CN101541016A - Method for predicting data and equipment - Google Patents
Method for predicting data and equipment Download PDFInfo
- Publication number
- CN101541016A CN101541016A CN200910135919A CN200910135919A CN101541016A CN 101541016 A CN101541016 A CN 101541016A CN 200910135919 A CN200910135919 A CN 200910135919A CN 200910135919 A CN200910135919 A CN 200910135919A CN 101541016 A CN101541016 A CN 101541016A
- Authority
- CN
- China
- Prior art keywords
- data
- period
- compensation
- compensation rate
- day
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a method for predicting data and an equipment. The method comprises the following steps of: reading a plurality of first data related to a first period which contains a plurality of first sub periods, wherein a plurality of first data and a plurality of first sub periods are in a one-to-one correspondence relationship; based on a plurality of first data and a plurality of first sub periods, determining a plurality of first trend values which are in a one-to-one correspondence relationship with a plurality of first sub periods; based on a plurality of first data and a plurality of first trend values, obtaining a compensation sequence constituted by a plurality of compensation amount, wherein a plurality of compensation amount and a plurality of first sub periods are in a one-to-one correspondence relationship; based on the compensation sequence, obtaining estimation values of the compensation dosage which corresponds to a second period by utilizing a support vector machine, wherein the time of the second period is later than that of the first period; based on the estimation values of compensation amount, predicting a second data related to the second period. The embodiment of the invention has effective prediction, thereby being capable of bringing considerable economic benefits for operators.
Description
Technical field
The embodiment of the invention relates to data predication method and data prediction equipment, more specifically, relates to and utilizes SVMs (SVM; Support vector machine) data predication method and data prediction equipment.
Background technology
In recent years, radio communication service has all obtained development at full speed in the whole world, and userbase constantly enlarges, and many new business also occur in succession.But the sharp increase day by day of telephone traffic (being obvious ascendant trend), and the user is to improving constantly that radio communication quality requires, all cordless communication network is proposed new requirement, increased difficulty at aspects such as the network planning, construction and adjustment for simultaneously mobile communication carrier.Especially in special festivals or holidays (as the Spring Festival, the mid-autumn, May Day, National Day, New Year's Day, Easter, Thanksgiving Day, the holy discipline day, New Year etc.) or generation special event (as Wenchuan violent earthquake, Shanghai car plate auction activity etc.), the wireless communication burst telephone traffic can be uprushed than usual, very easily cause the switching system overload, it is congested phone to occur, the speech call completing rate descends, traffic call drop ratio rises, even the phenomenon of switch large tracts of land paralysis appears, all cause irretrievable loss for mobile communication carrier and mobile subscriber, have influence on the degree of belief of user operator.Therefore, traffic forecast is significant for the operation management of cordless communication network.But increase rapidly incompatible be that traffic forecast technical development and imperfection to the networking of mobile communication carrier and adjust and lack validity and instruct, have influenced the performance of enterprises with the radio communication telephone traffic.
Traffic forecast not only can be used for instructing the dilatation of network, also can be equalizing traffic volume reference is provided.As everyone knows, the purpose of equalizing traffic volume is in order to improve the resource utilization of network carrier frequency, by the equilibrium of telephone traffic being avoided the waste of capacity, to reduce the telephone traffic of high congestion sub-district, improve network congestion rate index, improving user's impression.The method of equalizing traffic volume generally is the purpose that reaches equalizing traffic volume by the behavior of control cell coverage area and control mobile phone.Usually can wait balanced traffic by engineering parameter adjustment, power adjustment, handover optimization, cell selecting and reselection optimization, the adjustment of cell channel resource optimization, improve the resource distribution utilance.Certainly, more congested if the sub-district of covering the same area all exists, illustrate that at this moment network capacity is not enough, at this moment just need solve by dilatation.
Equally, for other data services, for example text message, multimedia messages, interactive game etc., data prediction also is that resource optimization institute is requisite.
With the wireless traffic amount is example, and present wireless traffic amount linear regression prediction scheme can be divided into one-variable linear regression and multiple linear regression.Wherein, one-variable linear regression only considers that wireless traffic amount data are the function of time, realizes simpler relatively; Multiple linear regression then is to consider many factors to the telephone traffic influence, but these factors are difficult to quantize in the practical application, have limited the application of multiple linear regression.Usually in the prediction of wireless traffic amount, adopt the strategies of one-variable linear regressions more.Fig. 1 is the flow chart of data predication method that the employing one-variable linear regression of prior art is shown.Usually the mathematic(al) representation of the one-variable linear regression forecast model that adopts is
y＝bt+a (1)
Wherein, t is time (being independent variable), and y is the radio communication telephone traffic (being dependent variable) of certain time point, and parameter a and b are estimated to obtain by the least square method that describes below.
The least square method problem that above-mentioned parameter is estimated can be described as: according to known data { (x
i, y
i), i=1,2 ..., n} chooses an approximate function f (x)=bt+a, makes
Minimum.Ask the method for extreme value by calculus, the least-squares estimation value that can get parameter a and b is respectively
With sequential { t corresponding in the time span of forecast
j| the one-variable linear regression forecast model that j＞n} substitution is obtained by (1) formula, (3) formula, (4) formula can obtain the predicted value of following telephone traffic.
Another kind of regression forecasting method adopts the index return model.Fig. 2 is the flow chart of data predication method that the employing index return of prior art is shown.Wireless traffic volume index regression forecasting scheme and linear regression prediction scheme are similar, and practical application is relatively simple, but only can describe the telephone traffic trend of index variation.
