CN104376716B - Method for dynamically generating bus timetables on basis of Bayesian network models - Google Patents

Abstract

The invention relates to a method for dynamically generating bus timetables on the basis of Bayesian network models. The method includes screening microscopic and macroscopic factors affecting dynamic generation of the bus timetables; building the double-layer microscopic and macroscopic Bayesian network models for dynamically generating the bus timetables, and in other words, building the Bayesian network models for forecasting dynamic variation of bus environments and the Bayesian network models for dynamically generating the bus timetables; predicting transport capacity and transport volume occurrence probabilities of various routes under the condition of random disturbance and reasons for unbalance of the transport capacity and the transport volumes of the various routes; combining scheduling policies with one another and generating possible timetable schemes around the target for timely evacuating passengers; computing various indexes for evaluating the quality of the timetables from the points of governments, enterprises and the passengers and evaluating the quality of the timetables. The method has the advantages that a function of dynamically adjusting the timetables according to variation of the bus environments can be implemented, and accordingly technical support can be provided for daily bus operation management.

Description

A kind of transit scheduling dynamic creation method based on Bayesian network model

Technical field

The present invention relates to Bus transit informization technical field, during a kind of public transport based on Bayesian network model Carve table dynamic creation method.

Background technology

Transit scheduling establishment is one of core missions of the daily operation of public transport, according to resident trip spatial-temporal distribution characteristic, Rationalization arranges departure frequency and the type thereof of day part, mainly solves transport power and the maximum matching problem of freight volume.Work as reality In random factor interference when causing bus passenger flow or running time change, this causes public transport transport power and freight volume unbalance, thus public Friendship scheduling scheme lost efficacy.Therefore, dynamically change dynamic time adjustment table according to public transport environment, have theory value and reality meaning Justice.

Directly determine that the factor that transit scheduling lost efficacy is transport power and the freight volume of up-downlink direction, they by Changes in weather, The external environment influence such as traffic congestion, large-scale activity, when detecting traffic events, assessing it affects passenger flow or running time respectively Intensity of variation, and then analyze the failure cause of timetable, dynamic time adjustment table accordingly.At present, numerous Chinese scholars according to Public transport environment dynamically changes dynamic time adjustment table, and main Research Thinking has two:

One, passenger flow forecast or running time change, on the one hand, utilize the method such as multiple linear regression, structural equation, qualitative Relatedness between the many factors that quantitatively further investigated affect passenger flow or running time changes, and carry out associated sensitivity analysis； On the other hand, utilizing time series method, be regarded as a black box, the evolution directly disclosing passenger flow or running time change becomes Gesture, provides data supporting for establishment timetable.

Two, establishment timetable, on the one hand, study passenger flow and the statistical law of running time on above-mentioned working foundation, when During detection traffic events, build timetable compiling model, utilize Optimum Theory to generate timetable；On the other hand, nerve is utilized The artificial analogue technique such as network, the thinking model of operation simulation personnel, according to environmental change, adjust time adjustment table.

From the foregoing, it will be observed that existing research way cannot solve the transit scheduling that random disturbances causes dynamically changes chain reaction Process, it should from entirety, disclosing affects how external environment condition change causes passenger flow or running time change, and then when affecting Quarter, how table dynamic adjustment process occurred, and mutually caused between them, interferes, converts and the complex relationship such as coupling, it was predicted that Transit scheduling under complicated traffic environment change situation generates and probability of happening.

Bayesian network is that one portrays causal probability graph model between things, is especially suitable for accident The chain reaction process occurred and cause is modeled analyzing.Based on this, analyzing influence transit scheduling of the present invention is dynamically given birth to Become transport power and freight volume between unbalance reason, external environmental factor is considered as input, analyze its how to affect passenger flow or travel time Between change, and then how to cause between public transport transport power and freight volume unbalance, according to existing conveyance equilibrium, output is maximum of timetable Joining the result of freight volume, control input and can control the change of partial status, build in accident Bayesian network accordingly is each External environment condition node input passenger flow or running time transport power freight volume calculate four etale topology nets of timetable decision node output Network structure, it is achieved the transit scheduling under prediction complicated traffic environment change situation and probability of happening thereof, for bus dynamic dispatching Reliable technical support is provided.

Summary of the invention

The invention provides a kind of transit scheduling dynamic creation method based on Bayesian network model, at analyzing influence On the basis of the various influence factors that transit scheduling dynamically changes, in conjunction with actual public transport dynamic data, portray between them Cause effect relation, the fortune when changing according to intelligent public transportation dispatching detection of platform external environment condition, under the various complicated traffic environment of reasoning Power and the unbalance reason of freight volume and probability of happening thereof, calculate departure frequency and scheduling type thereof accordingly, present invention is mainly used for basis The change of public transport external environment condition dynamically generates timetable, provides technical support for the daily operation management of public transport.

The present invention program is achieved through the following technical solutions:

The present invention provides a kind of transit scheduling dynamic creation method based on Bayesian network model, including following step Rapid:

(1) the method screening qualitatively and quantitatively combined is used to affect numerous external environment conditions that transit scheduling dynamically generates Factor, causes running time to change including vehicle accident, and then causes transport power deficiency or large-scale activity to cause passenger flow to fluctuate, and then Cause freight volume not enough；

(2) the two-layer Bayesian network model up and down that transit scheduling dynamically generates is built, wherein: upper layer model is portrayed and drawn Playing passenger flow or the external environment condition random disturbances of running time change, be the initial conditions of underlying model, underlying model describes and causes Transport power and the unbalance external environment condition random disturbances of freight volume, generate for timetable and provide data supporting；

(3) when traffic events occurs, the realtime running data of combined with intelligent bus dispatching platform, utilize step (2) micro- The various circuit transport power freight volume probability of happening of macromodel prediction and unbalance reason, in conjunction with scheduling strategy, around with minimum It is target that cost evacuates passenger in time, generates multiple timetable scheme；

(4) the passenger's Waiting time of every kind of timetable scheme in calculation procedure (3), website are detained situation, operation cost Indices, assesses it good and bad.

Improve as one, use the method qualitatively and quantitatively combined, screening to affect what transit scheduling dynamically generated Many factors process, including:

(1) from microcosmic, N number of external environment condition random factor X=(X is analyzed₁, X₂..., X_N) how to disturb visitor at time t Stream fluctuation pf_t(X) or running time change pt_t(X), such as: road type, road conditions, vehicle accident, large-scale activity, traffic control and Changes in weather etc., so needed for affecting timetable with car demand and available vehicle number；

(2) from macroscopically, the up direct determiner of circuit that current T period timetable generates is disclosedAnd line The descending direct acting factor in roadThe up car demand of current T periodDescending car demand Up available vehicle numberWith descending available vehicle numberAnd within the lower T+1 period up car demandDescending car demandUp available vehicle numberWith descending available vehicle number

As improving further, affect, according to filter out, microcosmic and the Macroscopic Factors that transit scheduling generates, build public affairs Hand over two layers of Bayesian network model that timetable dynamically generates, including:

(1) node abstraction definition,

In public transport environment dynamically changes forecast Bayesian network model, its condition node is that N number of external environment condition is random Factor X=(X₁, X₂..., X_N), including road type, road conditions, vehicle accident, large-scale activity, traffic control and Changes in weather Deng；Its decision node is passenger flow fluctuation or the running time change Y={pf of time t_t(X), pt_t(X)}。

Dynamically generating in Bayesian network model at transit scheduling, its condition node is on circuit Or down direction uses car demand, and their available vehicle number, i.e. within current T period and next T+1 period $Z=\{{\mathrm{pr}}_T^U\left(X\right),{\mathrm{pr}}_T^D\left(X\right),{\mathrm{pv}}_T^U\left(X\right),{\mathrm{pv}}_T^D\left(X\right);{\mathrm{pr}}_{T+1}^U\left(X\right),{\mathrm{pr}}_{T+1}^D\left(X\right),{\mathrm{Pv}}_{T+1}^U\left(X\right),{\mathrm{pv}}_{T+1}^D\left(X\right)\};$ Its decision-making knot Point is the departure frequency of current T period circuit up or down row

(2) Structure learning,

Utilize the conditional independence method of inspection, respectively public transport environment is dynamically changed forecast and transit scheduling dynamically generates All nodes of Bayesian network model, if any two node and between interdepend, there is directed edge and be connected, build one Individual directed acyclic graph, sets up their bayesian network structure figure S.

(3) parameter learning,

Utilize maximum Likelihood, respectively public transport environment is dynamically changed forecast and transit scheduling dynamically generates shellfish Ye Si network model, gives network topology structure S and training sample set D in each of which, utilizes priori, determine respective shellfish Conditional probability density at each node of Ye Si network model is:

Probability causal relation between external environment condition change and passenger flow or running time fluctuation is described

$p(Y/X)=\frac{p\left(Y\right)\×p\left(X\right|Y)}{p\left(X\right)};$

Portray state transfer relationship between the departure frequency of external environment condition random disturbances and circuit up-downgoing

$p(S/Z)=\frac{p\left(S\right)\×p\left(Z\right|S)}{p\left(Z\right)}.$

Preferred as one, it was predicted that various circuit transport power freight volume probability of happening under random disturbances and they are unbalance former Cause, including:

(1) for certain circuit, total K website and M car, utilize the actual number of intelligent public transportation dispatching platform According to, in conjunction with the latitude and longitude coordinates of every i car, estimate that the passenger flow arriving k website at time t arrives numberAnd this vehicle arrives The running time at first and last station

(2) when traffic events being detected, N number of external environment condition random factor X=(X is determined₁, X₂..., X_N) value, Utilize clique tree propagation algorithm, according to public transport environment dynamic change model X → Y={pf_t(X), pt_t(X) }, it was predicted that their fluctuation Time pt_t(X) and change passenger flow pf_tAnd their probability of happening (X)；

(3) on the basis of the above, collect circuit up-downgoing and use car demand in certain T period ${\mathrm{pr}}_T^U\left(X\right)=\underset{k=1}{\overset{K}{\mathrm{\Σ}}}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\left[p_{t-{\mathrm{at}}_t^i-{\mathrm{pt}}_t\left(X\right)}^k{+\mathrm{pf}}_{t-{\mathrm{at}}_t^i-{\mathrm{pt}}_t\left(X\right)}\left(X\right)\right]{,\mathrm{pr}}_T^D\left(X\right)=\underset{k=1}{\overset{K}{\mathrm{\Σ}}}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}[p_{t-{\mathrm{at}}_t^i-{\mathrm{pt}}_t\left(X\right)}^k+{\mathrm{pf}}_{t-{\mathrm{at}}_t^i-{\mathrm{pt}}_t\left(X\right)}\left(X\right)],$ Available Vehicle number ${\mathrm{pv}}_T^U\left(X\right)=\underset{i=1}{\overset{M}{\mathrm{\Σ}}}f({\mathrm{at}}_t^i+{\mathrm{pt}}_t\left(X\right)),{\mathrm{pv}}_T^D\left(X\right)=\underset{i=1}{\overset{M}{\mathrm{\Σ}}}f({\mathrm{at}}_t^i+{\mathrm{pt}}_t\left(X\right))$ And their probability, analyze fortune Whether mate between power and freight volume, provide data supporting for transit scheduling establishment.

Preferential as another kind, in conjunction with scheduling strategy, around evacuating passenger's target in time, generate possible timetable side Case, including:

(1) use uniformly equilibrium to dispatch a car strategy, it is considered to single scheduling method, meet the capacity consistency c of bus, crowded Degree γ ∈ [γ_min, γ_max], on the basis of government departure interval F, determine departure frequency scopeWith ${\mathrm{Trip}}_T^D\left(X\right)=\max (\frac{{\mathrm{pr}}_T^D\left(X\right)}{\mathrm{\γ}\·c},F)$ And their probability；

(2) according to unbalance between transport power freight volume, utilize clique tree propagation algorithm, utilize timetable to generate Bayesian network model $Z=\{{\mathrm{pr}}_T^U\left(X\right),{\mathrm{pr}}_T^D\left(X\right),{\mathrm{pv}}_T^U\left(X\right),{\mathrm{pv}}_T^D\left(X\right);{\mathrm{pr}}_{T+1}^U\left(X\right),{\mathrm{pr}}_{T+1}^D\left(X\right),{\mathrm{pv}}_{T+1}^U\left(X\right),{\mathrm{pv}}_{T+1}^D\left(X\right)\}\→$ $S=\{{\mathrm{Trip}}_T^D\left(X\right),{\mathrm{Trip}}_T^U\left(X\right)\},$ Generate possible timetable scheme.

As the most preferential, from government, enterprise and passenger's angle, calculate the various fingers of assessment timetable quality Mark, assesses their quality, including:

(1) on the basis of timetable scheme, it is considered to unit mileage running cost l, the total passenger calculating every kind of scheme waits for bus Time $\underset{k=1}{\overset{K}{\mathrm{\Σ}}}[\frac{T\·\underset{T}{\∫}{p}_t^k\mathrm{dt}}{{\mathrm{Trip}}_T^U\left(X\right)}+\frac{T\·\underset{T}{\∫}{p}_t^k\mathrm{dt}}{{\mathrm{Trip}}_T^D\left(X\right)}],$ Website is detained situation $\underset{k=1}{\overset{K}{\mathrm{\Σ}}}[{\mathrm{Trip}}_T^U\left(X\right)\·\mathrm{\γ}\·c-\underset{T}{\∫}{p}_t^k\mathrm{dt}],$ Operation cost $\left[\right[{\mathrm{Trip}}_T^U\left(X\right)+{\mathrm{Trip}}_T^D\left(X\right)]\·l$ Deng indices；

(2) from government, enterprise and passenger's angle, the quality of the above-mentioned each timetable scheme of overall merit, for public transport Administration section selects the best time table scheme to provide decision support.

Due to the fact that have employed above-mentioned several measure improves, utilize Bayesian network to portray external environment condition and change such as What causes passenger flow or running time change, and then affect the process of the dynamically adjustment of transit scheduling, it is to avoid existing method without Method solves transport power and the unbalance chain reaction process of freight volume that accident causes, it is possible to excavate outside public transport from raw sample data Portion's environmental change, passenger flow or running time fluctuation, timetable dynamically adjust between coupled relation, from advance, thing neutralize afterwards Reason that how the multi-faceted real-time analysis traffic events of overall process causes transit scheduling to change and development trend thereof, move for public transport State scheduling provides data supporting.

Accompanying drawing explanation

Fig. 1 is the structural representation that the transit scheduling that the present invention relates to generates Bayesian network；

Fig. 2 is the flow chart that the present invention implements.

Detailed description of the invention

It is described further below in conjunction with accompanying drawing provided by the present invention:

As it is shown in figure 1, the present invention provides a kind of transit scheduling dynamic creation method based on Bayesian network model, press The process occur according to traffic events, developing and developing, builds two layers of microcosmic and macroscopic view Bayes that transit scheduling dynamically generates Network model, unbalance reason between transport power and freight volume that analyzing influence transit scheduling dynamically generates, external environmental factor is regarded For input, analyze how it affects passenger flow or running time change, and then how to cause between public transport transport power and freight volume unbalance, defeated Going out is the result of timetable maximum match freight volume.Control input and can control the change of partial status, build accident accordingly Each external environment condition node input passenger flow or running time prediction transport power freight volume in Bayesian network calculate timetable decision-making Four etale topology network structures of node output, it is achieved the transit scheduling under prediction complicated traffic environment change situation and generation thereof Probability, provides reliable technical support for bus dynamic dispatching.

As in figure 2 it is shown, the present invention provides a kind of transit scheduling dynamic creation method based on Bayesian network model, bag Including four steps such as Analysis on Mechanism, modelling, modelling verification and model analysis utilization, detailed description of the invention is as follows.

Step 1: Analysis on Mechanism, uses the method screening that qualitatively and quantitatively combines to affect the crowd that transit scheduling dynamically generates Many external environmental factors, set up and affect the factor storehouse that transit scheduling lost efficacy.

Step 1.1: from microcosmic, analyzes N number of external environment condition random factor X=(X₁, X₂..., X_N) at time t, such as what connection Disturb passenger flow fluctuation pf_t(X) or running time change pt_t(X), such as: road type, road conditions, vehicle accident, large-scale activity, traffic pipe System and Changes in weather etc., and then the use car demand needed for affect timetable and available vehicle number；

Step 1.1.1: invitation expert has an informal discussion, selects all n possible shadows of passenger flow fluctuation or running time change respectively The factor of sound, the former relates to season, festivals or holidays, period, large-scale activity, traffic control, vehicle trouble, weather etc., and the latter is contained Road type, traffic flow, traffic congestion, type of site, passenger flow and weather etc..

Step 1.1.2: combined with intelligent bus dispatching platform, dynamically obtains certain j external environment influence factor when any Carve the numerical value a of i_ij, and they corresponding passenger flows or running time y_i, altogether as sample D, bar will be formed by m data record Part matrix A=(a_ij)_mnWith decision vector Y=(y_i)_m。

Step 1.1.3: according to AB=Y, based on method of least square, calculate B=(b₁, b₂..., b_n)=(A ' A)^-1(A ' Y), RightIf b_jMore than artificial threshold values σ, this factor determines passenger flow or running time, obtains n influence factor's variable.

Step 1.2: from macroscopically, disclose the up direct determiner of circuit that current T period timetable generatesWith line downstream direct acting factorThe up car demand of current T periodDescending car DemandUp available vehicle numberWith descending available vehicle numberAnd it is up within the lower T+1 period Use car demandDescending car demandUp available vehicle numberWith descending available vehicle number

Step 1.2.1: according to the service time scope of circuit, is divided into N number of period, calculating up-downlink direction respectively Car demand is used within current T period and next T+1 periodAnd their available vehicle number ${\mathrm{pv}}_T^U\left(X\right),{\mathrm{pv}}_T^D\left(X\right).$

Step 1.2.2: combined with intelligent bus dispatching platform, for certain circuit, total K website and M car, knot Close the latitude and longitude coordinates of every i car, estimate that the passenger flow arriving k website at time t arrives numberAnd this vehicle arrives first and last The running time stood

Step 1.2.3: collect circuit and use car demand in the T period ${\mathrm{pr}}_T^U\left(X\right)=\underset{k=1}{\overset{K}{\mathrm{\Σ}}}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\left[p_{t-{\mathrm{at}}_t^i-{\mathrm{pt}}_t\left(X\right)}^k{+\mathrm{pf}}_{t-{\mathrm{at}}_t^i-{\mathrm{pt}}_t\left(X\right)}\left(X\right)\right],$ ${\mathrm{pr}}_T^D\left(X\right)=\underset{k=1}{\overset{K}{\mathrm{\Σ}}}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}[p_{t-{\mathrm{at}}_t^i-{\mathrm{pt}}_t\left(X\right)}^k+{\mathrm{pf}}_{t-{\mathrm{at}}_t^i-{\mathrm{pt}}_t\left(X\right)}\left(X\right)],$ Available vehicle number ${\mathrm{pv}}_T^U\left(X\right)=\underset{i=1}{\overset{M}{\mathrm{\Σ}}}f\left({\mathrm{at}}_t^i{+\mathrm{pt}}_t\left(X\right)\right),$ Analyze and whether mate between transport power and freight volume, provide data to prop up for transit scheduling establishment Support.

Step 2: modelling, builds two layers of microcosmic up and down and macroscopic view Bayesian network that transit scheduling dynamically generates Model, including node variable definition, determine each condition and the span of decision variable and prior probability distribution, Structure learning and Parameter learning four part, its at the middle and upper levels model portray cause passenger flow or running time change external environment condition random disturbances, under being The initial conditions of layer model, underlying model describes and causes transport power and the unbalance external environment condition random disturbances of freight volume, raw for timetable Become to provide data supporting；.

Step 2.1: variable node defines,

In public transport environment dynamically changes forecast Bayesian network model, total n+2 node variable X={X₁, X₂..., X_n∪ Y={pf_t(X), pt_t(X) n condition and 2 decision variable nodes }, it are divided into.The former is that public transport external environment condition is done at random Disturbing input key element, the latter is passenger flow or running time output result, pays close attention between public transport external environment condition random disturbances input key element Influence each other, and how their change causes passenger flow or running time change.

Dynamically generate in Bayesian network model at transit scheduling, its condition node be on circuit or Down direction uses car demand, and their available vehicle number, i.e. within current T period and next T+1 period $Z=\{{\mathrm{pr}}_T^U\left(X\right),{\mathrm{pr}}_T^D\left(X\right),{\mathrm{pv}}_T^U\left(X\right),{\mathrm{pv}}_T^D\left(X\right);{\mathrm{pr}}_{T+1}^U\left(X\right),{\mathrm{pr}}_{T+1}^D\left(X\right),{\mathrm{pv}}_{T+1}^U\left(X\right),{\mathrm{pv}}_{T+1}^D\left(X\right)\};\→$ Its decision-making knot Point is the departure frequency of current T period circuit up or down row

Step 2.2: determine respectively microcosmic and the above-mentioned condition of macromodel and the span of decision variable node and it Prior probability distribution between.

Step 2.2.1: micro-(grand) is seen any node of modelThe span of (S ∪ Z)By its discretization K feature value state space it is $x_i^j=X_{\min }^i+(j-1)*[X_{\max }^i-X_{\min }^i]/K.$

Step 2.2.2: add up micro-(grand) and see any node of model(S ∪ Z) value stateProbability $p\left(x_i^j\right)=\mathrm{Count}\left(x_i^j\right)/\underset{j=1}{\overset{K}{\mathrm{\Σ}}}\mathrm{Count}\left(x_i^j\right),$ And two variablees(S ∪ Z) and(S∪Z) Prior probability distribution between they state values differentWherein Count () The number of times that expression event occurs at sample set D.

Step 2.3: build topological structure between microcosmic and the macroscopic view each node of Bayesian network respectively.

Step 2.3.1: use K2 algorithm, carry out unsupervised machine learning in training set, respectively obtains micro-(grand) and sees The initial network structure of model.

Step 2.3.2: utilize the priori of expert, based on the conditional independence method of inspection, if micro-(grand) sees model Any two node(S ∪ Z) andInterdepend between (S ∪ Z), there is directed edge and be connected ConnectThe network structure of micro-(grand) model is finely adjusted.

Step 2.3.3: detection obtains micro-(grand) after adjusting and sees whether prototype network structure meets the requirements, wants if meeting Ask, the bayesian network structure figure S of output public transport dynamic environment forecast (timetable generation)；Otherwise return step 2.3.2, continue Micro-(grand) sees the network structure of model.

Step 2.4: on the basis of above-mentioned network structure, utilizes maximum Likelihood, estimates microcosmic and macromodel In conditional probability distribution table between each node.

Step 2.4.1: any node to micro-macromodel(S ∪ Z), by prior distribution and likelihood letter Number combines, and estimates parameter

Step 2.4.2: assume that θ is the random distribution of Dirichlet function, the likelihood function making θ is $=\underset{k=1}{\overset{r_i}{\mathrm{\Π}}}\underset{i=1}{\overset{n}{\mathrm{\Π}}}P\left(X_i\right|{\mathrm{\θ}}_i)=\underset{i=1}{\overset{n}{\mathrm{\Π}}}\underset{j=1}{\overset{K}{\mathrm{\Π}}}P\left(x_i^j\right|{\mathrm{\θ}}_{\mathrm{ij}}),$ Because $L\left({\mathrm{\θ}}_{\mathrm{ij}}\right|X)=P\left(X\right|{\mathrm{\θ}}_{\mathrm{ij}})=\underset{k=1}{\overset{r_i}{\mathrm{\Π}}}{{\mathrm{\θ}}_{\mathrm{ijk}}}^{N_{\mathrm{ijk}}},$ According to $\frac{\∂\left(L\left(\mathrm{\θ}\right|X)\right)}{\∂\mathrm{\θ}}=0,$ DeriveThus estimate any node of micro-macromodel(S∪ Z) p (X between_i|X_n+1).Wherein: r_iFor X_iEigenvalue number；N_ijkFor nodes X_iWhen taking kth eigenvalue, take at father node The quantity of value jth eigenvalue.

Step 2.4.3: according to above-mentioned formulation process, can calculate in microcosmic and macromodel condition between each node general Conditional probability between rate distribution table, i.e. public transport external environment condition random disturbances key element and passenger flow or running time fluctuation, and then impact Conditional probability between transport power and freight volume fluctuation (timetable dynamically adjusts).

Probability causal relation between description external environment condition change with passenger flow or running time fluctuation:

$p(Y/X)=\frac{p\left(Y\right)\×p\left(X\right|Y)}{p\left(X\right)};$

Portray state transfer relationship between the departure frequency of external environment condition random disturbances and circuit up-downgoing:

$p(S/Z)=\frac{p\left(S\right)\×p\left(Z\right|S)}{p\left(Z\right)}.$

Step 3: using part training set as test sample, microcosmic and the precision of macromodel in inspection, if model knot Fruit does not reaches target, returns step 2.

Step 4: model analysis is used, the transit scheduling dynamic generative process of the various complicated traffic environment of Inference Forecast.

Step 4.1: combined with intelligent bus dispatching platform, when public transport external environment condition changes, each influence factor of dynamic monitoring Measured value x=(x₁, x₂..., x_n), rightCalculate its significant conditionIf (x_i?WithBetween), determine the current institute of network model There is node state

Step 4.2: utilize clique tree propagation algorithm, according to micromodel X → Y={pf_t(X), pt_t(X) }, for certain line For K the website on road and M car, in conjunction with the latitude and longitude coordinates of every i car, estimate that the passenger flow arriving k website at time t arrives NumberAnd this vehicle arrives the running time at first and last station

Step 4.3: on the basis of the above, collects circuit up-downgoing and uses car demand in certain T period ${\mathrm{pr}}_T^U\left(X\right)=\underset{k=1}{\overset{K}{\mathrm{\Σ}}}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}\left[p_{t-{\mathrm{at}}_t^i-{\mathrm{pt}}_t\left(X\right)}^k{+\mathrm{pf}}_{t-{\mathrm{at}}_t^i-{\mathrm{pt}}_t\left(X\right)}\left(X\right)\right]{,\mathrm{pr}}_T^D\left(X\right)=\underset{k=1}{\overset{K}{\mathrm{\Σ}}}\underset{i=1}{\overset{M}{\mathrm{\Σ}}}[p_{t-{\mathrm{at}}_t^i-{\mathrm{pt}}_t\left(X\right)}^k+{\mathrm{pf}}_{t-{\mathrm{at}}_t^i-{\mathrm{pt}}_t\left(X\right)}\left(X\right)],$ Available Vehicle number ${\mathrm{pv}}_T^U\left(X\right)=\underset{i=1}{\overset{M}{\mathrm{\Σ}}}f({\mathrm{at}}_t^i+{\mathrm{pt}}_t\left(X\right)),{\mathrm{pv}}_T^D\left(X\right)=\underset{i=1}{\overset{M}{\mathrm{\Σ}}}f({\mathrm{at}}_t^i+{\mathrm{pt}}_t\left(X\right))$ And their probability of happening, point Whether mate between analysis transport power and freight volume, provide data supporting for analyzing the unbalance reason of transit scheduling.

Step 4.4: use and uniformly equalize strategy of dispatching a car, it is considered to single scheduling method, meeting the capacity consistency of bus C, degree of crowding γ ∈ [γ_min, γ_max], on the basis of government departure interval F, primarily determine that departure frequency scopeWithAnd its probability, utilize clique tree propagation to calculate Method, according to $Z=\{{\mathrm{pr}}_T^U\left(X\right),{\mathrm{pr}}_T^D\left(X\right),{\mathrm{pv}}_T^U\left(X\right),{\mathrm{pv}}_T^D\left(X\right);{\mathrm{pr}}_{T+1}^U\left(X\right),{\mathrm{pr}}_{T+1}^D\left(X\right),{\mathrm{pv}}_{T+1}^U\left(X\right),{\mathrm{pv}}_{T+1}^D\left(X\right)\}\→$ $S=\{{\mathrm{Trip}}_T^D\left(X\right),{\mathrm{Trip}}_T^U\left(X\right)\},$ Generate possible timetable scheme.

Step 4.5: on the basis of timetable scheme, it is considered to unit mileage running cost l, calculates total passenger of every kind of scheme Waiting time $\underset{k=1}{\overset{K}{\mathrm{\Σ}}}[\frac{T\·\underset{T}{\∫}{p}_t^k\mathrm{dt}}{{\mathrm{Trip}}_T^U\left(X\right)}+\frac{T\·\underset{T}{\∫}{p}_t^k\mathrm{dt}}{{\mathrm{Trip}}_T^D\left(X\right)}],$ Website is detained situation $\underset{k=1}{\overset{K}{\mathrm{\Σ}}}[{\mathrm{Trip}}_T^U\left(X\right)\·\mathrm{\γ}\·c-\underset{T}{\∫}{p}_t^k\mathrm{dt}],$ Operation costEtc. index, from government, enterprise and passenger's angle, the above-mentioned each timetable of overall merit The quality of scheme, and carry out the reason of changes that the inefficacy of its timetable of backward reasoning is possible, when selecting optimal for public traffic management department Carve table scheme and decision support is provided.

Listed above is only the specific embodiment of the present invention.It is clear that the invention is not restricted to above example, it is also possible to have Many deformation, such as: the present invention changes structure design and the parametric learning method of Bayesian network, can expand and affect passenger flow or traveling The influence factor of time, such as: road type etc., uses different scheduling strategies and appraisal procedure.The ordinary skill people of this area All deformation that member can directly derive from present disclosure or associate, are all considered as protection scope of the present invention.

Claims (6)

1. a transit scheduling dynamic creation method based on Bayesian network model, it is characterised in that comprise the following steps:

(1) use the method screening qualitatively and quantitatively combined to affect numerous external environment conditions that transit scheduling dynamically generates because of Element, causes running time to change including vehicle accident, and then causes transport power deficiency or large-scale activity to cause passenger flow to fluctuate, Jin Eryin Inducement is not enough；

(2) the two-layer Bayesian network model up and down that transit scheduling dynamically generates is built, wherein: upper layer model is portrayed and caused visitor Stream or the external environment condition random disturbances of running time change, be the initial conditions of underlying model, and underlying model describes and causes transport power The external environment condition random disturbances unbalance with freight volume, generates for timetable and provides data supporting；

(3) when traffic events occurs, the realtime running data of combined with intelligent bus dispatching platform, utilize pattra leaves in step (2) The various circuit transport power freight volume probability of happening of this network model prediction and unbalance reason, in conjunction with scheduling strategy, around with It is target that little cost evacuates passenger in time, generates multiple timetable scheme；

(4) the passenger's Waiting time of every kind of timetable scheme in calculation procedure (3), that website is detained situation, operation cost is every Index, assesses it good and bad.

A kind of transit scheduling dynamic creation method based on Bayesian network model the most according to claim 1, it is special Levy and be: the method qualitatively and quantitatively combined in described step (1), including on microcosmic, analyze N number of external environment condition random because of Element X=(X₁, X₂..., X_N) how to disturb passenger flow fluctuation pf at time t_t(X) or running time change pt_t(X)；From macroscopically, Disclose the up direct determiner of circuit that current T period timetable generatesWith line downstream direct acting factorThe up car demand of current T periodDescending car demandUp available vehicle numberWith descending available vehicle numberAnd within the lower T+1 period up car demandDescending car DemandUp available vehicle numberWith descending available vehicle number

Transit scheduling dynamic creation method based on Bayesian network model the most according to claim 2, it is special Levy and be: what step (1) was filtered out by described step (2) affects microcosmic and the Macroscopic Factors that transit scheduling generates, respectively Build public transport environment and dynamically change Bayesian network model X → Y={pf_t(X), pt_t(X) } and transit scheduling generate pattra leaves This network model Probability causal relation between external environment condition change and passenger flow or running time fluctuation is describedAnd then portray external environment condition random disturbances and circuit with car demand and available vehicle number it Between state transfer relationshipWherein P represents that probability, Y/X represent the probability of Y, X/Y table in the case of X Showing the probability of X in the case of Y, Z/S represents the probability of Z in the case of S, and S/Z represents the probability of S under Z case.

Transit scheduling dynamic creation method based on Bayesian network model the most according to claim 2, it is special Levying and be: step (3) utilizes the real data of intelligent public transportation dispatching platform, total K website and M car, when friendship being detected During interpreter's part, according to public transport environment dynamic change model, arrive the passenger flow of k website to intelligent in conjunction with every i car at time t NumberAnd i-th vehicle arrive first and last station running timePredict their wave time pt_t(X) and change Passenger flow pf_t(X), generate model according to transit scheduling accordingly, collect circuit up-downgoing and use car demand in certain T periodAvailable Vehicle numberAnd their probability, during for public transport Carve table establishment and data supporting is provided.

Transit scheduling dynamic creation method based on Bayesian network model the most according to claim 2, its feature exists In: step (4), on the basis of the data of step (3), is target around evacuating passenger in time, uses and uniformly equalizes strategy of dispatching a car, Consider single scheduling method, meet the capacity consistency c of bus, degree of crowding γ ∈ [γ_min, γ_max], government's departure interval On the basis of F, determine up-downgoing departure frequency scopeWithWith And their probability, generate possible timetable scheme.

Transit scheduling dynamic creation method based on Bayesian network model the most according to claim 2, its feature exists In: step (5) is on the basis of the timetable scheme of step (4), it is considered to unit mileage running cost l, calculates the total of every kind of scheme Passenger's Waiting timeWebsite is detained situationOperation costIndices, and analyze each timetable scheme whether be suitable to traffic environment change, for Public traffic management department selects the best time table scheme to provide decision support, whereinRepresent that the passenger flow at time t arrival k website arrives Intelligent's number, c represents that the capacity consistency of bus, γ represent the degree of crowding.