CN105704335A - Predictive form calling-out algorithm based on dynamic statistics process, switch dialing method and device - Google Patents

Abstract

The invention relates to a predictive form calling-out algorithm based on a dynamic statistics process, a switch dialing method and device. The device includes a data input module, a predictive form calling-out algorithm module, a switch, a dynamic statistics module, a data storage module, and a data output display module. The data input module inputs initialization parameter data to the predictive form calling-out algorithm module. The predictive form calling-out algorithm module is connected to the switch for information interaction. The dynamic statistics module is connected to the data storage module and receives the data of the data storage module. The dynamic statistics module counts the statistics data with different frequencies and transmits the data to the predictive form calling-out algorithm module. At the same time the data display is performed through a data output interface. According the invention, statistics and prediction are carried out according to dialed historical data, the dialing number next moment is dynamically determined, the service utilization rate is improved when a target phone call loss rate is not exceeded as far as possible, and optimizing balance is reached and the artificial dialing is completely replaced.

Description

Based on the predictive outbound algorithm of dynamic statistics process, switch dialing method and device

Technical field

The present invention relates to telephone outbound call technical field, be specifically related to a kind of based on the predictive outbound algorithm of dynamic statistics process, switch dialing method and device。Make use of statistics, data mining, queueing theory, Markov chain, the raw technology such as opinion and Algorithmic Design & Analysis of going out, the technology that especially real time resources allocation schedule and feedback control optimize。

Background technology

Call center can be divided into incoming call type and call-out type two kinds。Incoming call type call center provides the service of passive type, and to provide be actively a kind of mode of customer service in call-out type call center, and carrying out the business such as the marketing, questionnaire survey for enterprise provides a kind of efficient mode。Development along with computer and communication technology, call center system based on conventional art is difficult to meet turn of the market demand, automatic outer call is given birth to therefrom, automatic outer call system actively initiates calling, will be unable to the clients filter connected, the extremely idle customer service of client's distribution that will be successfully connected, thus improving the utilization rate of contact staff。

Existing outgoing call algorithm mainly has three classes: experience outgoing call prediction algorithm, the application in prediction algorithm of integral model algorithm and queuing theory。

Experience outgoing call prediction algorithm is by feedback rates value is dynamically regulated by outer group of recent result, and adjustment outgoing call adjustment parameter (during outer group of hypervelocity>1, when dialling outside deceleration<1), make the adaptive capacity of algorithm become by force。Shortcoming is duration distribution instability of on average leaving unused, and call loss can not be controlled preferably。

First integral model algorithm examines or check the timing distribution situation that continues of calling: percent of call completed distribution V (t), and rate of taking out stitches distribution F (t) then hasExamine or check the time series of the duration of call again: duration of call distribution h (t), meetThe then duration of call probability more than time t, H (O)=l, H (+∞)=0。Assume system exhalation sequence C (t), then at the access calling number of t, the current talking calls being somebody's turn to do when t, the calls that expression is currently talking, including oneself through own through the unallocated calling to operator in wait of off-hook with the calling and user of operator's call。It is g to seek the instant percent of call completed after breathing out user's off-hook, namely the available free proportion of calls attended a banquet after exhalation user's off-hook, then just have N (t) * g<A, simultaneously for making full use of resource of attending a banquet, the number of attending a banquet having again N (t)>=A, A to be current registration。Namely can obtain from above: A≤N (t) < A/g。From then on constraints can calculate exhalation time sequence C (t) of system。According to this exhalation prediction algorithm can according to the result of history call and conventional call time sequence and current system resource situation, carry out self feed back study, dynamically adjust follow-up exhalation。Although as above algorithm is really feasible, and play a significant role in actual applications, but find by analysis to still have the place being worth improvement, being divided in such as cycle T that n criticizes is optimum selection, it is truth or ideal state that average handling time calculates L/2, it is unalterable or has fluctuation, fluctuate how many？Or the present situation can or can not affect the prediction at lower a moment extremely, stability how？

Queuing theory application in prediction algorithm: queuing theory (queuingtheory) is introduced when solving automatic telephone design problem by A.K. angstrom of erlang the earliest。Prediction algorithm is summarized as and to solve two relevant problems: 1, when the busy factor (utilization rate of attending a banquet) is attended a banquet in maximization, do not allow the percent of call lost more than an acceptable maximum ARmax。2, when reducing the percent of call lost as far as possible, do not allow and attend a banquet the busy factor less than an acceptable minimum level Umin。Algorithm is exactly meet two above conditions to produce the number N when this calling of cause。Some parameters that this algorithm can be collected or reception system is relevant: total number m of attending a banquet, throw phone arrival rate p, average T service time into, hit rate ρ。This algorithm assumes that exhalation stream obeys the Poisson distribution of parameter lambda；Service time is within the system for obeying the general distribution of parameter μ。In this prediction algorithm, the distribution of exhalation stream being decided to be Poisson distribution, namely the time between twice phone of adjacent dial-out just obeys exponential。This has very big discrepancy with the phenomenon under reality。It addition, obey general distribution service time, finding under study for action, the time of service is concentrated in two stages, and one is 3-10s, between another 140-160s, is not the general distribution of a standard, and the deviation of this result of calculation that will result in this model is bigger。Additionally, the state attended a banquet at present is not joined in consideration by this algorithm, this just will cause when suddenly all attend a banquet that all the line is busy, appoint and so call by former data calculating number, until T average time is retrieved calculating (taking long enough), one section of upheaval for a long time therefore will be produced。When a period of time, service time reduced suddenly, will also result in the upheaval that utilization rate of attending a banquet diminishes suddenly。