The mathematic(al) representation of index return forecast model is
y＝AB
t (5)
Wherein, t is time (being independent variable), and y is the radio communication telephone traffic (being dependent variable) of following certain time point, and A and B are parameter to be estimated.
Be estimated parameter A and B, earlier (5) formula taken from right logarithmic transformation usually, be converted into the situation of one-variable linear regression
lny＝lnA+tlnB (6)
At this moment, can obtain the value of lnA and lnB, do the estimated value that index variation can obtain A and B again by (3) formula and the least-squares estimation of (4) formula
With
3. the index access regression model is to the telephone traffic predicted value
With sequential { t corresponding in the time span of forecast
j| j＞n} substitution by
The one-variable linear regression forecast model that obtains can obtain the predicted value of following telephone traffic.
But all there are following shortcoming in one-variable linear regression and index return data predication method:
1. only can characterize the development of linear trend of telephone traffic, the local detail that can't portray telephone traffic changes;
2. only considered time factor, and the wireless traffic amount also is subjected to all multifactor influences in the reality, except that tendency, also has seasonal and periodicity.
Above-mentioned shortcoming causes one-variable linear regression and index return to have significant limitation on prediction accuracy.
Summary of the invention
Therefore, the embodiment of the invention purpose provides the data prediction scheme with higher prediction accuracy.
Another purpose of the embodiment of the invention provides the data prediction scheme based on SVMs.
Another purpose of the embodiment of the invention provides the neighbor point that utilizes SVMs to make up and search and makes up part/segmentation forecast model.
Another purpose of the embodiment of the invention provides the data prediction scheme of portraying radio communication telephone traffic seasonal effect in time series strong nonlinearity characteristic better.
For realizing above-mentioned and/or other purposes, the embodiment of the invention provides a kind of data predication method.This method comprises: read a plurality of first data that are associated with first period, wherein comprise a plurality of first period of the day from 11 p.m. to 1 a.m phases first period, a plurality of first data are one by one corresponding to a plurality of first period of the day from 11 p.m. to 1 a.m phases; Based on a plurality of first data and a plurality of first period of the day from 11 p.m. to 1 a.m phase, determine a plurality of first Trend value of corresponding one by one a plurality of first period of the day from 11 p.m. to 1 a.m phases; Based on a plurality of first data and a plurality of first Trend value, the compensation sequence that acquisition is made of a plurality of compensation rates, wherein a plurality of compensation rates are one by one corresponding to a plurality of first period of the day from 11 p.m. to 1 a.m phases; Based on the compensation sequence, utilize SVMs to obtain and corresponding compensation rate estimated value in second period, wherein be later than for first period second period in time; Based on the compensation rate estimated value, second data that prediction was associated with second period.
For realizing above-mentioned and/or other purposes, the embodiment of the invention provides a kind of data prediction equipment, comprise: reading device, read a plurality of first data that are associated with first period, wherein comprise a plurality of first period of the day from 11 p.m. to 1 a.m phases first period, a plurality of first data are one by one corresponding to described a plurality of first period of the day from 11 p.m. to 1 a.m phases; Trend value is determined device, based on a plurality of first data and a plurality of first period of the day from 11 p.m. to 1 a.m phase, determines one by one a plurality of first Trend value corresponding to a plurality of first period of the day from 11 p.m. to 1 a.m phases; The compensation sequence is determined device, based on a plurality of first data and described a plurality of first Trend value, and the compensation sequence that acquisition is made of a plurality of compensation rates, wherein a plurality of compensation rates are one by one corresponding to a plurality of first period of the day from 11 p.m. to 1 a.m phases; The compensation rate estimated value is determined device, based on the compensation sequence, utilizes SVMs to obtain and corresponding compensation rate estimated value in second period, wherein is later than described first period second period in time; And prediction unit, based on the compensation rate estimated value, second data that prediction was associated with second period.
The embodiment of the invention has taken into full account the seasonal effect in time series nonlinear characteristic (as tendency, seasonality and periodically etc.) of data (as the radio communication telephone traffic etc.), utilize phase space reconfiguration and neighbor point to search technology, and in local prediction in conjunction with effective Predicting Technique-SVMs SVM, data are carried out effective prediction of short-term, and then instruct the balanced management of operator to existing network, improve network congestion rate index, improve the resource utilization of network carrier frequency, avoid directly carrying out the economy input that dilatation needs, also can increase the degree of belief of user indirectly operator.In a word, the technical scheme of the embodiment of the invention is effectively predicted for the short-term such as the data of wireless traffic amount etc., is brought considerable economic directly or indirectly for operator.
Description of drawings
According to following description in conjunction with the accompanying drawings, other advantages, feature and the details that invention will be more fully understood.
Fig. 1 is the flow chart of data predication method that the employing one-variable linear regression of prior art is shown.
Fig. 2 is the flow chart of data predication method that the employing index return of prior art is shown.
Fig. 3 illustrates the flow chart of data predication method according to an embodiment of the invention.
Fig. 4 is the schematic diagram that the stable state compensation sequence of obtaining the traffic sequence is shown.
Fig. 5 is the figure that is shown specifically according to the compensation rate estimated value deterministic process of the embodiment of the invention.
Fig. 6 shows a kind of forecast model structure chart of SVMs.
Fig. 7 illustrates the block diagram of data prediction equipment according to an embodiment of the invention.
Fig. 8 is the block diagram that the data prediction equipment of another embodiment of the present invention is shown.
Embodiment
Describe each embodiment of the present invention with reference to the accompanying drawings in detail.
