CN106102167A - Real-time on-demand data broadcast scheduling adaptive channel divides and distribution system and method - Google Patents

Abstract

The invention discloses a kind of real-time on-demand data broadcast scheduling adaptive channel to divide and distribution system and method, system is made up of a server and multiple user, comprises a up channel and several down channels, and upstream channel bandwidth is less than down channel；When user needs data, upload a request to server by up channel；After upload request, user's monitors downlink channel obtains data；Server is for safeguarding a request queue RQ, a preparation queue PQ and multiple broadcast queue BQ_i0≤i≤N；Request queue RQ is responsible for collecting user's request, and server obtains data genaration from data base according to request queue RQ and prepares queue PQ；Server is periodically from preparing queue PQ extraction data tissue broadcast.The present invention adjusts channel dynamically according to real time data request environment, it is possible to effectively adapt to hot spot data dispersion, focus change is fast and data item spy changes fast broadcast environment.Can effectively improve the broadcasting efficiency of real-time on-demand Radio Data System.

Description

Real-time on-demand data broadcast scheduling adaptive channel divides and distribution system and method

Technical field

The invention belongs to computer science and real-time on-demand data broadcast scheduling technical field, relate to a kind of real-time on-demand number Divide and distribution system and method OCSM (Optimized Channel Split Method) according to broadcast scheduling adaptive channel.

Background technology

Owing to mobile network's technology and data broadcast scheduling support that a large number of users accesses hot spot data, real-time on-demand number simultaneously It is widely used in the transmission of real-time hot spot data according to broadcast scheduling, as stock information transmission, Traffic information demonstration etc. are real-time System.Data broadcast packets includes quiescent period broadcast and two kinds of broadcast modes of on-demand broadcasting.Owing to information is taken by mobile network user Improving constantly of business quality requirement, the broadcast of real-time on-demand data becomes the focus of current research, and ageing is the broadcast of on-demand data The important restrictions condition of scheduling, designs efficient data broadcast scheduling algorithm, and it is on-demand for meeting user's request to greatest extent The most important thing of data broadcast research.

On-demand digital broadcasting dispatch system real-time reception user data requests, and utilize corresponding dispatching algorithm tissue data Broadcast scheduling sequence.On-demand data broadcast scheduling has on-demand and real-time.It is wide that on-demand is embodied in each broadcast cycle institute The data broadcast are asked to determine by user, and real-time is embodied in the service of user's request must be expired in a broadcast cycle Foot.On-demand data broadcast scheduling is effectively reduced average latency and the tuning period of mobile terminal of user's request, because of And be widely used in dynamic, the broadcast of large-scale data and transmission.Existing technology have studied the on-demand number of single channel According to broadcast, it is proposed that the most outstanding algorithm, such as SIN-α, RxW algorithm etc., these algorithms are reducing system request crash rate, fall The aspects such as harmonic(-)mean waiting time, minimizing client tuning period and reduction client energy consumption have outstanding performance.But along with shifting The development of dynamic network technology, the variation of user's request makes broadcasting dispatch system being supported, the demand of many data item request is continuous Improving, single channel on-demand data broadcast scheduling can not play multichannel broadcast in parallel advantage.For this problem, existing skill Art have studied fixing multichannel many data item dispatching algorithm, and prove that fixing multichannel many data item request scheduling is one The problem of NP-hard, it is proposed that substantial amounts of dispatching distribution algorithm.Such as TOSA (Near-Optimal Scheduling Algorithm) algorithm etc. can well combine the feature of multichannel broadcast in parallel, improves data broadcast scheduling efficiency.Gu but Determining multicasting dispatching algorithm can only be for the network of particular demands, it is impossible to adapt to changeable demand environment.

Either single channel broadcast or fixing multicasting all cannot the dispersion of self-adaptive hot pot data, focus changes Fast and that data feature difference is big broadcast environment.First, in a mobile network, physical restriction based on structure, such as client End may have different communication capacities, constrains the feasibility of single channel high-speed transfer；Secondly, the need of user oriented application Asking, channel can merge or coordinate the service quality variable with offer；Meanwhile, along with the dispersion of real-time hot spot data, multichannel Broadcast and have higher ageing relative to single channel broadcast；Finally, broadcast letter is adjusted in real time according to broadcast data feature Road can preferably adapt to changeable mobile network environment.Therefore, based on the changeable reality of active user's demand and above-mentioned calculation Some limitation that method exists, are studied adaptive multi-channel broadcast scheduling herein, it is proposed that a kind of real-time on-demand number Divide and distribution method OCSM (Optimized Channel Split Method) according to the adaptive channel of broadcast.

Summary of the invention

Adaptive multi-channel broadcast scheduling is studied by the present invention, it is proposed that a kind of real-time on-demand data broadcast from Adaptive channel divides and distribution system and method OCSM.

