CN115892039A - Method for predicting scene fitness of automatic driving automobile - Google Patents

Method for predicting scene fitness of automatic driving automobile Download PDF

Info

Publication number
CN115892039A
CN115892039A CN202211513165.XA CN202211513165A CN115892039A CN 115892039 A CN115892039 A CN 115892039A CN 202211513165 A CN202211513165 A CN 202211513165A CN 115892039 A CN115892039 A CN 115892039A
Authority
CN
China
Prior art keywords
scene
automatic driving
elements
fitness
risk
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.)
Pending
Application number
CN202211513165.XA
Other languages
Chinese (zh)
Inventor
郭柏苍
韩卓桐
王胤霖
刘星辰
雒国凤
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.)
Yanshan University
Original Assignee
Yanshan University
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 Yanshan University filed Critical Yanshan University
Priority to CN202211513165.XA priority Critical patent/CN115892039A/en
Publication of CN115892039A publication Critical patent/CN115892039A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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

Landscapes

  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention relates to a method for predicting scene fitness of an automatic driving automobile. And then considering two dimensions of safety and efficiency, and calculating the adaptability of the automatic driving automobile to the high-risk scene by using an analytic hierarchy process according to the selected evaluation index and grading. And finally, inputting the obtained scene key elements and the scene fitness grade into a prediction model constructed based on a recurrent neural network, predicting the scene fitness of the automatic driving system at the next moment, and obtaining the prediction model of the scene fitness of the automatic driving automobile. The method can improve the safety of the automatic driving technology and provide theoretical basis and technical support for high-grade automatic driving quick landing.

Description

Method for predicting scene fitness of automatic driving automobile
Technical Field
The disclosure relates to the field of testing of an automatic driving automobile, in particular to a method for predicting scene fitness of the automatic driving automobile.
Background
Automobile driving automation grading and SAE J3016 issued by the Ministry of industry and belief of the people's republic of China: the driving automation hierarchy all defines 6 levels of automation levels, and 4 dimensions of horizontal/longitudinal motion control, environmental monitoring and object taking over, dynamic driving task assistance (DDT) and design operating range (ODD) define dividing elements of different levels. Focusing on an automatic driving automobile, 3-5 levels of automatic driving systems have independent automatic driving permission, and the difference is that dynamic driving task assistance is different from a designed operation range which is an important factor for distinguishing the automatic driving automobile subordinate automation level.
However, the domestic and foreign automatic driving classification standards are limited to conceptually defining responsibility distribution relations of automatic systems of various levels, and the design operation range of the automatic driving systems of various levels cannot be quantitatively analyzed and clearly defined, so that the response capability of the automatic driving systems of various levels to driving scenes cannot be accurately described. In addition, although there are many documents on the test methods for automatically driving automobiles at home and abroad, there are few relevant researches on the designed operating range of the automatically driving automobiles, so that the objective test, evaluation and classification of the automatic driving level of the automobile is still in a blank stage at present.
Disclosure of Invention
Aiming at the problem that the adaptability of the automatic driving system to the scene is not considered in the current method for testing the capability of the automatic driving system, the invention aims to provide an automatic driving scene adaptability prediction model, select a proper evaluation index, calculate the weighted value of each index by using an analytic hierarchy process, input the adaptability of the automatic driving system in a dangerous scene and dangerous scene elements into a prediction model built based on a recurrent neural network to effectively predict and grade the adaptability of the automatic driving scene at the next moment, and improve the reliability, the safety and the traffic efficiency of the automatic driving vehicle in different road scenes.
In order to solve the above technical problems, the technical solution of the present invention is as follows.
In a first aspect, the present invention provides a method for predicting a scene fitness of an autonomous vehicle, the method comprising the following steps:
extracting and reducing dimensions of elements in the dangerous scene detected at the current moment, and taking the former s characteristic elements as high-risk elements;
and analyzing and acquiring the state of each high-risk element at the current moment and the time sequence characteristics of the corresponding scene fitness thereof by adopting a trained recurrent neural network model based on the high-risk elements at the current moment and the scene fitness calculated at the moment, and predicting the scene fitness at the next moment.
According to the technical scheme, the recurrent neural network capable of processing time sequence data of any length is adopted, and the time sequence characteristics of high-risk elements in historical dangerous scenes are learned, so that the scene adaptability of the next time can be effectively predicted on the basis of analyzing the high-risk elements and the adaptability of the current time.
In the above technical solution, after extracting and performing dimension reduction processing on elements in a dangerous scene detected at the current time, taking the first s characteristic elements as high risk elements, including:
s101, extracting scene elements from the dangerous scene data set to obtain a scene dangerous element matrix P mn Wherein: m is a dangerous element serial number, and n represents the number of the extracted scene segments;
s102, standardizing the dangerous element matrix to obtain a standardized matrix P * ：
P * ＝[Q m1 ，Q m2 ，…，Q mn ]
Wherein:
Figure BDA0003970268680000031
in the formula: sigma n The calculation method is that the standard deviation of the characteristic parameters of the nth scene segment is as follows:
Figure BDA0003970268680000032
s103, standardizing a matrix P * Calculating a correlation coefficient matrix R among all elements;
s104, calculating a characteristic value and a characteristic vector of the R, and arranging the characteristic value and the characteristic vector in a descending order;
wherein: the characteristic values are: lambda [ alpha ] 1 ≥λ 2 ≥…λ n The corresponding feature vector is more than or equal to 0: e.g. of the type j ＝(l 1 ，l 2 …，l n )。
S105, calculating the cumulative contribution rate of the elements:
Figure BDA0003970268680000033
s106, according to the set principal component contribution rate value threshold, when the cumulative contribution rate of the elements is larger than or equal to the set principal component contribution rate value threshold, the first S characteristic elements are reserved, and the matrix P after dimensionality reduction is obtained sn 。
In the above technical solution, the scene fitness is obtained by calculating through the following steps:
constructing a high-risk scene based on the obtained first s characteristic elements;
calculating a set evaluation index based on the high risk scene;
calculating the weight of each evaluation index by adopting an analytic hierarchy process;
based on the obtained index weight, calculating the adaptability of the automatic driving to the scene, wherein the calculating method comprises the following steps:
Figure BDA0003970268680000041
in the formula: d is the adaptability of the automatic driving scene,
Figure BDA0003970268680000042
is the ith index weight, S i Is the attribute value of the ith evaluation index, and belongs to [0, 100 ]]，S i ∈[0，100]。
In the technical scheme, the evaluation index weight is calculated based on a judgment matrix, and the judgment matrix is constructed by adopting a 1-9 scale method;
and recording the judgment matrix as A, then:
Figure BDA0003970268680000043
wherein: a is a ij And the importance comparison relationship between the ith evaluation index and the jth evaluation index is shown.
In the above technical solution, the threshold range of the principal component contribution value is 75% to 90%.
In the technical scheme, the scene fitness is divided into four grades of extreme poor, good and excellent according to the value of the scene fitness from low to high, and the effectiveness of the automatic driving system is evaluated through the grades.
In the above technical solution, the evaluation index includes: the automatic driving accumulated test mileage, the automatic driving accumulated test duration, the risk avoidance and separation rate, the traffic capacity influence rate and the running speed difference rate; wherein:
the automatic driving accumulated testing mileage (km. (car-year) -1 ) The average total mileage of each automatic driving vehicle in the same enterprise in one year when the driving mode is automatic driving;
the automatic driving accumulated test duration (h · (vehicle-year) - 1 ) The method refers to the total running time of each automatic driving vehicle in the same enterprise in the same year when the driving mode is automatic driving;
the risk avoidance separation rate (second hundred kilometers) -1 ) The number of times that the unit mileage (hundred kilometers) in one year is averagely separated from an automatic driving mode and switched into an artificial driving mode with an automatic driving vehicle of an enterprise due to risk avoidance is determined;
the traffic capacity influence rate (%) refers to a change rate of road traffic capacity before and after the automatic driving road test, and the calculation formula is as follows:
Figure BDA0003970268680000051
in the formula:
Figure BDA0003970268680000052
the road traffic capacity before the automatic driving drive test is indicated; />
Figure BDA0003970268680000053
The road traffic capacity after the automatic driving drive test is indicated;
the running speed difference (%) refers to a difference between an average running speed of an autonomous vehicle and an average running speed of a manually-operated vehicle, and is calculated as follows:
Figure BDA0003970268680000054
in the formula:
Figure BDA0003970268680000055
is the average running speed of the autonomous vehicle at the time of the autonomous drive test; />
Figure BDA0003970268680000056
Is the average running speed of the surrounding manually-driven vehicles during the test.
In the above technical solution, the recurrent neural network model is one of the following: LSTM, RNN, GRU, BPTT.
In a second aspect, the present invention provides an apparatus for predicting a scene fitness of an autonomous vehicle, comprising a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and execute any of the above methods.
In a third aspect, the present invention provides a computer-readable storage medium, characterized in that: a computer program is stored which can be loaded by a processor and which performs any of the methods described above.
Drawings
FIG. 1 is a schematic flow diagram of a method in one embodiment;
FIG. 2 is a schematic diagram of a scene adaptive judgment index hierarchy model in one embodiment;
FIG. 3 is a schematic diagram of an LSTM cell structure in one embodiment;
FIG. 4 is a diagram illustrating the structure of the prediction model based on LSTM in one embodiment.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments.
The automatic driving test and evaluation technology is a necessary means for accelerating the intelligent automobile to mature and apply, in order to endow an automatic driving system with an objective and effective driving scene coping level (credibility of a safe passing scene) self-evaluation function,
in one embodiment, a method for predicting the adaptability of an automatic driving automobile to a scene is implemented, and referring to fig. 1, the method comprises the following steps:
and S100, extracting and reducing dimensions based on elements in the dangerous scene data set to obtain a high-risk scene data set.
The dimension reduction method is a PCA dimension reduction method, and the dimension reduction process is as follows:
(1) selecting a car following scene as a dangerous scene to obtain a scene main element matrix P mn Wherein: m is the number of the main element, and n represents the number of the extracted scene segments.
In the embodiment, specifically, as shown in table 1, the scene body element has a body element number m of 16 and n of 1010 according to 1010 extracted following scenes.
The obtained car following yard Jing Zhuti element matrix:
Figure BDA0003970268680000071
TABLE 1 scene subject elements
Figure BDA0003970268680000072
(2) Normalizing the main element matrix:
Figure BDA0003970268680000081
in the formula: sigma n The calculation method is that the standard deviation of the characteristic parameters of the nth scene segment is as follows:
Figure BDA0003970268680000082
after each parameter in the matrix is normalized by a formula, the finally obtained main element normalized matrix is as follows:
P * ＝[Q m1 ，Q m2 ，…，Q mn ]
wherein: q mn The value of the m row and n column in the normalized matrix; p * Is a standardized matrix of subject elements.
(3) Calculating a correlation coefficient matrix R:
calculating a correlation coefficient between elements, namely a correlation coefficient between rows by using a pearson correlation coefficient, and obtaining a correlation coefficient matrix as follows:
Figure BDA0003970268680000083
Figure BDA0003970268680000084
in the formula: x is a radical of a fluorine atom i 、y i Are respectively two characteristic elements to be calculated, which are P sn Corresponding row vector of (1), and (2) to be calculatedElement correlation of correlation coefficients;
(4) calculating the characteristic value and the characteristic vector of the R, and arranging the characteristic value and the characteristic vector from large to small respectively;
wherein: the characteristic values are: lambda [ alpha ] 1 ≥λ 2 ≥…λ n The corresponding feature vector is more than or equal to 0: e.g. of the type j ＝(l 1 ，l 2 …，l n )。
(5) Calculating the cumulative contribution rate of the first s elements:
Figure BDA0003970268680000091
(6) screening elements to obtain a matrix P after dimension reduction sn
In the principal component analysis process of the PCA algorithm, the accumulated contribution rate of the task principal components reaches more than 85%, and the principal components can represent original information. Therefore when a sj When the characteristic elements of the first s items are not less than 85 percent, the characteristic elements of the first s items are reserved as high risk elements, and the matrix P after dimension reduction is obtained sn ：
Figure BDA0003970268680000092
And S200, obtaining the scene adaptability of the automatic driving system based on the high-risk scene data set.
The scene fitness calculation process is as follows:
(1) constructing a high-risk scene based on the main elements obtained in the step (1)
(2) Setting evaluation index in the obtained high risk scene
In the embodiment, 5 evaluation indexes capable of representing safety and efficiency are selected from the automatic driving scene fitness prediction model. In the aspect of safety, accident rate statistical data in a drive test stage is lacked at present, so that a risk avoidance and disengagement rate index capable of representing safety is selected. If the calculated risk avoidance separation rate is difficult to objectively reflect the performance of the automatic driving drive test when the accumulated test mileage or the accumulated test duration is too low, the accumulated test mileage and the accumulated test duration are considered at the same time. Along with the increase of the accumulated test mileage and the accumulated test duration, the safety of the automatic driving vehicle is also improved to a certain degree. Therefore, the accumulated test mileage and the accumulated test duration are also selected as the safety evaluation index of the suitability of the automatic driving scenario. And in the aspect of efficiency, selecting the running speed difference rate representing the individual efficiency and the traffic capacity influence rate representing the road section traffic efficiency.
Automatic driving cumulative testing mileage (Km (vehicle year) -1 ) The average total mileage of each automatic driving vehicle in the same enterprise in one year when the driving mode is automatic driving.
Automatic driving cumulative test duration (h (vehicle year) -1 ) The method refers to the total running time of each automatic driving vehicle in the driving mode of automatic driving in the same enterprise in one year.
Escape rate of risk -1 ) The number of times that the unit mileage (hundred kilometers) in one year is averagely separated from an automatic driving mode and switched into a manual driving mode with an enterprise automatic driving vehicle due to risk avoidance is determined.
The traffic capacity influence rate (%) indicates a change rate of road traffic capacity before and after the automatic driving test, and is calculated as follows:
Figure BDA0003970268680000101
in the formula:
Figure BDA0003970268680000102
the road traffic capacity before the automatic driving drive test is indicated; />
Figure BDA0003970268680000103
The road traffic capacity after the automatic driving drive test is indicated.
The operation speed difference rate (%) indicates a difference rate between the average operation speed of the autonomous vehicle and the average operation speed of the manually driven vehicle, and is calculated as follows:
Figure BDA0003970268680000104
in the formula:
Figure BDA0003970268680000105
the average running speed of the autonomous vehicle in the autonomous driving test; />
Figure BDA0003970268680000106
Is the average running speed of the surrounding manually-driven vehicles during the test.
(3) Calculating the weight of each evaluation index
1) Building a hierarchical model
The evaluation index weight calculation only considers an automatic driving scene fitness evaluation target layer and an index layer containing 5 evaluation indexes, and a specific hierarchical structure model is shown in fig. 2 and comprises five evaluation indexes of 'automatic driving accumulated test mileage', 'automatic driving accumulated test duration', 'risk avoidance and disengagement rate', 'traffic capacity influence rate' and 'operation capacity difference rate'.
2) The evaluation indexes are calculated, and are shown in table 2:
table 2 evaluation indexes
Figure BDA0003970268680000111
3) Constructing a judgment matrix, and calculating index weight:
and on the basis of the established hierarchical structure model, performing evaluation index importance comparison through expert scoring, and determining the weight of each evaluation index relative to the target. For the significance comparisons in this study, the judgment matrix A was constructed using a 1-9 scale method, as follows:
Figure BDA0003970268680000112
wherein: a is ij The importance comparison relationship between the ith evaluation index and the jth evaluation index is represented, and the value taking method is shown in table 3:
TABLE 3 evaluation index importance Scale
Figure BDA0003970268680000121
For the comparison of importance in this study, the judgment matrix A was constructed using a 1-9 scale method as follows:
Figure BDA0003970268680000122
based on the matrix a, the importance of each evaluation index is calculated as follows:
Figure BDA0003970268680000123
in the formula:
Figure BDA0003970268680000124
is the weight of the i-th evaluation criterion>
Figure BDA0003970268680000125
The calculation results are shown in table 4:
TABLE 4 weight of each evaluation index
Figure BDA0003970268680000126
Figure BDA0003970268680000131
4) Based on the obtained index weight, calculating the adaptability of the automatic driving to the scene, wherein the calculating method comprises the following steps:
Figure BDA0003970268680000132
in the formula: d is the adaptability of the automatic driving scene,
Figure BDA0003970268680000133
is the ith index weight, si is the attribute value of the ith evaluation index, and D belongs to [0, 100 ]]，S i ∈[0，100]。
Using the values in tables 2 and 3, substituting the above equation:
D＝0.1086×298.9+0.0416×10.4+0.5048×13.0+0.1725×7.0+0.1725×6.0＝41.698
s300, analyzing and acquiring the state of each high-risk element at the current moment and the time sequence characteristics of the corresponding scene fitness thereof by adopting a trained recurrent neural network model based on the risk elements in the high-risk scene data set at the current moment and the scene fitness calculated at the moment, and predicting the scene fitness at the next moment.
The recurrent neural network model is one of the following: LSTM, RNN, GRU, BPTT. The following embodiments use LSTM.
The LSTM is a variation of a circulating neural network model, inherits the excellent characteristics of the circulating neural network, and simultaneously relieves the problem of gradient disappearance in the back propagation process of the neural network. And as the computational power of computers increases, LSTM has a strong advantage in processing time series information. Therefore, the LSTM is selected to solve the problem of prediction of the scene adaptability of the automatic driving system. The basic structure of the LSTM cell is shown in fig. 3. In FIG. 3, b i 、b f 、b c 、b o 、h t-1 Respectively an input gate, a forgetting gate, a cell unit, a bias item corresponding to the output gate and a network output at the time t-1. The upper left corner unit body f of the rectangular frame is an output gate, the middle right unit gate f is a forgetting gate, the lower left corner unit gate f is an input gate, the middle lower unit gate g is a tanh () function calculation unit, and the middle upper unit gate h is a tanh () function calculation unit.
In this embodiment, a prediction model structure based on LSTM is used, which includes an input layer, a hidden layer, and an output layer, where the input is the states of the feature elements in the scene detected at time t and the field at that timeAnd (3) outputting a scene adaptability predicted value at the t +1 moment, wherein the hidden layer is provided with one or more layers of neural networks consisting of LSTM cells. As shown in fig. 4, the input values are the state value and fitness D (t) of the feature elements after dimensionality reduction, and the feature elements P of s items are obtained after dimensionality reduction 1 (t)，P 2 (t)，...，P s (t)。
Selection of LSTM model parameters:
selecting an appropriate number of hidden layers may improve the performance of the training model. At this stage, the parameters of the predictive model are determined empirically and by the amount of data in the model, with no particular strategy.
The input and the output of the model constructed by the embodiment are both 600 time steps, the model is staggered by one time step, and the predicted time interval is 0.1s. The basic parameters of the training model are shown in table 6, and the remaining parameters are set to default values.
TABLE 5 basic parameters and values for LSTM fitness model training
Figure BDA0003970268680000141
Figure BDA0003970268680000151
Wherein the time step is 0.1s; the meaning of s +1 is: and the sum of the number of the characteristic elements subjected to dimensionality reduction and the fitness at the moment is recorded as an input dimensionality.
And finally, determining the hidden layer to be 30 neurons, optimizing by adopting an Adam algorithm, selecting a Tanh function as an activation function in an LSTM neural network model with the learning rate of 0.001, and training for 200 times.
The input is the state of the characteristic elements in the scene detected at the time t and the scene fitness calculated at the time t, the output is a predicted value D (t + 1) of the scene fitness at the time t +1, and a neural network composed of a layer of LSTM cells is arranged in the hidden layer and comprises 30 neuron cells.
With the prediction model constructed, the output predicted value of D (t + 1) is 39.852.
And S400, obtaining a predicted scene fitness grade based on the output scene fitness predicted value.
The method includes the steps that the opinions of experts are obtained through questionnaire collection, the adaptability of an automatic driving system to a scene is divided into 5 levels, and the adaptability is extremely poor, good and excellent from low to high. The scores corresponding to the 4 grades were set to 0-25, 25-50, 50-75, 75-100, respectively, as shown in table 6:
TABLE 6 grade of adaptability of automatic driving system to scene
Figure BDA0003970268680000152
Based on the classification of the scene fitness level in table 6, the predicted scene fitness D (t + 1) is 39.852, and the level is "poor" and output.
By grading the scene fitness, the effectiveness of the automatic driving system in the scene, namely the adaptability to the scene, can be visually embodied.
Through the above description of the embodiments, those skilled in the art will clearly understand that the present disclosure may be implemented by software plus necessary general hardware, and certainly may also be implemented by special hardware including special integrated circuits, special CPUs, special memories, special components and the like. Generally, functions performed by computer programs can be easily implemented by corresponding hardware, and specific hardware structures for implementing the same functions may be various, such as analog circuits, digital circuits, or dedicated circuits. However, more often than not for the purposes of this disclosure, software program implementations are preferred embodiments.
To sum up: in the embodiment, the scene adaptability (scienios fitness) of the automatic driving system is innovatively provided, and an evaluation index system of the performance of the automatic driving system is established, and an automatic driving vehicle scene adaptability prediction method is adopted, so that effective technical support and theoretical support are provided for testing and evaluating the membership grade of the automatic driving system and enabling the system to autonomously judge the driving capability.
Although the embodiments of the present invention have been described above with reference to the accompanying drawings, the present invention is not limited to the above-described embodiments and application fields, and the above-described embodiments are illustrative, instructive, and not restrictive. Those skilled in the art, having the benefit of this disclosure, may effect numerous modifications thereto without departing from the scope of the invention as defined by the appended claims.

Claims (10)

1. A method for predicting the adaptability of an automatic driving automobile to a scene is characterized by comprising the following steps:
extracting and reducing dimensions of elements in the dangerous scene detected at the current moment, and taking the former s characteristic elements as high-risk elements;
and analyzing and acquiring the state of each high-risk element at the current moment and the time sequence characteristics of the corresponding scene fitness thereof by adopting a trained recurrent neural network model based on the high-risk elements at the current moment and the scene fitness calculated at the moment, predicting the scene fitness at the next moment and grading.
2. The method according to claim 1, wherein after extracting and dimension-reducing the elements in the dangerous scene detected at the current time, taking the first s characteristic elements as high-risk elements comprises:
s101, extracting scene elements from the dangerous scene data set to obtain a scene dangerous element matrix P mn Wherein: m is a dangerous element serial number, and n represents the number of the extracted scene segments;
s102, standardizing the risk element matrix to obtain a standardized matrix P * ：
P * ＝[Q m1 ，Q m2 ，…，Q mn ]
Wherein:
Figure FDA0003970268670000011
in the formula: sigma n The calculation method is that the standard deviation of the characteristic parameters of the nth scene segment is as follows:
Figure FDA0003970268670000012
s103, standardizing a matrix P * Calculating a correlation coefficient matrix R among all elements;
s104, calculating a characteristic value and a characteristic vector of the R, and arranging the characteristic value and the characteristic vector in a descending order;
wherein: the characteristic values are: lambda [ alpha ] 1 ≥λ 2 ≥…λ n The corresponding feature vector is more than or equal to 0: e.g. of the type j ＝(l 1 ，l 2 …，l n )。
S105, calculating the cumulative contribution rate of the elements:
Figure FDA0003970268670000021
s106, according to the set principal component contribution rate value threshold, when the cumulative contribution rate of the elements is larger than or equal to the set principal component contribution rate value threshold, the first S characteristic elements are reserved, and the matrix P after dimensionality reduction is obtained sn 。
3. The method of claim 1, wherein the scene fitness is calculated by:
constructing a high-risk scene based on the obtained first s characteristic elements;
calculating a set evaluation index based on the high risk scene;
calculating the weight of each evaluation index by adopting an analytic hierarchy process;
based on the obtained index weight, calculating the adaptability of the automatic driving to the scene, wherein the calculating method comprises the following steps:
Figure FDA0003970268670000022
/>
in the formula: d is the fitness of the automatic driving scene,
Figure FDA0003970268670000023
is the ith index weight, S i Is the attribute value of the ith evaluation index, and belongs to [0, 100 ]]，S i ∈[0，100]。
4. The method of claim 3, wherein:
the evaluation index weight is calculated based on a judgment matrix, and the judgment matrix is constructed by adopting a 1-9 scale method;
and recording the judgment matrix as A, then:
Figure FDA0003970268670000031
wherein: a is ij And the importance comparison relationship between the ith evaluation index and the jth evaluation index is shown.
5. The method of claim 1, wherein the principal component contribution value threshold range is 75% to 90%.
6. The method according to claim 1, wherein the scene fitness is classified into four grades of poor, good and excellent according to the value of the scene fitness, and the effectiveness of the automatic driving system is evaluated according to the grades.
7. The method according to claim 3, wherein the evaluation index includes: the automatic driving accumulated test mileage, the automatic driving accumulated test duration, the risk avoidance and separation rate, the traffic capacity influence rate and the running speed difference rate; wherein:
the total driving mileage (km. (car. Year) -1 ) The average total mileage of each automatic driving vehicle in the same enterprise in one year when the driving mode is automatic driving;
the said autopilot accumulated test duration (h (vehicle year) -1 ) The total driving time of each automatic driving vehicle in the driving mode of automatic driving in the same enterprise in one year is indicated;
the risk avoidance separation rate (second hundred kilometers) -1 ) The number of times that a unit mileage (hundred kilometers) in one year is averagely switched from an automatic driving mode to an artificial driving mode from an automatic driving mode of each vehicle of an enterprise due to risk avoidance;
the traffic capacity influence rate (%) refers to a change rate of road traffic capacity before and after the automatic driving road test, and the calculation formula is as follows:
Figure FDA0003970268670000041
in the formula:
Figure FDA0003970268670000042
the road traffic capacity before the automatic driving drive test is indicated; />
Figure FDA0003970268670000043
The road traffic capacity after the automatic driving drive test is indicated;
the running speed difference rate (%) is a difference rate between the average running speed of the autonomous vehicle and the average running speed of the manually driven vehicle, and is calculated as follows:
Figure FDA0003970268670000044
in the formula:
Figure FDA0003970268670000045
the average running speed of the autonomous vehicle in the autonomous driving test; />
Figure FDA0003970268670000046
Is the average running speed of the surrounding manually-driven vehicles during the test.
8. The method of claim 1, wherein the recurrent neural network model is one of: LSTM, RNN, GRU, BPTT.
9. The utility model provides a prediction unit of autodrive car to scene fitness which characterized in that: comprising a memory and a processor, said memory having stored thereon a computer program which can be loaded by the processor and which performs the method of any of claims 1 to 8.
10. A computer-readable storage medium, characterized in that: a computer program which can be loaded by a processor and which performs the method according to any one of claims 1 to 8.
CN202211513165.XA 2022-11-29 2022-11-29 Method for predicting scene fitness of automatic driving automobile Pending CN115892039A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211513165.XA CN115892039A (en) 2022-11-29 2022-11-29 Method for predicting scene fitness of automatic driving automobile

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211513165.XA CN115892039A (en) 2022-11-29 2022-11-29 Method for predicting scene fitness of automatic driving automobile

Publications (1)

Publication Number Publication Date
CN115892039A true CN115892039A (en) 2023-04-04

Family

ID=86492289

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211513165.XA Pending CN115892039A (en) 2022-11-29 2022-11-29 Method for predicting scene fitness of automatic driving automobile

Country Status (1)

Country Link
CN (1) CN115892039A (en)

Similar Documents

Publication Publication Date Title
CN112085947B (en) Traffic jam prediction method based on deep learning and fuzzy clustering
Dereli et al. Industrial applications of type-2 fuzzy sets and systems: A concise review
CN111915059B (en) Attention mechanism-based Seq2Seq berth occupancy prediction method
Zhong et al. Genetically designed models for accurate imputation of missing traffic counts
CN111882869B (en) Deep learning traffic flow prediction method considering adverse weather
Zhao et al. A fuzzy system for concrete bridge damage diagnosis
CN112270355B (en) Active safety prediction method based on big data technology and SAE-GRU
CN114015825B (en) Method for monitoring abnormal state of blast furnace heat load based on attention mechanism
CN113723010A (en) Bridge damage early warning method based on LSTM temperature-displacement correlation model
CN116976857B (en) Predictive maintenance method and system for railway power equipment based on deep learning
CN111009122A (en) Tunnel traffic operation risk assessment method
CN116383685A (en) Vehicle lane change detection method based on space-time interaction diagram attention network
CN107688556B (en) Real-time travel time calculation method based on functional principal component analysis
CN114596726B (en) Parking berth prediction method based on interpretable space-time attention mechanism
CN111907523A (en) Vehicle following optimization control method based on fuzzy reasoning
CN112580254B (en) Multi-dimensional analysis and early warning method for converter transformer based on multi-level decision fusion
CN115892039A (en) Method for predicting scene fitness of automatic driving automobile
Dui et al. Reliability Evaluation and Prediction Method with Small Samples.
CN116128108A (en) Traffic accident severity prediction method, system and computer storage medium
CN116631203A (en) Self-adaptive traffic signal control method combining traffic flow prediction and reinforcement learning
CN116451322A (en) Bayesian optimization-based LSTM deep learning network mechanical prediction method
CN114463978B (en) Data monitoring method based on track traffic information processing terminal
CN116434531A (en) Short-time traffic flow prediction method based on Conv1D-LSTM model
CN116110219A (en) Traffic accident prediction method
CN112991765B (en) Method, terminal and storage medium for updating road high-emission source recognition model

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