Fig. 3 illustrates the flow chart of data predication method 100 according to an embodiment of the invention.
As shown in Figure 3, data predication method 100 is from S105.In S105, read a plurality of historical datas (" first data ") that are associated with the period of history (" first period ").The long measure of period of history (granularity) can be the sky, also can hour, minute etc.Period of history can comprise a plurality of historical period of the day from 11 p.m. to 1 a.m phases, and for example, if the length of period of history is year, the length of historical period of the day from 11 p.m. to 1 a.m phase is the sky, then can set 1 year and comprise 365 days.Historical data is the data that recorded in the period of history, for example radio communication telephone traffic, fixed line communication traffic amount, note amount, video flow, number of users etc.Historical data can for example can be the telephone traffic of every day in 1 year one by one corresponding to the historical period of the day from 11 p.m. to 1 a.m phase.For ease of understanding, be example with the radio communication traffic data below.Suppose in the S105 of method 100, read historical telephone traffic sequence y (t): t=1,2 ..., n}.Wherein t represents the historical period of the day from 11 p.m. to 1 a.m phase, and n is a positive integer.Understand easily, if the granularity of historical period of the day from 11 p.m. to 1 a.m phase be the sky, then n represents total fate of comprising in the period of history.It should be noted that this only is exemplary, should not be construed as scope of the present invention is construed as limiting.
Then, at S110, based on historical data and historical period of the day from 11 p.m. to 1 a.m phase, specified data Trend value (" first Trend value ").The data trend value can be one by one corresponding to the historical period of the day from 11 p.m. to 1 a.m phase.Can obtain corresponding tendency equation at first according to historical data and the one-to-one relationship of historical period of the day from 11 p.m. to 1 a.m phase.There is multiple tendency equation, one-variable linear regression tendency equation for example recited above and index return tendency equation or the like.Under the situation of the Linear Regression Model in One Unknown that adopts least square method, to historical telephone traffic sequence y (t): t=1,2 ..., n} utilizes above-mentioned equation (3) and (4) to determine its tendency equation y
tThe parameter value of=a+bt
With
Afterwards, with in each historical period of the day from 11 p.m. to 1 a.m phase t substitution tendency equation, obtain the telephone traffic Trend value of each time point successively again
It should be noted that tendency equation is not limited to above-mentioned concrete form.Also can adopt other any possible tendency equations.
After obtaining the data trend value, in S115, based on historical data and the data trend value that in S110, obtains, obtain the compensation sequence x (t): t=1,2 ..., n}.Each compensation rate x (t) in the compensation sequence is also corresponding one by one with historical period of the day from 11 p.m. to 1 a.m phase t.The method that obtains the compensation sequence is a lot, and commonly used have subtraction and a ratio method etc.The radio communication telephone traffic presents certain tendency, seasonality, if directly carry out phase space reconfiguration with original traffic data, and carry out the most contiguous searching mutually for the basis according to this, because the acquisition of the pattern that is complementary is covered in the growth of trend, can cause obtaining the most contiguous correct point mutually.Because original traffic data is except that having on the whole the tendency, along with the variation of its amplitude of evolution of time also in continuous increase, if so adopt the mode of subtraction remove on original traffic data basis the trend amount obtain new time series still might right and wrong stably.For overcoming above-mentioned deficiency, obtain to compensate sequence more stably, preferably, can take the strategy of ratio method, this also meets the thought of seaconal model design.Fig. 4 is the schematic diagram that the stable state compensation sequence of obtaining the traffic sequence is shown.As shown in Figure 4, can be divided by corresponding data trend value (promptly with historical traffic data
), thereby obtain stable state the compensation sequence x (t): t=1,2 ..., n}.It should be noted that ratio method does not constitute for restriction of the present invention, might adopt other to obtain the method for compensation sequence.
Subsequently, at S120, based on the compensation sequence that in S110, obtains x (t): t=1,2 ..., n} utilizes SVMs to obtain and the corresponding compensation rate estimated value of prediction period (" second period ").Prediction should be later than the period of history period in time,, predicts that the initial moment in period should equal or be later than the finish time of period of history that is.N is the numbering of historical period of the day from 11 p.m. to 1 a.m phase t, supposes that prediction step (prediction period and the distance of the corresponding historical period of the day from 11 p.m. to 1 a.m between the phase are long measure with the granularity of historical period of the day from 11 p.m. to 1 a.m phase) is T, and can be expressed as t the period of then predicting
j=n+T, corresponding compensation rate estimated value can be expressed as
Hereinafter also will describe compensation rate estimated value deterministic process S120 in detail.
In S120 after the amount of the being compensated estimated value, at S125, based on resulting compensation rate estimated value
The prediction data (" second data ") that prediction and prediction are associated period.Can utilize the tendency equation of in S110, determining
(the t that for example will predict period
j=n+T) this tendency equation of substitution is to obtain corresponding to this prediction (t in period
j=n+T) anticipation trend value (" second Trend value ") y
t(n+T).Like this, just can be according to anticipation trend value (" second Trend value ") y
t(n+T) and the compensation rate estimated value
And obtain final prediction data.Under the situation that adopts the ratio method strategy, prediction data equals the product of anticipation trend value and compensation rate estimated value, promptly
After this, method 100 finishes.
Fig. 5 is the figure that is shown specifically according to the compensation rate estimated value deterministic process of the embodiment of the invention.Fig. 6 shows a kind of forecast model structure chart of SVMs.
