any historical time;
calculating a vehicle speed limit value to be predicted according to a preset prediction time domain and the environment data set by adopting an adaptive window search algorithm;
training a preset Bayesian network according to a training sample set based on a expectation maximization algorithm to obtain a trained Bayesian network; the training samples in the training sample set comprise sample environment data sets corresponding to the vehicles and vehicle driving intentions at the same time; the vehicle driving intention includes a vehicle lateral travel and a vehicle longitudinal travel;
based on the environment data set, determining a vehicle speed prediction output by adopting a joint tree algorithm according to the trained Bayesian network and the vehicle speed limit value to be predicted;
and determining the final predicted speed of the self-vehicle according to the predicted output of the vehicle speeds corresponding to the preset prediction time domains by adopting an adaptive weight algorithm.
A vehicle speed prediction system in a self-vehicle highway scene, comprising:
the environment model construction module is used for constructing a dynamic environment model in a highway scene; the dynamic environment model comprises a vehicle of a lane where a self vehicle is currently located, an adjacent lane of the lane where the self vehicle is currently located and a vehicle of the adjacent lane;
the environment data acquisition module is used for acquiring an environment data set corresponding to the self-vehicle in the historical period based on the dynamic environment model; the environment data set comprises a self-vehicle speed acceleration data set, a self-vehicle adjacent lane data set and a self-vehicle relative distance data set under any historical time;
the vehicle speed limit value calculation module to be predicted is used for calculating the vehicle speed limit value to be predicted according to a preset prediction time domain and the environment data set by adopting an adaptive window search algorithm;
the Bayesian network training module is used for training the preset Bayesian network according to the training sample set based on the expectation maximization algorithm so as to obtain a trained Bayesian network; the training samples in the training sample set comprise sample environment data sets corresponding to the vehicles and vehicle driving intentions at the same time; the vehicle driving intention includes a vehicle lateral travel and a vehicle longitudinal travel;
the single prediction module is used for determining vehicle speed prediction output according to the trained Bayesian network and the vehicle speed limit value to be predicted by adopting a joint tree algorithm based on the environment data set;
and the final prediction module is used for determining the final predicted speed of the self-vehicle according to the vehicle speed prediction output corresponding to the preset prediction time domains by adopting an adaptive weight algorithm.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention discloses a vehicle speed prediction method and a system under a self-vehicle expressway scene, which use the thought of self-vehicle multi-mode speed prediction of a multi-dynamic Bayesian network of a self-adaptive window search algorithm, and specifically comprise the following steps: based on the constructed dynamic environment model in the expressway scene, the environment data set corresponding to the self-vehicle in the historical period is acquired, wherein the environment data set comprises a speed acceleration data set of the self-vehicle, a neighboring lane data set of the self-vehicle and a relative distance data set of the self-vehicle and the neighboring vehicle at any historical moment, so that effective work in the expressway scene with dynamically-changed road constraint and flexible quantity (the quantity of the neighboring vehicles can be set) of surrounding vehicles is realized. And calculating a vehicle speed limit value to be predicted according to a preset prediction time domain and the environment data set by adopting an adaptive window search algorithm, so that the search area is limited by using historical data to achieve the aim of meeting the real-time performance of prediction. Then training a preset Bayesian network based on a expectation maximization algorithm according to a training sample set to obtain a trained Bayesian network, and further determining the speed prediction output of the vehicle to be predicted; because the DBN model is based on a first-order Markov assumption, historical data cannot be fully utilized, and considering the fact, the method and the device adopt an adaptive weight algorithm finally, according to the vehicle speed prediction output corresponding to a plurality of preset prediction time domains, the final prediction speed of the self vehicle is determined, namely, a plurality of DBNs are combined to obtain the mDBN, the weight is adjusted on line according to the prediction performance of the DBN for a period of time, if the speed prediction error of one model is increased, the weight of the model is reduced, and therefore the robustness of the model is improved, and the accuracy and the stability of the whole vehicle speed prediction are finally improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the drawings that are needed in the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a method for predicting vehicle speed in a self-vehicle highway scene according to the present invention;
FIG. 2 is a schematic diagram of a dynamic environmental model in a highway scene according to the present invention;
FIG. 3 is a schematic diagram of a Bayesian network in accordance with the present invention;
fig. 4 is a schematic structural diagram of a vehicle speed prediction system in a self-vehicle expressway scene according to the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.
The invention provides a vehicle speed prediction method and a vehicle speed prediction system in a self-vehicle expressway scene, which can improve the real-time performance, accuracy and stability of vehicle speed prediction.
In order that the above-recited objects, features and advantages of the present invention will become more readily apparent, a more particular description of the invention will be rendered by reference to the appended drawings and appended detailed description.
Example 1
As shown in fig. 1, the present invention provides a vehicle speed prediction method in a self-vehicle expressway scene, including:
step 100, constructing a dynamic environment model under a highway scene to obtain a dynamic environment of the self-vehicle EV under any highway scene for modeling; the dynamic environment model comprises a vehicle of a lane where a self vehicle is currently located, an adjacent lane of the lane where the self vehicle is currently located and a vehicle of the adjacent lane.
As shown in fig. 2, in a specific example, the data in the dynamic environment model includes data of the right front vehicle PV, the left front vehicle LPV, the right left vehicle LAV, the left rear vehicle LFV, the right front vehicle RPV, the right vehicle RAV, the right rear vehicle RFV, and whether or not a condition exists in the neighboring lane. Fig. 2 is a three-lane example, in other examples, introducing neighbor lane presence information may be generalized to dynamic environments such as two lanes, four lanes, etc. The invention does not consider the speed prediction influence of the rear vehicle on the self-vehicle, namely, the rear vehicle is supposed to react in time according to the change of the front vehicle.
Step 200, acquiring an environment data set corresponding to the self-vehicle in a history period based on the dynamic environment model; the environmental data set includes a self-vehicle speed acceleration data set, a neighboring lane data set of the self-vehicle, a relative distance data set of the self-vehicle and the neighboring vehicle at any one of the historical time points.
In one embodiment, based on the dynamic environment model, the self-vehicle perception system is used to collect the environment data set in real time, thus obtaining the sequence, wherein ,/>The total sequence input at the time t is represented, and the observation variable at the time t corresponds to the observation variable at the time t; />Representing the total sequence input at the t-obs moment, and corresponding to the observation variable at the t-obs moment; />The total sequence input at time t-obs+1 is represented, and the observation variable at time t-obs+1 is corresponding to each time in seconds.
The environmental data set includes a self-vehicle speed acceleration data set, a neighboring lane data set of the self-vehicle, a relative distance data set of the self-vehicle and the neighboring vehicle. The self vehicle speed acceleration data set E t Comprising the longitudinal speed of the own vehicleLateral speed->Longitudinal acceleration->Lateral acceleration->I.e. +.>。
The set of relative distance data between the self-vehicle and the adjacent vehicle comprises a relative distance between the self-vehicle and any vehicle in the set of adjacent vehicles; the adjacent vehicle set comprises a front vehicle, a left front vehicle, a front left vehicle, a left rear vehicle, a right front vehicle, a front right vehicle and a right rear vehicle, and all vehicles in the adjacent vehicle set take the self vehicle as a reference center; the adjacent vehicles are gathered intoThe data in the set represent the longitudinal relative distance between the time t { PV, LPV, LAV, LFV, RPV, RAV, RFV } and the EV, respectively.
The adjacent lane dataset of the own vehicle includes a first lane variable and a second lane variable; the first lane variable is used for representing whether a left lane exists in a lane where the self-vehicle is currently located; the second lane variable is used to characterize whether a right lane exists in the lane in which the own vehicle is currently located. That is, the adjacent lane data set L of the own vehicle t Comprises two variables, wherein the two variables are Boolean variables, and the first lane variable LLE t Indicating whether the current lane of the self-vehicle has a left lane at the time t, wherein the existence of the left lane is 1, the nonexistence of the left lane is 0, and the second lane variable RLE t And (5) indicating whether the current lane of the self-vehicle has a right lane at the time t, wherein the existence of the right lane is 1, and the nonexistence of the right lane is 0.
Step 300, calculating a vehicle speed limit value to be predicted according to a preset prediction time domain and the environment data set by adopting an adaptive window search algorithm; the vehicle speed limit to be predicted comprises a longitudinal speed upper bound, a longitudinal speed lower bound, a transverse speed upper bound and a transverse speed lower bound.
The adaptive window search algorithm has the function of limiting the search area within a certain range and guaranteeing the prediction accuracy and the real-time performance. Specifically, firstly, calculating weighted variance based on historical longitudinal and transverse acceleration, then, calculating longitudinal and transverse self-adaptive window widths based on least square linear fitting, and finally, calculating a search range by combining the historical acceleration and the self-adaptive window width.
(1) And calculating the weighted variance based on the historical longitudinal acceleration and the historical transverse acceleration, wherein the calculation formula is as follows:
。
。
wherein , and />The weighted variances of the longitudinal and lateral historic accelerations are shown, respectively. The reason for introducing the weighted variance is to take into account that even though both are historical time information, the information closer to the present time is more important than the more distant historical time information, and the weighted variance represents the difference in the importance of the historical data. />The length of the historical data in the environment data set, namely the length of the observed variable of the incoming predictive model, is expressed in seconds. /> and />Respectively representing the variance of the acceleration in the longitudinal i-th history time and the variance of the acceleration in the transverse i-th history time,/respectively> and />Respectively is、/>Corresponding weight parameters and satisfies the following conditions:
。
。
to determine and />The value of (1) is required to extract data from the HighD natural driving data set in advance as a training set for analysis, and calculate the variance of acceleration in the i-th history time of the longitudinal and transverse directions of the human driver in the HighD natural driving data set> and />And calculates the width of the adaptive window in the ideal longitudinal and transverse directions +.> and />The superscript train represents the training set for the superparameter +.> and />The calculation formula is as follows:
。
。
determines and />After the value of (2), the weighted variance of the longitudinal and transverse historical acceleration under the real driving scene can be calculated> and />。
