CN106384219A - Warehouse partition assisted analysis method and device - Google Patents

Abstract

The invention relates to a warehouse partition assisted analysis method. The method comprises a step of obtaining the order information of an area in a historical period of time and obtaining the single product information in the historical period of time from the order information, a step of establishing a historical three-dimensional correlation matrix, wherein any point Hi (xi, yi and zi) of the historical three-dimensional correlation matrix is used for expressing that the correlation of a single product xi and a single product yi at a historical time zi is gi, a step of fitting the time sequences of the correlations of any two single products and calculating a future three-dimensional correlation matrix in a future time period according to a fitting result, a step of adding the future three-dimensional correlation matrix to the historical three-dimensional correlation matrix and fusing the added three-dimensional correlation matrix in the time dimensional to obtain a two-dimensional correlation matrix, a step of carrying out spectral clustering on the two-dimensional correlation matrix and carrying out graph segmentation on the two-dimensional correlation matrix according to a spectral clustering result, and a step of distributing a single-product set to each warehouse according to the result of the graph segmentation. According to the method, the operating cost can be reduced.

Description

Storage point storehouse aided analysis method and device

Technical field

It relates to storage technique field, in particular to a kind of storage point storehouse aided analysis method and one kind Storage point storehouse assistant analysis device.

Background technology

Electric business in evolution, with the expansion in market, using the continuous expansion of crowd, single sku (store Keeping unit, keeper unit) sales volume steeply rises.In order to ensure that stock rate (has existing in warehouse when i.e. client places an order Goods), adopt pin system and have to increase the stock amount of sku, this brings unprecedented pressure to storage operation.Restriction in capacity Under, some warehouses have to separately deposit existing storage commodity, i.e. so-called point of storehouse.Point storehouse process has expedited the emergence of in operation An important indicator, i.e. singulated rate (singulated quantity on order/blanket order quantity=singulated rate), described singulated rate refers to：One Customer order comprises N part commodity, leaves the individual warehouse of M (M≤N) respectively in, completes the storage production of this order and dispensing needs M times Production cost.For example, in above process, if M>1, then need " singulated "；That is, a customer order, it is split as many Individual sub- order；Therefore also just improve cost；Because after a customer order is split as M sub- order, electric business is had to Pay M times of cost and complete storage production and dispensing；But client only pays 1 cost, thus greatly increased entirety Cost.

When planning storage stock deposits, commonplace way is electric business at present：As far as possible identical category commodity are put In same warehouse；Wherein, commodity category for example can include：Mother and baby's category, clothing category, outdoor goods, personal nursing, 3C Electronic product and books, phonotapes and videotapes etc..But when carrying out warehouse fractionation, traditional work storage Planning Model is no longer applicable, because Fractionation for warehouse inevitably will carry out category fractionation.Therefore, it is related to category during point storehouse at present to divide, generally require Divided by force according to artificial experience, the same category in same warehouse is assigned to different warehouses and deposits.

The category planning in warehouse at present, when in the face of dividing storehouse problem, needs substantial amounts of artificial experience to carry out category fractionation, and And the situation of different regions difference storehouse is not quite similar, this is higher to personnel qualifications；If warehouse category splits unreasonable By the great singulated rate increasing customer order, increase the operation cost stored in a warehouse.Accordingly, it is desirable to provide a kind of new storage divides storehouse Aided analysis method and device.

It should be noted that information is only used for strengthening the reason of background of this disclosure disclosed in above-mentioned background section Solution, therefore can include not constituting the information to prior art known to persons of ordinary skill in the art.

Content of the invention

The purpose of the disclosure is to provide a kind of storage point storehouse aided analysis method and a kind of storage point storehouse assistant analysis Device, and then at least overcome one or more that lead to due to restriction and the defect of correlation technique to ask to a certain extent Topic.

According to an aspect of this disclosure, provide a kind of storage point storehouse aided analysis method, including：

Obtain an area and obtain described historical time in the sequence information in historical time section and from described sequence information Single product information in section；

Set up history three-dimensional incidence matrix, any point H of described history three-dimensional incidence matrix_i(x_i,y_i,z_i) be used for representing Single product x_iWith single product y_iIn historical time z_iThe relatedness at place is g_i；Wherein, g_iIt is according to described single product x_iWith single product y_iIn history Time z_iDescribed single product information be calculated；

The time serieses of each described relatedness of product single described in any two are fitted, and are calculated according to fitting result Following three-dimensional incidence matrix in future time section；

Described following three-dimensional incidence matrix is increased on described history three-dimensional incidence matrix, and to the three-dimensional association after increasing Matrix carries out fusion on time dimension and obtains two-dimentional incidence matrix；

To described two dimension incidence matrix carry out spectral clustering, and according to spectral clustering result with the subgraph after splitting between relatedness It is that target carries out figure segmentation to described two dimension incidence matrix less；

According to the result of described figure segmentation, single product set is assigned in each warehouse.

In a kind of exemplary embodiment of the disclosure, methods described also includes：

By the seasonal effect in time series first order and second order moments of described relatedness, the time serieses of described relatedness are gone Peel off Value Operations.

In a kind of exemplary embodiment of the disclosure, methods described also includes：

According to described spectral clustering result, described two dimension incidence matrix is clustered further, and cluster result is entered to advance One step evaluation.

Time sequence in a kind of exemplary embodiment of the disclosure, to each described relatedness of product single described in any two Row be fitted including：

$S_t=\underset{i=1}{\overset{T}{\Σ}}{w}_i{x}_{t+1-i}=w_1{x}_t+w_2{x}_{t-1}+\mathrm{...}+w_t{x}_{t-T+1};$

Wherein, x_iFor the value in time serieses, w_iFor seasonal effect in time series weights, T is the size of time window.