Be subjected to the influence and the restriction of all multifactor (as the market behavior, the density of population, economic development situation, festivals or holidays etc.) such as the data of radio communication telephone traffic etc., make it present characteristics such as certain tendency, seasonality, periodicity and randomness, can think the strongly non-linear system of a complexity.For instructing equalizing traffic volume effectively, improve the resource utilization of network carrier frequency, improve network congestion rate index, the embodiment of the invention provides the SVMs data prediction scheme of searching based on phase space reconfiguration and neighbor point.
To the stable state compensation sequence that obtains, after phase space reconfiguration and neighbor point are searched, directly utilize local methods such as averaging method or distance weighted method to come the local function of match/approach, can produce and make estimated performance relatively more responsive embedding parameters such as dimension, neighbor point number, extensive poor performance is when phase space neighborhood point is difficult to accurately be separated during at conllinear etc. not enough.And SVMs is based on the machine learning method of structure risk optimization criterion, its regression technique is compared neural net and is had the global optimum of converging on, extensive performance and reach advantages such as sparse solution well, makes its very suitable small sample, non-linear and higher-dimension problem of solving.At the deficiency of existing local Forecasting Methodology, the advantage of combination supporting vector machine is applied to SVMs to realize the approaching and match of local subclass of smooth mapping f, designs the local prediction new that can be convenient to practical application.Specific as follows:
The first step, to the make-up time sequence x (t): t=1,2 ..., n} selects suitable embedding dimension m and delay time T, carries out phase space reconfiguration.Can carry out phase space reconfiguration according to for example Takens theorem,, obtain following reconstructed vector for arbitrary compensation rate x (i):
X(i)＝[x(i)，x(i+τ)，…，x(i+(m-1)τ)]，i＜n-(m-1)τ (7)
Embedding dimension m and delay time T can select as required.For example, for the radio communication telephone traffic, with the sky be the radio communication traffic data of time granularity represented usually with 7 days (i.e. a week) be the periodicity in cycle, this also meets the daily life rule.In the case, can select to embed dimension m and delay time T is respectively 7 and 1.
In second step, according to top phase space reconfiguration track, establishing n the reconstructed vector of putting is X (n), calculate it and front n-1 reconstructed vector X (i=1,2 ..., Euclidean distance n-1)
Obtain p minimum distance according to the ascending power order.P is a positive integer, can determine as required.For example can be 6.Extract reconstructed vector successively, the neighbor point of ordering as n corresponding to this p minimum range:
Extract the known future value of these neighbor points simultaneously,
D＝x(t
r+T)，r＝1，2，…，p (10)
Wherein, T is a prediction step, and p is the neighbor point number that finds.
It should be noted that above-mentioned Euclidean distance is exemplary, should not be construed as limitation of the scope of the invention.Also can adopt the distance of other types.
The 3rd step is X
n (r)As the training in SVM prediction model input, corresponding future value D is as training output, trains SVMs (its structure can as shown in Figure 6), obtains each parameter (as supporting vector and weights coefficient) of corresponding SVMs.This SVMs is predicted because of the strategy that adopts piecewise approximation, thereby also be can be described as the local SVMs.
In the 4th step, the local SVMs that training obtains above utilizing as input, is predicted output with reconstructed vector X (n) accordingly, is above-mentioned compensation rate estimated value
Repeat above-mentioned four steps of the first step to the, can calculate all compensation rate estimated values.
SVMs local predicted method in this stage causes it insensitive to the selection that embeds dimension, neighbor point number, so make this method have better practicability in practice because SVMs has the sparse property of separating.
At this, at the data with strong nonlinearity characteristic (as wireless traffic amount etc.), the embodiment of the invention from dynamic nonlinear model angle design the small particle size of searching based on phase space reconfiguration and neighbor point that is applicable to equalizing traffic volume (my god) the SVMs data predication method.This method considers that local predicted method has that pliability is good, match speed fast and the operational precision advantages of higher, introduces the support vector machine technology, to obtain more excellent prediction effect under this framework.In addition, the characteristic of SVMs sparse solution can make new algorithm overcome the deficiency that traditional local method prediction effect is subjected to embedding dimension, the influence of neighbor point number, particularly, has very important using value for the situation of not knowing in practice to embed dimension and do not know how to select the neighbor point number.
Fig. 7 illustrates the block diagram of data prediction equipment 200 according to an embodiment of the invention.As shown in Figure 7, data prediction equipment 200 comprises that reading device 210, Trend value determine that device 220, compensation sequence determine that device 230, compensation rate estimated value determine device 240 and prediction unit 250.Reading device 210 reads a plurality of historical datas (" first data ") that are associated with the period of history (" first period ").Period of history can comprise a plurality of historical period of the day from 11 p.m. to 1 a.m phases.Historical data is one by one corresponding to the historical period of the day from 11 p.m. to 1 a.m phase.Trend value is determined device 220 based on historical data and historical period of the day from 11 p.m. to 1 a.m phase, determines one by one a plurality of first Trend value corresponding to the historical period of the day from 11 p.m. to 1 a.m phase.The compensation sequence is determined device 230 based on the historical data and first Trend value, obtains the compensation sequence.The compensation sequence is made of a plurality of compensation rates corresponding to a plurality of historical period of the day from 11 p.m. to 1 a.m phases one by one.The compensation rate estimated value determines that device 240 based on the compensation sequence, utilizes SVMs to obtain and the corresponding compensation rate estimated value of prediction period (" second period ").Prediction unit 250 is based on the compensation rate estimated value, the prediction data (" second data ") that prediction and prediction are associated period.The operation of each assembly of data prediction equipment 200 can for avoiding repetition, not repeat them here with reference to each step of preceding method 100.