(2) And calculating the width of the longitudinal and transverse self-adaptive window based on the least square method linear fitting. After calculating the variance of the weighted longitudinal and lateral acceleration, they are matched with the width of the longitudinal and lateral adaptive windows and />Fitting to a linear function as follows:
。
。
wherein , and />Fitting the longitudinal weighted variance and the longitudinal adaptive window width to coefficients of a linear function, respectively +.> and />The lateral weighted variance and the lateral adaptive window width are fitted to coefficients of a linear function, respectively. Their values are obtained based on the least squares method using training set data.
(3) The search range is calculated in combination with the historical acceleration and the adaptive window width. After the longitudinal and transverse self-adaptive windows are calculated, the prediction ranges of the longitudinal and transverse speeds are given by combining the prediction time domain, and the calculation formulas of the speed limit values of the vehicle to be predicted are respectively as follows:
。
。
。
。
wherein ,、/>、/>、/>respectively represent the upper boundary of the longitudinal speed,A longitudinal speed lower bound, a transverse speed upper bound, and a transverse speed lower bound; /> and />The longitudinal acceleration and the lateral acceleration at the present time are respectively indicated.
Step 400, training a preset Bayesian network according to a training sample set based on a expectation maximization algorithm to obtain a trained Bayesian network; the training samples in the training sample set comprise sample environment data sets corresponding to the vehicles and vehicle driving intentions at the same time; the vehicle driving intention includes a vehicle lateral travel and a vehicle longitudinal travel.
The training sample set is obtained by extracting data from a preset HighD data set. Specifically, the data extracted from the HighD natural driving data set is history data including, wherein ,/>The observation variable at the i-th history time is represented, and T represents the extracted sequence length. After obtaining a plurality of groups of sequences, firstly performing downsampling processing, wherein the sampling frequency of HighD original data is 25Hz, downsampling is 5Hz, and then at each time t, the longitudinal relative distance GP between a self-vehicle and adjacent vehicles in 7 interesting positions in a dynamic environment model t The following treatment is carried out:
。
other longitudinal distance GLP t 、GLA t 、GLF t 、GRP t 、GRA t 、GPF t And the same is true. 200m is the vision distance set by the invention, the vision distance exceeds the vision distance or no adjacent vehicles exist in the target area, and the characteristic value at the position is correspondingly set to be 200m, thereby conforming to the external environment of a human driverIs a perceived range and trend of (a).
Establishing a reasonable DBN network structure is a precondition for achieving excellent prediction performance, so that the preset bayesian network is obtained by performing structural design on the DBN based on expert priori driving experience, as shown in fig. 3. In the view of figure 3 of the drawings,representing generalized vehicle driving intent, M x 、M y The vehicle lateral running and the vehicle longitudinal running respectively refer to real change amounts of longitudinal and transverse speeds after prediction time domain, and the outline dotted line indicates that the node is hidden and cannot be observed. The meaning of the remaining nodes has been described in detail above, except that the network structure is clear, and the time t is uniformly written in the upper left corner.
Subject to self-vehicle history handling behavior (lateral longitudinal acceleration, etc.) and adjacent lane line constraint informationInteractive constraint information with surrounding vehicles +.>The former symbolizes the speed +.>The latter two symbolize the speed +.>Indicating a correction of the surrounding traffic flow to the pure physical information, where v x Representing a vehicle transverse speed parameter, v y Representing a vehicle longitudinal speed parameter. Finally, there is a certain time of action for the steering behavior at the current moment to change the state of the vehicle, so that the influence on the state is reflected in the next time slice, as the connection line crossing the time slices is shown in fig. 3.
Based on the foregoing, step 400 specifically includes:
(1) Based on the training sample set and model parameters of a preset Bayesian network, calculating the expectation of hidden variables. The expected calculation formula of the hidden variable is as follows:
。
(2) And carrying out maximum likelihood estimation on the model parameters of the preset Bayesian network according to the expectation of the hidden variables so as to obtain the maximum parameters. The calculation formula of the maximum parameter is as follows:
。
wherein ,log likelihood function for complete data->Regarding the observation variable given +.>And the current parameters->Lower pair hidden variable->Conditional probability distribution->Is (are) desirable to be (are)>Representing the desire to hide the variable, +.>Log likelihood function representing complete data, +.>Expressed in the given observation variable +.>And the current parameters->Lower pair hidden variable->Is a conditional probability distribution of (a),representing hidden variables in the mth training sample, namely driving intention of the vehicle; />Hidden variable representing the t-obs moment in the mth training sample,/o>Hidden variable representing the t-obs+1 time in the mth training sample,/o>Representing the hidden variable at the t-th moment in the mth training sample.
Representing the observed variable in the mth training sample; />Representing the observed variable at time t-obs in the mth training sample, +.>Representing the observation variable at time t-obs+1 in the mth training sample, +.>An observation variable representing the t time in the mth training sample; each observation variable includes vehicle speed data, vehicle acceleration data, adjacent lane data of the vehicle, the vehicleRelative distance data to adjacent vehicles; />Representing model parameters of a preset Bayesian network, and representing the conditional probability density represented by each side of the DBN network; />Representing the maximum parameter, which is the optimization variable; obs denotes the length of time of the training samples, i.e. the observed data length of time.