In a kind of exemplary embodiment of the disclosure, the three-dimensional incidence matrix after increasing is melted on time dimension Conjunction obtains two-dimentional incidence matrix and includes：

$H^{*}(x_i,y_i)=\underset{j=1}{\overset{n}{\Σ}}H(x_i,y_i,z_j);$

Wherein, n is described length of time series, H^*(x_i,y_i) be described two dimension incidence matrix on any point,For any point on described following three-dimensional incidence matrix.

In a kind of exemplary embodiment of the disclosure, spectral clustering is carried out to described two dimension incidence matrix and includes：

Set up similar diagram, the weighted adjacency matrix of similar diagram is W, wherein, W=H*, H* are described two dimension incidence matrix；

Calculate Laplacian Matrix L, whereinw_ijFor appointing in described weighted adjacency matrix W A bit, j is the columns in described weighted adjacency matrix W to meaning；

Decompose described Laplacian Matrix L=U Λ U^-1, wherein, U=[u₁,u₂,...,u_r],

$\mathrm{\Λ}=\left[\begin{array}{cccc}{\mathrm{\λ}}_1& & & \\ & {\mathrm{\λ}}_2& & \\ & & ...& \\ & & & {\mathrm{\λ}}_r\end{array}\right],$

[u₁,u₂,...,u_r] for L characteristic vector value, λ_iEigenvalue for L, and in Λ, λ₁≤λ₂≤...≤λ_r, R is the order of W；

The front k characteristic vector value choosing described Laplacian Matrix L forms the matrix of a r*k, by described matrix Every a line as one of k dimension space vector, using clustering algorithm, described characteristic vector is clustered, k be spectral clustering Pre- number of clusters.

In a kind of exemplary embodiment of the disclosure, according to described spectral clustering result, described two dimension incidence matrix is carried out Cluster includes further：

Wherein, S (k, m) is the amount of projects of described single product set distribution, C is binomial coefficient, and m is the number in the warehouse needing to participate in calculating, and k is the pre- number of clusters of spectral clustering, and has m ＜ k.

According to another aspect of the disclosure, provide a kind of storage point storehouse assistant analysis device, including：

Data obtaining module：For obtaining an area in the sequence information in historical time section and from described sequence information Obtain the single product information in described historical time section；

First incidence matrix module：For setting up history three-dimensional incidence matrix, described history three-dimensional incidence matrix arbitrary Point H_i(x_i,y_i,z_i) be used for representing single product x_iWith single product y_iIn historical time z_iThe relatedness at place is g_i；Wherein, g_iIt is according to described Single product x_iWith single product y_iIn historical time z_iDescribed single product information be calculated；

Fitting module：Time serieses for each described relatedness to product single described in any two are fitted；

Second incidence matrix module：For calculating the following three-dimensional association square in future time section according to fitting result Battle array；

3rd association matrix module：For described following three-dimensional association square is increased on described history three-dimensional incidence matrix Battle array；

Fusion Module：Obtain two dimension association square for the three-dimensional incidence matrix after increasing is carried out on time dimension with fusion Battle array；

Spectral clustering module：For spectral clustering is carried out to described two dimension incidence matrix；

Figure segmentation module：For according to relatedness between the subgraph after to split for the spectral clustering result minimum for target to described two Dimension incidence matrix carries out figure segmentation；

Single product distribute module：For the result split according to described figure, single product set is assigned in each warehouse.

In a kind of exemplary embodiment of the disclosure, described device also includes：

Go outlier module：For by the seasonal effect in time series first order and second order moments of described relatedness to described relatedness Time serieses carry out peeling off Value Operations.

In a kind of exemplary embodiment of the disclosure, described device also includes：

First cluster module：For being clustered further to described two dimension incidence matrix according to described spectral clustering result；

Evaluation module：For being evaluated further to cluster result.

Time sequence in a kind of exemplary embodiment of the disclosure, to each described relatedness of product single described in any two Row are fitted, and described matching includes：

$S_t=\underset{i=1}{\overset{T}{\Σ}}{w}_i{x}_{t+1-i}=w_1{x}_t+w_2{x}_{t-1}+\mathrm{...}+w_t{x}_{t-T+1};$

Wherein, x_iFor the value in time serieses, w_iFor seasonal effect in time series weights, T is the size of time window.

In a kind of exemplary embodiment of the disclosure, the three-dimensional incidence matrix after increasing is melted on time dimension Conjunction obtains two-dimentional incidence matrix, and described fusion includes：

$H^{*}(x_i,y_i)=\underset{j=1}{\overset{n}{\Σ}}H(x_i,y_i,z_j);$

Wherein, n is described length of time series, H^*(x_i,y_i) be described two dimension incidence matrix on any point,For any point on described following three-dimensional incidence matrix.

In a kind of exemplary embodiment of the disclosure, described spectral clustering module includes：

Similar diagram sets up module：For setting up similar diagram, the weighted adjacency matrix of similar diagram is W, and wherein, W=H*, H* are Described two dimension incidence matrix；

Laplacian Matrix computing module：For calculating Laplacian Matrix L, wherein,w_ij For any point in described weighted adjacency matrix W, j is the columns in described weighted adjacency matrix W；

Laplacian Matrix decomposing module：For decomposing described Laplacian Matrix L=U Λ U^-1, wherein, U=[u₁, u₂,...,u_r],

$\mathrm{\Λ}=\left[\begin{array}{cccc}{\mathrm{\λ}}_1& & & \\ & {\mathrm{\λ}}_2& & \\ & & ...& \\ & & & {\mathrm{\λ}}_r\end{array}\right],$

[u₁,u₂,...,u_r] for L characteristic vector value, λ_iEigenvalue for L, and in Λ, λ₁≤λ₂≤...≤λ_r, R is the order of W；

Second cluster module：Front k characteristic vector value for choosing described Laplacian Matrix L forms a r*k's Matrix, using the every a line in described matrix as one of k dimension space vector, is entered to described characteristic vector using clustering algorithm Row cluster, k is the pre- number of clusters of spectral clustering.

In a kind of exemplary embodiment of the disclosure, according to described spectral clustering result, described two dimension incidence matrix is carried out Cluster further, described cluster includes：

Wherein, S (k, m) is the amount of projects of described single product set distribution, C is binomial coefficient, and m is the number in the warehouse needing to participate in calculating, and k is the pre- number of clusters of spectral clustering, and has m ＜ k.

A kind of storage of exemplary embodiment of the disclosure divides in storehouse aided analysis method and device, on the one hand passes through user Sequence information abstract for commodity association matrix so that commodity distribution calculating be based entirely on incidence matrix, without traverse user Order, improves computational efficiency；On the other hand incidence matrix in a short time can be predicted by seasonal effect in time series analysis, from And the singulated situation behind commodity point storehouse can be predicted, auxiliary enterprises realize more excellent point storehouse, and then can drop to a great extent The cost that low " high singulated rate " is brought to enterprise operation.

It should be appreciated that above general description and detailed description hereinafter are only exemplary and explanatory, not The disclosure can be limited.

Brief description

Accompanying drawing herein is merged in description and constitutes the part of this specification, shows the enforcement meeting the disclosure Example, and be used for explaining the principle of the disclosure together with description.It should be evident that drawings in the following description are only the disclosure Some embodiments, for those of ordinary skill in the art, on the premise of not paying creative work, can also basis These accompanying drawings obtain other accompanying drawings.

Fig. 1 schematically shows a kind of storage point storehouse aided analysis method flow chart in disclosure exemplary embodiment.

Fig. 2 schematically shows single product information acquiring portion flow chart in disclosure exemplary embodiment.

Fig. 3 (a) schematically shows a kind of history three-dimensional incidence matrix figure in disclosure exemplary embodiment.

Fig. 3 (b) schematically shows a kind of history three-dimensional incidence matrix figure in disclosure exemplary embodiment.

Fig. 4 (a) schematically shows a kind of relatedness time serieses in disclosure exemplary embodiment.

Fig. 4 (b) schematically shows a kind of relatedness seasonal effect in time series fitting result in disclosure exemplary embodiment.

Fig. 5 schematically shows the three-dimensional incidence matrix after increasing following three-dimensional incidence matrix in disclosure exemplary embodiment Figure.

Fig. 6 schematically shows a kind of Spectral Clustering flow chart in disclosure exemplary embodiment.

Fig. 7 schematically shows one kind in disclosure exemplary embodiment and removes outlier method flow diagram.

Fig. 8 schematically shows a kind of two dimension incidence matrix figure in disclosure exemplary embodiment.

Fig. 9 schematically shows a kind of storage point storehouse assistant analysis device block diagram in disclosure exemplary embodiment.

Specific embodiment

It is described more fully with example embodiment referring now to accompanying drawing.However, example embodiment can be with multiple shapes Formula is implemented, and is not understood as limited to example set forth herein；On the contrary, these embodiments are provided so that the disclosure will more Fully and completely, and by the design of example embodiment comprehensively convey to those skilled in the art.Described feature, knot Structure or characteristic can combine in one or more embodiments in any suitable manner.In the following description, provide perhaps Many details are thus provide fully understanding of embodiment of this disclosure.It will be appreciated, however, by one skilled in the art that can Omit one of described specific detail or more to put into practice the technical scheme of the disclosure, or other sides can be adopted Method, constituent element, device, step etc..In other cases, be not shown in detail or describe known solution a presumptuous guest usurps the role of the host avoiding and The each side making the disclosure thicken.

Additionally, accompanying drawing is only the schematic illustrations of the disclosure, it is not necessarily drawn to scale.In figure identical accompanying drawing mark Note represents same or similar part, thus will omit repetition thereof.Some block diagrams shown in accompanying drawing are work( Energy entity, not necessarily must be corresponding with physically or logically independent entity.These work(can be realized using software form Energy entity, or realize these functional entitys in one or more hardware modules or integrated circuit, or at heterogeneous networks and/or place These functional entitys are realized in reason device device and/or microcontroller device.

A kind of storage point storehouse aided analysis method is provide firstly, this storage divides storehouse assistant analysis in this example embodiment Method, can go out from the purchasing behavior of user for different regions, different time sections from the data analysiss of customer order Send out, the purchase relatedness of analysis commodity, by planning article layout, n warehouse is split as m warehouse.With reference to shown in Fig. 1, A described storage point storehouse aided analysis method may comprise steps of：

In step s 110, obtain an area to obtain in the sequence information in historical time section and from described sequence information Single product information in described historical time section.With reference to shown in Fig. 2, under in this example embodiment, step S110 for example can include State step S112～S116.Wherein：

In step S112, obtain the sequence information in time period from hadoop cluster.

Hadoop cluster is a distributed system architecture by the exploitation of Apache fund club, and user can be not Distributed program is developed, the advantage making full use of cluster carries out high-speed computation and deposits in the case of understanding distributed low-level details Storage.In this example embodiment, the sequence information obtaining from hadoop cluster can include：Father's O/No., sub- order is compiled Number, buy single product (sku or category) information, (warehouse information includes warehouse title to warehouse information, when described region and outbound Between), place an order the time, order effectively mark etc., but the disclosure is not limited.Additionally, other the exemplary enforcements in the disclosure It is also possible to according to circumstances otherwise obtain sequence information in example, in this exemplary embodiment, this is not done with particular determination.

In step S114, taxonomic revision is carried out to the above order information.

In this example embodiment, invalid order exclusion, order class such as can be included to the taxonomic revision of sequence information Type identifies one-level attribute information mark etc., and the disclosure is not limited.In this example embodiment, permissible by step S114 Realize：Filter out invalid order (it is invalid that order is effectively designated)；Filter out the order not having warehouse information；According to order industry Service type is identified to order：Such as can be designated：Self-operation order, third party's order, self-operation and third party's mixing order Deng, order type of service can be inferred from O/No., or in sequence information add order type of service mark；And, right Order increases attribute information：Attribute information such as can be increased and distinguish trans-regional order and across category order etc..

In step S116, arrange single product information.

In this example embodiment, single product item_iCan be the commodity in order, it can be sku (store Keeping unit, keeper unit) or category etc.；Sku is the minimum memory unit in storehouse management, and category is according to business Difference can be divided into multistage：Such as Yanjing Brewery and Tsingtao beer, calculate as two single product by sku, but by three-level category both For medicated beer, it is a kind of single product；Such as Yanjing Brewery and Tsingtao beer and flying apsaras Maotai again, is three single product by sku, but three-level product Class is medicated beer and the single product of Chinese liquor two, calculates single product by seconds class for Chinese wine.When arranging single product information according to reality Application demand, select processing granularity (sku or category).

After obtaining each regional sequence information, can be for each area, by this area in this example embodiment Sequence information synthesize a single product set I, I comprises all single product that this area occurs in interior order for a period of time, such as： I={ item_i}.Additionally, in this example embodiment, can also integrate to sequence information, the single son of same father is single to be needed Merge, generate order set；Such as order o₁Including sub- orderWith sub- orderWhereinOrder includes single product (item₁, item₂,item₃),Order includes single product ((item₄,item₅), that is, Order o after integration₁For o₁{item₁,item₂,item₃,item₄,item₅}；Wherein, item₁,item₂, item₃,item₄,item₅For single product.

In the step s 120, history three-dimensional incidence matrix, any point H of described history three-dimensional incidence matrix are set up_i(x_i, y_i,z_i) be used for representing single product x_iWith single product y_iIn historical time z_iThe relatedness at place is g_i.Wherein, g_iIt is according to described single product x_iWith Single product y_iIn historical time z_iDescribed single product information be calculated.

With reference to shown in Fig. 3 (a), in this example embodiment, for each area, can start with from sequence information and be closed Connection analysis, statistics granularity is to sky；Obtain three-dimensional incidence matrix data cube H (X, Y, Z), this data cube is with sky For granularity, the relatedness of commodity is recorded；With reference to shown in Fig. 3 (b), each XY layer H of each of which incidence matrix (*, * z) represent point at the same time.

In step s 130, the time serieses of each described relatedness of product single described in any two are fitted, and root Calculate the following three-dimensional incidence matrix in future time section according to fitting result.

With reference to shown in Fig. 4 (a), each point H (x on matrix₁,y₁,z₁) A of in figure (as) represent certain two single product and exist The relatedness of certain time point, this along z-axis data be certain two single product relatednesss time serieses H (x₁,y₁,*).Originally show In example embodiment, the time serieses of relatedness are fitted to calculate by formula below：

$S_t=\underset{i=1}{\overset{T}{\Σ}}{w}_i{x}_{t+1-i}=w_1{x}_t+w_2{x}_{t-1}+\mathrm{...}+w_t{x}_{t-T+1};$

Wherein, x_iFor the value in time serieses, w_iFor seasonal effect in time series weights, T is the size of time window.

With reference to shown in Fig. 4 (b), Fig. 4 (b) represents the time serieses of any two commodity associations, and in figure N1 represents truly several According to N2 represents fitting data, the prediction expection of N3 region representation.In step S140, described history three-dimensional incidence matrix increases Plus described following three-dimensional incidence matrix, and to the three-dimensional incidence matrix after increasing, fusion is carried out on time dimension and obtains two dimension closing Connection matrix.But skilled addressee readily understands that, in other exemplary embodiments of the disclosure, it would however also be possible to employ its His mode calculates the following three-dimensional incidence matrix in future time section, and in this example embodiment, this is not done with particular determination.

With reference to shown in Fig. 5, Fig. 5 and Fig. 3 (a) is distinguished as, and increased the incidence matrix of predicted time section；Wherein C1 is not Carry out three-dimensional incidence matrix, history three-dimensional incidence matrix during C2；Melting on time dimension is carried out to the three-dimensional incidence matrix in Fig. 5 Close, new two-dimentional incidence matrix H can be formed^*.In this example embodiment, amalgamation mode can be as follows：

Wherein n is length of time series.

Matrix has been carried out summation on time dimension and has formed a new matrix H by above formula^*, and unlike matrix H, square Battle array H^*It is a two-dimensional matrix, H^*Each of matrix element represents the total correlation of certain two single product in a period of time.

In step S150, to described two dimension incidence matrix carry out spectral clustering, and according to spectral clustering result to split after Between subgraph, relatedness is minimum carries out figure segmentation for target to described two dimension incidence matrix.With reference to shown in Fig. 6, this example embodiment In, spectral clustering can include step S602～S608.Wherein：

In step S602, set up similar diagram (similarity graph).In this example embodiment, the band of similar diagram Power adjacency matrix is W, and wherein, W=H*, H* are above-mentioned two dimension incidence matrix.

In step s 604, calculate non-standard figure (unnormalized graph) Laplacian Matrix L, whereinFor any point in described weighted adjacency matrix W, j is in described weighted adjacency matrix W Columns.

In step S606, SVD decomposition is carried out to described Laplacian Matrix L.For example：

L=U Λ U^-1；

Wherein,[u₁,u₂,...,u_r] for L characteristic vector value, λ_iEigenvalue for L, and in Λ, λ₁≤λ₂≤...≤λ_r, r is the order of W.

In step S608, the front k characteristic vector value choosing described Laplacian Matrix L forms the matrix of a r*k, Using the every a line in described matrix as one of k dimension space vector, using clustering algorithm, described characteristic vector is gathered Class, k is the pre- number of clusters of spectral clustering.

Choose the first k minimum characteristic vector (identical with pre- number of clusters k of spectral clustering) of L, k value is chosen to be needed fully Consider a point storehouse condition；For example, in advance 2 storehouses are divided into 3 storehouses (i.e. a newly-built storehouse needs the warehouse participating in calculating to be 3), then K value selects 15；Again for example, (i.e. newly-built 2 storehouses need the warehouse participating in calculating to be 6 in advance 4 storehouses to be changed into 6 storehouses Individual), then need for k value to be scheduled on 30 about；The selection of general k value is at least 5 times of storehouse number and is not more than order r of H*, but this Disclosure is not limited.

By the matrix of one r*k of composition arranged together for above-mentioned k characteristic vector, using each of which row as k dimension space One of vector, and clustered using k-means algorithm；It is also possible to root in other exemplary embodiments of the disclosure Clustered according to other modes, in this exemplary embodiment, this is not done with particular determination.

In step S160, according to the result of described figure segmentation, single product set is assigned in each warehouse.

A kind of storage point storehouse aided analysis method and the device of the disclosure, on the one hand passing through will be abstract for the sequence information of user For commodity association matrix so that the calculating of commodity distribution is based entirely on incidence matrix, without traverse user order, improve calculating Efficiency；On the other hand incidence matrix in a short time can be predicted by seasonal effect in time series analysis, thus predicting commodity point storehouse Singulated situation afterwards.

Another kind of storage point storehouse aided analysis method and the device of the disclosure, by going outlier, mistake to seasonal effect in time series Filter the impact to single product relatedness for the sales promotion, improve the accuracy rate of relatedness, reduce singulated rate, greatly reduce into This.

In other embodiments of the disclosure, above-mentioned storage point storehouse aided analysis method also includes：By described association The seasonal effect in time series first order and second order moments of property are carried out peeling off to the time serieses of described relatedness Value Operations.

With reference to shown in Fig. 7, go the step of outlier can include：

In step S702, meet T distribution using the relatedness of any two list product, according to any two in the unit interval The time serieses of single product, calculate the first order matrix M within the above-mentioned unit interval₁With second-order matrix M₂.

In step S704, select confidence level 95%, i.e. α=0.05, degree of freedom subtracts 1 for sample size, and that is, p-1 is (when p is Between sequence extreme difference, in units of sky).

In above-mentioned steps S704, choose α=0.05, but the disclosure is not limited, and voluntarily can be selected according to reality Take.

In step S706, calculate marginal value c=t_(1-α/2)(p-1), wherein t_(1-α/2)(p-1) can be by looking into t-distribution table Obtain.

In step S708, chooseFor confidence interval, and each of time serieses are worth X enters line translation,If x' is in intervalIn then retain, otherwise remove.

By removing outlier, eliminate the sequence information with regard to sales promotion, increased the accuracy rate of relatedness.Additionally, it is above-mentioned Outlier method is gone to be only used as demonstrating, the outlier scheme of going realized by other means also belongs to the protection model of the disclosure Enclose.

In other embodiments of the disclosure, above-mentioned storage point storehouse aided analysis method also includes：Above-mentioned two dimension is closed Connection matrix is carried out.For example：

With reference to shown in Fig. 8, the relatedness of the first behavior list product a and other single product (a, b, c, d, e, f), shade represents The power of relatedness, deeper relatedness is stronger；Now, this programme only retains the front m associated data (m of often row>K, k are that spectrum is poly- The desire number of clusters of class), remaining zero setting.In addition, the H after cleaning^*Remain symmetrical matrix.

By wash the not strong data of relatedness it is ensured that cluster effect.

In other embodiments of the disclosure, above-mentioned storage point storehouse aided analysis method also includes：Poly- according to described spectrum Class result is clustered further to described two dimension incidence matrix, and cluster result is evaluated further.

In such scheme, involved single product collection is divided into k class, needs the warehouse participating in calculating to be m；By It is more than m in k, then needs to cluster further, k is birdsed of the same feather flock together to m class, then have S (k, m) to plant possible, be shown below, last m The category deposited in warehouse behind each of class class representative point storehouse.

Wherein C is binomial coefficient.

In above-mentioned steps, by clustering again, generate S (k, m) and plant possible single product placement scheme, in this step In, need the individual scheme to S (k, m) to evaluate, select optimal case；Outstanding single product placement scheme can effectively reduce to be torn open The generation of one-state, is therefore used herein as article layout's scheme to singulated contribution degree as evaluation criterion.

Evaluation procedure includes：

S (k, m)={ s₁,s₂,s₃...,s_l, l represents l assembled scheme, i.e. the length of set S (k, m)；

s_i={ m₁,m₂,m₃,...,m_m}；

Wherein, s_iFor one of individual scheme of S (k, m) list product placement scheme, each s_iIn all have m classification, wherein often One classification m_iIt is made up of single product, and a s_iM classification in single product be not repeat, that is,

Represent item_i,item_jRelatedness,Represent H^*On a line, represent item_iAssociation with all commodity Property；

As described above, item_iIn H^*In have a corresponding rowValue on corresponding row represents item_iWith all commodity Relatedness；If item_iAnd item_jIt is placed in same warehouse, then can reduce singulated；Contrary, if item_iWith item_jIt is placed in different warehouses, then will necessarily increase singulated amount.

Therefore, evaluation of programme s_iFine or not when, need calculate s_iEach of class m_iTo singulated contribution degree score_ m_i, to m_iMiddle single product item_iCorrespondingRow is sued for peace.item_iAnd item_jAll in m_iIn, then have：score_m_iAddIllustrate this two single product put together can reduce singulated；

Contrary, if item_iAnd item_jNot in m_iIn, then have：score_m_iAddIllustrate that this two single product do not have Put together, singulated rate can be increased.

By the score_m calculating_iCarry out summation and can obtain s_iEvaluation of estimate s_{i_score}；s_{i_score}Represent single product to divide The impact to order for the cloth, this value is bigger, illustrates that this article layout's scheme is more to singulated contribution；Conversely, then the explanation program is got over Good, lower to singulated contribution；If s_{i_score}Value be negative, then show that the program can effectively reduce singulated.In all s_i Middle selection minimum s_{i_score}It is final single product placement scheme.Clustered again and combining assessment parameter by this, can to point Storehouse scheme is further optimized, and then realizes optimal point storehouse scheme.

Following for disclosure device embodiment, can be used for executing method of disclosure embodiment.Real for disclosure device Apply the details not disclosed in example, refer to method of disclosure embodiment.

A kind of storage point storehouse assistant analysis device is additionally provided, this storage divides storehouse assistant analysis dress in this example embodiment Put the data analysiss from customer order, for different regions, different time sections, from the purchasing behavior of user, analyze N warehouse is split as m warehouse by planning article layout by the purchase relatedness of commodity.With reference to shown in Fig. 9, described storehouse A storage point storehouse assistant analysis device includes：Data obtaining module 900, the first incidence matrix module 910, the second incidence matrix module 920, the 3rd association matrix module 930, Fusion Module 940, spectral clustering module 950, figure segmentation module 960 and single product distribution mould Block 970；Wherein,

Data obtaining module 900：For obtain sequence information in historical time section for the area and from described order letter The single product information in described historical time section is obtained in breath；

First incidence matrix module 910：For setting up history three-dimensional incidence matrix, the appointing of described history three-dimensional incidence matrix 1 point of H_i(x_i,y_i,z_i) be used for representing single product x_iWith single product y_iIn historical time z_iThe relatedness at place is g_i；Wherein, g_iIt is according to institute State single product x_iWith single product y_iIn historical time z_iDescribed single product information be calculated；

Fitting module：Time serieses for each described relatedness to product single described in any two are fitted；

Second incidence matrix module 920：For calculating the following three-dimensional association in future time section according to fitting result Matrix；

3rd association matrix module 930：For described following three-dimensional association is increased on described history three-dimensional incidence matrix Matrix；

Fusion Module 940：Obtain two dimension pass for fusion is carried out on time dimension to the three-dimensional incidence matrix after increasing Connection matrix；

Spectral clustering module 950：For spectral clustering is carried out to described two dimension incidence matrix；

Figure segmentation module 960：For according to relatedness between the subgraph after to split for the spectral clustering result minimum for target to institute State two-dimentional incidence matrix and carry out figure segmentation；

Single product distribute module 970：For the result split according to described figure, single product set is assigned in each warehouse.

In other embodiments of the disclosure, described device also includes：

Go outlier module：For by the seasonal effect in time series first order and second order moments of described relatedness to described relatedness Time serieses carry out peeling off Value Operations.

In other embodiments of the disclosure, described device also includes：

First cluster module：For being clustered further to described two dimension incidence matrix according to described spectral clustering result；

Evaluation module：For being evaluated further to cluster result.

In other embodiments of the disclosure, the time serieses of each described relatedness of product single described in any two are entered Row matching, described matching includes：

$S_t=\underset{i=1}{\overset{T}{\Σ}}{w}_i{x}_{t+1-i}=w_1{x}_t+w_2{x}_{t-1}+\mathrm{...}+w_t{x}_{t-T+1};$

Wherein, x_iFor the value in time serieses, w_iFor seasonal effect in time series weights, T is the size of time window.

In other embodiments of the disclosure, the three-dimensional incidence matrix after increasing is carried out merging on time dimension Arrive two-dimentional incidence matrix, described fusion includes：

$H^{*}(x_i,y_i)=\underset{j=1}{\overset{n}{\Σ}}H(x_i,y_i,z_j);$

Wherein, n is described length of time series, H^*(x_i,y_i) be described two dimension incidence matrix on any point,For any point on described following three-dimensional incidence matrix.

In other embodiments of the disclosure, described spectral clustering module includes：

Similar diagram sets up module：For setting up similar diagram, the weighted adjacency matrix of similar diagram is W, and wherein, W=H*, H* are Described two dimension incidence matrix；

Laplacian Matrix computing module：For calculating Laplacian Matrix L, wherein,w_ij For any point in described weighted adjacency matrix W, j is the columns in described weighted adjacency matrix W；

Laplacian Matrix decomposing module：For decomposing described Laplacian Matrix L=U Λ U^-1, wherein, U=[u₁, u₂,...,u_r],

$\mathrm{\Λ}=\left[\begin{array}{cccc}{\mathrm{\λ}}_1& & & \\ & {\mathrm{\λ}}_2& & \\ & & ...& \\ & & & {\mathrm{\λ}}_r\end{array}\right],$

[u₁,u₂,...,u_r] for L characteristic vector value, λ_iEigenvalue for L, and in Λ, λ₁≤λ₂≤...≤λ_r, R is the order of W；

Second cluster module：Front k characteristic vector value for choosing described Laplacian Matrix L forms a r*k's Matrix, using the every a line in described matrix as one of k dimension space vector, is entered to described characteristic vector using clustering algorithm Row cluster, k is the pre- number of clusters of spectral clustering.

In other embodiments of the disclosure, traveling one is entered to described two dimension incidence matrix according to described spectral clustering result Step cluster, described cluster includes：

Wherein, S (k, m) is the amount of projects of described single product set distribution, C is binomial coefficient, and m is the number in the warehouse needing to participate in calculating, and k is the pre- number of clusters of spectral clustering, and has m ＜ k.

Because the storage of embodiment of the present invention divides each functional module of storehouse assistant analysis device and said method to invent Identical in embodiment, therefore will not be described here.

Although it should be noted that being referred to some modules or the list of the equipment for action executing in above-detailed Unit, but this division is not enforceable.In fact, according to embodiment of the present disclosure, above-described two or more The feature of module or unit and function can embody in a module or unit.Conversely, an above-described mould The feature of block or unit and function can be to be embodied by multiple modules or unit with Further Division.

Although additionally, describe each step of method in the disclosure in the accompanying drawings with particular order, this does not really want Ask or imply and must execute these steps according to this particular order, or having to carry out all shown step just enables Desired result.Additional or alternative, it is convenient to omit some steps, multiple steps are merged into a step execution, and/ Or a step is decomposed into execution of multiple steps etc..

Through the above description of the embodiments, those skilled in the art is it can be readily appreciated that example described herein is implemented Mode can be realized by software it is also possible to be realized with reference to by way of necessary hardware by software.Therefore, according to the disclosure The technical scheme of embodiment can be embodied in the form of software product, this software product can be stored in one non-volatile Property storage medium (can be CD-ROM, USB flash disk, portable hard drive etc.) in or network on, including some instructions so that a calculating Equipment (can be personal computer, server, mobile terminal or network equipment etc.) executes according to disclosure embodiment Method.

Those skilled in the art, after considering description and putting into practice invention disclosed herein, will readily occur to its of the disclosure Its embodiment.The application is intended to any modification, purposes or the adaptations of the disclosure, these modifications, purposes or Person's adaptations are followed the general principle of the disclosure and are included the undocumented common knowledge in the art of the disclosure Or conventional techniques.Description and embodiments be considered only as exemplary, the true scope of the disclosure and spirit by appended Claim is pointed out.

Claims (14)

1. a kind of storage point storehouse aided analysis method is it is characterised in that include：

Obtain an area and obtain in described historical time section in the sequence information in historical time section and from described sequence information Single product information；

Set up history three-dimensional incidence matrix, any point H of described history three-dimensional incidence matrix_i(x_i,y_i,z_i) be used for representing single product x_i With single product y_iIn historical time z_iThe relatedness at place is g_i；Wherein, g_iIt is according to described single product x_iWith single product y_iIn historical time z_i Described single product information be calculated；

The time serieses of each described relatedness of product single described in any two are fitted, and are calculated not according to fitting result Carry out the following three-dimensional incidence matrix in the time period；

Described following three-dimensional incidence matrix is increased on described history three-dimensional incidence matrix, and to the three-dimensional incidence matrix after increasing Fusion is carried out on time dimension and obtains two-dimentional incidence matrix；

Spectral clustering is carried out to described two dimension incidence matrix, and according to spectral clustering result with the subgraph after splitting between relatedness is minimum is Target carries out figure segmentation to described two dimension incidence matrix；

According to the result of described figure segmentation, single product set is assigned in each warehouse.

2. method according to claim 1 is it is characterised in that methods described also includes：

By the seasonal effect in time series first order and second order moments of described relatedness, the time serieses of described relatedness are carried out peeling off Value Operations.

3. method according to claim 1 is it is characterised in that methods described also includes：

According to described spectral clustering result, described two dimension incidence matrix is clustered further, and cluster result is carried out further Evaluate.

4. method according to claim 1 it is characterised in that to each described relatedness of product single described in any two when Between sequence be fitted including：

$S_t=\underset{i=1}{\overset{T}{\Σ}}{w}_i{x}_{t+1-i}=w_1{x}_t+w_2{x}_{t-1}+\mathrm{...}+w_t{x}_{t-T+1};$

Wherein, x_iFor the value in time serieses, w_iFor seasonal effect in time series weights, T is the size of time window.

5. method according to claim 1 it is characterised in that to increase after three-dimensional incidence matrix enterprising in time dimension Row fusion obtains two-dimentional incidence matrix and includes：

$H^{*}(x_i,y_i)=\underset{j=1}{\overset{n}{\Σ}}H(x_i,y_i,z_j);$

Wherein, n is described length of time series, H^*(x_i,y_i) be described two dimension incidence matrix on any point,For any point on described following three-dimensional incidence matrix.

6. method according to claim 1 includes it is characterised in that carrying out spectral clustering to described two dimension incidence matrix：

Set up similar diagram, the weighted adjacency matrix of similar diagram is W, wherein, W=H*, H* are described two dimension incidence matrix；

Calculate Laplacian Matrix L, wherein L=D-W,w_ijFor any point in described weighted adjacency matrix W, j For the columns in described weighted adjacency matrix W；

Decompose described Laplacian Matrix L=U Λ U^-1, wherein, U=[u₁,u₂,...,u_r],

$\mathrm{\Λ}=\left[\begin{array}{cccc}{\mathrm{\λ}}_1& & & \\ & {\mathrm{\λ}}_2& & \\ & & \mathrm{...}& \\ & & & {\mathrm{\λ}}_r\end{array}\right],$

[u₁,u₂,...,u_r] for L characteristic vector value, λ_iEigenvalue for L, and in Λ, λ₁≤λ₂≤...≤λ_r, r is W Order；

The front k characteristic vector value choosing described Laplacian Matrix L forms the matrix of a r*k, will be every in described matrix A line, as one of k dimension space vector, is clustered to described characteristic vector using clustering algorithm, and k is the pre- of spectral clustering Number of clusters.

7. method according to claim 3 it is characterised in that according to described spectral clustering result to described two dimension incidence matrix Clustered further including：

Wherein, S (k, m) is the amount of projects of described single product set distribution, and C is Binomial coefficient, m is the number in the warehouse needing to participate in calculating, and k is the pre- number of clusters of spectral clustering, and has m ＜ k.

8. a kind of storage point storehouse assistant analysis device is it is characterised in that include：

Data obtaining module：Obtain in the sequence information in historical time section and from described sequence information for obtaining an area Single product information in described historical time section；

First incidence matrix module：For setting up history three-dimensional incidence matrix, any point H of described history three-dimensional incidence matrix_i (x_i,y_i,z_i) be used for representing single product x_iWith single product y_iIn historical time z_iThe relatedness at place is g_i；Wherein, g_iIt is according to described single product x_iWith single product y_iIn historical time z_iDescribed single product information be calculated；

Fitting module：Time serieses for each described relatedness to product single described in any two are fitted；

Second incidence matrix module：For calculating the following three-dimensional incidence matrix in future time section according to fitting result；

3rd association matrix module：For described following three-dimensional incidence matrix is increased on described history three-dimensional incidence matrix；

Fusion Module：Obtain two-dimentional incidence matrix for fusion is carried out on time dimension to the three-dimensional incidence matrix after increasing；

Spectral clustering module：For spectral clustering is carried out to described two dimension incidence matrix；

Figure segmentation module：For closing for target to described two dimension according to relatedness between the subgraph after to split for the spectral clustering result is minimum Connection matrix carries out figure segmentation；

Single product distribute module：For the result split according to described figure, single product set is assigned in each warehouse.

9. device according to claim 8 is it is characterised in that described device also includes：

Go outlier module：For by the seasonal effect in time series first order and second order moments of described relatedness to described relatedness when Between sequence carry out peeling off Value Operations.

10. device according to claim 8 is it is characterised in that described device also includes：

First cluster module：For being clustered further to described two dimension incidence matrix according to described spectral clustering result；

Evaluation module：For being evaluated further to cluster result.

11. devices according to claim 8 are it is characterised in that to each described relatedness of product single described in any two Time serieses are fitted, and described matching includes：

$S_t=\underset{i=1}{\overset{T}{\Σ}}{w}_i{x}_{t+1-i}=w_1{x}_t+w_2{x}_{t-1}+\mathrm{...}+w_t{x}_{t-T+1};$

Wherein, x_iFor the value in time serieses, w_iFor seasonal effect in time series weights, T is the size of time window.

12. devices according to claim 8 it is characterised in that to increase after three-dimensional incidence matrix on time dimension Carry out fusion and obtain two-dimentional incidence matrix, described fusion includes：

$H^{*}(x_i,y_i)=\underset{j=1}{\overset{n}{\Σ}}H(x_i,y_i,z_j);$

Wherein, n is described length of time series, H^*(x_i,y_i) be described two dimension incidence matrix on any point,For any point on described following three-dimensional incidence matrix.

13. devices according to claim 8 are it is characterised in that described spectral clustering module includes：

Similar diagram sets up module：For setting up similar diagram, the weighted adjacency matrix of similar diagram is W, and wherein, W=H*, H* are described Two-dimentional incidence matrix；

Laplacian Matrix computing module：For calculating Laplacian Matrix L, wherein,w_ijFor institute State any point in weighted adjacency matrix W, j is the columns in described weighted adjacency matrix W；

Laplacian Matrix decomposing module：For decomposing described Laplacian Matrix L=U Λ U^-1, wherein, U=[u₁,u₂,..., u_r],

$\mathrm{\Λ}=\left[\begin{array}{cccc}{\mathrm{\λ}}_1& & & \\ & {\mathrm{\λ}}_2& & \\ & & \mathrm{...}& \\ & & & {\mathrm{\λ}}_r\end{array}\right],$

[u₁,u₂,...,u_r] for L characteristic vector value, λ_iEigenvalue for L, and in Λ, λ₁≤λ₂≤...≤λ_r, r is W Order；

Second cluster module：Front k characteristic vector value for choosing described Laplacian Matrix L forms the matrix of a r*k, Using the every a line in described matrix as one of k dimension space vector, using clustering algorithm, described characteristic vector is gathered Class, k is the pre- number of clusters of spectral clustering.

14. devices according to claim 10 are it is characterised in that associate square according to described spectral clustering result to described two dimension Battle array is clustered further, and described cluster includes：

Wherein, S (k, m) is the amount of projects of described single product set distribution, and C is Binomial coefficient, m is the number in the warehouse needing to participate in calculating, and k is the pre- number of clusters of spectral clustering, and has m ＜ k.