Fig. 8 is the block diagram that the data prediction equipment 300 of another embodiment of the present invention is shown.Data prediction equipment 300 comprises that reading device 310, Trend value determine that device 320, compensation sequence determine that device 330, compensation rate estimated value determine device 340 and prediction unit 350.The structure of data prediction equipment 300 and the function of each assembly and data prediction equipment 200 are basic identical, for avoiding repetition, do not repeat them here.
In addition, Trend value determines that device 320 can comprise the tendency equation module 3205 and the first Trend value module 3210.Tendency equation module 3205 can obtain tendency equation from historical data and historical period of the day from 11 p.m. to 1 a.m phase according to said method.For example, can adopt methods such as least square method, index return.The first Trend value module 3210 can be with historical period of the day from 11 p.m. to 1 a.m phase difference substitution tendency equation to obtain the data trend value.
Prediction unit 350 can comprise the second Trend value module 3505 and prediction module 3510.The tendency equation that the second Trend value module 3505 can utilize tendency equation module 3205 to obtain obtains and prediction corresponding anticipation trend value in period.For example, can be with prediction substitution in period tendency equation to obtain the anticipation trend value.Prediction module 3510 can be according to above-described method, based on the compensation rate estimated value of determining device 340 from the compensation rate estimated value with from the anticipation trend value of the second Trend value module 3505, obtains prediction data.
The compensation rate estimated value determines that device 340 can comprise first module 3405, second module 3410, three module 3415 and four module 3420.First module 3405 can be selected to embed dimension and time of delay, determines that to coming the self compensation sequence compensation sequence of device 330 carries out phase space reconfiguration, to obtain the reconstructed vector corresponding to each compensation rate in the compensation sequence.For example can select to embed dimension and be respectively 7 and 1 time of delay according to above-described method.Can certainly select other numerals as required.Second module 3410 can be searched the neighbor point of a certain specific compensation rate.For example, can calculate the Euclidean distance of reconstructed vector, extract wherein minimum several distances, the neighbor point of corresponding reconstructed vector as this specific compensation rate according to above-described method.Three module 3415 can use the training input of the reconstructed vector of the neighbor point that second module 3410 finds as SVMs, with the training output as SVMs of the known future value of this specific compensation rate, training SVMs.But the reconstructed vector of four module 3420 using compensation sequences is as the input variable of the SVMs that is obtained by three module 3410 training, and estimated value is measured in the output of calculating SVMs by way of compensation.
The said structure that it should be noted that data prediction equipment 200 and 300 is exemplary, the invention is not restricted to such concrete structure.Can add other assemblies, also can merge, split some assembly wherein.It should be noted that each components/functions module annexation each other is not construed as limiting scope of the present invention in the embodiment of the invention, one or more functional module can comprise or be connected to other functional module or outer member arbitrarily.
It should be noted that embodiments of the invention can realize that by the combination of hardware, software or hardware and software its implementation is not construed as limiting scope of the present invention.One of ordinary skill in the art will appreciate that, the all or part of step of the data predication method of the realization embodiment of the invention can be finished by the relevant hardware of program command, described program can be stored in the read/write memory medium, and this program is carried out step corresponding in the above-mentioned data predication method when carrying out.Described storage medium can be: ROM/RAM, magnetic disc, CD etc.
The embodiment of the invention has taken into full account the seasonal effect in time series nonlinear characteristic (as tendency, seasonality and periodically etc.) of data (as the radio communication telephone traffic etc.), utilize phase space reconfiguration and neighbor point to search technology, and in local prediction in conjunction with effective Predicting Technique-SVMs SVM, data are carried out effective prediction of short-term, and then instruct the balanced management of operator to existing network, improve network congestion rate index, improve the resource utilization of network carrier frequency, avoid directly carrying out the economy input that dilatation needs, also can increase the degree of belief of user indirectly operator.In a word, the technical scheme of the embodiment of the invention is effectively predicted for the short-term such as the data of wireless traffic amount etc., is brought considerable economic directly or indirectly for operator.
Though illustrated and described in detail some embodiments of the present invention above in conjunction with the accompanying drawings, those skilled in the art is to be understood that, under the situation that does not depart from principle of the present invention and spirit, can make changes and modifications these embodiment, and still drop in the scope of claims and equivalent thereof.
Claims (20)
1, a kind of data predication method is characterized in that, comprising:
Read a plurality of first data that are associated with first period, comprise a plurality of first period of the day from 11 p.m. to 1 a.m phases wherein said first period, described a plurality of first data are one by one corresponding to described a plurality of first period of the day from 11 p.m. to 1 a.m phases;
Based on described a plurality of first data and described a plurality of first period of the day from 11 p.m. to 1 a.m phase, determine one by one a plurality of first Trend value corresponding to described a plurality of first period of the day from 11 p.m. to 1 a.m phases;
Based on described a plurality of first data and described a plurality of first Trend value, the compensation sequence that acquisition is made of a plurality of compensation rates, wherein said a plurality of compensation rates are one by one corresponding to described a plurality of first period of the day from 11 p.m. to 1 a.m phases;
Based on described compensation sequence, utilize SVMs to obtain and corresponding compensation rate estimated value in second period, be later than described first period wherein said second period in time; And
Based on described compensation rate estimated value, second data that prediction was associated with described second period.
