CN102664805B - Predictive routing method for bus delay tolerant network - Google Patents

Abstract

The invention discloses a predictive routing method for a bus delay tolerant network (DTN). The method specifically comprises the following steps of: (1) disclosing an interval algebra-based extract network topological representation method for the semidefiniteness of a bus net node motion mode; (2) quantitatively calculating the possibility of future contact by utilizing the historical contact information of a node and adopting Bayesian estimation to obtain the probability of the future contact and a density function thereof; and (3) calculating an optimal decision sequence for a future communication path by adopting an iteration and recursion algorithm according to obtained future contact information. In a bus net scenario, a delivery rate is higher than those of most of other DTN routing, a high overhead rate and a good average delay are ensured, and the requirements of the bus DTN can be met.

Description

A kind of prediction method for routing for bus Delay Tolerant Network

Technical field

The invention belongs to communication technical field, relate to Delay Tolerant Network routing decision, be used in the lifting to overall performance of network in Delay Tolerant Network.

Background technology

Delay Tolerant Network (DTN, Delay Tolerant Networking) be the communication technology under a kind of challenge network (Challenged Networks) environment, its application scenarios has high latency, low data rate, there is not stable end-to-end connection, in network architecture, lack interactivity, the features such as system resource, life-span finite sum low duty cycle operation.By the communication mechanism of " store-carry-forward ", Delay Tolerant Network can utilize the mobility of node to carry out the transmission of message at challenge net environment.Therefore, the routing performance in Delay Tolerant Network depends critically upon the Move Mode of node.

Public traffic network (Public Transportation Networks, PTN) mainly refers to networks that public transport was formed such as bus.Each research institution has researched and developed a series of representational bus network prototype systems in recent years, as DieselNet and DakNet etc.The deployment of these application prototype systems shows that bus network has important application prospect.But, consider that the research work of the Move Mode particularity of node in public traffic network also comparatively lacks.

In public traffic network, the particularity of Move Mode is: 1) between two websites, may have the automobile of many different order of classes or grades at school to arrive, but not can go directly between any two; 2) bus has fixing Move Mode, can carry out regular cycle movement according to set timetable.And in the different moment, i.e. rush-hour or non-peak period, the probability that bus arrives at a station is on time often not identical.For the dynamic change of such network topology, node characteristics of motion nondeterministic network scene, general method for analyzing and modeling has evolution figure, time diagram and Shi Biantu etc., and these methods all have limitation separately, cannot comprehensively describe the contact in bus network.On the other hand, in traditional Routing Protocol, ignorant route is as Epidemic, and SW etc. do not utilize route knowledge that routing performance is provided, and network resource consumption is larger; And the another kind of route of route knowledge based on contact prediction of utilizing be as PRoPHET, Maxprop, RCM etc. have simplified the mobility of node in network scenarios, are not node motor pattern is considered as to completely random, exactly node motion are considered as to strict rule.These agreements all do not consider in bus network the node cycle change, uncertain situation of the moment of meeting, and above-mentioned agreement do not utilize the node motion characteristic of bus network to improve routing performance, lacks specific aim.

At present, about a not unified description of the contact in DTN.2003, the definition that the people such as Kevin Fall contact provide DTN first in document (Adelay-tolerant network architecture for challenged internets) in.In literary composition, DTN network is defined as to a multiple directed graph, contact is defined as to the chance of a data transfer.More specifically, contact is the limit of a time dependent description transmittability in the drawings, and traditional contact describing method has evolution figure, time diagram and Shi Biantu etc.

The people such as Chen Y and Borrel V of Georgia Institute of Technology proposes first to evolve in document (A framework for characterizing the wireless and mobile network continuum) and schemes this method, and the method is described the dynamic change of DTN network topology with a subgraphs sequence.The method is generally applicable to contacting in the middle of the satellite network that can accurately predict future, and the scope of application is less.Document (Computing shortest, fastest, and foremost journeys in dynamic networks) definition of evolution figure is expanded, and provide an attribute based on this definition, provide the evaluation attributes of a network performance from statistical property.But in these two kinds of methods, all lack the expression method of the randomness to contact in future, and lack foresight activity.Time diagram (Temporal Graphs) (Statistical Mechanics and its Applications) is proposed in 2009 by the researcher Vassilis Kostakos of CMU.The method object is time data to be converted to figure, thereby can utilize the existing algorithm based on graph theory to find the dynamic time characteristic of data.The method has been placed on a time relationship on the summit of figure, does not consider continuation and the uncertainty of time relationship, only can be used in the analysis to historical data.Whether Shi Biantu (Time-Varying Graphs) has contact to judge by a function to this moment, and judges the duration of contact with a function.The method has been considered the uncertainty of contact, but does not consider the duration characteristic of contact.Can find out that existing DTN contacts describing method and cannot describe the characteristics such as the generation moment of contact, duration, probability of happening simultaneously, thereby can not well embody the uncertainty of node contact in bus network.

DTN route aspect, the difference of the network topological information of grasping according to node, DTN route can be divided into certainty route and probability route (Routing in intermittently connected mobile ad hoc networks and delay tolerant networks:overview and challenges).In certainty Routing Protocol, movement and the connection in node future are predictable, know in advance moving situation and the junctor meeting of network future, and whole topology of networks; And in the network environment of considering at probability Routing Protocol, network topology is dynamic, uncertain.If node lacks the understanding to network state, so all nodes can only be randomly to neighbor node forwarding data, and this situation belongs to routing diffusion; If a node can dope its forwarding probability to other neighbor nodes, and according to good path of this probability selection, this just belongs to based on forwarding historical or prediction.In probability route, each node of infectious disease route (Epidemic Routing) can be preserved the Hash table of No. ID, a record buffer memory district packet, is called summary vector.In the time that two nodes connect, first node intercourses summary vector, and then passes to its packet not having of the other side.This routing policy based on inundation can ensure higher data delivery rate, but shortcoming is also clearly, be exactly the massive duplication of packet very fast by buffering area approach exhaustion.

Spray-wait for (SW, Spray and Wait) route and be divided into injection phase and two stages of loitering phase.At injection phase, source node is disposable to be copied n part by packet and passes to n different neighborhood of nodes.At loitering phase, if do not arrive object node at injection phase packet, this n node is directly passed to packet the node that the next one runs into and does not do any replication work.

Above route is only worked in routing forwarding, does not utilize route knowledge to improve routing performance, but exchanges delivery ratio for by consuming a large amount of Internet resources.And what this type of route was random chooses via node, easily occur that the via node of selecting does not also have the source node node that more easily achieves the goal, or the via node of those nodes that more easily achieve the goal does not obtain the unreasonable situations such as message copy and occurs.

In view of inscience route limitation is larger, lack extensibility and adaptivity, researcher proposes to utilize the knowledge such as historical information to contact prediction route.PRoPHET route has been introduced communicating predicted (delivery predictability) this concept, the transfer probability that each node comprises other all nodes.Each node in the time of forwarding data bag, according to the transfer probability information table of oneself by data packet delivery the node to maximum probability.The Maxprop route utilization history information of meeting is carried out route optimization decision-making.In agreement, each node is saved in the collision probability vector of the each node of neighbours, when node meets, the collision probability of the other side's node is set to 1, and remaining probable value is reduced by half, and then probability is converted into limit weights calculating shortest path and delivers.The average account form of this employing increment has extremely strong volatility, easily produces the problem that uses out-of-date probability vector to calculate shortest path.RCM(Routing in Cyclic Mobispace) scene considered of route is the scene that node motion has the fixed cycle, agreement adopts this index of minimum transfer time delay as the weights of weighing link, adopts on this basis the method for Markovian decision to estimate and route node transmission duration.But for the scene of node motion change, this agreement is very not applicable.If the period of motion of node changes, agreement just needs decision-making again, lacks dynamic adaptable.

By the analysis to above several Routing Protocols, we find, in existing Routing Protocol, ignorant route is as Epidemic, SW etc. do not utilize route knowledge that routing performance is provided, network resource consumption is larger; And the another kind of route of route knowledge based on contact prediction of utilizing be as PRoPHET, Maxprop, RCM etc. have simplified the mobility of node in network scenarios, are not node motor pattern is considered as to completely random, exactly node motion are considered as to strict rule.These agreements all do not consider in bus network the node cycle change, uncertain situation of the moment of meeting, and above-mentioned agreement do not utilize the node motion characteristic of bus network to improve routing performance, lacks specific aim.

Summary of the invention

This paper is for the feature of public traffic network, from the angle of contact, a kind of abstract network topological model based on interval algebra has been proposed, provide a kind of contact Forecasting Methodology in conjunction with Bayesian Estimation, and on this basis, a kind of prediction method for routing for bus Delay Tolerant Network is proposed, especially a kind of routing algorithm---IABR(Interval Algebra based Bus net Routing algorithm based on contact prediction).

For a prediction method for routing for bus Delay Tolerant Network, it is characterized in that, comprise the steps:

(1), based on interval algebra, the uncertainty in bus arrival moment is described:

A time interval IT is an interval number, is the closed interval being made up of two real numbers, and two real numbers are respectively the high-low limit of time interval, and limes superiors correspondence the time the latest in uncertain moment; Limit inferior correspondence the earliest time in uncertain moment; The bus arrival moment is a random variable of continuous type, and we represent this event of bus arrival with X; If its probability density function is f (x), if the time interval that event X occurs is (0 ,+∞), the interval of its generation is intercepted, and then obtains finite time interval, definition: for event X is at time interval [t ^-, t ⁺] in the probability that occurs, represent with alphabetical P, that is: $P(t^{-}\≤x\≤t^{+})={\∫}_{t^{-1}}^{t^{+}}f\left(x\right)\mathrm{dx};$

(2), to bus Delay Tolerant Network abstract modeling:

Describe definition according to contact, provide the definition of bus network: G::=(V, E), Website Hosting V is made up of bus stop, represents by the sequence number at station, and sequence number represents with a natural number; E is the set of contact, in the time there is a public bus network between two stations, has a contact e between two stations, represents i.e. e=(id with a five-tuple _s, id _d, bus, IT _s, f (x)); Wherein, id _sbus stop, expression source, id _drepresent object bus stop, id _s, id _d∈ V, bus represents public transport license number, IT _sbe a uncertain moment, represent that vehicle bus arrives id _dthe uncertain moment, f (x) represents the arrival time probability density function of this vehicle; N represents natural number set, and the formal definitions of model is as follows:

(3), based on Bayesian Estimation, the contact of future time is estimated:

By Bayesian formula: h (μ | x') ∝ π (μ) p (x'| μ), the contact interval time posterior density of μ is: wherein, ${\mathrm{\σ}}_1^2={(\frac{1}{{{\mathrm{\σ}}_0}^2}+\frac{n}{{\mathrm{\σ}}^2})}^{-1}=\frac{{\mathrm{\σ}}^2{\mathrm{\σ}}_0^2}{n{\mathrm{\σ}}_0^2+{\mathrm{\σ}}^2},$ ${\mathrm{\μ}}_1=(\frac{{\mathrm{\μ}}_0}{{{\mathrm{\σ}}_0}^2}+\frac{n\stackrel{\‾}{x}}{{{\mathrm{\σ}}_0}^2}){\mathrm{\σ}}_1^2;$ Obtaining after a posteriori distribution density of contact generation, we define: establish x ₁, x ₂..., x _nthe sample from overall X, X=(x ₁, x ₂..., x _n), t is unknown parameter, t ₁=t ₁(x), t ₂=t ₂(x) be two statistics, for given α ∈ (0,1), if there is P (t ₁≤ t≤t ₂)>=1-α, just claims [t ^-, t ⁺] be that the confidence level of parametric t is the confidential interval of (1-α); Expect μ interval time ₁confidence level for the confidential interval of (1-α) be:

(4), according to the contact information that dopes, link Metric value is quantized to calculate and carry out routing decision: the quality in path is weighed by a Metric value, and Metric is less, and path is more excellent; The Metric value computing formula in path 1 is as follows: ${\mathrm{Metric}}_l={\mathrm{\μ}}_1-\mathrm{\Δ}{t}_1+{\mathrm{dis}}_1+\underset{i=2}{\overset{n}{\mathrm{\Σ}}}(M_{\mathrm{delay}}+M_{\mathrm{wait}});$ Suppose for i article of limit, current time and last time arrival time difference be Δ t _i, interval number after treatment is IT _i, for node i, if IT _i> IT _i-1, representing can directly give next node after data arrive node i, now its delay is M _delay=P (IT _i> IT _i-1) × (μ _i-Δ t _i+ dis _i); Otherwise, need to wait for the arrival of next car, therefore, its stand-by period is M _wait=(1-P (IT _i> IT _i-1)) × (2 μ _i-Δ t _i+ dis _i)

Described step 4 further comprises:

(4.1), the time interval of the first jumping is processed, add the dis value of this jumping: [l ₁, r ₁] ← [l ₁+ dis ₁, r ₁+ dis ₁], upgrade Metric value simultaneously;

(4.2) time interval and upper hop when front jumping are compared, if P is ([l _i, r _i] > [l _i-1, r _i-1])=0, [l _i, r _i] ← [l _i+ μ _i, r _i+ μ _i], until P ([l _i, r _i] > [l _i-1, r _i-1]) ≠ 0, and according to the cumulative Metric value of the computing formula of Metric, and will add the dis value of this jumping when the time interval of front jumping: [l _i, r _i] ← [l _i+ dis _i, r _i+ dis _i];

(4.3) repeating step 4.2 to the last one is jumped, simultaneously the final Metric value of storing path;

(4.4) repeating step 4.1-4.3, until all paths are disposed;

(4.5) choosing the minimum front n paths of Metric value forwards.

Brief description of the drawings

Fig. 1 is schematic flow sheet of the present invention;

Fig. 2 is that path delay of the present invention calculates schematic diagram;

Fig. 3 is that the present invention is about the routing algorithm false code of estimating based on contact;

Fig. 4 is ONE simulating scenes schematic diagram of the present invention;

Fig. 5 is that the cache size of the present invention and traditional route affects performance comparison diagram to delivery ratio;

Fig. 6 is that the cache size of the present invention and traditional route affects performance comparison diagram to transmission average retardation;

Fig. 7 is that the cache size of the present invention and traditional route affects performance comparison diagram to offered load rate.

Embodiment

1. the uncertainty in pair bus arrival moment is described

The bus arrival moment is a uncertain moment, and we define: being a time interval IT(Interval Time) interval number, is the closed interval being made up of two real numbers, and two real numbers are respectively the high-low limit of time interval.Limes superiors correspondence the time the latest in uncertain moment; Limit inferior correspondence the earliest time in uncertain moment.

The bus arrival moment is a random variable of continuous type, and we represent this event of bus arrival with X.If its probability density function is f (x), if the time interval that event X occurs is (0 ,+∞), we need to intercept the interval of its generation, and then obtain finite time interval.We define: for event X is at time interval [t ^-, t ⁺] in the probability that occurs, represent with alphabetical P, that is: $P(t^{-}\≤x\≤t^{+})={\∫}_{t^{-1}}^{t^{+}}f\left(x\right)\mathrm{dx}.$

2. bus network abstract modeling

Bus network is abstract in by gathering the multiple directed graph G::=(V, E) that the contact set between station forms in station, and Website Hosting V is made up of bus stop, represents by the sequence number at station; E is the set of contact, in the time there is a public bus network between two stations, has a contact e between two stations, represents i.e. e=(id with a five-tuple _s, id _d, bus, IT _s, f (x)).Wherein, id _sbus stop, expression source, id _drepresent object bus stop, id _s, id _d∈ V, bus represents public transport license number, IT _sbe a uncertain moment, represent that vehicle bus arrives id _dthe uncertain moment, f (x) represents the arrival time probability density function of this vehicle.The formal definitions of model is as follows:

3. based on bayesian theory, bus is contacted and estimated

The estimation of the contact to two nodes refers to and utilizes its history information of meeting to draw the time of origin interval that this two node meets next time.Here, we directly do not ask the time of contact of bus, but bus time of contact are converted into μ interval time of twice contact.

According to Bayesian formula, the posteriority that obtain the bus contact generation moment distributes, and need to know its prior distribution and sample information.Our hypothesis: taking interval time of each bus arrival as sample, in the situation that sample size is enough large, Dui Ge road bus, the prior distribution Normal Distribution of its interval time of arriving at a station, expect μ ₀equal the headway of each road bus, variance can be obtained by historical experience information.Sample information aspect, the sample distribution density under specified criteria μ meets p (x _i| μ) ~ N (μ, σ ²), σ ²can adopt sample variance S ²replace.By Bayesian formula: h (μ | x') ∝ π (μ) p (x'| μ), the posterior density that finally can obtain contact interval time μ is: x ' represents stochastic variable, refers to the variable in sample here.

$h\left(\mathrm{\μ}\right|x^{\′})~N({\mathrm{\μ}}_1,{\mathrm{\σ}}_1^2)---\left(0.1\right)$

Wherein, ${\mathrm{\σ}}_1^2={(\frac{1}{{{\mathrm{\σ}}_0}^2}+\frac{n}{{\mathrm{\σ}}^2})}^{-1}=\frac{{\mathrm{\σ}}^2{\mathrm{\σ}}_0^2}{n{\mathrm{\σ}}_0^2+{\mathrm{\σ}}^2},$ ${\mathrm{\μ}}_1=(\frac{{\mathrm{\μ}}_0}{{{\mathrm{\σ}}_0}^2}+\frac{n\stackrel{\‾}{x}}{{{\mathrm{\σ}}_2}^2}){\mathrm{\σ}}_1^2,$ N refers to sample number, is obtaining after a posteriori distribution density of contact generation, and we define: establish x ₁, x ₂..., x _nthe sample from overall X, X=(x ₁, x ₂..., x _n), t is unknown parameter, t ₁=t ₁(x), t ₂=t ₂(x) be two statistics, for given α ∈ (0,1), if there is P (t ₁≤ t≤t ₂)>=1-α, just claims [t ^-, t ⁺] be that the confidence level of parametric t is the confidential interval of (1-α).

Providing after the concept of confidence in interval time of contact, we can obtain: if the posteriority of contact interval time is distributed as expect μ interval time ₁confidence level be the confidential interval of (1-α), contact occur time interval be:

$[\stackrel{\‾}{x_n}-\frac{z_{a/2}{\mathrm{\σ}}_1}{\sqrt{n}},\stackrel{\‾}{x_n}+\frac{z_{a/2}{\mathrm{\σ}}_1}{\sqrt{n}}]---\left(0.2\right)$

Wherein, for sample average, z _α/2represent upper α/2 quantile of standardized normal distribution.

Can find out the length of confidential interval α is less, and L is larger, and the time interval obtaining is just wider; α is larger, and L is less, and the time interval obtaining is just narrower.Like this, we just can be according to different situations, by regulating confidence level to reach the object of controlling confidential interval.After confidence level reduces, confidential interval broadens, and on path, former and later two contacts exist the possibility of overlapping time to become large, therefore qualified number of path can be more, but the order of accuarcy of prediction can reduce, easily cause the failure of forwarding, otherwise, confidence level is higher, and confidential interval is narrower, and qualified number of path is fewer, but result is more accurate, the possibility of retransmission failure is less.

4. carry out link prediction according to contact estimated value

Because message path is made up of multiple contacts, therefore need different contact time of origins to sort, the time interval of supposing contact A is [a, b], and the time interval of contact B is [c, d], and the posteriority of A is distributed as the posteriority of B is distributed as the sequential relationship that we define A and B is as follows:

1) if a-d >=0, P (A > B)=1;

2) if b-c < 0, P (A > B)=0;

3) if a-d≤0, b-c > 0, $P(A>B)={\&Integral;}_0^{b-c}{f}_{A-B}\left(x\right)\mathrm{dx},f_{A-B}\left(x\right)=N({\mathrm{\&mu;}}_A-{\mathrm{\&mu;}}_B\&pound;\&not;{\mathrm{\&sigma;}}_A^2+{\mathrm{\&sigma;}}_B^2).$

Because all bus routes are fixed, therefore we can know all possible path between any two stations in advance.The available information in station comprises: interval IT interval time of the route of each road bus, the historical juncture that bus arrives each website and each contact _i=[l _i, r _i] and posterior probability distribution by the historical juncture record of bus arrival, we define the distance dis between station _aBfor the time difference of the last this road bus from station A to station B, μ _iit is the expected elapsed time (Fig. 2 (a)) on i article of limit.The quality in path is weighed by a Metric value, and Metric is less, and path is more excellent.The Metric value computing formula of path l is as follows:

${\mathrm{Metric}}_l={\mathrm{\&mu;}}_1-\mathrm{\&Delta;}{t}_1+{\mathrm{dis}}_1+\underset{i=2}{\overset{n}{\mathrm{\&Sigma;}}}(M_{\mathrm{delay}}+M_{\mathrm{wait}})---\left(0.3\right)$

Suppose for i article of limit, current time and last time arrival time difference be Δ t _i, interval number after treatment is IT _i, for node i, if IT _i> IT _i-1, representing can directly give next node after data arrive node i, now its delay is M _delay=P (IT _i> IT _i-1) × (μ _i-Δ t _i+ dis _i); Otherwise, need to wait for the arrival of next car, therefore, its stand-by period is M _wait=(1-P (IT _i> IT _i-1)) × (2 μ _i-Δ t _i+ dis _i).

Taking Fig. 2 (a) as example, node A has data retransmission to F.Wherein the information of each contact of a paths as shown in the figure.First to contact Contact _aBprocess (Fig. 2 (b)), Metric _aF=4-3+3=4; Next to contact Contact _bDprocess (Fig. 2 (c)), upgrade Metric value simultaneously:

Metric _AF＝4+P([4,8]＞[6,8])×(6-4+4)+(1-P([4,8]＞[6,8]))×(12-4+4)

＝16-6P([4,8]＞[6,8])

Finally to contact Contact _dFprocess (Fig. 2 (d)), obtain final Metric value and be:

Metric _AF＝24-6P([4,8]＞[6,8])-3P([8,10]＞[8,12])

Final routing algorithm is described below:

1) time interval of the first jumping is processed, added the dis value of this jumping: [l ₁, r ₁] ← [l ₁+ dis ₁, r ₁+ dis ₁], upgrade Metric value simultaneously;

2) time interval and upper hop when front jumping are compared, if P is ([l _i, r _i] > [l _i-1, r _i-1])=0, to [l _i, r _i] carry out assignment again: [l _i, r _i] ← [l _i+ μ _i, r _i+ μ _i], until P ([l _i, r _i] > [l _i-1, r _i-1]) ≠ 0, and according to the cumulative Metric value of the computing formula of Metric, and will add the dis value of this jumping when the time interval of front jumping: [l _i, r _i] ← [l _i+ dis _i, r _i+ dis _i];

3) repeating step 2 to the last one is jumped, simultaneously the final Metric value of storing path;

4) repeating step 1-3, until all paths are disposed;

5) choosing the minimum front n paths of Metric value forwards.

The false code of algorithm as shown in Figure 3.

Final routing decision selects the minimum front n paths of Metric value as forward-path, if n=1, this route is single copy route; If n>1, this route is many copies route, and agreement can regulate according to concrete environment the size of copy number n.In addition, we have also adopted some simple measures to improve the performance of route, and such as notify mutually the successfully data ID of payment, the data of delivered in node Automatic clearance buffer memory, improve the utilance of buffer memory with this.

Algorithm complex aspect, supposes that number of path is m, and the average number of hops of every paths is k, and the complexity of algorithm is O (mk) so.In fact, because the account form of path metric is depth-first search, the metric value in path is along with the increase monotonic increase of jumping figure, therefore in the time calculating the metric value of a paths and be greater than n paths, algorithm can proceed to the calculating of next paths immediately, and therefore the time overhead of calculating path can be less than the complexity of theory analysis.

Effect of the present invention can further illustrate by following simulation result:

1. simulated conditions

The selected ONE of the present invention simulation software is as imitative experiment platform.We are by IABR agreement in this paper and FirstContact(FC), Epidemic(ED), Spray and Wait(SW), Maxprop(Maxp) and PRoPHET(Prop) five kinds of typical DTN routes carry out the impact of simulated experiment delivery ratio, average delay and the expense ratio of the different cache size of comparative analysis on different routes.We have built simulating scenes as shown in Figure 4 using area, the Olympic Sports Center, Beijing as simulated environment according to actual mileage chart.Simulating scenes relevant parameter information is as shown in the table:

Table 1ONE simulation parameter

Map size	5×5km 2
		Bus number	20 tunnel × 5/road
Bus stop number	85
		Mobility model	MapBasedMovement
The speed of a motor vehicle	5～10m/s
		The emulation cycle	8000s
Number-of-packet	1600
		Data package size	1MB
Communication radius	150m

In scene, node is all in initial position, the communication radius of hollow ring representation node.The bus track of every circuit node immobilizes, and each node is doing back and forth movement on route separately.Packet produces at random at each website.Due to for most of DTN networks, the cache size of node is the key factor that affects network performance, and therefore to gather respectively the buffer memory of bus be the data of 10MB to 1000MB in experiment.

2. simulation result

Fig. 6 has provided the delivery success rate under 6 kinds of different cache size of route.Deliver success rate relatively in, IABR route is a little less than Maxprop route, but far above other routes.Wherein, FC and SW route belong to inscience route, and therefore delivery ratio is lower, and these two kinds of agreements rely on for the buffer memory of node and not quite, therefore the increase of buffer memory does not almost affect its routing performance.The transfer rate of all the other routes is along with the increase of nodal cache all has lifting in various degree.Can find out from the trend promoting, the transmission success rate of Epidemic route and PRoPHET route is larger for the dependence of buffer memory, and IABR route and Maxprop route only have transmission success rate under the condition of 10MB at buffer memory be 60% left and right, and in the time that bringing up to 50MB, nodal cache reaches very high transmission success rate, illustrate that these two kinds of routes are less to the dependence of buffer memory, the utilance of buffer memory is higher.

Fig. 7 has provided the mean transit delay under 6 kinds of different cache size of route.Because the transmission mechanism of FC route is to be transmitted to first node oneself running into, this extreme routing policy cannot ensure effective transmission of data, and our emulation experiment also shows that the average transfer delay of FC route is the highest.IABR route and Maxprop agreement have comparatively been stablized routing performance owing to showing after reaching a certain size at buffer memory, and therefore the lasting increase of buffer memory does not almost affect their average delay.And the performance of Epidemic route is subject to the impact of cache size larger, be subject in limited time at buffer memory, because most messages is dropped processing, therefore mean transit delay is now lower; And in the time that buffer memory increases, packet loss number reduces, the message number that success is transmitted increases, and causes mean transit delay increase to a certain extent; And along with the further increase of buffer memory, no longer dropping packets of node, message success transfer rate reaches a high value, and now mean transit delay also declines thereupon.

Duty ratio refers to the message amount that starts transmission but do not arrive destination in network, with the ratio of successful message amount of paying.For algorithm, its expense, than larger, means more overhead of message needs of its every transmission.Can find out, IABR route higher than SW route lower than other routes.SW route is because the number of copies of its generation is fixed, and therefore its duty ratio is not affected by buffer memory.Epidemic route is because belong to the routing policy of inundation formula, therefore when buffer memory hour, because delivery ratio is lower, cause its expense ratio will be far above all the other routes, but along with the increase of buffer memory, the message amount that success is paid increases, and its duty ratio is declined rapidly.In like manner, along with the increase of buffer memory, the duty ratio of Maxprop route and PRoPHET route also decreases, but all the time higher than IABR route.

Experiment shows, under bus network scene, the delivery ratio of IABR route has reached 90% left and right, higher than most of DTN routes such as PRoPHET, Epidemic and SW, have simultaneously good expense than and average delay.It is larger that the performance of PRoPHET route, SW route is affected by nodal cache and simulating scenes: because experiment scene is larger, internodal average number of hops is more, therefore SW route its transmission success rate that can only ensure feature limits that limited redirect is sent out; And PRoPHET route based on be the network environment of node random motion, experiment shows that PRoPHET route can not well be applicable to the move network environment of the relative rule feature such as fixing relative to network topology of node.

IABR route and Maxprop route all belong to the routing algorithm based on historical knowledge.This type of route takes full advantage of the information that topological model provides, and topology information and effectiveness parameter have been carried out to preliminary treatment and pre-stored, is therefore relatively suitable for bus network.And emulation experiment also shows, in bus network scene, have certain cache size at node, IABR route and Maxprop route can reach higher transmission performance.Than Maxprop route, the delivery ratio of IABR and mean transit delay are slightly high, but aspect duty ratio, performance is more excellent, respectively has quality.

Claims (2)

1. for a prediction method for routing for bus Delay Tolerant Network, it is characterized in that, comprise the steps:

(1), based on interval algebra, the uncertainty in bus arrival moment is described: a time interval IT is an interval number, it is the closed interval being formed by two real numbers, two real numbers are respectively the high-low limit of time interval, and limes superiors correspondence the time the latest in uncertain moment; Limit inferior correspondence the earliest time in uncertain moment;

The bus arrival moment is a random variable of continuous type, represents this event of bus arrival with X; The arrival time probability density function of this vehicle is f (x), if the time interval that event X occurs is (0 ,+∞), the interval of its generation is intercepted, and then obtain finite time interval, definition: for event X is at time interval [t ^-, t ⁺] in the probability that occurs, represent with alphabetical P, that is:

(2), to bus Delay Tolerant Network abstract modeling: describe definition according to contact, provide the definition of bus network: G::=(V, E), Website Hosting V is made up of bus stop, represents by the sequence number at station, and sequence number represents with a natural number; E is the set of contact, in the time there is a public bus network between two stations, has a contact e between two stations, represents i.e. e=(id with a five-tuple _s, id _d, bus, IT _s, f (x)); Wherein, id _sbus stop, expression source, id _drepresent object bus stop, id _s, id _d∈ V, bus represents public transport license number, IT _sbe a uncertain moment, represent that vehicle bus arrives id _dthe uncertain moment, f (x) represents the arrival time probability density function of this vehicle; N represents natural number set, and the formal definitions of model is as follows:

(3), based on Bayesian Estimation, the contact of future time is estimated: by Bayesian formula: h (μ | x') ∝ π (μ) p (x'| μ), the contact interval time posterior density of μ is: wherein, wherein, in the situation that sample size is enough large, Dui Ge road bus, the prior distribution Normal Distribution of its interval time of arriving at a station, sample information aspect, the sample distribution density under specified criteria μ meets p (x _i| μ)～N (μ, σ ²), x ' represents stochastic variable, refers to the variable in sample here; σ ²can adopt sample variance S ²replace, expect μ ₀equal the headway of each road bus, variance can be obtained by historical experience information;

Obtaining after a posteriori distribution density of contact generation definition: establish x ₁, x ₂..., x _nthe sample from overall X, X=(x ₁, x ₂..., x _n), t is unknown parameter, t ₁=t ₁(x), t ₂=t ₂(x) be two statistics, for given α ∈ (0,1), if there is P (t ₁≤ t≤t ₂)>=1-α, just claims [t ₁, t ₂] be that the confidence level of parametric t is the confidential interval of (1-α); Expect μ interval time ₁confidence level for the confidential interval of (1-α) be: represent normal distribution 's ^α/ ₂quantile;

(4), according to the contact information that dopes, link Metric value is quantized to calculate and carry out routing decision: the quality in path is weighed by a Metric value, and Metric is less, and path is more excellent; The Metric value computing formula of path l is as follows: by the historical juncture record of bus arrival, we define the distance dis between station _aBfor the time difference of the last this road bus from station A to station B, μ _iit is the expected elapsed time on i article of limit; Suppose for i article of limit, current time and last time arrival time difference be Δ t _i, interval number after treatment is IT _i, for node i, if IT _i> IT _i-1, representing can directly give next node after data arrive node i, now its delay is M _delay=P (IT _i> IT _i-1) × (μ _i-Δ t _i+ dis _i); Otherwise, need to wait for the arrival of next car, therefore, its stand-by period is M _wait=(1-P (IT _i> IT _i-1)) × (2 μ _i-Δ t _i+ dis _i).

2. method according to claim 1, is characterized in that, described step 4 further comprises:

(4.1), the interval time of each contact interval IT _i=[l _i, r _i], the time interval of the first jumping is processed, add the dis value of this jumping: [l ₁, r ₁] ← [l ₁+ dis ₁, r ₁+ dis ₁], upgrade Metric value simultaneously;

(4.2) time interval and upper hop when front jumping are compared, if P is ([l _i, r _i] > [l _i-1, r _i-1])=0, [l _i, r _i] ← [l _i+ μ _i, r _i+ μ _i], until P ([l _i, r _i] > [l _i-1, r _i-1]) ≠ 0, and according to the cumulative Metric value of the computing formula of Metric, and will add the dis value of this jumping when the time interval of front jumping: [l _i, r _i] ← [l _i+ dis _i, r _i+ dis _i]; r _irepresent i article of limit time of contact interval section the upper bound;

(4.3) repeating step 4.2 to the last one is jumped, simultaneously the final Metric value of storing path;

(4.4) repeating step 4.1-4.3, until all paths are disposed;

(4.5) choosing the minimum front n paths of Metric value forwards.