The system of the present invention be the technical scheme is that a kind of real-time on-demand data broadcast scheduling adaptive channel divides With distribution system, it is characterised in that: be made up of a server and multiple user, comprise a up channel and several under Row channel, upstream channel bandwidth is less than down channel；When user needs data, upload a request by up channel To server；After upload request, user's monitors downlink channel obtains data；Described server is for safeguarding a request queue RQ, a preparation queue PQ and multiple broadcast queue BQ_i0≤i≤N；Described request queue RQ is responsible for collecting user's request, Server obtains data genaration from data base according to request queue RQ and prepares queue PQ；Described server is periodically from preparation Queue PQ extracts the broadcast of data tissue.

The method of the present invention be the technical scheme is that a kind of real-time on-demand data broadcast scheduling adaptive channel divides With distribution method, it is characterised in that comprise the following steps:

Step 1: initialize；

N number of data item is divided into N class, and define arrays T={t₁,t₂,...,t_i,...,t_NAs grader, wherein t_iFor Data item d_iClass label；

Step 2: user sends request Req by up channel；

Step 3: judge currently whether have request Req_i；

The most then perform following step 4；

If it is not, server continues to monitor up channel, and turn round the described step 3 of execution；

Step 4: take out Req_i, for data item d of this request_iPerform operations described below；

If d_iIt is present in broadcast queue BQ_iIn, then by this Req_iAdd this d to_iRequest list in；

If d_iIt is present in preparation queue PQ, then by this Req_iAdd this d to_iRequest list in；

Otherwise, by this Req_iJoin in request queue RQ；

Step 5: judge upper broadcast cycle K_i-1Whether terminate；

The most then perform following step 6；

If it is not, then revolution performs above-mentioned steps 2；

Step 6: use the broadcast of OCSM algorithm organization；

Step 7: broadcasted a d_i, then by this d_iSuccessful request information write file, delete this broadcast item institute ask the visitor in Asking, delete this broadcast item, advance a time point；

Step 8: judge whether to exceed running time T；

The most then perform following step 9；

If it is not, then revolution performs above-mentioned steps 2；

Step 9: read successful request information, adds up solicited message.

As preferably, using the broadcast of OCSM algorithm organization described in step 6, it implements process and includes following sub-step Rapid:

Step 6.1: utilize data mining clustering algorithm RxW/SL, determines each data item d in preparation queue PQ_iPower Weight values, provides clustering target for next step；

Step 6.2: utilize data item to equalize clustering algorithm WASC, according to data item d_iPriority and data item sizePreparation queue PQ is divided into n classification FG={g₁,g₂,…g_n, wherein g_iRepresent i-th packet；

Step 6.3: utilize channel to divide and allocation algorithm CSA, is divided into corresponding n of correspondence according to FG by channel C Channel set SC={c₁,c₂,...,c_i,...c_n, and by g_iThe data item of packet is placed in subchannel c_iBroadcast.

As preferably, the process that implements of step 6.1 is:

Assume subsequent time broadcast data item d_i, draw broadcast-end-time, thus calculate broadcast d_iTo cause system its The failure number that his data item request is totalThe least d_iPriority is the highest；Consider data item number of request, the longest wait simultaneously Two factors of time, are calculated by formula (1)Value weighs the urgency of data item, determines the priority of data item with this；

${\mathrm{Wf}}_{d_i}=\frac{R_{d_i}\×W_{d_i}}{{\mathrm{SL}}_{d_i}}---\left(1\right);$

Wherein,Represent data item d_iNumber of request,Represent data item d_iThe high latency of request,For extensively Broadcast d_iBy causing system, other ask total failure number；It is worth the biggest, data item priorityThe highest；

In formula (1), ifIt is 0, then usesWeigh d_iPriority；Therefore formula (1) is revised as (2)；

${\mathrm{Wf}}_{d_i}=\left\{\begin{array}{ccc}\frac{R_{d_i}\×W_{d_i}}{{\mathrm{SL}}_{d_i}}& if& {\mathrm{SL}}_{d_i}\≠0\\ R_{d_i}\×W_{d_i}& if& {\mathrm{SL}}_{d_i}=0\end{array}\right.---\left(2\right).$

As preferably, implementing of step 6.2 includes following sub-step:

Step 6.2.1: data item size in queue PQ will be preparedIt is stored in vector S；

Step 6.2.2: N number of data item sample is divided into N class, data item d_iCorresponding i-th class, class is designated as i, is stored in vector In T；

Step 6.2.3: sample carries out 10 times of cross validations, sample is divided into 10 points, carries out 10 to every a sample and takes turns Secondary KNN algorithm classification obtains 9 labels of initially presorting, and uses ballot method to determine the label z that finally presorts of this sample_i；

Step 6.2.4: the label of presorting of all samples is collected generation prediction classification samples Z_T；

Step 6.2.5: utilize formulaCalculate accurate angle value A of sample classification_T；

Step 6.2.6: the sample of a part of classification error of random choose, by its tag along sort t_iTag along sort is predicted with it z_iExchange, generates new tag along sort V；

Step 6.2.7: V is replaced T, obtains A_V；

Step 6.2.8: if AV >=AT, then T=V, Z_T=Z_V, A_T=A_V, R=R-1；

Step 6.2.8: judge；

If A_TEqual to 1 or reach maximum cycle, then perform following step 6.2.9；

Otherwise, revolution performs above-mentioned steps 6.2.3；

Step 6.2.9: adjust tag along sort numerical value in tag along sort T so that it is by { 1,2,3...} is incremented by, by item set Close D and be divided into categorieN subclass G={g₁,g₂,...g_i,...g_categorieN}；

Step 6.2.10: initialize empty collection of data items

FG={g_1,1,g_1,2,…g_i,1,g_i,2,…g_categorieN,1,g_categorieN,2, wherein g_i,1,g_i,2By the g in G_iWarp Step 6.2.12-6.2.13 generates；

Step 6.2.11: from the beginning of i=1, is sequentially executable following step 6.2.12-step 6.2.13, until i= categorieN；

Step 6.2.12: initialize

Step 6.2.13, for set g_iIn each data item d_j, it is judged thatWhether set up；

The most then by data item d_iAdd to_gi,1In set；WhereinFor data Item d_jSize；

If it is not, then by data item d_iAdd to_gi,2In set；

Step 6.2.14: output FG.

As preferably, implementing of step 6.2.9 includes following sub-step:

(1) definition comprises array Tag{-1 of N number of element, and-1 ... ,-1}, categorieN=0；

(2) from the beginning of i=1, it is sequentially executable following (3)-(6), until i=N；

(3) if Tag [i]=-1, Tag [i] represents that in Tag array, i-th element, i.e. tag along sort are not adjusted, Then n=t_i；categorieN++；t_i=categorieN；

(4) from the beginning of j=i+1, it is sequentially executable following (5), until j=N；

(5) if t_j=n&&Tag [j]=-1, then t_i=categorieN, Tag [j]=0；

(6) Tag [i]=0；

(7) T class mark is assigned to G, exports G, categorieN.

As preferably, implementing of step 6.3 includes following sub-step:

Step 6.3.1: initialize array Sumdata [categorieN]；Wherein collection of data items D is divided into categorieN Individual subclass G={g₁,g₂,...g_i,...g_categorieN}；

Step 6.3.2: from the beginning of i=1, is sequentially executable following step 6.3.3, until i=N；

Step 6.3.3: scan collection of data items D to be broadcast, if its class label t_i=j, then $Sumdata\[j\]=Sumdata\[j\]+{\mathrm{Size}}_{d_i};$

Step 6.3.4: initialize categorieN sub-broadcast channel；

Step 6.3.5: from the beginning of i=1, is sequentially executable following step 6.3.6, until i=categorieN；

Step 6.3.6: ask for subchannel c_iAmount of bandwidth

Step 6.3.7: output subchannel { c₁,c₂,...,c_i,...,c_categorieN}；

Step 6.3.8: initialize categorieN Ge Zi broadcast queue；

Step 6.3.9: from the beginning of i=1, is sequentially executable following step 6.3.10, until i=N；

Step 6.3.10: scan data item in collection of data items D to be broadcast, if d_iTag along sort be t_i, then by data item d_iJoin sub-broadcast queueIn；

Step 6.3.11: from the beginning of i=1, is sequentially executable following step 6.3.12, until i=categorieN；

Step 6.3.12: by sub-broadcast queue bd_iBy data item weighted valueSequence；

Step 6.3.13: export sub-broadcast queue { bd₁,bd₂,...,bd_i,...,bd_categorieN}。

The present invention is according to real time data request feature self-adaptative adjustment broadcast channel and size, thus effectively improves broadcast The broadcasting efficiency of system, system robustness, and effectively reduce crash rate and the average latency of user's request.

Accompanying drawing explanation

Fig. 1 is the system structure schematic diagram of the embodiment of the present invention；

Fig. 2 is the method flow diagram of the embodiment of the present invention；

Fig. 3 is the data Partition Analysis figure of the embodiment of the present invention.

Detailed description of the invention

Understand and implement the present invention for the ease of those of ordinary skill in the art, below in conjunction with the accompanying drawings and embodiment is to this Bright it is described in further detail, it will be appreciated that enforcement example described herein is merely to illustrate and explains the present invention, not For limiting the present invention.

Along with the development of mobile data broadcast, it faces challenges as follows: (1) data content and the variation of scale； (2) user request real-time and diversified demand, cause the increase of hot spot data, directly increased severely broadcast data total amount；(3) Service quality and the raising of level.

Adaptive multi-channel broadcast scheduling is studied by the present invention, it is proposed that a kind of real-time on-demand data broadcast from Adaptive channel divides and distribution method OCSM (Optimized Channel Split Method).

The main research work of the present invention includes following aspects:

(1) (wherein R (Request) is request of data number, W to propose a kind of RxW/SL data item priority evaluation algorithm (Wait) for ask high latency, SL (System Lose) be that system will failure number).This algorithm considers data simultaneously Number of request, request of data high latency and system will invalidation request number these three index, by accurately evaluating single number According to the urgency level of item, provide clustering target for equilibrium clustering algorithm.

(2) consider the factors such as data item size, channel size and data item urgency level, propose broadcast data Xiang Jun Weighing apparatus clustering algorithm WASC (Weight Average and Size Cluster algorithm), for adaptive channel division side Method OCSM provides partitioning standards.

(3) algorithm CSA (Channel Split Algorithm) is proposed, its result produced according to WASC algorithm, will letter Road is divided into many sub-channels, and scheduling data request sequence is broadcasted to corresponding subchannel simultaneously.

Asking for an interview Fig. 1, the real-time on-demand data broadcast scheduling adaptive channel of one that the present invention provides divides and distribution system, Including a server and multiple user.One request queue RQ of server maintenance (Rquest Queue), a preparation team Row PQ (Pending Queue) and multiple broadcast queue BQ_i0≤i≤N(Broadcast Queue).RQ is responsible for collecting use Family is asked, and server obtains data genaration PQ according to RQ from data base.Server periodically uses OCSM algorithm to carry from PQ The tissue that fetches data is broadcasted.When user needs data, upload a request to server, this paper vacation by up channel If each user the most only produces a request, and does not have relatedness before and after same user between the data of Twice requests. After upload request, user's monitors downlink channel obtains data, and data item size is the most unique.System comprises a up letter Road, several down channels, upstream channel bandwidth is much smaller than down channel, assumes that user can intercept multiple descending letter simultaneously herein Road.Concrete, data broadcast scheduling model based on OCSM is as shown in Figure 1.System is made up of four parts:

(1) (request receiving mobile client) is received；

(2) (obtaining request data from data base) is obtained；

(3) (data prepared in queue are divided into multiple classification) is divided；

(4) (channel is split into multiple channel) is split.

Data broadcast scheduling flow process based on OCSM method is as shown in table 1:

Table 1 data based on OCSM broadcast scheduling flow process

Data item priority and data item size are the most important features of data item, excavate the feature of broadcast data item, look for It is the key that adaptive channel divides to most suitable broadcast channel.Ask for an interview Fig. 2, the real-time on-demand data of one that the present invention provides Broadcast scheduling adaptive channel divides and distribution method OCSM (Optimized Channel Split Method), uses data The method mining data item feature clustered in excavation, proposes data item equilibrium clustering algorithm WASC, and proposes channel division and divide Join algorithm CSA, according to cluster result, find most suitable channel number and corresponding size.This chapter is discussed in detail OCSM, such as Fig. 2 Shown in, OCSM comprises RxW/SL algorithm, WASC algorithm and CSA algorithm.RxW/SL algorithm determines each data item d in PQ_i's Weighted value, provides clustering target for next step；WASC algorithm according toAnd data item sizePQ is divided into n Classification FG={g₁,g₂,…g_n}.Wherein g_iRepresent i-th packet；Channel C is divided into the corresponding of correspondence according to FG by CSA algorithm N sets of sub-channels SC={c₁,c₂,...,c_i,...c_n, and by g_iThe data item of packet is placed in subchannel c_iBroadcast.

The algorithm RxW/SL of the present embodiment described in detail below；

Data item priority determines that the key factor of data item broadcast order, is also simultaneously feature comprehensive of data item Index.Represent data item d_iPriority,The biggest, priority is the highest.Request crash rate LR (Request Lost Rate, LR), average latency AAT (Average Access Time, AAT) be weigh real-time data broadcast system service matter The index of amount.In order to improve the service quality of system, the strategy generally used is to determine data item priority by algorithm, chooses The data item that priority is the highest is broadcasted.In the case of not considering data item size, SIN-α algorithm and L × R × W calculate Method is proposed effective method to weigh the priority of data item, and wherein L × R × W algorithm considers data item request i.e. simultaneously By actual effect number, data item number of request with the longest wait until the time, can comprehensively weigh the priority of data item.And L × R × W algorithm And it is not suitable for the unfixed situation of data item size, how in the case of data item size is unfixed, accurately to weigh data item Priority is the problem that this section is endeavoured to solve.

In L × R × W algorithm, data item size is fixed, and in broadcast queue, front choosing of i-1 data item does not affects i-th Individual data item d_iBroadcast start time t_i, L represents t_iLost efficacy before about d_iNumber of request.Do not fix in data item size Situation, it is impossible to determine i-1 data item size before broadcast queue, therefore t cannot be determined_iOccurrence, therefore L × R × W loses Effect.Because cannot ensure that the queue that the single optimal data item calculated according to L × R × W algorithm forms is that system optimal broadcasts team Row.In L × R × W algorithm, L is it is considered that d_iAt t_iThe produced failure number of moment broadcast, if L is the biggest, represents and pushes away over time After, d_iProduced failure number is the most, therefore d_iPriority the highest.For overcoming L × R × W algorithm not fix feelings in data item size Inadaptability under condition, it is proposed that RxW/SL algorithm, its use system will replace L by failure number SL (System Lose), its meter Calculation method is for assuming subsequent time broadcast data item d_i, draw broadcast-end-time, thus calculate broadcast d_iTo cause system its The failure number that his data item request is totalIt is appreciated that intuitivelyThe least d_iPriority is the highest.Consider data item simultaneously Number of request, two factors of high latency.Calculated by formula (1)Value weighs the urgency of data item, determines with this The priority of data item.(1) in formulaRepresent data item d_iNumber of request,Represent data item d_iThe high latency of request,For broadcast d_iBy causing system, other ask total failure number.

${\mathrm{Wf}}_{d_i}=\frac{R_{d_i}\×W_{d_i}}{{\mathrm{SL}}_{d_i}}---\left(1\right)$

It is worth the biggest, data item priorityThe highest.In (1),May be 0, now just with now just usingWeigh d_iPriority, therefore formula (1) is revised as (2).

${\mathrm{Wf}}_{d_i}=\left\{\begin{array}{ccc}\frac{R_{d_i}\×W_{d_i}}{{\mathrm{SL}}_{d_i}}& if& {\mathrm{SL}}_{d_i}\≠0\\ R_{d_i}\×W_{d_i}& if& {\mathrm{SL}}_{d_i}=0\end{array}\right.---\left(2\right)$

The algorithm WASC of the present embodiment described in detail below；

In on-demand data broadcast scheduling algorithm, needing three principal elements considered is data item sizeData The weighted value of itemAnd channel size Bw.Relation between lower surface analysis three above factor.For the ease of analyzing, make Following symbol definition: channel C is divided into n sub-channels, SC={c₁,c₂,...,c_i,...c_n, subchannel c_iSize be I.e.Broadcast cycle K subchannel c_iBroadcast be worth be defined asShown in solution formula such as formula (3):

$\begin{array}{c}{V}_{c_i,k}={\mathrm{Wf}}_{d_1,c_i,k}+{\mathrm{Wf}}_{d_2,c_i,k}+\mathrm{...}+{\mathrm{Wf}}_{d_n,c_i,k}\\ =\underset{i=1}{\overset{i=n}{\Σ}}{\mathrm{Wf}}_{d_i,c_i,k}\end{array}---\left(3\right)$

WhereinC is believed for broadcast cycle K_iThe weighted value of middle i-th data item.Therefore the broadcast value that channel C is total SumV_kSolve such as formula (4).Think that the interior broadcast of a cycle is worth SumV intuitively_kThe highest, the efficiency of broadcast system is the highest.

$\begin{array}{c}{\mathrm{SumV}}_k=V_{c_1,k}+V_{c_2,k}+\mathrm{...}+V_{c_N,k}\\ =\underset{j=1}{\overset{j=N}{\Σ}}{V}_{c_j,k}=\underset{j=1}{\overset{j=N}{\Σ}}\underset{i=1}{\overset{i=n}{\Σ}}{W}_{d_i,c_j,k}\end{array}---\left(4\right)$

(1) size of data: the analytical data size impact on broadcast scheduling.Assuming that divide the channel into two sub-channels Bw₁=3MB, Bw₂=1MB, data item d₁…d₅Size is 3MB, d₆…d₁₀Size is 1MB, and each data item weighted value is 1, extensively The cycle of broadcasting is 5S.According to the cluster broadcast of data item size as shown in Fig. 3 (a), (b).Combine formula (6) from Fig. 3 (a) can calculate Go out SumV_k=10, and if data item is not positioned in the channel corresponding with its size broadcast, bigger channel wave can be there is Taking, Fig. 3 (b) illustrates a kind of more extreme situation, and channel exists the bandwidth waste of 3M, simultaneously SumV_k=9, less than by data Item size cluster broadcast.