(3) And updating the model parameters of the preset Bayesian network to the maximum parameters.
(4) If the preset Bayesian network after the model parameter update does not reach the preset convergence condition, returning to the expected step of calculating the hidden variable based on the training sample set and the model parameter of the preset Bayesian network so as to carry out iterative repetition; the preset convergence condition refers to model parameters iterated in two times before and after, and the relative variation of the size of a likelihood function of the complete data is smaller than a certain threshold value.
(5) And if the preset Bayesian network after the model parameter updating reaches the preset convergence condition, marking the preset Bayesian network after the model parameter updating as a trained Bayesian network.
And 500, determining a vehicle speed prediction output according to the trained Bayesian network and the vehicle speed limit value to be predicted by adopting a joint tree algorithm based on the environment data set. Specifically, based on the trained bayesian network and the vehicle speed limit to be predicted, a probability distribution is calculated for the predicted speed within the search range, and a probability calculation is performed using Junction Tree Algorithm (joint tree algorithm) based on message passing.
Step 500 specifically includes:
(1) Converting the trained Bayesian network into a joint tree to obtain a plurality of nodes (clusters) and a plurality of separation sets (separators); two adjacent nodes are connected by a separation set; the random variables in the separate set are the intersection of the random variables of the two adjacent nodes;
(2) Initializing potential functions for each node and each separation set; specifically, the potential function for each node and separation set is initialized to 1.
。
(3) For each random variable V in any node, updating the potential function of the node using the random variable based on the conditional probability of the parent node of the random variable. The update formula is:
。
wherein ,a parent node variable representing variable V; />Representing the probability of choosing a random variable within its parent node variables, +.>Represents the potential function of cluster or isolate.
(4) Collecting messages from all neighbor nodes in parallel, and performing potential function calculation based on the received messages to obtain potential functions after the messages are received; the neighbor node is a node adjacent to the node where the random variable is located. Specifically, at each step, each node in the Junction Tree collects all messages from its neighbors in parallel and calculates:
。
wherein ,is a message sent by node j1 to node j2，/>Is the potential function of the initialized node j2, nbr (j 2) represents the node set consisting of neighbor nodes of node j2, +.>Is a potential function after message reception.
(5) Based on the potential function after the message reception, the node sends a message to each neighbor node. Specifically, once a node j2 receives messages from all its neighbors, it can send a piece of information to each neighbor:
。
wherein ,is the message sent by node j2 to node k, is->Characterizing node j1 represents traversing all nodes in nbr (j 2) except node k, +.>Is the message sent by node k to node j2,representing the potential function +.>Marginalizing to isolate set->On the sign->Meaning marginalization. The following are examples: sign->Meaning +.>，/>Representing the removal of elements in set S from elements in set T, i.e., marginalizing the probability from set S to set T.
(6) After the information transmission is carried out on all the nodes to update potential functions, the potential functions corresponding to the nodes containing the interesting variables in the joint tree are determined; the interesting variable is a speed value in the range of the speed limit value of the vehicle to be predicted; specifically, the environment data set is input into a trained Bayesian network to obtain a vehicle speed prediction feature set; and then, screening the data in the vehicle speed prediction feature set by adopting the vehicle speed limit value to be predicted, wherein the screened vehicle speed prediction feature is the interested variable.
After the potential function is updated through message transmission on all the variables, the interesting variables can be obtainedP (V) of (b), the calculation formula is as follows:
。
wherein X is any set containing a variable V node of interest, X\ { V } represents the set of elements in set X with the remainder of the elements in set V removed,representing the potential function of the updated set X.
(7) And normalizing potential functions corresponding to all the interesting variables to obtain a plurality of corresponding probability values. All the speeds in the self-adaptive window area are taken to be values through the methodThe probability is obtained after calculationFinally, obtaining normalized probability distribution by softmax operation>, wherein ,/>，/>Represents the lower bound of the longitudinal and lateral velocity search range, +.>Representing the upper bound of the cross-direction velocity search range.
。
(8) And taking the interested variable corresponding to the maximum probability value as the predicted output of the vehicle speed. The calculation formula of the vehicle speed prediction output v is:
。
and 600, determining the final predicted speed of the self-vehicle according to the predicted output of the vehicle speeds corresponding to the preset prediction time domains by adopting an adaptive weight algorithm.
In the step 500, the DBN prediction output of a single preset prediction time domain is obtained, in order to better utilize the historical data and improve the robustness of the prediction system, mDBN is obtained by combining DBNs of a plurality of different prediction time domains, the weights of a plurality of DBNs are adjusted on line in real time according to the prediction performance of each DBN in the historical 1s, and finally the weighted prediction value is returned as the final prediction speed. Taking 3 different preset prediction time domains as an example, the specific steps are as follows:
(1) The DBN network weights for 3 different prediction domains are initialized. The prediction result of the single DBN is given in the above stepsThe DBN model of 3 different prediction time domains is obtained by changing the size of the prediction time domains:、/>andwherein is h 1 =1、/>、/>. Considering that the prediction accuracy decreases as the prediction time domain increases, the prediction methodWith better prediction effect in the general case, considering this, the DBN network weights of 3 different prediction time domains are initialized first:
。
wherein ,、/> and />The initialization weights corresponding to the DBN models of 3 different prediction time domains are respectively represented, the superscript indicates the prediction time domain of the network, and the subscript indicates the iteration sequence number along with time.
(2) And carrying out linear weighting on the prediction results of the 3 prediction time domain DBNs to obtain a final prediction result. For the current time t, 3 DBN predictors are known、/> and />The corresponding weight is->、/>Andfinal prediction result->The calculation formula for the linear weighted sum of the 3 prediction time domain DBN predictors is as follows:
。
(3) As the prediction speed truth is learned, the Root Mean Square Error (RMSE) of the 3 prediction time domain DBNs is calculated. Speed truth value over time up to the predicted timeThe formula for calculating the root mean square error RMSE of 3 different prediction time domains DBN is known as follows:
。
。
。
wherein ,、/> and />The DBN models respectively representing 3 different prediction time domains predict performance indexes at the time t, the smaller the values, the more accurate the prediction is, and the +.>Mean that the root mean square error of a and B is calculated.
(4) According to root mean square error of 3 prediction time domain DBNs, corresponding weights are adjusted in real time based on an adaptive weight algorithm, and the method specifically comprises the following steps:
1) Calculating the DBN network with the best performance; wherein,the minimum value of root mean square error representing three prediction time domain DBNs is calculated as follows:
。
2) Increasing the weight of the best performing network, decreasing the weights of the remaining two networks:
。
。
。
and (3) circulating the steps (2) - (4), namely realizing that the mDBN with the self-adaptive weight adjusting function predicts the future speed of the self-vehicle.
In summary, the invention constructs a DBN network structure by using expert priori driving experience, extracts a training set based on HighD natural driving data set for training, predicts the future speed of the self-vehicle EV in a general dynamic environment in a highway scene by using a DBN model, designs an adaptive window searching algorithm based on historical driving data, limits the range of DBN searching to improve real-time computing efficiency, and solves the problem of low efficiency and poor real-time performance when the DBN model predicts continuous nodes. Based on single-chip DBN model prediction, an adaptive online weight algorithm is provided, DBNs of a plurality of prediction time domains are combined to obtain mDBNs, historical data is fully utilized, and the robustness of a prediction framework is improved.
Example two
As shown in fig. 4, in order to achieve the technical solution in the first embodiment to achieve the corresponding functions and technical effects, the present embodiment further provides a vehicle speed prediction system in a self-vehicle expressway scene, including:
the environment model construction module 101 is used for constructing a dynamic environment model in a highway scene; the dynamic environment model comprises a vehicle of a lane where a self vehicle is currently located, an adjacent lane of the lane where the self vehicle is currently located and a vehicle of the adjacent lane.
An environmental data obtaining module 201, configured to obtain an environmental data set corresponding to the self-vehicle in the history period based on the dynamic environmental model; the environmental data set includes a self-vehicle speed acceleration data set, a neighboring lane data set of the self-vehicle, a relative distance data set of the self-vehicle and the neighboring vehicle at any one of the historical time points.
The vehicle speed limit to be predicted calculating module 301 is configured to calculate a vehicle speed limit to be predicted according to a preset prediction time domain and the environmental data set by using an adaptive window searching algorithm.
The bayesian network training module 401 is configured to train the preset bayesian network according to the training sample set based on the expectation maximization algorithm, so as to obtain a trained bayesian network; the training samples in the training sample set comprise sample environment data sets corresponding to the vehicles and vehicle driving intentions at the same time; the vehicle driving intention includes a vehicle lateral travel and a vehicle longitudinal travel.
The single prediction module 501 is configured to determine a vehicle speed prediction output according to the trained bayesian network and the vehicle speed limit to be predicted by using a joint tree algorithm based on the environmental data set.
The final prediction module 601 is configured to determine a final predicted speed of the self-vehicle according to predicted vehicle speeds corresponding to a plurality of preset prediction time domains by using an adaptive weight algorithm.
Example III
The present embodiment provides an electronic device, including a memory and a processor, where the memory is configured to store a computer program, and the processor is configured to execute the computer program to cause the electronic device to execute the vehicle speed prediction method in the self-vehicle expressway scene of the first embodiment. Alternatively, the electronic device may be a server.
In addition, the embodiment of the invention also provides a computer readable storage medium storing a computer program, which when executed by a processor, implements the vehicle speed prediction method in the self-vehicle expressway scene of the first embodiment.
In the present specification, each embodiment is described in a progressive manner, and each embodiment is mainly described in a different point from other embodiments, and identical and similar parts between the embodiments are all enough to refer to each other. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant points refer to the description of the method section.
The principles and embodiments of the present invention have been described herein with reference to specific examples, the description of which is intended only to assist in understanding the methods of the present invention and the core ideas thereof; also, it is within the scope of the present invention to be modified by those of ordinary skill in the art in light of the present teachings. In view of the foregoing, this description should not be construed as limiting the invention.

Claims (8)

1. A method for predicting vehicle speed in a self-vehicle highway scene, the method comprising:
constructing a dynamic environment model under a highway scene; the dynamic environment model comprises a vehicle of a lane where a self vehicle is currently located, an adjacent lane of the lane where the self vehicle is currently located and a vehicle of the adjacent lane;
acquiring an environment data set corresponding to the self-vehicle in a history period based on the dynamic environment model; the environment data set comprises a self-vehicle speed acceleration data set, a self-vehicle adjacent lane data set and a self-vehicle relative distance data set under any historical time;
calculating a vehicle speed limit value to be predicted according to a preset prediction time domain and the environment data set by adopting an adaptive window search algorithm;
training a preset Bayesian network according to a training sample set based on a expectation maximization algorithm to obtain a trained Bayesian network; the training samples in the training sample set comprise sample environment data sets corresponding to the vehicles and vehicle driving intentions at the same time; the vehicle driving intention includes a vehicle lateral travel and a vehicle longitudinal travel;
based on the environment data set, determining a vehicle speed prediction output by adopting a joint tree algorithm according to the trained Bayesian network and the vehicle speed limit value to be predicted;
and determining the final predicted speed of the self-vehicle according to the predicted output of the vehicle speeds corresponding to the preset prediction time domains by adopting an adaptive weight algorithm.
2. The method of claim 1, wherein the self-vehicle speed acceleration data set includes longitudinal speed, lateral speed, longitudinal acceleration, and lateral acceleration of the self-vehicle;
the set of relative distance data between the self-vehicle and the adjacent vehicle comprises a relative distance between the self-vehicle and any vehicle in the set of adjacent vehicles; the adjacent vehicle set comprises a front vehicle, a left front vehicle, a front left vehicle, a left rear vehicle, a right front vehicle, a front right vehicle and a right rear vehicle, and all vehicles in the adjacent vehicle set take the self vehicle as a reference center;
the adjacent lane dataset of the own vehicle includes a first lane variable and a second lane variable; the first lane variable is used for representing whether a left lane exists in a lane where the self-vehicle is currently located; the second lane variable is used to characterize whether a right lane exists in the lane in which the own vehicle is currently located.
3. The method for predicting vehicle speed in a self-vehicle highway scene according to claim 2, wherein the vehicle speed limit to be predicted comprises a longitudinal speed upper bound, a longitudinal speed lower bound, a lateral speed upper bound, and a lateral speed lower bound;
the calculation formulas of the vehicle speed limit value to be predicted are respectively as follows:
；
；
；
；
wherein ,、/>、/>、/>respectively representing a longitudinal speed upper boundary, a longitudinal speed lower boundary, a transverse speed upper boundary and a transverse speed lower boundary; /> and />Respectively representing the longitudinal acceleration and the transverse acceleration at the current moment; />Representing a preset prediction time domain; /> and />Respectively representing the variance of the acceleration in the longitudinal i-th history time and the variance of the acceleration in the transverse i-th history time,/respectively> and />Respectively->、/>Corresponding weight parameters; /> and />Respectively representing the weighted variance of the longitudinal historical acceleration and the weighted variance of the transverse historical acceleration; t (T) h Representing the length of the historical data in the environmental data set; /> and />To fit the weighted variance of the longitudinal history acceleration to the adaptive window width to the coefficient of the linear function,/> and />To fit the weighted variance of the lateral history acceleration to the adaptive window width to coefficients of a linear function.
4. The method for predicting vehicle speed in a self-vehicle highway scene according to claim 1, wherein the training sample set is obtained by extracting data from a preset HighD data set;
the preset Bayesian network is obtained by structural design of the DBN based on expert priori driving experience.
5. The method for predicting vehicle speed in a self-vehicle highway scene according to claim 1, wherein training the preset bayesian network according to the training sample set based on the expectation maximization algorithm to obtain a trained bayesian network, specifically comprising:
calculating the expectation of hidden variables based on the training sample set and model parameters of a preset Bayesian network;
carrying out maximum likelihood estimation on the model parameters of the preset Bayesian network according to the expectation of the hidden variables so as to obtain maximum parameters;
updating the model parameters of the preset Bayesian network to the maximum parameters;
if the preset Bayesian network after the model parameter update does not reach the preset convergence condition, returning to the expected step of calculating the hidden variable based on the training sample set and the model parameter of the preset Bayesian network so as to carry out iterative repetition;
and if the preset Bayesian network after the model parameter updating reaches the preset convergence condition, marking the preset Bayesian network after the model parameter updating as a trained Bayesian network.
6. The method for predicting vehicle speed in a self-vehicle highway scene according to claim 5, wherein the expected calculation formula of the hidden variable is:
；
the calculation formula of the maximum parameter is as follows:
；
wherein ,log likelihood function for complete data->Regarding the observation variable given +.>And the current parameters->Lower pair hidden variable->Conditional probability distribution->Is (are) desirable to be (are)>Representing the desire to hide the variable, +.>Log likelihood function representing complete data, +.>Expressed in the given observation variable +.>And the current parameters->Lower pair hidden variable->Is a conditional probability distribution of (a),representing hidden variables in the mth training sample; />Hidden variable representing the t-obs moment in the mth training sample,/o>Hidden variable representing the t-obs+1 time in the mth training sample,/o>A hidden variable representing the t moment in the mth training sample;representing the observed variable in the mth training sample; />Representing the observed variable at time t-obs in the mth training sample, +.>Representing the observation variable at time t-obs+1 in the mth training sample, +.>An observation variable representing the t time in the mth training sample; each observation variable comprises vehicle speed data, vehicle acceleration data, adjacent lane data of the vehicle, and relative distance data between the vehicle and the adjacent vehicle; />Representing model parameters of a preset Bayesian network, and representing the conditional probability density represented by each side of the DBN network; />Representing the maximum parameter, obs represents the length of time the training sample.
7. The method for predicting vehicle speed in a self-vehicle highway scene according to claim 1, wherein determining a vehicle speed prediction output according to the trained bayesian network and the vehicle speed limit to be predicted by adopting a joint tree algorithm based on the environmental data set, specifically comprises:
converting the trained Bayesian network into a joint tree to obtain a plurality of nodes and a plurality of separation sets; two adjacent nodes are connected by a separation set; the random variables in the separate set are the intersection of the random variables of the two adjacent nodes;
initializing potential functions for each node and each separation set;
updating a potential function of any node for each random variable in the node using the random variable based on a conditional probability of a parent node of the random variable;
collecting messages from all neighbor nodes in parallel, and performing potential function calculation based on the received messages to obtain potential functions after the messages are received; the neighbor node is a node adjacent to the node where the random variable is located;
based on the potential function after the message is received, the node sends a message to each neighbor node;
after the information transmission is carried out on all the nodes to update potential functions, the potential functions corresponding to the nodes containing the interesting variables in the joint tree are determined; the interesting variable is a speed value in the range of the speed limit value of the vehicle to be predicted;
normalizing potential functions corresponding to all the interesting variables to obtain a plurality of corresponding probability values;
and taking the interested variable corresponding to the maximum probability value as the predicted output of the vehicle speed.
8. A vehicle speed prediction system in a self-vehicle highway scene, the system comprising:
the environment model construction module is used for constructing a dynamic environment model in a highway scene; the dynamic environment model comprises a vehicle of a lane where a self vehicle is currently located, an adjacent lane of the lane where the self vehicle is currently located and a vehicle of the adjacent lane;
the environment data acquisition module is used for acquiring an environment data set corresponding to the self-vehicle in the historical period based on the dynamic environment model; the environment data set comprises a self-vehicle speed acceleration data set, a self-vehicle adjacent lane data set and a self-vehicle relative distance data set under any historical time;
the vehicle speed limit value calculation module to be predicted is used for calculating the vehicle speed limit value to be predicted according to a preset prediction time domain and the environment data set by adopting an adaptive window search algorithm;
the Bayesian network training module is used for training the preset Bayesian network according to the training sample set based on the expectation maximization algorithm so as to obtain a trained Bayesian network; the training samples in the training sample set comprise sample environment data sets corresponding to the vehicles and vehicle driving intentions at the same time; the vehicle driving intention includes a vehicle lateral travel and a vehicle longitudinal travel;
the single prediction module is used for determining vehicle speed prediction output according to the trained Bayesian network and the vehicle speed limit value to be predicted by adopting a joint tree algorithm based on the environment data set;
and the final prediction module is used for determining the final predicted speed of the self-vehicle according to the vehicle speed prediction output corresponding to the preset prediction time domains by adopting an adaptive weight algorithm.
CN202311139549.4A 2023-09-06 2023-09-06 Speed prediction method and system in self-vehicle expressway scene Active CN116882505B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311139549.4A CN116882505B (en) 2023-09-06 2023-09-06 Speed prediction method and system in self-vehicle expressway scene