Being analyzed from above, existing predictive outbound algorithm is mostly coupled with dynamic statistics part, even without dynamic statistics part。And the most important determiner of predictive outbound algorithm is dynamic statistics part, existing outgoing call algorithm some be use theoretical formula method to obtain, it is desirable to setting up in advance a kind of mathematical model, parameter value afterwards is all by theoretical formula method gained；Although this method or need not seldom access data base, but it is more fixing to set up model, and for the business having, this model is no longer set up, and the adaptability of this model is poor in other words。Also some is realized existing outgoing call algorithm by dynamic statistics, but this method wants substantial amounts of access history data to carry out machine learning, training data or neutral net etc., this method accesses data base continually and needs substantial amounts of data, waste a lot of time, cause hydraulic performance decline, and do not have during mass data when initial and be difficult to reasonably predict。

Summary of the invention

In order to overcome the defect existed in above-mentioned prior art, it is an object of the invention to provide a kind of based on the predictive outbound algorithm of dynamic statistics process, switch dialing method and device。

To achieve these goals, according to the first aspect of the invention, the invention provides a kind of predictive outbound algorithm based on dynamic statistics process, it comprises the steps:

S1, starts, and arranges target and attends a banquet utilization rate AO, maximum percent of call lost AR_max, number of switches T, maximum waiting time TQ_max, it was predicted that interval time Δ t, waiting list greatest length qL, user is patient time 1/ η on average, and user is reluctant the Probability p adding waiting list₀, and outwards call the initial value of several λ, make the accumulated value al=0 of l。

S2, in t, obtains current customer service sum N by switch, and idle customer service sum f, just at called telephone number j, percent of call completed ρ, average jingle bell duration 1/ β, average call duration 1/ μ, and the numerical value of waiting list length q, described t is positive number；

S3, it may be judged whether meet q < qL, if meeting, goes to step S4, and otherwise dormancy to t+ Δ t goes to step S2 and carries out next round prediction, and to update current time be t=t+ Δ t；

S4, calculates required minimum several M that attends a banquet, and computing formula is:

S5, it may be judged whether meet N≤M, if meeting, going to step S17, otherwise going to step S6；

S6, computing system predicts maximum outgoing call value l, cumulative calculation al=al+l, wherein,

$l=\frac{\mathrm{\&Delta;}t*N*AO*\mathrm{\&mu;}}{\mathrm{\&rho;}};$

S7, the percent of call lost AR in dynamic statistics last moment also judge whether to meet AR < AR_maxIf meeting, going to step S8, otherwise go to step S16；

S8, it may be judged whether meet current idle and attend a banquet several f > al, if meeting, going to step S9, otherwise going to step S13；

S9, it was predicted that result s=(f-j)/ρ-q+al, if al > 1, al sets to 0；

S10, calculates period of reservation of number Tq, Tq=q/ (N* μ)；

S11, it may be judged whether meet Tq < TQ_maxIf meeting, going to step S12, otherwise dormancy to t+ Δ t goes to step S2 and carries out next round prediction, and to update current time be t=t+ Δ t；

S12, updates the exhalation that s=s*ARrate, ARrate are system that predicts the outcome and weakens ratio, ARrate=1-AR, and AR is the theoretical percent of call lost,

AR=p₀B(θ,N)+(1-p₀) λ β ρ η, go to step S18；

S13, it was predicted that result s=f-j-q+al, if al > 1, al sets to 0；

S14, calculates theoretical percent of call lost AR, AR=p₀B(θ,N)+(1-p₀) λ β ρ η,

B (θ, 0)=1,

$\begin{array}{cc}B(\mathrm{\&theta;},k)=\frac{\mathrm{\&theta;}B(\mathrm{\&theta;},k-1)}{k+\mathrm{\&theta;}B(\mathrm{\&theta;},k-1)}& 1\&le;k\&le;N\end{array};$

S15, it may be judged whether meet AR < AR_maxIf meeting, going to step S10, otherwise dormancy to t+ Δ t goes to step S2 and carries out next round prediction, and to update current time be t=t+ Δ t；

S16, it was predicted that result s=f-j-q, goes to step S18；

S17, it was predicted that result s=f-j, goes to step S18；

S18, calculates and finally calls number λ, λ=(int) min (max (s, 0), T)；

S19, outwards dials λ phone, and algorithm dormancy to t+ Δ t goes to step S2 and carries out next round prediction, and to update current time be t=t+ Δ t；Repeat above step, until application program terminates running。

To achieve these goals, according to the second aspect of the invention, the invention provides the switch dialing method of a kind of predictive outbound algorithm based on dynamic statistics process, it is characterised in that comprise the steps:

Step 1, predictive outbound algorithm predicts according to claim 1 goes out subsequent time will call quantity λ；