2, data predication method as claimed in claim 1 is characterized in that, describedly determines one by one to comprise corresponding to a plurality of first Trend value of described a plurality of first period of the day from 11 p.m. to 1 a.m phases:
Based on described a plurality of first data and described a plurality of first period of the day from 11 p.m. to 1 a.m phase, obtain tendency equation; And
With the described tendency equation of described a plurality of first period of the day from 11 p.m. to 1 a.m phases difference substitutions, to obtain described a plurality of first Trend value.
3, data predication method as claimed in claim 2 is characterized in that, described acquisition tendency equation comprises: utilize least square method to obtain described tendency equation.
4, data predication method as claimed in claim 2 is characterized in that, described prediction comprises with second data that are associated described second period:
Utilize described tendency equation, obtain and corresponding second Trend value in described second period; And
Based on described compensation rate estimated value and described second Trend value, obtain described second data.
5, data predication method as claimed in claim 4, it is characterized in that, describedly obtain described second data based on described compensation rate estimated value and described second Trend value and comprise: described second Trend value be multiply by described compensation rate estimated value to obtain described second data.
6, data predication method as claimed in claim 1 is characterized in that, described acquisition is comprised by the compensation sequence that a plurality of compensation rates constitute: with described a plurality of first data respectively divided by described a plurality of first Trend value, to obtain described compensation sequence.
7, data predication method as claimed in claim 1 is characterized in that, described utilize SVMs obtain with second period corresponding compensation rate estimated value comprise:
Select to embed dimension and time of delay, described compensation sequence is carried out phase space reconfiguration, to obtain reconstructed vector corresponding to each compensation rate in the described compensation sequence;
Search the neighbor point of specific compensation rate;
With the training input of the reconstructed vector of the described neighbor point that found,, train described SVMs with the training output of the known future value of described specific compensation rate as SVMs as SVMs; And
The reconstructed vector that uses described compensation sequence is as the input variable of training the SVMs that obtains in the 3rd step, and the output of calculating described SVMs is as described compensation rate estimated value.
8, data predication method as claimed in claim 7 is characterized in that, described selection embeds dimension and comprises time of delay: selecting described embedding dimension is 7, and to select described time of delay be 1.
9, data predication method as claimed in claim 7, it is characterized in that, the described neighbor point of searching specific compensation rate comprises: the distance of calculating the reconstructed vector of each compensation rate before the reconstructed vector of described specific compensation rate and the described specific compensation rate respectively, extract p minimum distance in these distances, and will be corresponding to the reconstructed vector of this p distance neighbor point as described specific compensation rate, wherein p is a positive integer.
10, data predication method as claimed in claim 1 is characterized in that, described first data and second data are at least one in the following type: communication traffic amount, note amount, multimedia message amount, video flow, number of users.
11, a kind of data prediction equipment is characterized in that, comprising:
Reading device reads a plurality of first data that are associated with first period, comprises a plurality of first period of the day from 11 p.m. to 1 a.m phases wherein said first period, and described a plurality of first data are one by one corresponding to described a plurality of first period of the day from 11 p.m. to 1 a.m phases;
Trend value is determined device, based on described a plurality of first data and described a plurality of first period of the day from 11 p.m. to 1 a.m phase, determines one by one a plurality of first Trend value corresponding to described a plurality of first period of the day from 11 p.m. to 1 a.m phases;
The compensation sequence is determined device, based on described a plurality of first data and described a plurality of first Trend value, and the compensation sequence that acquisition is made of a plurality of compensation rates, wherein said a plurality of compensation rates are one by one corresponding to described a plurality of first period of the day from 11 p.m. to 1 a.m phases;
The compensation rate estimated value is determined device, based on described compensation sequence, utilizes SVMs to obtain and corresponding compensation rate estimated value in second period, is later than described first period wherein said second period in time; And
Prediction unit, based on described compensation rate estimated value, second data that prediction was associated with described second period.
12, data prediction equipment as claimed in claim 11 is characterized in that, described Trend value determines that device comprises:
The tendency equation module is based on described a plurality of first data and described a plurality of first period of the day from 11 p.m. to 1 a.m phase and obtain tendency equation; And
The first Trend value module, with described a plurality of first period of the day from 11 p.m. to 1 a.m phases respectively the described tendency equation of substitution to obtain described a plurality of first Trend value.
13, data prediction equipment as claimed in claim 12 is characterized in that, described tendency equation module utilizes least square method to obtain described tendency equation.
14, data prediction equipment as claimed in claim 12 is characterized in that, described prediction unit comprises:
The second Trend value module utilizes described tendency equation to obtain and corresponding second Trend value in described second period; And
Prediction module obtains described second data based on described compensation rate estimated value and described second Trend value.
15, data prediction equipment as claimed in claim 14 is characterized in that, described prediction module multiply by described compensation rate estimated value to obtain described second data with described second Trend value.
16, data prediction equipment as claimed in claim 11 is characterized in that, described compensation sequence determine device with described a plurality of first data respectively divided by described a plurality of first Trend value, to obtain described compensation sequence.
17, data prediction equipment as claimed in claim 11 is characterized in that, described compensation rate estimated value determines that device comprises:
First module is selected to embed dimension and time of delay, and described compensation sequence is carried out phase space reconfiguration, to obtain the reconstructed vector corresponding to each compensation rate in the described compensation sequence;
Second module is searched the neighbor point of specific compensation rate;
Three module with the training input as SVMs of the reconstructed vector of the described neighbor point that found, with the training output as SVMs of the known future value of described specific compensation rate, is trained described SVMs; And
Four module uses the input variable of the reconstructed vector of described compensation sequence as the SVMs that is obtained by described three module training, and the output of calculating described SVMs is as described compensation rate estimated value.
18, data prediction equipment as claimed in claim 17 is characterized in that, it is 7 that described first module is selected described embedding dimension, and to select described time of delay be 1.
19, data prediction equipment as claimed in claim 17, it is characterized in that, described second module is calculated the distance of the reconstructed vector of each compensation rate before the reconstructed vector of described specific compensation rate and the described specific compensation rate respectively, extract p minimum distance in these distances, and will be corresponding to the reconstructed vector of this p distance neighbor point as described specific compensation rate, wherein p is a positive integer.
20, data prediction equipment as claimed in claim 11 is characterized in that, described first data and second data are at least one in the following type: communication traffic amount, note amount, multimedia message amount, video flow, number of users.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2009101359190A CN101541016B (en) | 2009-05-06 | 2009-05-06 | Method for predicting data and equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2009101359190A CN101541016B (en) | 2009-05-06 | 2009-05-06 | Method for predicting data and equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101541016A true CN101541016A (en) | 2009-09-23 |
| CN101541016B CN101541016B (en) | 2011-01-05 |
Family
ID=41123933
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2009101359190A Active CN101541016B (en) | 2009-05-06 | 2009-05-06 | Method for predicting data and equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101541016B (en) |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102065449A (en) * | 2010-12-13 | 2011-05-18 | 哈尔滨工业大学 | Method for predicting mobile communication telephone traffic based on clustered LS-SVM (Least Squares-Support Vector Machine) |
| CN102111284A (en) * | 2009-12-28 | 2011-06-29 | 北京亿阳信通软件研究院有限公司 | Method and device for predicting telecom traffic |
| CN102570509A (en) * | 2011-12-13 | 2012-07-11 | 哈尔滨工业大学 | Base-point power off-set setting method for BLR-type AGC unit |
| CN103178990A (en) * | 2011-12-20 | 2013-06-26 | 中国移动通信集团青海有限公司 | Network device performance monitoring method and network management system |
| CN103313307A (en) * | 2013-05-31 | 2013-09-18 | 中国联合网络通信集团有限公司 | Network resource allocation method and device |
| CN103746750A (en) * | 2013-08-23 | 2014-04-23 | 西华大学 | Radio monitoring electromagnetic situation prediction system |
| CN104463323A (en) * | 2013-09-17 | 2015-03-25 | 北京邮电大学 | Data prediction method and apparatus |
| CN104715292A (en) * | 2015-03-27 | 2015-06-17 | 上海交通大学 | City short-term water consumption prediction method based on least square support vector machine model |
| CN105095613A (en) * | 2014-04-16 | 2015-11-25 | 华为技术有限公司 | Method and device for prediction based on sequential data |
| CN105120487A (en) * | 2015-09-02 | 2015-12-02 | 中国联合网络通信集团有限公司 | Forecasting method and device for business data |
| CN105580407A (en) * | 2014-12-05 | 2016-05-11 | 华为技术有限公司 | Network resource deployment method and device |
| CN106817499A (en) * | 2015-11-30 | 2017-06-09 | 华为技术有限公司 | A kind of resources for traffic dispatching method and forecast dispatching device |
| CN107171848A (en) * | 2017-05-27 | 2017-09-15 | 华为技术有限公司 | A kind of method for predicting and device |
| CN107704962A (en) * | 2017-10-11 | 2018-02-16 | 大连理工大学 | A kind of smelter steam flow interval prediction method based on imperfect time series data collection |
| CN107730887A (en) * | 2017-10-17 | 2018-02-23 | 海信集团有限公司 | Realize method and device, the readable storage medium storing program for executing of traffic flow forecasting |
| CN109254550A (en) * | 2018-09-03 | 2019-01-22 | 深圳市智物联网络有限公司 | A kind of data processing method and processing equipment |
| CN109302719A (en) * | 2018-11-05 | 2019-02-01 | 北京市天元网络技术股份有限公司 | A kind of LTE cell capacity prediction analysis method and device |
| CN110381515A (en) * | 2019-08-12 | 2019-10-25 | 北京互联无界科技有限公司 | Based on the method for closing merotype realization subzone network floating resources index prediction |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI581591B (en) * | 2015-07-07 | 2017-05-01 | Chunghwa Telecom Co Ltd | Data usage forecasting system, method and computer program product |
-
2009
- 2009-05-06 CN CN2009101359190A patent/CN101541016B/en active Active
Cited By (27)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102111284A (en) * | 2009-12-28 | 2011-06-29 | 北京亿阳信通软件研究院有限公司 | Method and device for predicting telecom traffic |
| CN102065449A (en) * | 2010-12-13 | 2011-05-18 | 哈尔滨工业大学 | Method for predicting mobile communication telephone traffic based on clustered LS-SVM (Least Squares-Support Vector Machine) |
| CN102570509A (en) * | 2011-12-13 | 2012-07-11 | 哈尔滨工业大学 | Base-point power off-set setting method for BLR-type AGC unit |
| CN103178990A (en) * | 2011-12-20 | 2013-06-26 | 中国移动通信集团青海有限公司 | Network device performance monitoring method and network management system |
| CN103313307B (en) * | 2013-05-31 | 2016-03-02 | 中国联合网络通信集团有限公司 | Network resource configuration method and device |
| CN103313307A (en) * | 2013-05-31 | 2013-09-18 | 中国联合网络通信集团有限公司 | Network resource allocation method and device |
| CN103746750A (en) * | 2013-08-23 | 2014-04-23 | 西华大学 | Radio monitoring electromagnetic situation prediction system |
| CN103746750B (en) * | 2013-08-23 | 2016-06-01 | 西华大学 | The pre-examining system of radio monitoring Electromagnetic Situation |
| CN104463323A (en) * | 2013-09-17 | 2015-03-25 | 北京邮电大学 | Data prediction method and apparatus |
| CN105095613B (en) * | 2014-04-16 | 2018-05-18 | 华为技术有限公司 | A kind of method and device predicted based on sequence data |
| CN105095613A (en) * | 2014-04-16 | 2015-11-25 | 华为技术有限公司 | Method and device for prediction based on sequential data |
| CN105580407B (en) * | 2014-12-05 | 2019-09-13 | 华为技术有限公司 | A kind of Internet resources dispositions method and equipment |
| CN105580407A (en) * | 2014-12-05 | 2016-05-11 | 华为技术有限公司 | Network resource deployment method and device |
| WO2016086406A1 (en) * | 2014-12-05 | 2016-06-09 | 华为技术有限公司 | Network resource deployment method and device |
| CN104715292A (en) * | 2015-03-27 | 2015-06-17 | 上海交通大学 | City short-term water consumption prediction method based on least square support vector machine model |
| CN105120487A (en) * | 2015-09-02 | 2015-12-02 | 中国联合网络通信集团有限公司 | Forecasting method and device for business data |
| CN106817499A (en) * | 2015-11-30 | 2017-06-09 | 华为技术有限公司 | A kind of resources for traffic dispatching method and forecast dispatching device |
| CN107171848A (en) * | 2017-05-27 | 2017-09-15 | 华为技术有限公司 | A kind of method for predicting and device |
| CN107171848B (en) * | 2017-05-27 | 2020-07-07 | 华为技术有限公司 | Flow prediction method and device |
| CN107704962A (en) * | 2017-10-11 | 2018-02-16 | 大连理工大学 | A kind of smelter steam flow interval prediction method based on imperfect time series data collection |
| CN107704962B (en) * | 2017-10-11 | 2021-03-26 | 大连理工大学 | Steam flow interval prediction method based on incomplete training data set |
| CN107730887A (en) * | 2017-10-17 | 2018-02-23 | 海信集团有限公司 | Realize method and device, the readable storage medium storing program for executing of traffic flow forecasting |
| CN109254550A (en) * | 2018-09-03 | 2019-01-22 | 深圳市智物联网络有限公司 | A kind of data processing method and processing equipment |
| CN109302719A (en) * | 2018-11-05 | 2019-02-01 | 北京市天元网络技术股份有限公司 | A kind of LTE cell capacity prediction analysis method and device |
| CN109302719B (en) * | 2018-11-05 | 2022-02-08 | 北京市天元网络技术股份有限公司 | LTE cell capacity prediction analysis method and device |
| CN110381515A (en) * | 2019-08-12 | 2019-10-25 | 北京互联无界科技有限公司 | Based on the method for closing merotype realization subzone network floating resources index prediction |
| CN110381515B (en) * | 2019-08-12 | 2022-04-12 | 桔帧科技(江苏)有限公司 | Method for realizing cell network flow resource index prediction based on combination and division mode |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101541016B (en) | 2011-01-05 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN101541016B (en) | Method for predicting data and equipment | |
| US11410046B2 (en) | Learning-based service migration in mobile edge computing | |
| Zhang et al. | Citywide cellular traffic prediction based on densely connected convolutional neural networks | |
| CN101541030B (en) | Method for predicting data based on support vector machine and equipment thereof | |
| CN101800995B (en) | Method and equipment for determining cell relationship, and method and equipment for planning frequency | |
| CN105050170A (en) | Base station dormancy method based on flow prediction in heterogeneous network | |
| CN105531746A (en) | Management of data collected for traffic analysis | |
| CN101706888B (en) | Method for predicting travel time | |
| CN102377494A (en) | Frequency spectrum sensing method, device and system | |
| CN112203228B (en) | Double-layer cellular Internet of vehicles frequency spectrum allocation method based on vehicle number prediction | |
| Dangi et al. | 5G network traffic control: a temporal analysis and forecasting of cumulative network activity using machine learning and deep learning technologies | |
| CN105281959B (en) | A kind of mobile communication subscriber attrition prediction method based on hidden Markov model | |
| CN104254083A (en) | Method and device for predicting business hot spots | |
| Kirmaz et al. | Mobile network traffic forecasting using artificial neural networks | |
| Li et al. | GM-PAB: a grid-based energy saving scheme with predicted traffic load guidance for cellular networks | |
| CN117056478A (en) | Control method and device for electric power customer service system | |
| CN106162671A (en) | A kind of base station deployment location determining method and device | |
| Schoenen et al. | First survey results of quantified user behavior in user-in-the-loop scenarios for sustainable wireless networks | |
| CN112954726B (en) | Method, system, equipment and medium for predicting holiday wireless flow based on correlation clustering hybrid algorithm model | |
| CN114726463B (en) | Method and device for predicting space-time distribution of mobile communication users based on neural network | |
| Xiao et al. | 5G construction efficiency enhancement based on LSTM algorithm | |
| Teixeira et al. | Model-free predictor of signal-to-noise ratios for mobile communications systems | |
| CN112423327B (en) | Capacity prediction method and device and storage medium | |
| Nekrasova et al. | Forecasting of investments into wireless telecommunication systems | |
| CN102724687B (en) | Method for acquiring wireless network utilization rate and network management equipment |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant |