CN105472631A - Service data quantity and/or resource data quantity prediction method and prediction system - Google Patents

Abstract

The invention provides a service data quantity and/or resource data quantity prediction method and prediction system. The method comprises the following steps: establishing a service data quantity and/or resource data quantity original data set; carrying out dimension reduction preprocessing on the original data set to obtain a preprocessed data set; carrying out initial cluster processing on the preprocessed data set to obtain an initial cluster data set; carrying out at least one accurate clustering processing on the original data set to obtain an accurate clustering data set according to the initial cluster data set; determining a prediction model according to the accurate clustering data set; and obtaining service and/or resource expected data quantity according to the prediction model. Through the technical scheme above, the problem that an existing data analysis mode cannot be suitable for increasingly-rising service data quantity and/or resource data quantity is solved. The prediction method can carry out analysis and predication on the increasingly-rising service data quantity and/or resource data quantity, and the prediction result is scientific and accurate.

Description

The Forecasting Methodology of a kind of business datum amount and/or resource data amount and prognoses system

Technical field

The present invention relates to mobile communication technology field, particularly relate to Forecasting Methodology and the prognoses system of a kind of business datum amount and/or resource data amount.

Background technology

Along with the development of LTE (LongTermEvolution, Long Term Evolution) network and the universal of 4G business, kind and the flow of data service are all greatly improved, and therefore, analyze also with regard to more sophisticated the group behavior of user.

The user behavior data produced for the communication system of agreement with LTE is completely different from traditional 2G, 3G, it comprises the information of more business and resource.Under current spectral bandwidth, LTE protocol can provide up-downgoing peak rate faster, therefore the use amount of data service increases considerably, the data volume produced in wireless side and core-network side increases with exponential form, and thus traditional data analysis tool is no longer suitable for so big data volume.

Summary of the invention

The invention provides Forecasting Methodology and the prognoses system of a kind of business datum amount and/or resource data amount, solve available data analysis mode and cannot be applicable to growing business and/or the data volume of resource, cause the problem that cannot the data volume of business and/or resource is analyzed and be predicted.

For solving the problems of the technologies described above, the invention provides the Forecasting Methodology of a kind of business datum amount and/or resource data amount, comprising:

Build the initial data set of business datum amount and/or resource data amount;

Dimensionality reduction preliminary treatment is carried out to described initial data set, obtains preprocessed data set;

First an initial clustering process is carried out to described preprocessed data set, obtain initial clustering data acquisition system, according to described initial clustering data acquisition system, more accurately clustering processing is carried out at least once to described initial data set, obtain accurate cluster data set;

According to described accurate cluster data set, determine forecast model;

According to described forecast model, obtain the anticipatory data amount of described business and/or resource.

In an embodiment of the present invention, the initial data set building business datum amount and/or resource data amount specifically comprises:

Determine business to be predicted and/or resource;

Obtain at least one historical time section, the consumption data of described business and/or the consumption data of resource, using the consumption data of described business as described business datum amount, using the consumption data of described resource as described resource data amount;

According to the consumption data of described business and/or the consumption data of resource, build initial data set.

In an embodiment of the present invention, by PCA, dimensionality reduction preliminary treatment is carried out to described initial data set, obtains preprocessed data set.

In an embodiment of the present invention, dimensionality reduction preliminary treatment is being carried out to described initial data set, before obtaining preprocessed data set, is also comprising: described initial data set is being normalized;

And/or, dimensionality reduction preliminary treatment is being carried out to described initial data set, after obtaining preprocessed data set, is also comprising: described preprocessed data set is being normalized.

In an embodiment of the present invention, first an initial clustering process is carried out to described preprocessed data set, obtain initial clustering data acquisition system, according to described initial clustering data acquisition system, again accurately clustering processing is carried out at least once to described initial data set, obtains accurate cluster data set and specifically comprise:

According to initial clustering method, first an initial clustering process is carried out to described preprocessed data set, obtain initial clustering data acquisition system;

According to described initial clustering data acquisition system, calculate initial cluster center;

According to accurate clustering method, described initial cluster center, more accurately clustering processing is carried out once to described initial data set, obtain accurate cluster data set.

In an embodiment of the present invention, according to described accurate cluster data set, determine that forecast model specifically comprises:

In described accurate cluster data set, determine elementary item, item to be predicted;

According to described elementary item, item to be predicted, determine master data amount, data volume to be predicted;

According to gradient descent method, matching is carried out to described master data amount, data volume to be predicted, determines fitting function, using described fitting function as forecast model.

In an embodiment of the present invention, according to described forecast model, the anticipatory data measurer body obtaining described business and/or resource comprises:

According to different elementary items, select different fitting functions;

According to the fitting function of described selection, described business datum amount and/or resource data amount are predicted, obtain the anticipatory data amount of described business and/or resource.

In an embodiment of the present invention, according to described forecast model, after obtaining the anticipatory data amount of described business and/or resource, also comprise:

According to described anticipatory data amount, network is optimized.

A prognoses system for business datum amount and/or resource data amount, comprising:

Build module, for building the initial data set of business datum amount and/or resource data amount;

Pretreatment module, for carrying out dimensionality reduction preliminary treatment to the initial data set of described structure module construction, obtains preprocessed data set;

Cluster module, an initial clustering process is carried out in preprocessed data set for first obtaining described pretreatment module, obtain initial clustering data acquisition system, according to described initial clustering data acquisition system, again accurately clustering processing is carried out at least once to the initial data set of described structure module construction, obtain accurate cluster data set;

Determination module, for the accurate cluster data set obtained according to described cluster module, determines forecast model;

Prediction module, for the forecast model determined according to described determination module, obtains the anticipatory data amount of described business and/or resource.

In an embodiment of the present invention, also acquisition module is comprised;

Described determination module is also for determining business to be predicted and/or resource;

Described acquisition module, for obtaining at least one historical time section, the consumption data of the business that described determination module is determined and/or the consumption data of resource, using the consumption data of described business as described business datum amount, using the consumption data of described resource as described resource data amount;

Described structure module, specifically for the consumption data of business that obtains according to described acquisition module and/or the consumption data of resource, builds initial data set.

In an embodiment of the present invention, described pretreatment module also for carrying out dimensionality reduction preliminary treatment to the initial data set of described structure module construction, before obtaining preprocessed data set, is normalized described initial data set;

And/or described pretreatment module also for carrying out dimensionality reduction preliminary treatment to the initial data set of described structure module construction, after obtaining preprocessed data set, is normalized described preprocessed data set.

In an embodiment of the present invention, also comprise computing module, described cluster module comprises initial clustering submodule, accurately cluster submodule;

Described initial clustering submodule, for according to initial clustering method, an initial clustering process is carried out in the preprocessed data set first obtained described pretreatment module, obtains initial clustering data acquisition system;

Described computing module, for the initial clustering data acquisition system obtained according to described initial clustering submodule, calculates initial cluster center;

Described accurate cluster submodule, for the initial cluster center calculated according to accurate clustering method, described computing module, then carries out once accurately clustering processing to the initial data set of described structure module construction, obtains accurate cluster data set.

In an embodiment of the present invention, described determination module also in the accurate cluster data set that obtains at described accurate cluster submodule, determines elementary item, item to be predicted; Also for according to described elementary item, item to be predicted, determine master data amount, data volume to be predicted;

Described determination module, specifically for according to gradient descent method, carries out matching to described master data amount, data volume to be predicted, determines fitting function, using described fitting function as forecast model

In an embodiment of the present invention, also comprise:

Select module, for according to different elementary items, select different fitting functions;

Described prediction module, specifically for the fitting function according to described selection model choice, is predicted described business datum amount and/or resource data amount, obtains the anticipatory data amount of described business and/or resource.

Beneficial effect of the present invention:

The invention provides Forecasting Methodology and the prognoses system of a kind of business datum amount and/or resource data amount, clustering processing is carried out after preliminary treatment is carried out to initial data, realize predicting the various dimensions of user, business and resource, thus provide reference to the optimization of LTE network resource.By clustering processing, using the initial condition of the result of initial clustering as accurate cluster, the cluster result science more that distributes is made accurately, also more to meet the incidence relation between different dimensions data resource.In most cases, the prediction effect of the forecast model in the present invention is better than the prediction effect of the direct matching of initial data, and predicated error reduces more than 10%, and some resource can reach 25%.In addition, the present invention can embody overall feature and the effect of all initial data by low volume data, saves data resource analysis cost, and for data analysis reduces algorithm complex, the MRP that can be LTE network that predicts the outcome provides reference.The present invention is more suitable for the prediction to LTE network, i.e. the related algorithm of LTE data, realizes the analysis of the prediction to channel resource and the group behavior to user.

Accompanying drawing explanation

The flow chart of the Forecasting Methodology of the business datum amount that Fig. 1 provides for the embodiment of the present invention one and/or resource data amount;

The flow chart of the K-means algorithm that Fig. 2 provides for the embodiment of the present invention one;

The structural representation of the prognoses system of the business datum amount that Fig. 3 provides for the embodiment of the present invention two and/or resource data amount;

The flow chart of the Forecasting Methodology of the business datum amount that Fig. 4 provides for the embodiment of the present invention three and/or resource data amount;

The partial data set chosen from sample data that Fig. 5 provides for the embodiment of the present invention three;

The cluster data set by obtaining after clustering processing that Fig. 6 provides for the embodiment of the present invention three;

The Forecast effect that Fig. 7 provides for the embodiment of the present invention three and the direct prediction effect comparison diagram of sample data;

The evaluation parameter of the cluster result that Fig. 8 provides for the embodiment of the present invention three;

The MAPE prediction effect block diagram to channel utilization that Fig. 9 provides for the embodiment of the present invention three;

The algorithm complex comparison diagram that Figure 10 provides for the embodiment of the present invention three;

The flow chart of the Forecasting Methodology of the business datum amount that Figure 11 provides for the embodiment of the present invention four and/or resource data amount;

The partial data set chosen from sample data that Figure 12 provides for the embodiment of the present invention four;

The cluster data set by obtaining after clustering processing that Figure 13 provides for the embodiment of the present invention four;

The Forecast effect that Figure 14 provides for the embodiment of the present invention four and the direct prediction effect comparison diagram of sample data;

The evaluation parameter of the cluster result that Figure 15 provides for the embodiment of the present invention four;

The MAPE prediction effect block diagram to channel utilization that Figure 16 provides for the embodiment of the present invention four;

The algorithm complex comparison diagram that Figure 17 provides for the embodiment of the present invention four.

Embodiment

Below in conjunction with the accompanying drawing in the embodiment of the present invention, be clearly and completely described the technical scheme in the embodiment of the present invention, obviously, described embodiment is a part of embodiment in the present invention, instead of whole embodiments.Based on the embodiment in the present invention, those of ordinary skill in the art, not making the every other embodiment obtained under creative work prerequisite, belong to the scope of protection of the invention.

By reference to the accompanying drawings the present invention is described in further detail below by embodiment.

Embodiment one:

The flow chart of the business datum amount provided for the embodiment of the present invention one as Fig. 1 and/or the Forecasting Methodology of resource data amount, as shown in Figure 1, the Forecasting Methodology of this business datum amount and/or resource data amount comprises:

S101: the initial data set building business datum amount and/or resource data amount;

Particularly, along with the development of LTE network, business datum amount and resource data amount increase substantially, and need by analyzing the data volume increased with exponential form, predict the traffic-operating period of each regional LTE network, thus complete the optimization to LTE network further.

In the present embodiment, in order to predict the traffic-operating period of LTE network, need the initial data set building business datum amount and/or resource data amount, its building mode is including, but not limited to under type:

Mode one, according to actual prediction demand, determine business to be predicted, after having determined, obtain at least one historical time section, the consumption data of this business to be predicted, its obtain manner including, but not limited to being obtained the consumption data in LTE network base station by the network equipment etc., using this consumption data as business datum amount, according to this consumption data, build initial data set;

Mode two, according to actual prediction demand, determine resource to be predicted, after having determined, obtain at least one historical time section, the consumption data of this resource to be predicted, its obtain manner including, but not limited to being obtained the consumption data in LTE network base station by the network equipment etc., using this consumption data as resource data amount, according to this consumption data, build initial data set;

Mode three, according to actual prediction demand, determine business to be predicted and resource, after having determined, obtain at least one historical time section, the consumption data of this business to be predicted and the consumption data of resource, its obtain manner is including, but not limited to obtaining the consumption data in LTE network base station by the network equipment etc., using the consumption data of business as business datum amount, using the consumption data of resource as resource data amount, according to this consumption data, build initial data set.

By above-mentioned several mode, this initial data set is included at least one historical time section, according to the consumption data of business to be predicted and/or resource acquisition, if the number of the business to be predicted determined and/or resource is m, the number of historical time section is N number of, then this raw data set is combined into the matrix of a N*m, and wherein, m, N are positive integer.

In technique scheme, at least one historical time section above-mentioned, preferably, the granularity of each historical time section is all identical, if granularity is 1 hour, namely the duration of each historical time section is 1 hour, in addition, this historical time section can be selected according to the actual requirements, such as, multiple historical time sections that each week is interior on the same day, the multiple historical time sections in continuous one week every day, or continuous three week every day, 8 a.m. was to the multiple historical time sections etc. at 8 in evening.

S102: carry out dimensionality reduction preliminary treatment to initial data set, obtains preprocessed data set;

Particularly, when after the initial data set building business datum amount and/or resource data amount, dimensionality reduction preliminary treatment is carried out to this initial data set, obtains preprocessed data set.

In the present embodiment, pass through PCA, carry out dimensionality reduction preliminary treatment to initial data set, obtain preprocessed data set, this PCA mainly utilizes the thought of dimensionality reduction, be a few generalized variable by multiple variables transformations, i.e. principal component, wherein each principal component is the linear combination of original variable, uncorrelated mutually between each principal component, thus these principal components can embody most characteristic informations of original variable, and contained information non-overlapping copies.

Record for such scheme, if the number of the business to be predicted determined and/or resource is m, then adopts this m variable to carry out descriptive study object, use Z respectively ₁, Z ₂..., Z _mrepresent, it is Z=(Z that the m of this m variable formation ties up random vector ₁, Z ₂..., Z _m) t, if the average of random vector Z is μ, covariance matrix is Σ, carries out as shown in the formula the linear change shown in (1.1) to random vector Z, and consider the linear combination of original variable, can obtain principal component is incoherent linear combination Y ₁, Y ₂..., Y _k, wherein, m, k are positive integer, and k < m.

$\left\{\begin{array}{c}{Y}_1\left(x\right)={\mathrm{\&mu;}}_{11}{Z}_1\left(x\right)+{\mathrm{\&mu;}}_{12}{Z}_2\left(x\right)+...+{\mathrm{\&mu;}}_{\mathrm{lm}}{Z}_m\left(x\right)\\ Y_2\left(x\right)={\mathrm{\&mu;}}_{21}{Z}_1\left(x\right)+{\mathrm{\&mu;}}_{22}{Z}_2\left(x\right)+...+{\mathrm{\&mu;}}_{2m}{Z}_m\left(x\right)\\ ...\\ Y_k\left(x\right)={\mathrm{\&mu;}}_{k1}{Z}_1\left(x\right)+{\mathrm{\&mu;}}_{k2}{Z}_2\left(x\right)+...+{\mathrm{\&mu;}}_{\mathrm{km}}{Z}_m\left(x\right)\end{array}\right.$ Formula (1.1)

In formula (1.1), the characteristic vector corresponding to the characteristic value of the covariance matrix Σ that average μ is random vector Z, Y ₁(x), Y ₂(x) ..., Y _kx () is the principal component of original variable after linear combination.After PCA, obtain preprocessed data set, this preprocessed data set is the matrix of a N*k, initial data set by m dimension becomes the preprocessed data set of k dimension, the generalized variable tieed up by k embodies the characteristic information of the original variable of m dimension, wherein, k, N are positive integer.

In the present embodiment, in order to simplify calculating, and it is more accurate to make to predict the outcome, dimensionality reduction preliminary treatment is being carried out to initial data set, before obtaining preprocessed data set, be normalized each consumption data in this initial data set, its normalization formula is as shown in the formula shown in (1.2):

$Z\left(x\right)=\frac{x-\stackrel{\&OverBar;}{x}}{s\left(x\right)}=\frac{x-\stackrel{\&OverBar;}{x}}{\sqrt{\frac{{(x-\stackrel{\&OverBar;}{x})}^2}{N}}}$ Formula (1.2)

Wherein, x represents a certain business or the resource concrete consumption data amount in a historical time section in initial data set, represent this business or the mean value of resource in a historical time section, N represents the number of historical time section, and Z (x) represents the consumption data after normalization, after normalized, initial data set is still the matrix of a N*m, and wherein, m, N are positive integer.

And/or,

Carrying out dimensionality reduction preliminary treatment to initial data set, after obtaining preprocessed data set, be also normalized the data in preprocessed data set, its normalization formula is as shown in the formula shown in (1.3):

$Z\left(x\right)=\frac{Y\left(x\right)-\min Y\left(x\right)}{\max Y\left(x\right)-\min Y\left(x\right)}$ Formula (1.3)

Wherein, Y (x) represents the data in preprocessed data set, minY (x), maxY (x) represent minimum value, the maximum of all data in preprocessed data set respectively, and Z (x) represents the data after normalization.

S103: first carry out an initial clustering process to preprocessed data set, obtain initial clustering data acquisition system, according to initial clustering data acquisition system, then carries out at least once accurately clustering processing to initial data set, obtains accurate cluster data set;

Particularly, after obtaining preprocessed data set, first can carry out an initial clustering process to preprocessed data set, obtain initial clustering data acquisition system, according to this initial clustering data acquisition system, accurately clustering processing is carried out at least once to initial data set, obtains accurate cluster data set.

In technique scheme, the object of initial clustering process is to carry out initial analysis to data, and at least one times accurately clustering processing initial accurately cluster condition is provided, the result making accurate clustering processing science and accurately more, such as, in initial clustering process, first time cluster is carried out in pretreated initial data set, the mean value of each class data after calculating first time cluster, using the initial cluster center of this mean value as accurate clustering processing at least one times, according to this initial cluster center, accurately clustering processing is carried out at least one times to initial data, thus accurate cluster is carried out to initial data.

In the present embodiment, for an initial clustering process, once accurately clustering processing is described, according to initial clustering method, an initial clustering process is carried out to preprocessed data set, this preprocessed data set can for the preprocessed data set after being normalized, also for the preprocessed data set be not normalized, initial clustering data acquisition system can be obtained.

For this initial clustering method, if can carry out fast data, simply, cluster roughly, and obtain cluster centre, any algorithm, it is including, but not limited to Canopy algorithm, this Canopy algorithm is a kind of clustering method simply, fast but not too accurately, therefore can be used as aided algorithm.The algorithm principle of this Canopy algorithm is that each object is represented by the point of in multidimensional feature space, adopt a quick approximate distance tolerance and two distance threshold T1>T2 (T1 > 0, T2 > 0) to carry out clustering processing to data, its algorithm flow is:

(1) by data set vectorization, put into internal memory after obtaining a data point set, select two distance threshold T1 and T2, wherein the value of T1>T2, T1 and T2 can be determined with cross check;

(2) from data point set appoint get 1 P, (Canopy is there is not if current by the distance between the low method that assesses the cost quick calculation level P and all Canopy (class in Canopy cluster process here), then using a P as a Canopy), if fruit dot P and certain Canopy distance are within T1, then a P is joined this Canopy;

(3) as fruit dot P once with the distance of certain Canopy within T2, then need a some P to delete from data point set, this step thinks that a P is now enough near with this Canopy, and therefore it cannot do the center of other Canopy again;

(4) step (2), (3) are repeated, until data point set is empty end.

When by an initial clustering method, when obtaining initial clustering data acquisition system, then calculate the mean value of each class data in this initial clustering data acquisition system, using the initial cluster center of this mean value as accurate clustering processing.

According to accurate clustering method, initial cluster center, accurately clustering processing is carried out once to initial data set, obtains accurate cluster data set.For this accurate clustering method, only need to carry out accurate cluster to data, any algorithm, it is including, but not limited to K-means algorithm, as the flow chart of the K-means algorithm that Fig. 2 provides for the embodiment of the present invention one, this K-means algorithm is hard clustering algorithm, is typically based on the representative of the target function clustering method of prototype, it is data point to certain distance of prototype as the target function of optimization, utilizes function to ask the method for extreme value to obtain the regulation rule of interative computation.K-means algorithm is using Euclidean distance as similarity measure, and it is the optimal classification asking corresponding a certain initial cluster center, makes evaluation index minimum.The algorithm principle of this K-means algorithm is typically based on the clustering algorithm of distance, and adopt distance as the evaluation index of similitude, namely think that the distance of two objects is nearer, its similarity is larger, this distance.K-means algorithm to be thought bunch by forming apart from close object, therefore compact and independently bunch as final goal using obtaining.The iteration many times passed through, and each iteration is to the correction of central point, finally reaches convergence, realizes the polymerization classification of data.It should be noted that, after have passed through K-means algorithm cluster, analyze cluster result, if certain one dimension variable change minimum (excursion is no more than 5%), then illustrate that this one dimension variable does not have very large meaning in cluster result, this one dimension variable should be deleted.

After an above-mentioned initial clustering process, once accurate clustering processing, obtain accurate cluster data set, if the business of the m in initial data set and/or resource are divided into h class by this accurate cluster data set, then this accurate cluster data set is the matrix of a h*m, wherein, h, m are positive integer, and h < < N.

S104: according to accurate cluster data set, determine forecast model;

Particularly, after obtaining accurate cluster data set, forecast model can be determined.

In the present embodiment, in accurate cluster data set, determine elementary item, item to be predicted, according to this elementary item, item to be predicted, determine master data amount, data volume to be predicted, according to gradient descent method, matching is carried out to master data amount, data volume to be predicted, determine fitting function, using this fitting function as forecast model.It should be noted that, this forecast model can be one-dimensional forecast model, and namely determined elementary item has one, and can be also multidimensional prediction model, namely determined elementary item has two at least.

In technique scheme, this elementary item is business or resource, this master data amount is this business or the cluster centre of resource in all classes in accurate cluster set, this item to be predicted is business or resource, this data volume to be predicted is this business or the cluster centre of resource in all classes in accurate cluster set, and this fitting function really stable condition comprises:

If situation one builds initial data set for business, then accurate cluster data set is equally for business, now, elementary item, item to be predicted are business, namely elementary item is basic service, item to be predicted is business to be predicted, according to basic service, determine the cluster centre of this basic service in all classes, according to business to be predicted, determine the cluster centre of this business to be predicted in all classes, according to gradient descent method, matching is carried out to these two cluster centres, determines fitting function, using this fitting function as forecast model;

If situation two is for the set of resource construction initial data, then accurate cluster data set is equally for resource, now, elementary item, item to be predicted are resource, namely elementary item is basic resources, item to be predicted is resource to be predicted, according to basic resources, determine the cluster centre of this basic resources in all classes, according to resource to be predicted, determine the cluster centre of this resource to be predicted in all classes, according to gradient descent method, matching is carried out to these two cluster centres, determines fitting function, using this fitting function as forecast model;

Situation three, if for business and the set of resource construction initial data, then accurate cluster data set is equally for business and resource, now, elementary item both can be business, also can be resource, Xiang Jike to be predicted thinks business, also can be resource, namely when elementary item is basic service, item to be predicted can be business to be predicted, also can be resource to be predicted, when elementary item is basic resources, item to be predicted can be business to be predicted, also can be resource to be predicted, according to basic service or basic resources, determine this basic service or the cluster centre of basic resources in all classes, according to business to be predicted or resource to be predicted, determine this business to be predicted or the cluster centre of resource to be predicted in all classes, according to gradient descent method, matching is carried out to these two cluster centres, determine fitting function, using this fitting function as forecast model.

In technique scheme, for same Xiang Eryan to be predicted, the elementary item of selection is different, and its fitting function determined is also different, can according to the forecast assessment parameter of different fitting function, select most suitable fitting function, the forecast assessment parameter of the fitting function namely selected is less, then its predict the outcome better, if the forecast assessment parameter of a certain fitting function is minimum, elementary item so in this fitting function is best elementary item, and by this fitting function, it predicts the outcome more accurate.

S105: according to forecast model, obtains the anticipatory data amount of business and/or resource.

Particularly, after forecast model has been determined, can according to this forecast model, obtain the anticipatory data amount of business in required predicted time section and/or resource.

In the present embodiment, according to different elementary items, select different fitting functions, according to selected fitting function, business datum amount and/or resource data amount are predicted, obtain the anticipatory data amount of business and/or resource in required predicted time section.

In the present embodiment, when according to forecast model, after obtaining the anticipatory data amount of business and/or resource, can also according to this anticipatory data amount, for network carry out planning, optimize, dilatation etc. provides certain guidance, thus improves network to the bearing capacity of the data service become increasingly abundant.

After preliminary treatment is carried out to initial data, carry out clustering processing, using the initial condition of the result of initial clustering as accurate cluster, make the cluster result science more that distributes accurately, also more meet the incidence relation between different dimensions data resource.In most cases, the prediction effect of the forecast model in the present invention is better than the prediction effect of the direct matching of initial data, and predicated error reduces more than 10%, and some resource can reach 25%.In addition, the present invention can embody overall feature and the effect of all initial data by low volume data, saves data resource analysis cost, and for data analysis reduces algorithm complex, the MRP that can be LTE network that predicts the outcome provides reference.

Embodiment two:

The structural representation of the business datum amount provided for the embodiment of the present invention two as Fig. 3 and/or the prognoses system of resource data amount, as shown in Figure 3, this prognoses system comprises structure module 1, pretreatment module 2, cluster module 3, determination module 4 and prediction module 5, build module 1 for building the initial data set of business datum amount and/or resource data amount, pretreatment module 2 is for carrying out dimensionality reduction preliminary treatment to the initial data set building module 1 structure, obtain preprocessed data set, cluster module 3 carries out an initial clustering process for the preprocessed data set first obtained pretreatment module 2, obtain initial clustering data acquisition system, according to initial clustering data acquisition system, again accurately clustering processing is carried out at least once to the initial data set building module 1 structure, obtain accurate cluster data set, the accurate cluster data set of determination module 4 for obtaining according to cluster module 3, determine forecast model, the forecast model of prediction module 5 for determining according to determination module 4, obtain the anticipatory data amount of business and/or resource.

In technique scheme, also comprise acquisition module 6, determination module 4 is also for determining business to be predicted and/or resource, acquisition module 6 is for obtaining at least one historical time section, the consumption data of the business that determination module 4 is determined and/or the consumption data of resource, using the consumption data of business as business datum amount, using the consumption data of resource as resource data amount, build module 1 specifically for the consumption data of business that obtains according to acquisition module 6 and/or the consumption data of resource, the set of structure initial data.

In technique scheme, pretreatment module 2, by PCA, is carried out dimensionality reduction preliminary treatment to initial data set, is obtained preprocessed data set.

In technique scheme, pretreatment module 3 is also for carrying out dimensionality reduction preliminary treatment to the initial data set building module 1 structure, before obtaining preprocessed data set, initial data set is normalized, and/or, pretreatment module 3 also for carrying out dimensionality reduction preliminary treatment to the initial data set building module 1 structure, after obtaining preprocessed data set, is normalized preprocessed data set.

In technique scheme, also comprise computing module 7, cluster module 3 comprises initial clustering submodule 31, accurate cluster submodule 32, initial clustering submodule 31 is for according to initial clustering method, an initial clustering process is carried out in the preprocessed data set first obtained pretreatment module 2, obtain initial clustering data acquisition system, the initial clustering data acquisition system of computing module 7 for obtaining according to initial clustering submodule 31, calculate initial cluster center, accurate cluster submodule 32, for according to accurate clustering method, the initial cluster center that computing module 7 calculates, again accurately clustering processing is carried out once to the initial data set building module 1 structure, obtain accurate cluster data set.

In technique scheme, determination module 4 is also in the accurate cluster data set that obtains at accurate cluster submodule 32, determine elementary item, item to be predicted, also for according to elementary item, item to be predicted, determine master data amount, data volume to be predicted, determination module 4, specifically for according to gradient descent method, carries out matching to master data amount, data volume to be predicted, determine fitting function, using fitting function as forecast model.

In technique scheme, also comprise and select module 8, select module 8 for according to different elementary items, select different fitting functions, the fitting function of prediction module 9 specifically for selecting according to selection module 8, business datum amount and/or resource data amount are predicted, obtains the anticipatory data amount of business and/or resource.

In technique scheme, also comprise and optimize module 9, when prediction module 8 is according to forecast model, after obtaining the anticipatory data amount of business and/or resource, optimizing module 9 according to this anticipatory data amount, network being optimized, as adjusted etc. LTE network business and/or resource.

Embodiment three:

The flow chart of the business datum amount provided for the embodiment of the present invention three as Fig. 4 and/or the Forecasting Methodology of resource data amount, as shown in Figure 4, the Forecasting Methodology of this business datum amount and/or resource data amount comprises:

S201: according to weekly on the same day in the business that produces of multiple different time sections and resource data amount, the set of structure initial data;

Particularly, gathered the data volume of business and resource in LTE network base station by the network equipment in the present embodiment, after Preliminary screening, the business and the resource that relate to prediction have RRC (RadioResourceControl, wireless heterogeneous networks) connect number of users, up-downgoing average discharge, successful incoming call exhalation number of times, up-downgoing shared channel utilance, down control channel utilance etc., wherein, RRC connects number of users maximum active users and average active users.

These data are derive from the user in certain area in LTE existing network, resource, business three-dimensional data, its time granularity is one hour, namely the duration of time period is one hour, sample data acquisition time is two adjacent Mondays, specifically, that the N bar data of the Monday of adjacent two weeks are picked out, each data represents the business of a base station in one hour and resource consumption data volume, filter out m row business and the resource of Water demand, m is exactly the number of business and the resource that will predict, build initial data set P, P is the matrix of N*m;

$P=\left(X_{i,j}\right)=\left[\begin{array}{c}{x}_1\\ x_2\\ ...\\ x_N\end{array}\right]=\left[\begin{array}{cccc}{x}_{11}& x_{12}& ...& x_{1m}\\ x_{21}& x_{22}& ...& x_{2m}\\ ...& ...& ...& ...\\ x_{N1}& x_{N2}& ...& x_{\mathrm{Nm}}\end{array}\right]i\&Element;(1,N),j\&Element;(1,m)$

Wherein, N, m are natural numbers, x _n,mrepresent the data volume that a certain business and resource specifically consumed in a certain hour.As the partial data set chosen from sample data that Fig. 5 provides for the embodiment of the present invention three, as shown in Figure 5, in Fig. 5, every data line represents this area's base station business of a hour and resource service condition, this business or the resource concrete data volume consumed in a hour is shown in each list, and the object of its research is nine kinds of business and resource.

S202: preliminary treatment is carried out to this initial data set;

Particularly, the data volume in above-mentioned initial data set P be normalized, its normalization formula is as shown in the formula shown in (2.1):

$Z\left(x\right)=\frac{x-\stackrel{\&OverBar;}{x}}{s\left(x\right)}=\frac{x-\stackrel{\&OverBar;}{x}}{\sqrt{\frac{{(x-\stackrel{\&OverBar;}{x})}^2}{N}}}$ Formula (2.1)

Wherein, x be in initial data set P certain business or resource at the concrete numerical value of one hour internal consumption, be the mean value of the data that this business or resource consume for a hour, N is the number of multiple time period, and Z (x) represents the consumption data after normalization, after normalized, initial data set is still the matrix of a N*m, and wherein, m, N are positive integer.

After completing normalized, the m dimension business in the initial data set P after above-mentioned normalization and resource are carried out dimension-reduction treatment, and concrete grammar can use PCA, after dimensionality reduction, obtain preprocessed data set Q, Q is the matrix of N*k, k<m.

$Q=\left(Y_{i,j}\right)=\left[\begin{array}{c}{y}_1\\ y_2\\ ...\\ y_N\end{array}\right]=\left[\begin{array}{cccc}{y}_{11}& y_{12}& ...& y_{1k}\\ y_{21}& y_{22}& ...& y_{2k}\\ ...& ...& ...& ...\\ y_{N1}& y_{N2}& ...& y_{\mathrm{Nk}}\end{array}\right]i\&Element;(1,N),j\&Element;(1,k)$

Wherein, N, k are natural numbers, y _n,krepresent the data after dimensionality reduction.Each data in preprocessed data set Q be by normalization after initial data set P obtain through principal component analysis, for the first row after changing, its conversion formula is as shown in the formula shown in (2.2): $\begin{array}{c}Q(i,1)=P_1(i,1)*0.338+P_1(i,2)*0.333+P_1(i,3)*0.340+\\ P_1(i,4)*0.329+P_1(i,5)*0.176+P_1(i,6)*0.326+P_1(i,7)*0.319+\\ P_1(i,8)*0.317+P_1(i,9)*0.320\end{array}$ Formula (2.2)

S203: twice clustering processing is carried out to pretreated data acquisition system, initial data set, obtains cluster data set;

Particularly, be normalized each data in preprocessed data set Q, its normalization formula is as shown in the formula shown in (2.3):

$Z\left(x\right)=\frac{Y\left(x\right)-\min Y\left(x\right)}{\max Y\left(x\right)-\min Y\left(x\right)}$ Formula (2.3)

Wherein, Y (x) represents the data in preprocessed data set Q, minY (x), maxY (x) represent minimum value, the maximum of all data in preprocessed data set respectively, and Z (x) represents the data after normalization.

After completing normalized, preprocessed data set after normalization is divided into h class, clustering algorithm is used to process the preprocessed data set after normalization, this clustering algorithm is first through a Canopy cluster to business and resource data, using the initial cluster center of the result of Canopy cluster as second time K-means cluster, after the clustering processing of complete paired data, obtain cluster data set Q ₁, Q ₁be the matrix of a h*m, h<<N here.As the cluster data set by obtaining after clustering processing that Fig. 6 provides for the embodiment of the present invention three, as shown in Figure 6, in figure 6, h is 11, namely 11 class results are represented, every data line represents cluster centre, the business namely comprised in this class and the mean value of resource data of business or resource in each class after cluster.

S204: according to cluster data set, determine fitting function;

Particularly, at Q ₁in choose elementary item and data volume, item to be predicted and data volume thereof, adopt gradient descent method matching, carry out curve fitting to two data volumes chosen, fitting result is function y=f (x _n), n ∈ [1,8], wherein, y is item to be predicted, x _nelementary item, elementary item x _nbe used to business or the resource of predicting item y to be predicted, according to different elementary items, its fitting function is also different.Such as choosing average RRC connection number of users is elementary item, and choosing descending average discharge is item to be predicted, then fitting function is y _descending=f (x ₁), y is descending average discharge, x ₁that average RRC connects number of users.

S205: according to the time period of required prediction, prediction business and resource data amount.

Particularly, according to the fitting function obtained, namely in measurable required time section, business and resource data amount, the anticipatory data amount of business and resource in Zhou Zhouyi as a certain in future.

By above-mentioned Forecasting Methodology, according to the business of some day the last week and change in resources situation and a related service and change in resources situation, prediction needs the business of prediction and the consumption trend of resource.

For the effect of the present embodiment, further illustrate as follows:

For different elementary item x _n, to function y=f (x _n), n ∈ [1,8] computational prediction evaluate parameter, this forecast assessment parameter comprises MSE (MeanSquaredError, mean square error), MAPE (MeanAbsolutePercentageError, average percent absolute error), ME (meanerror, mean error).Forecast assessment parameter values is less, predicts the outcome better, and according to the best elementary item of forecast assessment parameter choose, in the present embodiment, the best elementary item of descending average discharge is that average RRC connects number of users, then fitting function y _descending=f (x ₁) be exactly predict the outcome.The computing formula of MSE, MAPE and ME is as shown in the formula shown in (2.4), (2.5), (2.6):

$\mathrm{MSE}=\frac{\mathrm{\&Sigma;}{e}^2}{n-1}=\frac{\mathrm{\&Sigma;}{(x-x\text{'})}^2}{n-1}$ Formula (2.4)

$\mathrm{MAPE}=\frac{1}{n}\mathrm{\&Sigma;}\left|\mathrm{PE}\right|=\frac{1}{n}\mathrm{\&Sigma;}\frac{\left|(x-x\text{'})\right|}{x}$ Formula (2.5)

$\mathrm{ME}=\stackrel{\&OverBar;}{e}=\frac{1}{n}{\mathrm{\&Sigma;}}_{i=1}^n{e}_i=\frac{1}{n}{\mathrm{\&Sigma;}}_{i=1}^n(x_i-x_i\text{'})$ Formula (2.6)

The Forecast effect provided for the embodiment of the present invention three as Fig. 7 and the direct prediction effect comparison diagram of sample data, as shown in Figure 7, during one-dimensional prediction, elementary item is that maximum RRC connects number of users, when item to be predicted is average RRC connection number of users, loose point in Fig. 7 is the scatter diagram of sample data, and darker curve is the function of matching after cluster analysis, and lighter curve is the function of the direct matching of sample data.The parameter can predicted as required during multivariate joint probability prediction is selected to arrange, and is predicted as example here with one-dimensional.

In order to make cluster result more directly perceived, as the evaluation parameter of the cluster result that Fig. 8 provides for the embodiment of the present invention three, as shown in Figure 8, in Fig. 8, every a line is exactly the value of a kind of evaluate parameter of corresponding a certain business and resource, and a kind of forecast assessment parameter of item to be predicted is shown in each list.

For the ease of observing, as the MAPE prediction effect block diagram to channel utilization that Fig. 9 provides for the embodiment of the present invention three, as shown in Figure 9, PDCCH-UTI is down control channel utilance, and PDSCH-UTI is DSCH Downlink Shared Channel utilance, and MAPE is error parameter of measurement, its numerical value is higher, represent that prediction is more inaccurate, as can be seen from the figure, the effect of Forecast is better than the effect that data are directly predicted.

Simultaneously, in order to show the optimization of algorithm to data processing aspect, this give the result block diagram that algorithm complex is taken the logarithm the end of for 10, as the algorithm complex comparison diagram that Figure 10 provides for the embodiment of the present invention three, as shown in Figure 10, the fit approach adopted in cluster matching is gradient descent method, during one-dimensional prediction, algorithm complex is N*k*a, N is sample data number, k is the business of research and the number of resource class, and a is the iterative computation number of times of gradient descent method.During multidimensional prediction, the elementary item number of complexity to be NW*K, W be multidimensional prediction.As can be seen from Figure 10, in algorithm complex, the calculating after clustering algorithm process will obviously be better than data and directly process, in associated prediction, input prediction dimension is more, complexity optimized more obvious, and this LTE for future is large, and data research has very great meaning.

Embodiment four:

The flow chart of the business datum amount provided for the embodiment of the present invention four as Figure 11 and/or the Forecasting Methodology of resource data amount, as shown in figure 11, the Forecasting Methodology of this business datum amount and/or resource data amount comprises:

S301: according to weekly on the same day in the business that produces of multiple different time sections and resource data amount, the set of structure initial data;

Particularly, after Preliminary screening, the business and the resource that relate to prediction have average user number, forward control channel average, preceding paragraph Traffic Channel average, Reverse Access Channel average, and up-downgoing flow, reverse CE takies average etc.This Data Source is the user in certain area in 3G existing network, resource, business three-dimensional data, its time granularity is one hour, namely the duration of time period is one hour, and sample data acquisition time is on July 2nd, 2012, and correction data acquisition time is on July 9th, 2012.Specifically, that the N bar data in 2 days July in 2012 are picked out, each data represents the business of a base station in one hour and resource consumption data volume, filter out m row business and the resource of Water demand, m is exactly the number of business and the resource that will predict, build initial data set P, P is the matrix of N*m;

$P=\left(X_{i,j}\right)=\left[\begin{array}{c}{x}_1\\ x_2\\ ...\\ x_N\end{array}\right]=\left[\begin{array}{cccc}{x}_{11}& x_{12}& ...& x_{1m}\\ x_{21}& x_{22}& ...& x_{2m}\\ ...& ...& ...& ...\\ x_{N1}& x_{N2}& ...& x_{\mathrm{Nm}}\end{array}\right]i\&Element;(1,N),j\&Element;(1,m)$

Wherein, N, m are natural numbers, x _n,mrepresent the data volume that a certain business and resource specifically consumed in a certain hour.As the partial data set chosen from sample data that Figure 12 provides for the embodiment of the present invention four, as shown in figure 12, in Figure 12, every data line represents this area's base station business of a hour and resource service condition, such business or the resource concrete numerical value consumed in one hour is shown in each list, and the object of its research is seven kinds of business and resource.

S302: preliminary treatment is carried out to this initial data set;

Particularly, the data volume in above-mentioned initial data set P be normalized, its normalization formula is as shown in the formula shown in (3.1):

$Z\left(x\right)=\frac{x-\stackrel{\&OverBar;}{x}}{s\left(x\right)}=\frac{x-\stackrel{\&OverBar;}{x}}{\sqrt{\frac{{(x-\stackrel{\&OverBar;}{x})}^2}{N}}}$ Formula (2.1)

Wherein, x be in initial data set P certain business or resource at the concrete numerical value of one hour internal consumption, be the mean value of the data that this business or resource consume for a hour, N is the number of multiple time period, and Z (x) represents the consumption data after normalization, after normalized, initial data set is still the matrix of a N*m, and wherein, m, N are positive integer.

After completing normalized, the m dimension business in the initial data set P after above-mentioned normalization and resource are carried out dimension-reduction treatment, and concrete grammar can use PCA, after dimensionality reduction, obtain preprocessed data set Q, Q is the matrix of N*k, k<m.

$Q=\left(Y_{i,j}\right)=\left[\begin{array}{c}{y}_1\\ y_2\\ ...\\ y_N\end{array}\right]=\left[\begin{array}{cccc}{y}_{11}& y_{12}& ...& y_{1k}\\ y_{21}& y_{22}& ...& y_{2k}\\ ...& ...& ...& ...\\ y_{N1}& y_{N2}& ...& y_{\mathrm{Nk}}\end{array}\right]i\&Element;(1,N),j\&Element;(1,k)$

Wherein, N, k are natural numbers, y _n,krepresent the data after dimensionality reduction.Each data in preprocessed data set Q be by normalization after initial data set P obtain through principal component analysis, for the first row after changing, its conversion formula is as shown in the formula shown in (3.2):

$\begin{array}{c}Q(i,1)=P_1(i,1)*0.338+P_1(i,2)*0.333+P_1(i,3)*0.340+P_1(i,4)*0.329+\\ P_1(i,5)*0.176+P_1(i,6)*0.326+P_1(i,7)*0.319\end{array}$ Formula (3.2)

S303: twice clustering processing is carried out to pretreated data acquisition system, initial data set, obtains cluster data set;

Particularly, be normalized each data in preprocessed data set Q, its normalization formula is as shown in the formula shown in (3.3):

$Z\left(x\right)=\frac{Y\left(x\right)-\min Y\left(x\right)}{\max Y\left(x\right)-\min Y\left(x\right)}$ Formula (3.3)

Wherein, Y (x) represents the data in preprocessed data set Q, minY (x), maxY (x) represent minimum value, the maximum of all data in preprocessed data set respectively, and Z (x) represents the data after normalization.

After completing normalized, preprocessed data set after normalization is divided into h class, clustering algorithm is used to process the preprocessed data set after normalization, this clustering algorithm is first through a Canopy cluster to business and resource data, using the initial cluster center of the result of Canopy cluster as second time K-means cluster, after the clustering processing of complete paired data, obtain cluster data set Q ₁, Q ₁be the matrix of a h*m, h<<N here.As the cluster data set by obtaining after clustering processing that Figure 13 provides for the embodiment of the present invention four, as shown in figure 13, in fig. 13, h is 10, namely 10 class results are represented, every data line represents cluster centre, the business namely comprised in this class and the mean value of resource data of business or resource in each class after cluster.

S204: according to cluster data set, determine fitting function;

Particularly, at Q ₁in choose elementary item and data volume, item to be predicted and data volume thereof, adopt gradient descent method matching, carry out curve fitting to two data volumes chosen, fitting result is function y=f (x _n), n ∈ [1,6], wherein, y is item to be predicted, x _nelementary item, elementary item x _nbe used to business or the resource of predicting item y to be predicted, according to different elementary items, its fitting function is also different.Such as choosing average user number is elementary item, and choosing reverse CE, to take average be item to be predicted, then fitting function is y _{reverse CE}=f (x ₁), y is that reverse CE takies average, x ₁it is average user number.

S205: according to the time period of required prediction, prediction business and resource data amount.

Particularly, according to the fitting function obtained, namely in measurable required time section, business and resource data amount, the anticipatory data amount of business and resource in Zhou Zhouyi as a certain in future.

For the effect of the present embodiment, further illustrate as follows:

For different elementary item x _n, to function y=f (x _n), n ∈ [1,6] computational prediction evaluate parameter, this forecast assessment parameter comprises MSE (MeanSquaredError, mean square error), MAPE (MeanAbsolutePercentageError, average percent absolute error), ME (meanerror, mean error).Forecast assessment parameter values is less, predicts the outcome better, and according to the best elementary item of forecast assessment parameter choose, in the present embodiment, the best elementary item that reverse CE takies average is average user number, then fitting function y _{reverse CE}=f (x ₁) be exactly predict the outcome.The computing formula of MSE, MAPE and ME is as shown in the formula shown in (3.4), (3.5), (3.6):

$\mathrm{MSE}=\frac{\mathrm{\&Sigma;}{e}^2}{n-1}=\frac{\mathrm{\&Sigma;}{(x-x\text{'})}^2}{n-1}$ Formula (3.4)

$\mathrm{MAPE}=\frac{1}{n}\mathrm{\&Sigma;}\left|\mathrm{PE}\right|=\frac{1}{n}\mathrm{\&Sigma;}\frac{\left|(x-x\text{'})\right|}{x}$ Formula (3.5)

$\mathrm{ME}=\stackrel{\&OverBar;}{e}=\frac{1}{n}{\mathrm{\&Sigma;}}_{i=1}^n{e}_i=\frac{1}{n}{\mathrm{\&Sigma;}}_{i=1}^n(x_i-x_i\text{'})$ Formula (3.6)

The Forecast effect provided for the embodiment of the present invention four as Figure 14 and the direct prediction effect comparison diagram of sample data, as shown in figure 14, during one-dimensional prediction, elementary item is average user number, item to be predicted is that reverse CE is when taking average, loose point in Figure 14 is the scatter diagram of sample data, and darker curve is the function of matching after cluster analysis, and lighter curve is the function of the direct matching of sample data.The parameter can predicted as required during multivariate joint probability prediction is selected to arrange, and is predicted as example here with one-dimensional.

In order to make cluster result more directly perceived, as the evaluation parameter of the cluster result that Figure 15 provides for the embodiment of the present invention four, as shown in figure 15, in Figure 15, every a line is exactly the value of a kind of evaluate parameter of corresponding a certain business and resource, and a kind of forecast assessment parameter of item to be predicted is shown in each list.

For the ease of observing, as the MAPE prediction effect block diagram to channel utilization that Figure 16 provides for the embodiment of the present invention four, as shown in figure 16, MAPE is error parameter of measurement, its numerical value is higher, represent that prediction is more inaccurate, as can be seen from the figure, the effect of Forecast is better than the effect that data are directly predicted.

Simultaneously, in order to show the optimization of algorithm to data processing aspect, this give the result block diagram that algorithm complex is taken the logarithm the end of for 10, as the algorithm complex comparison diagram that Figure 17 provides for the embodiment of the present invention four, as shown in figure 17, the fit approach adopted in cluster matching is gradient descent method, during one-dimensional prediction, algorithm complex is N*k*a, N is sample data number, k is the business of research and the number of resource class, and a is the iterative computation number of times of gradient descent method.During multidimensional prediction, complexity is N ^w* K, W are the elementary item numbers of multidimensional prediction.As can be seen from Figure 17, in algorithm complex, the calculating after clustering algorithm process will obviously be better than data and directly process, in associated prediction, input prediction dimension is more, complexity optimized more obvious, and this LTE for future is large, and data research has very great meaning.

Above content is in conjunction with concrete preferred implementation further description made for the present invention, can not assert that specific embodiment of the invention is confined to these explanations.For general technical staff of the technical field of the invention, without departing from the inventive concept of the premise, some simple deduction or replace can also be made, all should be considered as belonging to protection scope of the present invention.

Claims (14)

1. a Forecasting Methodology for business datum amount and/or resource data amount, is characterized in that, comprising:

Build the initial data set of business datum amount and/or resource data amount;

Dimensionality reduction preliminary treatment is carried out to described initial data set, obtains preprocessed data set;

First an initial clustering process is carried out to described preprocessed data set, obtain initial clustering data acquisition system, according to described initial clustering data acquisition system, more accurately clustering processing is carried out at least once to described initial data set, obtain accurate cluster data set;

According to described accurate cluster data set, determine forecast model;

According to described forecast model, obtain the anticipatory data amount of described business and/or resource.

2. the Forecasting Methodology of business datum amount according to claim 1 and/or resource data amount, is characterized in that, the initial data set building business datum amount and/or resource data amount specifically comprises:

Determine business to be predicted and/or resource;

Obtain at least one historical time section, the consumption data of described business and/or the consumption data of resource, using the consumption data of described business as described business datum amount, using the consumption data of described resource as described resource data amount;

According to the consumption data of described business and/or the consumption data of resource, build initial data set.

3. the Forecasting Methodology of business datum amount according to claim 1 and/or resource data amount, is characterized in that, by PCA, carries out dimensionality reduction preliminary treatment, obtain preprocessed data set to described initial data set.

4. the Forecasting Methodology of business datum amount according to claim 1 and/or resource data amount, it is characterized in that, dimensionality reduction preliminary treatment is being carried out to described initial data set, before obtaining preprocessed data set, is also comprising: described initial data set is being normalized;

And/or, dimensionality reduction preliminary treatment is being carried out to described initial data set, after obtaining preprocessed data set, is also comprising: described preprocessed data set is being normalized.

5. the business datum amount according to any one of claim 1-4 and/or the Forecasting Methodology of resource data amount, it is characterized in that, first an initial clustering process is carried out to described preprocessed data set, obtain initial clustering data acquisition system, according to described initial clustering data acquisition system, again accurately clustering processing is carried out at least once to described initial data set, obtains accurate cluster data set and specifically comprise:

According to initial clustering method, first an initial clustering process is carried out to described preprocessed data set, obtain initial clustering data acquisition system;

According to described initial clustering data acquisition system, calculate initial cluster center;

According to accurate clustering method, described initial cluster center, more accurately clustering processing is carried out once to described initial data set, obtain accurate cluster data set.

6. the Forecasting Methodology of business datum amount according to claim 5 and/or resource data amount, is characterized in that, according to described accurate cluster data set, determines that forecast model specifically comprises:

In described accurate cluster data set, determine elementary item, item to be predicted;

According to described elementary item, item to be predicted, determine master data amount, data volume to be predicted;

According to gradient descent method, matching is carried out to described master data amount, data volume to be predicted, determines fitting function, using described fitting function as forecast model.

7. the Forecasting Methodology of business datum amount according to claim 6 and/or resource data amount, is characterized in that, according to described forecast model, the anticipatory data measurer body obtaining described business and/or resource comprises:

According to different elementary items, select different fitting functions;

According to the fitting function of described selection, described business datum amount and/or resource data amount are predicted, obtain the anticipatory data amount of described business and/or resource.

8. the business datum amount according to any one of claim 1-3 and/or the Forecasting Methodology of resource data amount, is characterized in that, according to described forecast model, after obtaining the anticipatory data amount of described business and/or resource, also comprises:

According to described anticipatory data amount, network is optimized.

9. a prognoses system for business datum amount and/or resource data amount, is characterized in that, comprising:

Build module, for building the initial data set of business datum amount and/or resource data amount;

Pretreatment module, for carrying out dimensionality reduction preliminary treatment to the initial data set of described structure module construction, obtains preprocessed data set;

Cluster module, an initial clustering process is carried out in preprocessed data set for first obtaining described pretreatment module, obtain initial clustering data acquisition system, according to described initial clustering data acquisition system, again accurately clustering processing is carried out at least once to the initial data set of described structure module construction, obtain accurate cluster data set;

Determination module, for the accurate cluster data set obtained according to described cluster module, determines forecast model;

Prediction module, for the forecast model determined according to described determination module, obtains the anticipatory data amount of described business and/or resource.

10. the prognoses system of business datum amount according to claim 9 and/or resource data amount, is characterized in that, also comprise acquisition module;

Described determination module is also for determining business to be predicted and/or resource;

Described acquisition module, for obtaining at least one historical time section, the consumption data of the business that described determination module is determined and/or the consumption data of resource, using the consumption data of described business as described business datum amount, using the consumption data of described resource as described resource data amount;

Described structure module, specifically for the consumption data of business that obtains according to described acquisition module and/or the consumption data of resource, builds initial data set.

The prognoses system of 11. business datum amounts according to claim 9 and/or resource data amount, is characterized in that,

Described pretreatment module also for carrying out dimensionality reduction preliminary treatment to the initial data set of described structure module construction, before obtaining preprocessed data set, is normalized described initial data set;

And/or described pretreatment module also for carrying out dimensionality reduction preliminary treatment to the initial data set of described structure module construction, after obtaining preprocessed data set, is normalized described preprocessed data set.

The prognoses system of 12. business datum amounts according to any one of claim 9-11 and/or resource data amount, is characterized in that, also comprise computing module, and described cluster module comprises initial clustering submodule, accurately cluster submodule;

Described initial clustering submodule, for according to initial clustering method, an initial clustering process is carried out in the preprocessed data set first obtained described pretreatment module, obtains initial clustering data acquisition system;

Described computing module, for the initial clustering data acquisition system obtained according to described initial clustering submodule, calculates initial cluster center;

Described accurate cluster submodule, for the initial cluster center calculated according to accurate clustering method, described computing module, then carries out once accurately clustering processing to the initial data set of described structure module construction, obtains accurate cluster data set.

The prognoses system of 13. business datum amounts according to claim 12 and/or resource data amount, is characterized in that, described determination module also in the accurate cluster data set that obtains at described accurate cluster submodule, determines elementary item, item to be predicted; Also for according to described elementary item, item to be predicted, determine master data amount, data volume to be predicted;

Described determination module, specifically for according to gradient descent method, carries out matching to described master data amount, data volume to be predicted, determines fitting function, using described fitting function as forecast model.

The prognoses system of 14. business datum amounts according to claim 13 and/or resource data amount, is characterized in that, also comprise:

Select module, for according to different elementary items, select different fitting functions;

Described prediction module, specifically for the fitting function according to described selection model choice, is predicted described business datum amount and/or resource data amount, obtains the anticipatory data amount of described business and/or resource.