(2) data rights weight values: analytical data item weighted value is on the impact on scheduling.Assuming that divide the channel into two son letters Road Bw₁=2M, Bw₂=2M, requested data item d₁...d₈Size is 2M, wherein d₁...d₄Weighted value d₅...d₈Weighted valueBroadcast cycle is 3S.Broadcast as shown in Fig. 3 (c), (d) according to data item weighted value equalization. As shown in Fig. 3 (c), weighted value be 5 data item be distributed in two channels broadcast.After broadcast cycle terminates, data item d₇,d₈ Fail to broadcast in time, total channel broadcast can be calculated and be worth SumV_k=22.If not by dispersed for data item big for weighted value In each broadcast queue, it would be possible to the situation as shown in Fig. 3 (d) occurs.Data item that weighted value is big crowded with in a channel. Total channel broadcast is caused to be worth as SumV_k=18, leverage broadcasting efficiency.

Being found by analysis above, flock together broadcast by the data item that data item size is close, it is possible to fully Utilize channel width, reduce the waste of periodic intervals bandwidth.By dispersed to each channels broadcast for data item big for weighted value, It is beneficial to improve the total value that channel is broadcasted in the unit interval.

Analyzed by above, drawTwo clustering target.By collection of data items according toCluster, has It is beneficial to making full use of of bandwidth；By collection of data items according toDispersion broadcast, is conducive to improving channels broadcast in the unit interval It is worth.Propose WASC algorithm, collection of data items is pressedCluster, then foundationThe subclass G that will have clustered It is dispersed into FG.To specifically introduce the realization of equilibrium clustering algorithm below.WASC algorithm is broadly divided into three steps:

(1)Cluster: use the KODAMA algorithm improved, data are clustered by size.() initializes N number of data Item is divided into N class, and define arrays T={t₁,t₂,...,t_i,...,t_NAs grader, wherein t_iFor d_iClass label.Use KNN Algorithm carries out 10 times of cross validations to sample, draws the label Z that presorts_T={ z₁,z₂,...,z_i,...,z_N, wherein z_iFor d_i's Presort label.Calculate accurate angle value A of classifying_T.() hands over T and Z at random_TMiddle mistake classification samples label, generates new contingency table Sign array V={v₁,v₂,...,v_i,...,v_N}.Use KNN algorithm to carry out 10 times of cross validations V, draw the label Z that presorts_V ={ z₁,z₂,...,z_i,...,z_N, calculate accurate angle value A of classifying_VIf, A_V≥A_T, replace T with V, by Z_VReplace Z_T, A_TEqual to A_V。 Circulation performs () until A_TEqual to 1 or reach maximum cycle.

(2) tag along sort adjusts: the class label of each class discrete ordered series of numbers in the T that step (1) produces, for side The division in notelet road, needs (2) to be adjusted class label in T.Use tag along sort adjustable strategies TRS (Tag Rank Strategy), collection of data items D is divided into categorieN subclass G={g₁,g₂,...g_i,...g_categorieN}.Specifically Adjustable strategies be shown in Table 2.

Table 2 TRS strategy

(3)Equalization processing: be distributed to data item weighted value uniformly in each subchannel broadcast.Fori=0 To catagorieN, will generate g_i,1,g_i,2Two subclass, ifBy g_iMiddle data item d_jAdd subclass to g_i,1In, otherwise add g to_i,2In.Pass throughEqualization processing, generate final classification subclass FG.

To sum up three step, WASC algorithm pseudo code is as shown in table 3.

Table 3 WASC algorithm

The algorithm CSA of the present embodiment described in detail below；

Channel is divided by algorithm CSA according to classification subclass FG, and then distributes data item to corresponding subchannel Broadcast.It mainly comprises channel and divides and data item two parts of distribution.

(1) channel divides: broadcast channel bandwidth is determined by its data item feature to be broadcast.Broadcast channel and broadcast data item Adapt, efficiency of algorithm can be improved further.By equilibrium cluster, the data characteristics in data item to be broadcast is excavated. Data item size clusters, and decreases the waste of channel width, and data item is disperseed according to weighted value, improves channel cycle broadcast valency Value.Broadcast channel is divided according to its data adaptive to be broadcast, divides subchannel situation and determined by cluster result.

(2) channel distribution: will data item be assigned to each sub-channels in broadcast.According to each data item to be broadcast Tag along sort, distributed to corresponding subchannel form sub-broadcast queue.Then it is defined as weighted value by son according to algorithm Broadcast queue is ranked up, and the data item that weighted value is big has precedence over the data item broadcast that weighted value is little.

Concrete CSA algorithm is as shown in table 4:

Table 4 algorithm CSA pseudo-code

The real-time on-demand data broadcast scheduling adaptive channel that the present invention provides divides and distribution method OCSM (Optimized Channel Split Method) combines broadcast channel number, size and real-time broadcast environment, and data please Ask feature difference real-time adaptive to adjust channel number and size, thus improve system sensitivity, robustness and broadcasting efficiency.

It should be appreciated that the part that this specification does not elaborates belongs to prior art.

It should be appreciated that the above-mentioned description for preferred embodiment is more detailed, can not therefore be considered this The restriction of invention patent protection scope, those of ordinary skill in the art, under the enlightenment of the present invention, is weighing without departing from the present invention Profit requires under the ambit protected, it is also possible to make replacement or deformation, within each falling within protection scope of the present invention, this The bright scope that is claimed should be as the criterion with claims.

Claims (7)

1. a real-time on-demand data broadcast scheduling adaptive channel divides and distribution system, it is characterised in that: by a service Device and multiple user composition, comprise a up channel and several down channels, and upstream channel bandwidth is less than down channel； When user needs data, upload a request to server by up channel；After upload request, user intercepts descending Channel obtains data；Described server is for safeguarding a request queue RQ, a preparation queue PQ and multiple broadcast queue BQ_i0≤i≤N；Described request queue RQ is responsible for collecting user's request, and server obtains from data base according to request queue RQ Data genaration prepares queue PQ；Described server is periodically from preparing queue PQ extraction data tissue broadcast.

2. a real-time on-demand data broadcast scheduling adaptive channel divides and distribution method, it is characterised in that include following step Rapid:

Step 1: initialize；

N number of data item is divided into N class, and define arrays T={t₁, t₂..., t_i..., t_NAs grader, wherein t_iFor data Item d_iClass label；

Step 2: user sends request Req by up channel；

Step 3: judge currently whether have request Req_i；

The most then perform following step 4；

If it is not, server continues to monitor up channel, and turn round the described step 3 of execution；

Step 4: take out Req_i, for data item d of this request_iPerform operations described below；

If d_iIt is present in broadcast queue BQ_iIn, then by this Req_iAdd this d to_iRequest list in；

If d_iIt is present in preparation queue PQ, then by this Req_iAdd this d to_iRequest list in；

Otherwise, by this Req_iJoin in request queue RQ；

Step 5: judge upper broadcast cycle K_i-1Whether terminate；

The most then perform following step 6；

If it is not, then revolution performs above-mentioned steps 2；

Step 6: use the broadcast of OCSM algorithm organization；

Step 7: broadcasted a d_i, then by this d_iSuccessful request information write file, delete all requests of this broadcast item, Deleting this broadcast item, advance a time point；

Step 8: judge whether to exceed running time T；

The most then perform following step 9；

If it is not, then revolution performs above-mentioned steps 2；

Step 9: read successful request information, adds up solicited message.

Real-time on-demand data broadcast scheduling adaptive channel the most according to claim 2 divides and distribution method, its feature Being, using the broadcast of OCSM algorithm organization described in step 6, it implements process and includes following sub-step:

Step 6.1: utilize data mining clustering algorithm RxW/SL, determines each data item d in preparation queue PQ_iWeighted value, Clustering target is provided for next step；

Step 6.2: utilize data item to equalize clustering algorithm WASC, according to data item d_iPriority and data item size Preparation queue PQ is divided into n classification FG={g₁,g₂,…g_n, wherein g_iRepresent i-th packet；

Step 6.3: utilize channel to divide and allocation algorithm CSA, according to FG, channel C is divided into the corresponding n subchannel of correspondence Set SC={c₁, c₂..., c_i... c_n, and by g_iThe data item of packet is placed in subchannel c_iBroadcast.

Real-time on-demand data broadcast scheduling adaptive channel the most according to claim 3 divides and distribution method, its feature Being, the process that implements of step 6.1 is:

Assume subsequent time broadcast data item d_i, draw broadcast-end-time, thus calculate broadcast d_iOther numbers of system will be caused According to the failure number that item request is total The least d_iPriority is the highest；Consider data item number of request, high latency simultaneously Two factors, are calculated by formula (1)Value weighs the urgency of data item, determines the priority of data item with this；

${\mathrm{Wf}}_{d_i}=\frac{R_{d_i}\×W_{d_i}}{{\mathrm{SL}}_{d_i}}---\left(1\right);$

Wherein,Represent data item d_iNumber of request,Represent data item d_iThe high latency of request,For broadcast d_i By causing system, other ask total failure number；It is worth the biggest, data item priorityThe highest；

In formula (1), ifIt is 0, then usesWeigh d_iPriority；Therefore formula (1) is revised as (2)；

${\mathrm{Wf}}_{d_i}=\left\{\begin{array}{cc}\frac{R_{d_i}\×W_{d_i}}{{\mathrm{SL}}_{d_i}}& if{\mathrm{SL}}_{d_i}\≠0\\ R_{d_i}\×W_{d_i}& if{\mathrm{SL}}_{d_i}=0\end{array}\right.---\left(2\right).$

Real-time on-demand data broadcast scheduling adaptive channel the most according to claim 3 divides and distribution method, its feature Being, implementing of step 6.2 includes following sub-step:

Step 6.2.1: data item size in queue PQ will be preparedIt is stored in vector S；

Step 6.2.2: N number of data item sample is divided into N class, data item d_iCorresponding i-th class, class is designated as i, is stored in vector T；

Step 6.2.3: sample carries out 10 times of cross validations, sample is divided into 10 points, carries out 10 rounds to every a sample KNN algorithm classification obtains 9 labels of initially presorting, and uses ballot method to determine the label z that finally presorts of this sample_i；

Step 6.2.4: the label of presorting of all samples is collected generation prediction classification samples Z_T；

Step 6.2.5: utilize formulaCalculate accurate angle value A of sample classification_T；

Step 6.2.6: the sample of a part of classification error of random choose, by its tag along sort t_iTag along sort z is predicted with it_iHand over Change, generate new tag along sort V；

Step 6.2.7: V is replaced T, obtains A_V；

Step 6.2.8: if A_V≥A_T, then T=V, Z_T=Z_V, A_T=A_V, R=R-1；

Step 6.2.8: judge；

If A_TEqual to 1 or reach maximum cycle, then perform following step 6.2.9；

Otherwise, revolution performs above-mentioned steps 6.2.3；

Step 6.2.9: adjust tag along sort numerical value in tag along sort T so that it is by { 1,2,3...} is incremented by, by collection of data items D It is divided into categorieN subclass G={g₁,g₂,...g_i,...g_categorieN}；

Step 6.2.10: initialize empty collection of data items

FG={g_1,1,g_1,2,…g_i,1,g_i,2,…g_categorieN,1,g_categorieN,2}；

Step 6.2.11: from the beginning of i=1, is sequentially executable following step 6.2.12-step 6.2.13, until i= categorieN；

Step 6.2.12: initialize

Step 6.2.13, for set g_iIn each data item d_j, it is judged thatWhether set up；

The most then by data item d_iAdd g to_i,1In set；WhereinFor data item d_j Size；

If it is not, then by data item d_iAdd g to_i,2In set；

Step 6.2.14: output FG.

Real-time on-demand data broadcast scheduling adaptive channel the most according to claim 5 divides and distribution method, its feature Being, implementing of step 6.2.9 includes following sub-step:

(1) definition comprises array Tag{-1 of N number of element, and-1 ... ,-1}, categorieN=0；

(2) from the beginning of i=1, it is sequentially executable following (3)-(6), until i=N；

(3) if Tag [i]=-1, Tag [i] represents that in Tag array, i-th element, i.e. tag along sort are not adjusted, then n= t_i；categorieN++；t_i=categorieN；

(4) from the beginning of j=i+1, it is sequentially executable following (5), until j=N；

(5) if t_j=n&&Tag [j]=-1, then t_i=categorieN, Tag [j]=0；

(6) Tag [i]=0；

(7) T class mark is assigned to G, exports G, categorieN.

Real-time on-demand data broadcast scheduling adaptive channel the most according to claim 3 divides and distribution method, its feature Being, implementing of step 6.3 includes following sub-step:

Step 6.3.1: initialize array Sumdata [categorieN]；Wherein collection of data items D is divided into categorieN son Set G={g₁,g₂,...g_i,...g_categorieN}；

Step 6.3.2: from the beginning of i=1, is sequentially executable following step 6.3.3, until i=N；

Step 6.3.3: scan collection of data items D to be broadcast, if its class label t_i=j, then

$\mathrm{Sumdata}\left[j\right]=\mathrm{Sumdata}\left[j\right]+{\mathrm{Size}}_{d_i};$

Step 6.3.4: initialize categorieN sub-broadcast channel；

Step 6.3.5: from the beginning of i=1, is sequentially executable following step 6.3.6, until i=categorieN；

Step 6.3.6: ask for subchannel c_iAmount of bandwidth

Step 6.3.7: output subchannel { c₁, c₂..., c_i,...,c_categorieN}；

Step 6.3.8: initialize categorieN Ge Zi broadcast queue；

Step 6.3.9: from the beginning of i=1, is sequentially executable following step 6.3.10, until i=N；

Step 6.3.10: scan data item in collection of data items D to be broadcast, if d_iTag along sort be t_i, then by data item d_iAdd Enter to sub-broadcast queueIn；

Step 6.3.11: from the beginning of i=1, is sequentially executable following step 6.3.12, until i=categorieN；

Step 6.3.12: by sub-broadcast queue bd_iBy data item weighted valueSequence；

Step 6.3.13: export sub-broadcast queue { bd₁, bd₂..., bd_i..., bd_categorieN}。