Step 2, switch predicts the outcome according to this and dials λ phone；

Step 3, after average ring time β, described β is positive number, has the closing of the circuit of ρ ratio, and the phone access failure of remaining 1-ρ ratio is filtered, and described ρ is the positive number more than 0 less than 1；

Step 4, the number metering-in control system of connection, control system can according to currently whether available free attending a banquet is allocated；

Step 5, attends a banquet if available free, and distribution is attended a banquet and serviced for it, services complete successful for dialing；If not having idle seating, the client having p0 probability is ready that adding waiting list waits, the client of remaining 1-p0 probability is unwilling to add waiting list and abandons forming call loss；

Step 6, also some bears with time η owing to waiting more than it and abandons being formed call loss the client queued up, only remaining meeting distributes, according to idle the attending a banquet serviced at present, the service of attending a banquet by control system for it when not above its patient time η, and finally service completes。

To achieve these goals, according to the third aspect of the present invention, the invention provides and a kind of run the present invention device based on the predictive outbound algorithm of dynamic statistics process, it includes data input module, predictive outbound algoritic module, switch, dynamic statistics module, data memory module and data output display module, described data input module is to pretest outgoing call algoritic module input initialization supplemental characteristic, described pretest outgoing call algoritic module is connected with switch, the information of carrying out is mutual, described dynamic statistics module is connected with data memory module, receive the data of data memory module, described dynamic statistics module is added up the statistical data of different frequency and is transferred to pretest outgoing call algoritic module, carry out data by data output interface to show simultaneously；

Described predictive outbound algoritic module comprises configuration parameter unit, initiation parameter unit, Theoretical Calculation data cell, cycle statistical parameter unit, predictive outbound algorithm unit, predict the outcome answer unit；Described configuration parameter unit is used for desampler configuration parameter information, described initiation parameter unit is for receiving the various object parameter data of data input module input, described Theoretical Calculation data cell is for receiving Theoretical Calculation but wanting the data of cyclomorphosis, described cycle statistical parameter unit is for receiving the various data that the different statistic frequency of dynamic statistics module comes out, the described predictive outbound algorithm unit predictive outbound algorithm for realizing described in claim 1, described in predict the outcome interface for predicting the outcome to switch transmission；

Described dynamic statistics module includes obtaining historical data unit, selects statistics policy unit, statistics policy unit, transmits the unit of statistical data to predictive outbound algoritic module and transmit the unit of statistical data to output interface, in described acquisition historical data unit, storage has and includes obtaining the data in written data storehouse and switch preserves services but also have not enough time to the data in write into Databasce, described selection adds up policy unit is used for selecting to use which kind of statistics strategy, and described statistics policy unit includes multiple concrete statistics strategy。

The present invention has following Advantageous Effects:

1, the present invention carries out adding up and predicting according to the historical data called, the dynamic telephone number determining should dial lower a moment, improves, while as far as possible less than the target phone percent of call lost, utilization rate of attending a banquet, and reaches optimize balance and replace manual dial completely。

2, the outgoing call algorithm of the present invention sets up queuing theory model when attending a banquet not enough, ring time, service time and statistics strategy then make the statistical model of dial data by hand, the model set up is closer to practical situation, and utilization rate of attending a banquet under the same percent of call lost can be greatly improved。The outgoing call algorithm algorithm of the present invention is based on queuing theory, but initial parameter (such as service time, percent of call completed etc.) is very big for different business difference, so these parameters can be well adapted for miscellaneous service by dynamic statistics。Utilization rate of attending a banquet when number of attending a banquet is 10 has 65-70%, percent of call lost 3-4%；Utilization rate of attending a banquet when number of attending a banquet is 20 has 70-73%, the percent of call lost about 2%；Utilization rate of attending a banquet when number of attending a banquet is 30 has 72-75%, percent of call lost 1-2%；Utilization rate of attending a banquet when number of attending a banquet is 40 has 80-85%, the percent of call lost about 1%；Utilization rate of attending a banquet when number of attending a banquet is 50 has 82-85%, and the percent of call lost is less than 1%；And along with several increase of attending a banquet, more and more steadily, the percent of call lost just started with regard to only small even without, fluctuate also only small；When number of attending a banquet be 60 or above time occur that call loss maintenance are attended a banquet utilization rate 80% hardly。And have good effect for different business datums。

3, the present invention opens a kind of predictive outbound algorithm and is divided into two important modules: predictive outbound algoritic module and dynamic statistics module；Such that it is able to optimize from both direction, thus avoiding the bottleneck problem that an optimized algorithm there will be。And need not be concerned about that when optimized algorithm the concrete of parameter is added up, the realization of algorithm need not be considered during statistical parameter and whether that algorithm is useful。

4, add dynamic statistics module, improve the adaptability of algorithm so that it is adapt to multiple different business and the too big situation of data fluctuations；Adopt the thinking that combines with theoretical model of statistical model, it is to avoid theoretical model deviates bigger defect in extreme circumstances, make the theoretical model can be better close to statistical model。

5, the thinking adopting feedback to combine with prediction, is enable those to predict and becomes more accurate accurately, it was predicted that can be become accurately by feedback less accurately, thus the advantage combining prediction and feedback。

The additional aspect of the present invention and advantage will part provide in the following description, and part will become apparent from the description below, or is recognized by the practice of the present invention。

Accompanying drawing explanation

Above-mentioned and/or the additional aspect of the present invention and advantage are from conjunction with will be apparent from easy to understand the accompanying drawings below description to embodiment, wherein:

Fig. 1 is predictive outbound algorithm flow chart in one preferred implementation of the present invention；

Fig. 2 is the schematic flow sheet of switch dialing method in a kind of preferred implementation of the present invention；

Fig. 3 is the composition structural representation of the device of the predictive outbound algorithm in a kind of preferred implementation of invention based on dynamic statistics。

Detailed description of the invention

Being described below in detail embodiments of the invention, the example of described embodiment is shown in the drawings, and wherein same or similar label represents same or similar element or has the element of same or like function from start to finish。The embodiment described below with reference to accompanying drawing is illustrative of, and is only used for explaining the present invention, and is not considered as limiting the invention。

In describing the invention, unless otherwise prescribed and limit, it should be noted that term " installation ", " being connected ", " connection " should be interpreted broadly, for instance, can be mechanically connected or electrical connection, can also be the connection of two element internals, it is possible to be joined directly together, it is also possible to be indirectly connected to by intermediary, for the ordinary skill in the art, it is possible to understand the concrete meaning of above-mentioned term as the case may be。

Meaning of parameters in the present invention is explained as follows:

Ring duration: dial a phone, successfully connects or identifies be not turned on required time from starting to dial into a phone。

Service duration: the service time of the phone after successfully connecting。

Arrange duration: client is done the business record time (part that also can regard service time as)。

Waiting list length: connect phone when can not get servicing, is ready to wait online, this queue length just waits for queue length。

User's patience time: user is reluctant to wait the time actively abandoning hanging up the telephone to it Zi adding waiting list。

Period of reservation of number: user's actual time waited in queuing process。

Maximum waiting time: the client of all online waits be ready wait the maximum time upper limit。

Customer service utilization rate: the ratio of total duration that the service duration summation of all customer services is online with all customer services。

The phone percent of call lost: connect the telephone number still not obtaining service and account for the ratio connecting total call。

The present invention is known after carrying out data statistic analysis according to actual artificial outgoing call dial data: is similar to service time and obeys quantum condition entropy, and ring time is approximate obeys generalized Poisson distribution。

The invention provides a kind of predictive outbound algorithm based on dynamic statistics process, as it is shown in figure 1, it comprises the steps:

S1, starts, arranges parameter value, including configuration parameter and the initiation parameter of switch, for instance, it is specially target and attends a banquet utilization rate AO, maximum percent of call lost AR_max, number of switches T, maximum waiting time TQ_max, it was predicted that interval time Δ t, waiting list greatest length qL, user is patient time 1/ η on average, and user is reluctant the Probability p adding waiting list₀, and outwards call the initial value of several λ。

In the present embodiment, workload, maximum percent of call lost AR are set_max=0.02, number of switches T=10000, target is attended a banquet utilization rate AO=0.8, maximum waiting time TQ_max=15s, it was predicted that interval time Δ t=5s, waiting list greatest length (capacity of queue) qL=2, user is patient time 1/ η=10s on average, and user is reluctant the Probability p adding waiting list₀The target component such as=0.01。The numerical value of equal sign back is the default value of parameter, makes the accumulated value al=0 of l。

S2, in t, obtains current customer service sum N by switch, and idle customer service sum f, just at called telephone number j, percent of call completed ρ, average jingle bell duration 1/ β, average call duration 1/ μ, and the numerical value of waiting list length q, described t is positive number；

S3, it may be judged whether meet q < qL, if meeting, goes to step S4, and otherwise dormancy to t+ Δ t goes to step S2 and carries out next round prediction, and to update current time be t=t+ Δ t；

S4, calculates required minimum several M that attends a banquet, and computing formula is:

S5, it may be judged whether meet N≤M, if meeting, going to step S17, otherwise going to step S6；

S6, computing system predicts maximum outgoing call value l, cumulative calculation al=al+l, wherein,

$l=\frac{\mathrm{\&Delta;}t*N*AO*\mathrm{\&mu;}}{\mathrm{\&rho;}};$

S7, statistics policy module utilizes the percent of call lost AR in the statistical method dynamic statistics last moment prestored and judges whether to meet AR < AR_maxIf meeting, going to step S8, otherwise go to step S16；

S8, it may be judged whether meet current idle and attend a banquet several f > al, if meeting, going to step S9, otherwise going to step S13；

S9, it was predicted that result s=(f-j)/ρ-q+al, if al > 1, al sets to 0；

S10, calculates period of reservation of number Tq, Tq=q/ (N* μ), and wherein q is the current number queued up, if q is 0, represents currently without waiting client；

S11, it may be judged whether meet Tq < TQ_maxIf meeting, going to step S12, otherwise dormancy to t+ Δ t goes to step S2 and carries out next round prediction, and to update current time be t=t+ Δ t；

S12, updates the exhalation that s=s*ARrate, ARrate are system that predicts the outcome and weakens ratio, ARrate=1-AR, and AR is the theoretical percent of call lost, AR=p₀B(θ,N)+(1-p₀) λ β ρ η, go to step S18；

S13, it was predicted that result s=f-j-q+al, if al > 1, al sets to 0；

S14, calculates theoretical percent of call lost AR, AR=p₀B(θ,N)+(1-p₀) λ β ρ η, phone call loss can be divided into two parts that are that be reluctant entrance waiting list call loss and that be ready to enter waiting list call loss, and first half only occurs when i=N, it is assumed that θ=λ β ρ (1-p₀)/μ is successful connection phone input load, according to Erlang-B formula, it is possible to calculate probability π when N number of customer service does not have a free time_N, i.e. the percent of call lost, this probability can draw for crossing Erlang formula iteration:

B (θ, 0)=1,

$\begin{array}{cc}B(\mathrm{\&theta;},k)=\frac{\mathrm{\&theta;}B(\mathrm{\&theta;},k-1)}{k+\mathrm{\&theta;}B(\mathrm{\&theta;},k-1)}& 1\&le;k\&le;N\end{array};$

S15, it may be judged whether meet AR < AR_maxIf meeting, going to step S10, otherwise dormancy to t+ Δ t goes to step S2 and carries out next round prediction, and to update current time be t=t+ Δ t；

S16, it was predicted that result s=f-j-q, goes to step S18；

S17, it was predicted that result s=f-j, goes to step S18；

S18, calculates and finally calls number λ, λ=(int) min (max (s, 0), T)；

S19, outwards dials λ phone, and algorithm dormancy to t+ Δ t goes to step S2 and carries out next round prediction, and to update current time be t=t+ Δ t；Repeat above step, until application program terminates running。

In the present embodiment, in step S7, the percent of call lost AR in dynamic statistics last moment is all of phone number that the phone number of t1 moment before current time to current time stage call loss was connected divided by this stage, and described t1 is positive number。In the embodiment more preferably of the present invention, t1 is the time between 15 minutes to 45 minutes, it is preferred to 20 minutes。

If the interval chosen is too little, it is possible to the data in the time period very little, are not enough to reflect real situation。If the interval chosen is excessive, it is impossible to the change of real-time response parameter, and when data fluctuations is more apparent, also it is not enough to reflect real fluctuation situation。

The present invention is relative to traditional predictive outbound algorithm, have the advantage that and set up queuing theory model when attending a banquet not enough by raw opinion of going out with queueing theory, ring time, service time and statistics strategy then make the statistical model of dial data by hand, the model set up is closer to practical situation, and utilization rate of attending a banquet under the same percent of call lost can be greatly improved；The present invention opens a kind of predictive outbound algorithm and is divided into two important modules: predictive outbound algoritic module and dynamic statistics module。Such that it is able to optimize from both direction, both direction has a lot of prioritization scheme, thus avoiding the bottleneck problem that an optimized algorithm there will be；And need not be concerned about that when optimized algorithm the concrete of parameter is added up, the realization of algorithm need not be considered during statistical parameter and whether that algorithm is useful；Add dynamic statistics module, improve the adaptability of algorithm so that it is adapt to multiple different business and the too big situation of data fluctuations；Adopt the thinking that combines with theoretical model of statistical model, it is to avoid theoretical model deviates bigger defect in extreme circumstances, make the theoretical model can be better close to statistical model；The thinking adopting feedback to combine with prediction, is enable those to predict and becomes more accurate accurately, it was predicted that can be become accurately by feedback less accurately, thus the advantage combining prediction and feedback。

Automatic outer call system is a very complicated calling system, there are certain requirements in situation particularly in customer service utilization rate and the phone percent of call lost。So a predictive outbound algorithm not only to consider that above-mentioned affecting parameters it is also contemplated that running state parameter。Therefore the present invention proposes the prediction outgoing call algorithm of the prediction of a kind of binding isotherm modeling and system real-time running state, so that the customer service utilization rate of outer paging system and the phone percent of call lost reach certain requirement。

As in figure 2 it is shown, utilize the predictive outbound algorithm based on dynamic statistics process of the present invention, switch dialing method comprises the steps:

Step 1, goes out subsequent time according to the predictive outbound algorithm predicts of the present invention and will call quantity λ；

Step 2, switch predicts the outcome according to this and dials λ phone；

Step 3, after average ring time β, the only closing of the circuit of ρ ratio, the phone access failure of remaining 1-ρ ratio is filtered；

Step 4, the number metering-in control system of connection, control system can according to currently whether available free attending a banquet is allocated；

Step 5, attends a banquet if available free, and distribution is attended a banquet and serviced for it, services complete successful for dialing；If not having idle seating, the client having p0 probability is ready that adding waiting list waits, the client of remaining 1-p0 probability is unwilling to add waiting list and abandons forming call loss；

Step 6, also some bears with time η owing to waiting more than it and abandons being formed call loss the client queued up, only remaining meeting distributes, according to idle the attending a banquet serviced at present, the service of attending a banquet by control system for it when not above its patient time η, and finally service completes。

As shown in Figure 3, the present invention runs the device of the predictive outbound algorithm based on dynamic statistics process, it includes data input module, predictive outbound algoritic module, switch, dynamic statistics module, data memory module and data output display module, data input module is to pretest outgoing call algoritic module input initialization supplemental characteristic, pretest outgoing call algoritic module is connected with switch, the information of carrying out is mutual, dynamic statistics module is connected with data memory module, receive the data of data memory module, dynamic statistics module is added up the statistical data of different frequency and is transferred to pretest outgoing call algoritic module, carry out data by data output interface to show simultaneously。Predictive outbound algoritic module is for dialing result to lower a moment be predicted according to the historical data in a upper moment, the dynamic statistics result of intermediate data and default parameter (including target component) and gross data, and calls according to the decision that predicts the outcome or accumulative parameter is to lower a moment。Dynamic statistics module for carrying out the statistics of algorithm desired parameters or intermediate object program according to historical data, and the parameter of some measure algorithm indexs is carried out Real-time and Dynamic displaying。

Predictive outbound algoritic module comprises configuration parameter unit, initiation parameter unit, Theoretical Calculation data cell, cycle statistical parameter unit, predictive outbound algorithm unit, predict the outcome answer unit。Configuration parameter unit (includes number of switches for desampler configuration parameter information, maximum waiting time, current queue number, currently available number etc. of attending a banquet), initiation parameter unit is for receiving the various object parameter data of data input module input, Theoretical Calculation data cell is for receiving Theoretical Calculation but wanting the data of cyclomorphosis (to include as theoretical AR's etc, every data calculated by formula but do not use statistical method to obtain are all the data using Theoretical Calculation), cycle statistical parameter unit is for receiving the various data that the different statistic frequency of dynamic statistics module comes out, predictive outbound algorithm unit is used for realizing predictive outbound algorithm, predict the outcome interface for predicting the outcome to switch transmission。

Dynamic statistics module includes obtaining historical data unit, select statistics policy unit, statistics policy unit (prestores statistics strategy in statistics policy unit, such as according to week, month or festivals or holidays statistics, according to classification morning and afternoon, that add up according to each time period of every day and according to cumulative statistics strategy etc.), transmit the unit of statistical data to predictive outbound algoritic module and transmit the unit of statistical data to output interface, what obtain that storage in historical data unit has that the data including obtaining written data storehouse and switch preserve services but also has not enough time to the data in write into Databasce, select statistics policy unit is used for selecting to use which kind of statistics strategy, statistics policy unit includes multiple concrete statistics strategy。

The device of the predictive outbound algorithm based on dynamic statistics of the present invention, it is possible to be embedded in an existing switch, it is also possible to provide the interface of and switch individually to dispose。One of ordinary skill in the art will appreciate that the process of the predictive outbound algorithm based on dynamic statistics realizing the present invention, can be completed by programming, concrete program can be stored in the readable storage medium storing program for executing of the device of a kind of predictive outbound algorithm based on dynamic statistics, and this program performs the corresponding step in said method upon execution。Described storage medium can be as: ROM/RAM, magnetic disc, CD and hard disk etc.。

The present invention can regard a queuing model as outer paging system, telephone number in systems is regarded as time dependent function, so this system is exactly the Markov Chain of a Time Continuous, it is W that system arranges maximum quantity (workload) of dialing, it is assumed that call center have N number of identical attend a banquet and allow capacity be c wait in line queue length and T switch。Pass through the data mining to a large amount of manual dial's data of company and theory analysis, it is the Poisson distribution of λ that telephone outbound call number is modeled as obedience parameter, phone ringing duration obeys the Poisson distribution that average was 1/ β second, the average percent of call completed of phone is ρ, and it is p that connection client does not enter the probability of waiting list (directly hanging up) under can not get service scenario₀, the average patient time of client was 1/ η second simultaneously, and the service time connecting client obeys the quantum condition entropy that average was 1/ μ second。

Present outgoing call systematic procedure can be considered the Probability State transformation process under the Markov Chain of Time Continuous, and X (t) is the total telephone number in t system, including the telephone number serviced and in waiting list。The state space E={0 that then Markov Chain of Time Continuous now has, 1,2 ..., N+c}, for many queuing processes, the arrival (causing state to add 1) of one client is represented by " life ", and leave away (the causing state to subtract 1) of a client is represented by " extremely ", then during X (t)=k, can only becoming k ± 1 during the change of its subsequent time state, therefore whole process becomes " life " probability is λ_k, " extremely " probability is μ_kIndependent random process-birth and death process。

Wherein:

${\mathrm{\&lambda;}}_k=\left\{\begin{array}{cc}\mathrm{\&lambda;}\mathrm{\&beta;}\mathrm{\&rho;}(1-p_0)& 0\&le;k\&le;N+c-1\\ 0& k\&GreaterEqual;N+c\end{array}\right\}$

${\mathrm{\&mu;}}_k=\left\{\begin{array}{cc}k\mathrm{\&mu;}& 1\&le;k\&le;N-1\\ N\mathrm{\&mu;}+(k-N)\mathrm{\&eta;}& N\&le;k\&le;N+c\\ 0& k=0,k>N+c\end{array}\right\}$

So can carry out computing system reach the probability distribution { π of plateau according to " life is gone out " process₀,π₁,…,π_N+c, wherein ∑_i∈Eπ_i=1, enter this state during according to plateau in the unit interval and leave that the average time of this state is equal can list formula:

$\left\{\begin{array}{cc}{\mathrm{\&mu;}}_1{\mathrm{\&pi;}}_1={\mathrm{\&lambda;}}_0{\mathrm{\&pi;}}_0& \\ {\mathrm{\&lambda;}}_{k-1}{\mathrm{\&pi;}}_{k-1}+{\mathrm{\&mu;}}_{k+1}{\mathrm{\&pi;}}_{k+1}=({\mathrm{\&lambda;}}_k+{\mathrm{\&mu;}}_k){\mathrm{\&pi;}}_k& k\&GreaterEqual;1\end{array}\right\}$

After drawing system probability distribution under each state, it is possible to calculate some indexs weighing systematic function:

1. the expectation TL of system waiting list length can try to achieve according to system probability distribution in this condition:

TL=∑_{I ∈ E, i > N}(i-N)π_i

2. client is not obtaining servicing average latency E [W] expectation also do not abandoned when waiting:

$E\&lsqb;W\&rsqb;=\frac{TL}{\mathrm{\&lambda;}\mathrm{\&beta;}\mathrm{\&rho;}(1-p_0)}$

3. the utilization rate expectation U that attends a banquet of customer service:

$U=1-{\mathrm{\&Sigma;}}_{i\&Element;E,i<N}\frac{N-i}{N}{\mathrm{\&pi;}}_i$

The core of common phone predictive dialer system is predictive outbound algorithm, and native system opens a kind of predictive outbound algorithm and is divided into two important modules: predictive outbound algoritic module and dynamic statistics module。Such that it is able to optimize from both direction, both direction has a lot of prioritization scheme, thus avoiding the bottleneck problem that an optimized algorithm there will be。And need not be concerned about that when optimized algorithm the concrete of parameter is added up, the realization of algorithm need not be considered during statistical parameter and whether that algorithm is useful。

The telephone number dialed by predictive outbound algorithm is typically greater than idle number of seats at that time, this is because percent of call completed generally than 1 little a lot, so comparing the dialing dialling and being not previously predicted function by hand can be greatly enhanced seat utilization rate, if but dial many, will result in phone to get through but service for it of not attending a banquet yields, this situation is called " call loss ", reality must suppress the ratio of call loss, the i.e. percent of call lost, algorithm otherwise user can be caused great harassing and wrecking, so will be balanced between utilization rate and the phone percent of call lost attending a banquet。As long as while it is true, algorithm is reasonably predicted, still can reach significantly high utilization rate of attending a banquet, produce good prediction effect。But this just requires that algorithm can set up good model, the Changeement of various parameters is gone out a set of theory (including queueing theory), and then adapts to the situation of various complexity。Monochromatic light is by setting up good forecast model not enough, because prediction must be predicted according to historical data, need which and how many historical datas, and how to be analyzed drawing a more rational value etc. to data, here it is the effect of dynamic statistics module, the present invention pertains only to predictive outbound algorithm。

In the description of this specification, specific features, structure, material or feature that the description of reference term " embodiment ", " some embodiments ", " example ", " concrete example " or " some examples " etc. means in conjunction with this embodiment or example describe are contained at least one embodiment or the example of the present invention。In this manual, the schematic representation of above-mentioned term is not necessarily referring to identical embodiment or example。And, the specific features of description, structure, material or feature can combine in an appropriate manner in any one or more embodiments or example。

Although an embodiment of the present invention has been shown and described, it will be understood by those skilled in the art that: these embodiments can being carried out multiple change, amendment, replacement and modification when without departing from principles of the invention and objective, the scope of the present invention is limited by claim and equivalent thereof。

Claims (5)

1. the predictive outbound algorithm based on dynamic statistics process, it is characterised in that comprise the steps:

S1, starts, and arranges target and attends a banquet utilization rate AO, maximum percent of call lost AR_max, number of switches T, maximum waiting time TQ_max, it was predicted that interval time Δ t, waiting list greatest length qL, user is patient time 1/ η on average, and user is reluctant the Probability p adding waiting list₀, and outwards call the initial value of several λ, make the accumulated value al=0 of l；

S2, in t, obtains current customer service sum N by switch, and idle customer service sum f, just at called telephone number j, percent of call completed ρ, average jingle bell duration 1/ β, average call duration 1/ μ, and the numerical value of waiting list length q, described t is positive number；

S3, it may be judged whether meet q < qL, if meeting, goes to step S4, and otherwise dormancy to t+ Δ t goes to step S2 and carries out next round prediction, and to update current time be t=t+ Δ t；

S4, calculates required minimum several M that attends a banquet, and computing formula is:

S5, it may be judged whether meet N≤M, if meeting, going to step S17, otherwise going to step S6；

S6, computing system predicts maximum outgoing call value l, cumulative calculation al=al+l, wherein,

$l=\frac{\mathrm{\&Delta;}t*N*AO*\mathrm{\&mu;}}{\mathrm{\&rho;}};$

S7, the percent of call lost AR in dynamic statistics last moment also judge whether to meet AR < AR_maxIf meeting, going to step S8, otherwise go to step S16；

S8, it may be judged whether meet current idle and attend a banquet several f > al, if meeting, going to step S9, otherwise going to step S13；

S9, it was predicted that result s=(f-j)/ρ-q+al, if al > 1, al sets to 0；

S10, calculates period of reservation of number Tq, Tq=q/ (N* μ)；

S11, it may be judged whether meet Tq < TQ_maxIf meeting, going to step S12, otherwise dormancy to t+ Δ t goes to step S2 and carries out next round prediction, and to update current time be t=t+ Δ t；

S12, updates the exhalation that s=s*ARrate, ARrate are system that predicts the outcome and weakens ratio, ARrate=1-AR, and AR is the theoretical percent of call lost,

AR=p₀B(θ,N)+(1-p₀) λ β ρ η, go to step S18；

S13, it was predicted that result s=f-j-q+al, if al > 1, al sets to 0；

S14, calculates theoretical percent of call lost AR, AR=p₀B(θ,N)+(1-p₀) λ β ρ η,

B (θ, 0)=1,

$B(\mathrm{\&theta;},k)=\frac{\mathrm{\&theta;}B(\mathrm{\&theta;},k-1)}{k+\mathrm{\&theta;}B(\mathrm{\&theta;},k-1)},1\&le;k\&le;N;$

S15, it may be judged whether meet AR < AR_maxIf meeting, going to step S10, otherwise dormancy to t+ Δ t goes to step S2 and carries out next round prediction, and to update current time be t=t+ Δ t；

S16, it was predicted that result s=f-j-q, goes to step S18；

S17, it was predicted that result s=f-j, goes to step S18；

S18, calculates and finally calls number λ, λ=(int) min (max (s, 0), T)；

S19, outwards dials λ phone, and algorithm dormancy to t+ Δ t goes to step S2 and carries out next round prediction, and to update current time be t=t+ Δ t；Repeat above step, until application program terminates running。

2. the predictive outbound algorithm based on dynamic statistics process according to claim 1, it is characterized in that, in described step S7, the percent of call lost AR in dynamic statistics last moment is all of phone number that the phone number of t1 moment before current time to current time stage call loss was connected divided by this stage, and described t1 is positive number。

3. the predictive outbound algorithm based on dynamic statistics process according to claim 1, it is characterised in that described t1 is the time between 15 minutes to 45 minutes。

4. the switch dialing method according to the predictive outbound algorithm based on dynamic statistics process described in based on claim 1, it is characterised in that comprise the steps:

Step 1, predictive outbound algorithm predicts according to claim 1 goes out subsequent time will call quantity λ；

Step 2, switch predicts the outcome according to this and dials λ phone；

Step 3, after average ring time β, described β is positive number, has the closing of the circuit of ρ ratio, and the phone access failure of remaining 1-ρ ratio is filtered, and described ρ is the positive number more than 0 less than 1；

Step 4, the number metering-in control system of connection, control system can according to currently whether available free attending a banquet is allocated；

Step 5, attends a banquet if available free, and distribution is attended a banquet and serviced for it, services complete successful for dialing；If not having idle seating, the client having p0 probability is ready that adding waiting list waits, the client of remaining 1-p0 probability is unwilling to add waiting list and abandons forming call loss；

Step 6, also some bears with time η owing to waiting more than it and abandons being formed call loss the client queued up, only remaining meeting distributes, according to idle the attending a banquet serviced at present, the service of attending a banquet by control system for it when not above its patient time η, and finally service completes。

5. the device of the predictive outbound algorithm based on dynamic statistics process run described in claim 1, it is characterized in that, including data input module, predictive outbound algoritic module, switch, dynamic statistics module, data memory module and data output display module, described data input module is to pretest outgoing call algoritic module input initialization supplemental characteristic, described pretest outgoing call algoritic module is connected with switch, the information of carrying out is mutual, described dynamic statistics module is connected with data memory module, receive the data of data memory module, described dynamic statistics module is added up the statistical data of different frequency and is transferred to pretest outgoing call algoritic module, carry out data by data output interface to show simultaneously；

Described predictive outbound algoritic module comprises configuration parameter unit, initiation parameter unit, Theoretical Calculation data cell, cycle statistical parameter unit, predictive outbound algorithm unit, predict the outcome answer unit；Described configuration parameter unit is used for desampler configuration parameter information, described initiation parameter unit is for receiving the various object parameter data of data input module input, described Theoretical Calculation data cell is for receiving Theoretical Calculation but wanting the data of cyclomorphosis, described cycle statistical parameter unit is for receiving the various data that the different statistic frequency of dynamic statistics module comes out, the described predictive outbound algorithm unit predictive outbound algorithm for realizing described in claim 1, described in predict the outcome interface for predicting the outcome to switch transmission；

Described dynamic statistics module includes obtaining historical data unit, selects statistics policy unit, statistics policy unit, transmits the unit of statistical data to predictive outbound algoritic module and transmit the unit of statistical data to output interface, in described acquisition historical data unit, storage has and includes obtaining the data in written data storehouse and switch preserves services but also have not enough time to the data in write into Databasce, described selection adds up policy unit is used for selecting to use which kind of statistics strategy, and described statistics policy unit includes multiple concrete statistics strategy。