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311139549.4A CN116882505B (en) 2023-09-06 2023-09-06 Speed prediction method and system in self-vehicle expressway scene

Publications (2)

Publication Number Publication Date
CN116882505A true CN116882505A (en) 2023-10-13
CN116882505B CN116882505B (en) 2023-12-12

Family

ID=88271846

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311139549.4A Active CN116882505B (en) 2023-09-06 2023-09-06 Speed prediction method and system in self-vehicle expressway scene

Country Status (1)

Country Link
CN (1) CN116882505B (en)

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130085976A1 (en) * 2010-06-18 2013-04-04 Honda Motor Co., Ltd. System for predicting a driver's intention to change lanes
WO2015144410A1 (en) * 2014-03-24 2015-10-01 Bayerische Motoren Werke Aktiengesellschaft Device for predicting driving state transitions
CN110297494A (en) * 2019-07-15 2019-10-01 吉林大学 A kind of automatic driving vehicle lane-change decision-making technique and system based on rolling game
WO2022206079A1 (en) * 2021-04-01 2022-10-06 天津大学 Iterative optimization multi-scale fusion speed prediction algorithm for intelligent connected vehicle
US20230137489A1 (en) * 2021-11-02 2023-05-04 China Automotive Innovation Co., Ltd Method for detecting can bus intrusion of vehicle-mounted network based on gmm-hmm and system

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130085976A1 (en) * 2010-06-18 2013-04-04 Honda Motor Co., Ltd. System for predicting a driver's intention to change lanes
WO2015144410A1 (en) * 2014-03-24 2015-10-01 Bayerische Motoren Werke Aktiengesellschaft Device for predicting driving state transitions
CN110297494A (en) * 2019-07-15 2019-10-01 吉林大学 A kind of automatic driving vehicle lane-change decision-making technique and system based on rolling game
WO2022206079A1 (en) * 2021-04-01 2022-10-06 天津大学 Iterative optimization multi-scale fusion speed prediction algorithm for intelligent connected vehicle
US20230137489A1 (en) * 2021-11-02 2023-05-04 China Automotive Innovation Co., Ltd Method for detecting can bus intrusion of vehicle-mounted network based on gmm-hmm and system

Also Published As

Publication number Publication date
CN116882505B (en) 2023-12-12

Similar Documents

Publication Publication Date Title
CN107169567B (en) Method and device for generating decision network model for automatic vehicle driving
WO2022052406A1 (en) Automatic driving training method, apparatus and device, and medium
CN107229973B (en) Method and device for generating strategy network model for automatic vehicle driving
US20230280702A1 (en) Hybrid reinforcement learning for autonomous driving
CN112201069B (en) Deep reinforcement learning-based method for constructing longitudinal following behavior model of driver
CN110068302A (en) A kind of vehicle odometry method based on deep neural network
CN113538910A (en) Self-adaptive full-chain urban area network signal control optimization method
CN114358128A (en) Method for training end-to-end automatic driving strategy
He et al. Probabilistic intention prediction and trajectory generation based on dynamic bayesian networks
CN112141098B (en) Obstacle avoidance decision method and device for intelligent driving automobile
Wang et al. Cognitive competence improvement for autonomous vehicles: A lane change identification model for distant preceding vehicles
CN113561995B (en) Automatic driving decision method based on multi-dimensional reward architecture deep Q learning
Wei et al. End-to-end vision-based adaptive cruise control (ACC) using deep reinforcement learning
CN112462759B (en) Evaluation method, system and computer storage medium of rule control algorithm
CN116882505B (en) Speed prediction method and system in self-vehicle expressway scene
Hao et al. Aggressive lane-change analysis closing to intersection based on UAV video and deep learning
CN112085101A (en) High-performance and high-reliability environment fusion sensing method and system
US11794780B2 (en) Reward function for vehicles
CN116166642A (en) Spatio-temporal data filling method, system, equipment and medium based on guide information
CN114926823A (en) WGCN-based vehicle driving behavior prediction method
CN115009291A (en) Automatic driving aid decision-making method and system based on network evolution replay buffer area
Chen et al. Imitating driver behavior for fast overtaking through bagging Gaussian process regression
CN114444597B (en) Visual tracking method and device based on progressive fusion network
Li et al. Research on vehicle lane change recognition based on ABC-SVM algorithm
Yang et al. A Knowledge Distillation-based Ground Feature Classification Network with Multiscale Feature Fusion in Remote Sensing Images

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant