CN102609599B - Method for designing emulational underground road alignment and transverse clear distance based on multiple intelligent agents - Google Patents
Method for designing emulational underground road alignment and transverse clear distance based on multiple intelligent agents Download PDFInfo
- Publication number
- CN102609599B CN102609599B CN201210104967.5A CN201210104967A CN102609599B CN 102609599 B CN102609599 B CN 102609599B CN 201210104967 A CN201210104967 A CN 201210104967A CN 102609599 B CN102609599 B CN 102609599B
- Authority
- CN
- China
- Prior art keywords
- operator
- state
- rule
- represent
- clear distance
- 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.)
- Active
Links
Images
Abstract
The invention relates to road traffic safety and design and aims at providing an ideal method for designing road alignment and transverse clear distance. For achieving the purpose, the technical scheme is that the method for designing the emulational underground road alignment and the transverse clear distance based on multiple intelligent agents adopts an SOAR cognitive framework to model driver intelligent agents driving in underground roads and comprises working memory, long-term memory, a procedural operator selecting method and a learning mechanism sub-module. The method for designing emulational underground road alignment and transverse clear distance based on the multiple intelligent agents is mainly applied to the design of the road alignment and the transverse clear distance.
Description
Technical field
The present invention relates to traffic safety and design, specifically relate to the linear and horizontal clear distance method for designing of road tunnel based on multiple agent emulation.
Background technology
Reality shows, road tunnel is due to its closure and hide the limitation in space, the occurrence frequency of traffic accidents is several times of Ordinary Rd, the order of severity is very high and very easily cause the generation of follow-up accident, and the road tunnel traffic engineering that several aspects such as Ren,Che, road and environment are comprehensively considered design is considered to one of important measures that solve highway road tunnel traffic safety problem.In the various measures that relate in road tunnel traffic engineering, road tunnel optimal design linear and horizontal clear distance can help driver to make more effective driving behavior decision-making, reducing and drive anxiety and pressure, is one of means that really can guarantee road tunnel traffic safety, the raising traffic capacity.
Nederlandse Centrale Organisatie Voor Toegepast-natuurwetenschappelijk Onderzoek (TNO) thinks, the particular design in tunnel comprises the length in tunnel, type, width, and number, curvature and the fluctuation etc. of tunnel route curve affect driver's behavior and the subjective sense of security to a great extent.But these researchs are all the research for vcehicular tunnel, the research of road tunnel is mainly concentrated on to the qualitative analysis etc. of design, gateway speed change lane length and the traffic safety of transversal section, the less driving behavior to Impacts Evaluation of Urban Tunnel is studied.
Summary of the invention
The present invention is intended to solve and overcomes the deficiencies in the prior art, provides a kind of comparatively ideal road tunnel linear and horizontal clear distance method for designing.For achieving the above object, the technical scheme that the present invention takes is, linear and the horizontal clear distance method for designing of road tunnel based on multiple agent emulation, adopt SOAR cognitive frame to carry out modeling to the driver's intelligent body in underground road travel, comprise intelligent body working memory, long-term memory, procedural operator system of selection and study mechanism submodule;
SOAR cognitive frame working memory comprises the multidate information all about the world and internal reasoning of SOAR intelligent body, it comprises perception information, intermediate computations, the state of classification and relevant operator and target etc., the reasoning that intelligent body is all and decision-making are all carried out in working memory;
SOAR cognitive frame long-term memory characterizes with production rule through the procedural memory of road tunnel driving behavior, first, by " if " part of each rule, remove to mate the element in working memory, if " if " of rule part is mated completely with working memory element, this rule will be triggered, then by sending one to the information of kinematic system or producing the suggestion that changes current state, cause that " then " part triggers, any rule of mating with current goal, state and operator can change current target and state;
Program mode operator is selected: through in the SOAR cognitive frame of road tunnel, adopt numerical value preference, every rule in long-term memory rule base comprises that matching condition and matching condition meet the lower operator that can advise, and the numerical value preference value of advising this operator under this condition, new regulation of every interpolation need to judge the initial value preference size of operator in this rule, and according to extraneous feedback, the size of this value is upgraded so that it more approaches truth in decision process;
Study mechanism: at the processing procedure learning chunking rule of sub-state.When predicament produces, meaning in the long-term memory of current system does not have utilizable operator that problem solving process is moved forward in problem space, need automatically to create a new rule and solve current predicament, the foundation of chunking rule need to be analyzed production rule relevant to reaching result in long-term memory and episodic memory clue.
The initial long-term memory rule of SOAR cognitive frame long-term memory
In upper table, r1, r2, r3, r4, r5 represent rule numbers, and IF/THEN represents trigger condition and triggers result, [I] expression condition, [T] represents action, R-A1, R-A2, C-D, S-D, P-D, Tem, Des, D-S1, D-S2 is the abbreviation of object properties, represents respectively linear, linear eigenwert, lateral separation, blocks up, predicted congestion, mood, destination, transport condition, transport condition eigenwert.R-A1 (a) represents linear for a class, a=1 wherein, and 2,3 represent respectively straight-line segment, left-turning pathways and right-hand rotation bend; R-A2 (b) represents that the turning radius of linear eigenwert is b level, only have when for bend just has feature when linear, b=1,2,3,4 represent respectively turning radius be little, in, large, especially big; C-D (c) represents that horizontal clear distance value is c level, c=1 wherein, 2,3 represent horizontal clear distance be little, in, large; S-D (d1, d2, d3), P-D (d1, d2, d3) represent that the perception degree of blocking up of current road junction ahead clockwise direction corresponding road section and predicted congestion perception degree are d1, d2, d3 level, wherein parameter occurrence 1,2,3,4 represent unobstructed, generally block up and seriously block up; D-S1 (e) represents that transport condition classification is e, e=1 wherein, and 2,3 represent to keep straight on, turn left, turn right; D-S2 (f1, f2) represents that transport condition eigenwert rank is f1 and f2, f1=1 wherein, and 2 represent travel speed and angles, f2=1,2,3,4 expression speed and angle rank from small to large, occurrence is demarcated by experiment; Tem (g) represents that current driver's mood classification is g, g=1 wherein, 2 represent respectively happy and irritable; Des (h) represents that current destination direction is h, h=1 wherein, 2,3 represent respectively left, front, right.
Program mode operator is selected to be specially:
1) make O (s) represent the candidate's operator set under state s, if its radix | O (s) | equal 1, select o ∈ O (s) to enter working memory, otherwise continue;
2) if | O (s) | > 1, p[o
max(s)]-p[o
sec(s)]>=τ (s), selects from O (s) with roulette mechanism to enter working memory when pre-operator, otherwise continues; P[o wherein
max(s)], p[o
sec(s)] the preference of optimum operator and suboptimum operator under expression state s, τ (s) represents that under state s, operator is directly selected threshold value;
3) if O (s)=Φ or | O (s) | > 1, p[o
max(s)]-p[o
sec(s)] < τ (s), can not directly carry out operator selection, produces predicament, enters chunk learning phase.
Study mechanism comprises solution and the chunking rule submodule of predicament:
When program mode memory can not be selected operator to current state, adopt following steps to solve predicament:
1) order is from original state s
0through i next state, shifting the current state obtaining is s
iif, O (s
i)=Φ or | O (s
i) | > 1, p[o
max(s
i)]-p[o
sec(s
i)] < τ (s), meet chunk condition for study, make j=i+1, continue; Otherwise adopt program mode memory to carry out operator selection;
2) if j=0 turns 5); Otherwise j=j-1;
3) comprise s find state transition path in all episodic memories storehouse in
jthe operator set of episodic memory, be designated as O
q(s
j), if O
q(s
j)=Φ, turns 2), otherwise continue;
4) at O
q(s
j) in select optimum operator to enter working memory, solve current predicament;
5) if O is (s
i)=Φ, changes the matching precision of current state with step-length 0.1, until meet current state operator set O (s '
i) appearance, wherein s '
irepresent to change the new state after matching precision;
6) at O (s
i) or O (s '
i) in adopt roulette mechanism to select operator to enter working memory, solve current predicament;
Chunking rule:
If intelligent body ran into predicament in state transitions, illustrate in decision process and have partial status there is no operator or can not directly select, after this time decision-making of intelligent body, need to adopt chunk learning method to create rule and preference renewal, if G is (s so
e)-G
e(s
e)≤η
z, to solving the operator of predicament, carry out chunk renewal, wherein G one time
e(s
e) represent that intelligent body is at state s
eunder expectation drive target, G (s
e) represent the actual driving target of intelligent body action, η
zfor regular chunk upgrades threshold value; If same rule is double, by chunk, upgraded, corresponding operator is added in decision process and run into predicament state, chunk is successful;
No matter be that chunk forms or the rule having existed, the operator that all will relate to this decision-making after intelligent body carries out feedback learning, because decision process relates to a plurality of states and operator, therefore adopt each state and the distance of dbjective state to distribute the feedback preference of corresponding operator; Decision-making period t state transition path in k the operator o (s that state is corresponding
k) feedback preference value be λ [d (s
k, s
e)] p
r(s
e), p wherein
r(s
e)=[G (s
e)-G
e(s
e)]
αfor end-state s
eunder total preference value of feedback, parameter alpha gets 0.5 herein; λ [d (s
k, s
e), r (s
k)] be p
r(s
e) be assigned to o (s
k) on weight, it is state s
kto s
edistance d (s
k, s
e) and s
kstate transition path r (the s at place
k) function, herein
Wherein | r (s
k) | be s
kthe number of states that path, place comprises.
Technical characterstic of the present invention and effect:
Adopt simulation frame of the present invention, can obtain highway layout parameter and the mutual relationship of conflicting, obtain the linear and horizontal clear distance of comparatively ideal road tunnel, reduce the cost of road construction, improved again travel safety.
Accompanying drawing explanation
The structure of Fig. 1 SOAR framework.
Fig. 2 driver's intelligent body working memory.
Linear and the horizontal clear distance optimization Simulation framework of Fig. 3 road tunnel.
Linear and the horizontal clear distance scene 1 of Fig. 4 road tunnel.
Linear and the horizontal clear distance scene 2 of Fig. 5 road tunnel.
Fig. 6 road tunnel simulated scenario 1.
Fig. 7 road tunnel simulated scenario 2.
Fig. 8 straight-line segment traffic conflict figure (car) under the horizontal clear distance of middle difference that travels.
Fig. 9 straight-line segment traffic conflict figure (truck) under the horizontal clear distance of middle difference that travels.
Traffic conflict figure (car) under the different horizontal clear distances of Figure 10 negotiation of bends.
Traffic conflict figure (truck) under the different horizontal clear distances of Figure 11 negotiation of bends.
Embodiment
The present invention mainly sets up oversize vehicle multiple agent model based on SOAR cognitive frame, to different road alignments and laterally traffic behavior and the traffic flow character under clear distance carry out emulation and analysis, propose the linear and horizontal clear distance of the road tunnel optimized and design suggested design.
6.1.1 the impact analysis of general section to driving behavior
The general linear impact on driving behavior in section mainly contains:
1) long straight line road can make driver produce in the process of moving hypnosis phenomenon, and reaction slows up, dispersion attention, thus cause error in judgement in driving;
2) horizontal curve of correct radial, gives the suitable tension of driver, is conducive to traffic safety.Irrational Horizontal Curve can make driver drive in conjecture and anxiety, makes them be easy to fatigue by health, and cardiac load is excessive, and judgment declines, slow in reacting.The stopping sight distance that too small sweep can reduce driver makes them can not observe in advance turning, the place ahead, can not accomplish to know what's what to the tendency of road alignment;
3) long and steep downgrade easily causes driver's frequently braking, feels fatigue, anxiety;
4) convex vertical curve easily causes driver's sighting distance not enough, can't see the place ahead traffic conditions and causes mood dysphoria; Night is not enough without the vertical sag curve place sighting distance on illuminating road, easily causes driver's nervous psychology.
6.1.2 the impact analysis of road tunnel to driving behavior
Road tunnel linear on driving behavior except having the impact similar to general section, also there is its singularity, wherein linear change is more obvious on the impact of the speed of a motor vehicle and changes in heart rate.
1) speed of a motor vehicle: road tunnel gateway is that driver psychology changes significantly and section fast.Research finds, general vehicle has obvious deceleration entering the last set a distance in tube crossing, and in road tunnel, the speed of a motor vehicle generally fluctuates up and down in reduced levels scope, and has obvious accelerator going out tube crossing.The main cause that causes road tunnel to import and export region speed of a motor vehicle variation is visual adaptation and the road alignment factor that the inside and outside light difference of road tunnel causes.According to No. two, Qinling Mountains road tunnel and Zhonnanshan Mountain road tunnel investigation research are found to vehicle 5s design speed stroke range operating speed before and after road tunnel is imported and exported is all subject to linear impact.The security of operation that Guo Zhongyin etc. import and export road tunnel is studied, and thinks that importing and exporting linear transition is one of major influence factors of road tunnel safety.
Different linear road tunnel gateways produce Different Effects to internal environment, straight line road tunnel is unfavorable for light transition, can not effectively regulate driver psychology, the deceleration that cannot avoid " black-hole effect " and " white hole effect " to cause enters hole, the appearance of the phenomenon of accelerating to appear.The highway road tunnel overwhelming majority of Switzerland and Northern European countries makes curved, and this is conducive to improve driver's notice, and road tunnel gateway segment of curve can effectively solve " black and white hole " problem, the generation of minimizing accident.
In road tunnel, travel, driver's the visual field is subject to the impact of factors, and the visual field changes with the variation of the speed of a motor vehicle, and same, the speed of a motor vehicle changes with the variation in the visual field, always driver selects travel speed according to its visual field size.When the place ahead occurs that minor radius horizontal curve, longitudinal gradient, the Deng Shi visual field, road tunnel hole diminish, driver can underspeed accordingly; When the place ahead is for long straight line or large radius horizontal curve intervisibility are when good, visual field increase, driver can improve speed.
2) heart rate: research is found, the larger correlativity of turning radius existence of Variation of Drivers ' Heart Rate increment and road tunnel.Minor radius road tunnel increases because the reasons such as centrifugal force, sighting distance, improper ventilation easily cause Variation of Drivers ' Heart Rate, and driver's perception, mood are impacted.In addition, the confined space of road tunnel easily causes driver's constriction, sensation road tunnel inside lane width diminishes, curve road tunnel can cause the fear bumping against with road tunnel inwall to driver, produce so-called " abutment wall effect ", motoring pressure increases greatly, and likely loses one's temper, misoperation, causes accident.The road tunnel of one-way trip should be set to sweeping curve, and what the suitable heart rate of driver increased, psychological stress is beneficial to notice concentrates, and sweeping curve road tunnel illumination transitionality is good, is beneficial to driver's light process.
According to the investigation data in road tunnel group section, the Western Han Dynastry of the Chang An University high speed Qinling Mountains, road tunnel longitudinal gradient changes because of climb and fall to some extent on the impact of driver psychology, up section upward slope process heart rate increment rate changes with the gradient is different, in the longitudinal gradient process of changing from small to big, driver is in order to rush the more gradient, general speed promotes, and it is large that heart rate increment rate becomes, and in longitudinal gradient 2.7% left and right heart rate increment rate, reaches peak; The gradient continues to increase, and driver takes conservative measure, low-grade speed stabilizing climbing, and heart rate declines to some extent; Along with the gradient continues to increase, climbing difficulty increases, and driver psychology is nervous, and heart rate raises again.Descending process changes in heart rate is subject to speed of a motor vehicle control effect, in the gradient, is that augmentation of heart rate is maximum between 3～4% time, and the gradient continues to increase, driver's regulation speed that brakes, and heart rate reduces on the contrary.
3) horizontal clear distance: < < highway road tunnel design specifications > > Zhong Dui highway road tunnel stopping sight distance has been made clear and definite regulation, and the safe-stopping sight distance that wherein adopted is consistent with the stopping sight distance of common roadbed.Due to the restriction of profile in road tunnel, the horizontal clear distance in hole (viewpoint is to the distance of the barriers such as hole wall or maintaining roadway) is much smaller than the horizontal clear distance value of common roadbed.The horizontal clear distance of road tunnel is less, requires minimum radius of horizontal curve larger; Unidirectional two-way traffic road tunnel, the horizontal clear distance of left-hand lane is less than the horizontal clear distance of right-hand lane; According to vehicle heading, the horizontal clear distance of left-hand bend road tunnel is less than right-hand bend, and the road tunnel left-hand bend curve road tunnel curve road tunnel of turning right is unfavorable for safety.Therefore,, when road tunnel radius of horizontal curve is less than minimum sighting distance radius, should take measures to strengthen the horizontal clear distance of road tunnel.
The general introduction of 6.2SOAR cognitive frame
The full name of SOAR is state, operator and target (State, Operator and Result), in a kind of framework that is called " universal intelligent " of nineteen eighty-three exploitation by people such as Allen Newell, main discussion knowledge, thinking, intelligence and memory are cognitive structures that range of application is very wide.Wherein, state is that current situation about will deal with problems characterizes, and operator is the operation that can change state, produce new state, and target is the desired result that need to deal with problems.The operation of SOAR is exactly to apply constantly operator and select next operator until the process of this problem realization of goal at problem space.
SOAR framework mainly comprises I/O interface, long-term memory district, working memory district three parts, also has some other potential mechanism as decision-making period, learning process etc.Be illustrated in fig. 1 shown below:
As shown in Figure 1, SOAR must alternately, be mapped to the external world in working memory by perception by perception/action interface and extraneous generation, by moving the sign returning to external environment of working memory inside and producing action.SOAR inside has working memory district and the long-term memory district of different forms of characterization, is used for respectively describing current state and the long-term memory of problem solving.Working memory represents result, moving target and the movable operator etc. of the perception data relevant to current state, middle reasoning with the state/object diagram with hierarchical organization.Long-term memory comprises procedural memory, Semantic memory and episodic memory.SOAR, by a fixing treatment mechanism---decision-making period, completes the functions such as choice and application operator of SOAR.Being accompanied by decision-making period SOAR has four kinds of different study mechanisms, is respectively intensified learning, chunk, episodic study and Semantic study.
6.3 are applied to road tunnel SOAR intelligent body linear and horizontal clear distance emulation designs
This section adopts SOAR cognitive frame to carry out modeling to the driver's intelligent body in underground road travel, comprises intelligent body working memory, long-term memory, procedural operator system of selection and study mechanism etc.
6.3.1 the driver's intelligent body working memory by road tunnel designs
Working memory comprises the multidate information all about the world and internal reasoning of SOAR intelligent body, and it comprises perception information, intermediate computations, and the state of classification and relevant operator and target etc., the reasoning that intelligent body is all and decision-making are all carried out in working memory.Information in SOAR in all working memory is organized into graph structure, and Fig. 2 is some graph structure through road tunnel driver intelligent body working memory.
As shown in Figure 2, the state S1 in working memory comprises five attributes, and wherein the value of vehicle, driver, IO attribute is object V1, D1 and I1, and all the other two attributes represent that respectively the name of this state is called s1, and it does not have father's state.The relevant information of the V1 attribute representation intelligent body component units vehicle of state S1, comprises that expression vehicle is of a size of little, and the purposes of vehicle is private car, and present speed is very fast.The D1 attribute representation driver's of S1 relevant information, comprises that driver's sex is man, and the age is 45, and the driving age is 20 years, and monthly income is 8000, and personality is steady type, very familiar to road network, and the matching precision of current driver's is 1, and current location is initial position.IO attribute is the input and output of S1, respectively by input-link and output-link Interface realization, wherein the identifier I1 of input-link has four attributes, represent respectively the information relevant to section and surrounding enviroment that intelligent body perceives from the external world, comprise that attribute Road-alignment, Clear-distance, Sense-density, Velocity, Turn-angle represent respectively that perception when vehicle in front place road is linear, lateral separation, perception are blocked up degree, speed, angle of turn.The identifier O1 of output interface output-link represents the output under this state, such as operator and application operator that selection will be applied are exported by the attribute under I3 the result that affects of working memory and surrounding enviroment.
The input and output of table 6-1 working memory
6.3.2 driver's intelligent body long-term memory design in road tunnel driving
Long-term memory district is an achievement memory district.Most important and procedural knowledge the most effectively in long-term memory, episodic knowledge only just works when procedural knowledge is not enough to support decision-making, does not relate to Semantic knowledge in driving behavior.SOAR intelligent body characterizes with production rule through the procedural memory of road tunnel driving behavior.First, by " if " part of each rule, remove to mate the element in working memory, if " if " of rule part is mated completely with working memory element, this rule will be triggered, then by sending one to the information of kinematic system or producing the suggestion that changes current state, cause that " then " part triggers, any rule of mating with current goal, state and operator can change current target and state.
SOAR intelligent body is specific experience and the memory of intelligent body through the episodic memory of road tunnel, it is the Knowledge Source of episodic study, once driving behavior decision-making completes, just current behavior, corresponding decision-making state transfer path and feedback preference value are carried out to record, in order to using when run into decision-making predicament next time.
The initial long-term memory rule of part of certain driver's intelligent body of table 6-2
In upper table, r1, r2, r3, r4, r5 represent rule numbers, and IF/THEN represents trigger condition and triggers result, [I] expression condition, [T] represents action, R-A1, R-A2, C-D, S-D, P-D, Tem, Des, D-S1, D-S2 is the abbreviation of object properties, represents respectively linear, linear eigenwert, lateral separation, blocks up, predicted congestion, mood, destination, transport condition, transport condition eigenwert.R-A1 (a) represents linear for a class, a=1 wherein, and 2,3 represent respectively straight-line segment, left-turning pathways and right-hand rotation bend; R-A2 (b) represents that the turning radius of linear eigenwert is b level, only have when for bend just has feature when linear, b=1,2,3,4 represent respectively turning radius be little, in, large, especially big; C-D (c) represents that horizontal clear distance value is c level, c=1 wherein, 2,3 represent horizontal clear distance be little, in, large; S-D (d1, d2, d3), P-D (d1, d2, d3) represent that the perception degree of blocking up of current road junction ahead clockwise direction corresponding road section and predicted congestion perception degree are d1, d2, d3 level, wherein parameter occurrence 1,2,3,4 represent unobstructed, generally block up and seriously block up; D-S1 (e) represents that transport condition classification is e, e=1 wherein, and 2,3 represent to keep straight on, turn left, turn right; D-S2 (f1, f2) represents that transport condition eigenwert rank is f1 and f2, f1=1 wherein, and 2 represent travel speed and angles, f2=1,2,3,4 expression speed and angle rank (occurrence is demarcated by experiment) from small to large; Tem (g) represents that current driver's mood classification is g, g=1 wherein, 2 represent respectively happy and irritable; Des (h) represents that current destination direction is h, h=1 wherein, 2,3 represent respectively left, front, right.Above-mentioned rule is a part for certain driver's intelligent body initial rules, and length is limit, and does not list in detail.In addition, driver's intelligent body rule can increase and delete by study mechanism is dynamic in learning process.
The episodic memory of the induction obedience behavior of SOAR intelligent body is specific experience and the memory of intelligent body, it is the Knowledge Source of episodic study, once the decision-making of induction obedience behavior completes, just current behavior, corresponding decision-making state transfer path and feedback preference value are carried out to record, in order to using when run into decision-making predicament next time.
6.3.3 program mode operator is selected
Through in the SOAR intelligent body of road tunnel, adopt numerical value preference.Every rule in long-term memory rule base comprises that matching condition and matching condition meet the lower operator that can advise, and the numerical value preference value of advising this operator under this condition.New regulation of every interpolation need to judge the initial value preference size of operator in this rule, and according to extraneous feedback, the size of this value is upgraded so that it more approaches truth in decision process, for driver's decision-making provides information more accurately.
1) make O (s) represent the candidate's operator set under state s, if its radix | O (s) | equal 1, select o ∈ O (s) to enter working memory, otherwise continue;
2) if | O (s) | > 1, p[o
max(s)]-p[o
sec(s)]>=τ (s), selects from O (s) with roulette mechanism to enter working memory when pre-operator, otherwise continues; P[o wherein
max(s)], p[o
sec(s)] the preference of optimum operator and suboptimum operator under expression state s, τ (s) represents that under state s, operator is directly selected threshold value;
3) if O (s)=Φ or | O (s) | > 1, p[o
max(s)]-p[o
sec(s)] < τ (s), can not directly carry out operator selection, produces predicament, enters chunk learning phase.
6.3.4 study mechanism
The main chunk mode of learning that adopts is described driver through the learning functionality of road tunnel driving behavior.Chunk is the mechanism that SOAR learns when predicament solves, at the processing procedure learning chunking rule of sub-state.When predicament produces, meaning in the long-term memory of current system does not have utilizable operator that problem solving process is moved forward in problem space, need automatically to create a new rule and solve current predicament, the foundation of chunking rule need to be analyzed production rule relevant to reaching result in long-term memory and episodic memory clue.
6.3.4.1 the solution of predicament
When program mode memory can not be selected operator to current state, adopt following steps to solve predicament.
1) order is from original state s
0through i next state, shifting the current state obtaining is s
iif, O (s
i)=Φ or | O (s
i) | > 1, p[o
max(s
i)]-p[o
sec(s
i)] < τ (s), meet chunk condition for study, make j=i+1, continue; Otherwise adopt program mode memory to carry out operator selection, see 3.4;
2) if j=0 turns 5; Otherwise j:=j-1;
3) comprise s find state transition path in all episodic memories storehouse in
jthe operator set of episodic memory, be designated as O
q(s
j).If O
q(s
j)=Φ, turns 2, otherwise continues;
4) at O
q(s
j) in select optimum operator to enter working memory, solve current predicament;
5) if O is (s
i)=Φ, changes the matching precision of current state with step-length 0.1, until meet current state operator set O (s '
i) appearance, wherein s '
irepresent to change the new state after matching precision;
6) at O (s
i) or O (s '
i) in adopt roulette mechanism to select operator to enter working memory, solve current predicament.
6.3.4.2 chunking rule
The intelligent body of process road tunnel is except initial program memory rule, and most rules create by chunk.If intelligent body ran into predicament in state transitions, illustrate in decision process and have partial status there is no operator or can not directly select, after this time decision-making of intelligent body, need to adopt chunk learning method to create rule and preference renewal so.If G is (s
e)-G
e(s
e)≤η
z, to solving the operator of predicament, carry out chunk renewal, wherein G one time
e(s
e) represent that intelligent body is at state s
eunder expectation drive target, G (s
e) represent the actual driving target of intelligent body action, η
zfor regular chunk upgrades threshold value.If same rule is double, by chunk, upgraded, corresponding operator is added in decision process and run into predicament state, chunk is successful.
No matter be that chunk forms or the rule having existed, the operator that all will relate to this decision-making after intelligent body carries out feedback learning, because decision process relates to a plurality of states and operator, therefore adopt each state and the distance of dbjective state to distribute the feedback preference of corresponding operator.Decision-making period t state transition path in k the operator o (s that state is corresponding
k) feedback preference value be λ [d (s
k, s
e)] p
r(s
e), p wherein
r(s
e)=[G (s
e)-G
e(s
e)]
αfor end-state s
eunder total preference value of feedback, parameter alpha gets 0.5 herein; λ [d (s
k, s
e), r (s
k)] be p
r(s
e) be assigned to o (s
k) on weight, it is state s
kto s
edistance d (s
k, s
e) and s
kstate transition path r (the s at place
k) function, herein
Wherein | r (s
k) | be s
kthe number of states that path, place comprises.
6.4 simulation frame
As shown in Figure 3.
6.5 emulation experiment and analysis
First this emulation experiment designs the SOAR intelligent body of and horizontal clear distance linear towards road tunnel, then under and horizontal clear distance condition linear for difference, carry out emulation experiment, experimental result under different simulated conditions is analyzed, provided the optimum suggested design under this experiment scene.
6.5.1 simulating scenes setting
Simulating scenes is Tianjin road tunnel, unidirectional three tracks, linear in simulating scenes, horizontal clear distance etc. carries out after adjustment to a certain degree on former design proposal basis, carry out respectively emulation experiment, on the simulation experiment result analysis foundation, obtain horizontal clear distance and linear prioritization scheme under different condition.Fig. 4,5 is part scene photo.
Based on this road tunnel design proposal, adopt uc win/road to carry out three-dimension virtual reality structure, be used for carrying out the basis of experiment parameter demarcation, road tunnel is set to three tracks, and design speed is 40 kilometers, and partial simulation scene is as shown in Figure 6.
6.5.2 emulation experiment
Under road tunnel straight-line segment and negotiation of bends condition, under different rows vehicle speed, standard lateral clear distance is carried out to certain amplitude adjustment, the horizontal clear distance under this chapter general condition and sighting distance radius are as shown in following table 6-3 and 6-4.
Table 6-3 road tunnel general condition bottom rail clear distance
Sighting distance radius under table 6-4 road tunnel general condition
6.5.2.1 the horizontal clear distance emulation of road tunnel straight-line segment
Under road tunnel straight-line segment car driving conditions, horizontal clear distance step-length with 5% on general basis is reduced or increases by 15%, adopt this chapter emulation mode by straight-line segment, to carry out emulation to vehicle, following table is traffic conflict data corresponding to different emulation experiments under the different speed of a motor vehicle on 200m section in 300 minutes.
The different speed of a motor vehicle of table 6-5 and road tunnel straight-line segment under the different horizontal clear distance traffic conflict data (car) of travelling
Under road tunnel straight-line segment truck driving conditions, horizontal clear distance step-length with 5% on general basis is reduced or increases by 20%, adopt this chapter emulation mode by straight-line segment, to carry out emulation to vehicle, following table is traffic conflict data corresponding to different emulation experiments under the different speed of a motor vehicle in 300 minutes.
Road tunnel straight-line segment under the different speed of a motor vehicle of the table 6-6 traffic conflict data (truck) of travelling
Traffic conflict curve map under friction speed and under horizontal clear distance is as shown below.
6.5.2.2 the horizontal clear distance emulation of road tunnel bend
Under road tunnel bend section car driving conditions, horizontal clear distance step-length with 5% on general basis is reduced or increases by 20%, adopt this chapter emulation mode by straight-line segment, to carry out emulation to vehicle, following table is traffic conflict data corresponding to different emulation experiments under the different speed of a motor vehicle in 30 minutes.
Road tunnel negotiation of bends traffic conflict data (car) under the different speed of a motor vehicle of table 6-7
Traffic conflict curve map under friction speed and under horizontal clear distance is as shown below.
Under road tunnel bend truck driving conditions, horizontal clear distance step-length with 5% on general basis is reduced or increases by 20%, adopt this chapter emulation mode by straight-line segment, to carry out emulation to vehicle, following table is traffic conflict data corresponding to different emulation experiments under the different speed of a motor vehicle in 30 minutes.
Road tunnel negotiation of bends traffic conflict data (truck) under the different speed of a motor vehicle of table 6-8
Traffic conflict curve map under friction speed and under horizontal clear distance as shown in figure 11.
6.5.3 interpretation of result
1) from showing 6-5 to 6-8 and Fig. 6-8 to 6-11, when horizontal clear distance increases, the traffic conflict number under friction speed is all on a declining curve, and this illustrates that horizontal clear distance increase is conducive to the lifting of road tunnel traffic safety.
2) see on the whole, increase along with horizontal clear distance, the speed that traffic conflict reduces slows down, the horizontal clear distance of this explanation road tunnel has certain sensitivity interval to the impact of traffic safety, when horizontal clear distance increases to 110% under general standard when above, continue to increase the impact of traffic conflict is significantly reduced.
3), for take the traffic flow that car is chief component, in straight-line segment travels, when the speed of a motor vehicle is greater than 60km/h, laterally clear distance increases to 5% when above of general standard, continues increase the impact of traffic conflict is significantly reduced; When the speed of a motor vehicle is less than or equal to 40km/h, even if laterally clear distance is reduced to 90% of general standard, also little on traffic conflict impact.Therefore under this paper experiment condition, when car straight-line segment travels, when design speed is more than or equal to 60 kilometers, the horizontal clear distance standard of recommendation is general standard 105%, when design speed is less than or equal to 40 kilometers, the horizontal clear distance standard of recommendation is general standard 90%.
4), for take the traffic flow that car is chief component, in negotiation of bends, when the speed of a motor vehicle is greater than 60km/h, laterally clear distance increases to 15% of general standard, still remarkable on the impact of traffic conflict; When the speed of a motor vehicle is less than or equal to 40km/h, even if laterally clear distance is reduced to 105% of general standard, also little on traffic conflict impact.Therefore under this paper experiment condition, during car negotiation of bends, when design speed is more than or equal to 60 kilometers, the horizontal clear distance standard of recommendation is general standard 115%, when design speed is less than or equal to 40 kilometers, the horizontal clear distance standard of recommendation is general standard 105%.
5), for take the traffic flow that truck is chief component, in straight-line segment travels, when the speed of a motor vehicle is greater than 60km/h, laterally clear distance increases to 5% when above of general standard, continues increase the impact of traffic conflict is significantly reduced; When the speed of a motor vehicle is less than or equal to 40km/h, on general standard basis, increase traffic conflict impact not remarkable.Therefore under this paper experiment condition, when truck straight-line segment travels, when design speed is more than or equal to 60 kilometers, the horizontal clear distance standard of recommendation is general standard 105%, when design speed is less than or equal to 40 kilometers, the horizontal clear distance standard of recommendation is general standard recommendation.
6) for take the traffic flow that truck is chief component, in negotiation of bends, horizontal clear distance is all remarkable to all velocity shootings, therefore under this paper experiment condition, during truck negotiation of bends, the horizontal clear distance of recommendation increases under feasible condition as far as possible, to improve travel safety.
Claims (3)
1. the linear and horizontal clear distance method for designing of the road tunnel based on multiple agent emulation, it is characterized in that, adopt SOAR cognitive frame to carry out modeling to the driver's intelligent body in underground road travel, comprise SOAR cognitive frame working memory, long-term memory, procedural operator system of selection and study mechanism submodule: then under and horizontal clear distance condition linear for difference, carry out emulation experiment, experimental result under different simulated conditions is analyzed, provided the optimum suggested design under experiment scene;
SOAR cognitive frame working memory comprises the multidate information all about the world and internal reasoning of SOAR intelligent body, it comprises perception information, intermediate computations, the state of classification and relevant operator and target, the reasoning that intelligent body is all and decision-making are all carried out in working memory;
SOAR cognitive frame working memory adopts the procedural memory of road tunnel driving behavior to characterize with production rule, first, by " if " part of each rule, remove to mate the element in working memory, if " if " of rule part is mated completely with working memory element, this rule will be triggered, then by sending one to the information of kinematic system or producing the suggestion that changes current state, cause that " then " part triggers, any rule of mating with current goal, state and operator can change current target and state;
Program mode operator is selected: in the SOAR cognitive frame of road tunnel, adopt numerical value preference, every rule in long-term memory rule base comprises that matching condition and matching condition meet the lower operator that can advise, and the numerical value preference value of advising this operator under this condition, new regulation of every interpolation need to judge the initial value preference size of operator in this rule, and according to extraneous feedback, the size of initial value preference is upgraded so that it more approaches truth in decision process;
Study mechanism: at the processing procedure learning chunking rule of sub-state, when predicament produces, meaning in the long-term memory of current system does not have utilizable operator that problem solving process is moved forward in problem space, need automatically to create a new rule and solve current predicament, the foundation of chunking rule need to be analyzed production rule relevant to reaching result in long-term memory and episodic memory clue;
The initial long-term memory of SOAR cognitive frame working memory is regular as following table:
In upper table, r1, r2, r3, r4, r5 represent rule numbers, IF/THEN represents trigger condition and triggers result, [I] expression condition, and [T] represents action, R-A1, R-A2, C-D, S-D, P-D, Tem, Des, D-S1, D-S2 is the abbreviation of object properties, represent respectively linear, linear eigenwert, lateral separation, block up, predicted congestion, mood, destination, transport condition, transport condition eigenwert, R-A1 (a) represents that linear is a class, wherein a=1,2,3 represent respectively straight-line segment, left-turning pathways and right-hand rotation bend; R-A2 (b) represents that the turning radius of linear eigenwert is b level, only have when for bend just has feature when linear, b=1,2,3,4 represent respectively turning radius be little, in, large, especially big; C-D (c) represents that horizontal clear distance value is c level, c=1 wherein, 2,3 represent horizontal clear distance be little, in, large; S-D (d1, d2, d3), P-D (d1, d2, d3) represent that the perception degree of blocking up of current road junction ahead clockwise direction corresponding road section and predicted congestion perception degree are d1, d2, d3 level, wherein parameter occurrence 1,2,3,4 represent unobstructed, generally block up and seriously block up; D-S1 (e) represents that transport condition classification is e, e=1 wherein, and 2,3 represent to keep straight on, turn left, turn right; D-S2(f1, f2) represent that transport condition eigenwert rank is f1 and f2, f1=1 wherein, 2 represent travel speed and angles, f2=1,2,3,4 expression speed and angle rank from small to large, occurrence is demarcated by experiment; Tem (g) represents that current driver's mood classification is g, g=1 wherein, 2 represent respectively happy and irritable; Des(h) represent that current destination direction is h, h=1 wherein, 2,3 represent respectively left, front, right.
2. the linear and horizontal clear distance method for designing of the road tunnel based on multiple agent emulation as claimed in claim 1, is characterized in that, program mode operator is selected to be specially:
1) make O (s) represent the candidate's operator set under state s, if its radix | O (s) | equal 1, select o ∈ O (s) to enter working memory, otherwise continue;
2) if | O (s) | >1, p[o
max(s)]-p[o
sec(s)]>=τ (s), selects from O (s) with roulette mechanism to enter working memory when pre-operator, otherwise continues; P[o wherein
max(s)], p[o
sec(s)] the preference of optimum operator and suboptimum operator under expression state s, τ (s) represents that under state s, operator is directly selected threshold value;
3) if O (s)=Φ or | O (s) | >1, p[o
max(s)]-p[o
sec(s)] < τ (s), can not directly carry out operator selection, produce predicament, enter chunk learning phase, s comprises five attributes, Vehicle, Driver, Super-State, Name, IO, the relevant information of Vehicle attribute representation intelligent body component units vehicle wherein, Driver attribute representation driver's relevant information, the input and output that IO attribute is S1, Name represents the title of this state, and Super-State represents whether this state has father's state.
3. the linear and horizontal clear distance method for designing of the road tunnel based on multiple agent emulation as claimed in claim 1, is characterized in that, study mechanism comprises solution and the chunking rule submodule of predicament:
When program mode memory can not be selected operator to current state, adopt following steps to solve predicament:
1) order is from original state s
0through i next state, shifting the current state obtaining is s
iif, O (s
i)=Φ or | O (s
i) | >1, p[o
max(s
i)]-p[o
sec(s
i)] < τ (s), meet chunk condition for study, make j=i+1, continue; Otherwise adopt program mode memory to carry out operator selection;
2) if j=0 turns 5); Otherwise j=j-1;
3) comprise s find state transition path in all episodic memories storehouse in
jthe operator set of episodic memory, be designated as O
q(s
j), if O
q(s
j)=Φ, turns 2), otherwise continue;
4) at O
q(s
j) in select optimum operator to enter working memory, solve current predicament;
5) if O is (s
i)=Φ, with the matching precision of step-length 0.1 change current state, until meet the operator set O (s' of current state
i) appearance, wherein s'
irepresent to change the new state after matching precision;
6) at O (s
i) or O (s'
i) in adopt roulette mechanism to select operator to enter working memory, solve current predicament;
Chunking rule:
If intelligent body ran into predicament in state transitions, illustrate in decision process and have partial status there is no operator or can not directly select, after this time decision-making of intelligent body, need to adopt chunk learning method to create rule and preference renewal, if G is (s so
e)-G
e(s
e)≤η
z, to solving the operator of predicament, carry out chunk renewal, wherein G one time
e(s
e) represent that intelligent body is at state s
eunder expectation drive target, G (s
e) represent the actual driving target of intelligent body action, η
zfor regular chunk upgrades threshold value; If same rule is double, by chunk, upgraded, corresponding operator is added in decision process and run into predicament state, chunk is successful;
No matter be that chunk forms or the rule having existed, the related operator of process that intelligent body all will be selected operator carries out feedback learning, because decision process relates to a plurality of states and operator, therefore adopt each state and the distance of dbjective state to distribute the feedback preference of corresponding operator, decision-making period t state transition path in k the operator o (s that state is corresponding
k) feedback preference value be λ [d (s
k, s
e)] p
r(s
e), p wherein
r(s
e)=[G (s
e)-G
e(s
e)]
αfor end-state s
eunder total preference value of feedback, parameter alpha gets 0.5; λ [d (s
k, s
e), r (s
k)] be p
r(s
e) be assigned to o (s
k) on weight, it is state s
kto s
edistance d (s
k, s
e) and s
kstate transition path r (the s at place
k) function, get
Wherein | r (s
k) | be s
kthe number of states that path, place comprises.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210104967.5A CN102609599B (en) | 2012-04-11 | 2012-04-11 | Method for designing emulational underground road alignment and transverse clear distance based on multiple intelligent agents |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201210104967.5A CN102609599B (en) | 2012-04-11 | 2012-04-11 | Method for designing emulational underground road alignment and transverse clear distance based on multiple intelligent agents |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN102609599A CN102609599A (en) | 2012-07-25 |
| CN102609599B true CN102609599B (en) | 2014-03-12 |
Family
ID=46526967
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201210104967.5A Active CN102609599B (en) | 2012-04-11 | 2012-04-11 | Method for designing emulational underground road alignment and transverse clear distance based on multiple intelligent agents |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102609599B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103198182B (en) * | 2013-03-20 | 2015-10-28 | 天津市市政工程设计研究院 | The pedestrian guide sign design method of view-based access control model perception simulation technology |
| CN103793570B (en) * | 2014-02-08 | 2016-09-21 | 东南大学 | A kind of road alignment safety analytical method based on dynamics of multibody systems |
| CN104239636B (en) * | 2014-09-16 | 2017-05-03 | 北京航空航天大学 | Fire emergency evacuation simulation method |
| CN111612162B (en) * | 2020-06-02 | 2021-08-27 | 中国人民解放军军事科学院国防科技创新研究院 | Reinforced learning method and device, electronic equipment and storage medium |
| CN113239566B (en) * | 2021-05-27 | 2021-11-09 | 大连理工大学 | Emotion-fused post-disaster three-level hierarchical simulation system and method based on intelligent agent |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101604204B (en) * | 2009-07-09 | 2011-01-05 | 北京科技大学 | Distributed cognitive technology for intelligent emotional robot |
| CN101814235B (en) * | 2010-01-22 | 2011-12-21 | 天津市市政工程设计研究院 | Method for assigning traffic in port area in reservation congregation mode |
-
2012
- 2012-04-11 CN CN201210104967.5A patent/CN102609599B/en active Active
Also Published As
| Publication number | Publication date |
|---|---|
| CN102609599A (en) | 2012-07-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110597245B (en) | Automatic driving track-changing planning method based on quadratic planning and neural network | |
| CN102609599B (en) | Method for designing emulational underground road alignment and transverse clear distance based on multiple intelligent agents | |
| WO2021077725A1 (en) | System and method for predicting motion state of surrounding vehicle based on driving intention | |
| CN109684702B (en) | Driving risk identification method based on trajectory prediction | |
| Nie et al. | Real-time dynamic predictive cruise control for enhancing eco-driving of electric vehicles, considering traffic constraints and signal phase and timing (SPaT) information, using artificial-neural-network-based energy consumption model | |
| CN106371439B (en) | Unified automatic driving transverse planning method and system | |
| Shi et al. | Application and evaluation of the reinforcement learning approach to eco-driving at intersections under infrastructure-to-vehicle communications | |
| CN102505591A (en) | Method for determining lengths of upstream transition regions of construction operation area | |
| CN102505644B (en) | Method for confirming set positions of speed limit signs and size of speed limit during road construction | |
| Kononov et al. | Relationship between freeway flow parameters and safety and its implications for hard shoulder running | |
| Xu et al. | New design method for horizontal alignment of complex mountain highways based on “trajectory–speed” collaborative decision | |
| DE102020205310A1 (en) | Computer-implemented method for providing a data structure for the route complexity detection and validation of functionalities of an automated driving system, such data structure, computer for validating functionalities of an automated driving system, computer program for providing such a data structure and computer-readable data carriers | |
| CN113120003B (en) | Unmanned vehicle motion behavior decision method | |
| CN114715193A (en) | Real-time trajectory planning method and system | |
| Zhang et al. | Dynamic control cycle speed limit strategy for improving traffic operation at freeway bottlenecks | |
| Kou et al. | Dual-objective intelligent vehicle lane changing trajectory planning based on polynomial optimization | |
| Dayi et al. | Research on coordinated control of vehicle’s speed in new mixed traffic flow | |
| CN110308718A (en) | A kind of pilotless automobile behaviour decision making method based on two type fuzzy comprehensive evoluations | |
| CN113538940A (en) | Real-time optimal lane selection method suitable for vehicle-road cooperative environment | |
| CN113688561A (en) | Neural network-based method for determining optimal early warning distance of expressway construction area | |
| Zheng et al. | Flow-balanced-based collaborative strategies and simulation of vehicle group behaviors for on-ramp areas | |
| Zhou et al. | Construction of real-time dynamic reversible lane safety control model in intelligent vehicle infrastructure cooperative system | |
| Liu et al. | A model prediction control based framework for optimization of signaled intersection: A cyber-physical perspective | |
| Shang et al. | Off-Ramp Vehicle Mandatory Lane-Changing Duration in Small Spacing Section of Tunnel-Interchange Section Based on Survival Analysis | |
| Wang et al. | QPNet: Lane‐changing trajectory planning combining quadratic programming and neural network under the convex optimization framework |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CP01 | Change in the name or title of a patent holder |
Address after: 300051 No. 239, Yingkou Road, Heping District, Tianjin Patentee after: Tianjin municipal engineering design and Research Institute Co.,Ltd. Address before: 300051 No. 239, Yingkou Road, Heping District, Tianjin Patentee before: TIANJIN MUNICIPAL ENGINEERING DESIGN & Research Institute |
|
| CP01 | Change in the name or title of a patent holder |