A method and device for selecting a site for a big data forward warehouse

Through the big data forward warehouse site selection method, the convex hull algorithm and machine learning model are used to solve the problem that traditional site selection methods rely on personal experience and computational overhead, achieving more efficient site selection evaluation and stronger applicability.

CN117350631BActive Publication Date: 2024-11-15STONE TECH CO LTD
View PDF 6 Cites 0 Cited by

Patent Information

Application Number
CN202311319196.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-10-11
Publication Date
2024-11-15
Estimated Expiration
2043-10-11

AI Technical Summary

Technical Problem

The traditional forward warehouse site selection method relies on personal experience and lacks standardization, and the calculation overhead is large, making it difficult to effectively reduce the amount of model calculations, and lack of applicability.

Method used

Using the big data forward warehouse site selection method, by obtaining basic data and converting data to reduce capacity, using convex hull algorithm to optimize boundaries and judge data points within the region, combined with machine learning models for comprehensive evaluation, reduce calculation overhead and improve model performance.

Benefits of technology

A more objective and effective forward warehouse site selection evaluation is achieved, which reduces calculation overhead, improves the applicability and computing efficiency of the model, and forms a complete site selection evaluation system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117350631B_ABST
    Figure CN117350631B_ABST
Patent Text Reader

Abstract

The present invention discloses a method and device for selecting a site for a big data forward warehouse, which belongs to the field of data processing technology, wherein the method comprises: step S1: obtaining basic data; step S2: converting the basic data to reduce the data capacity; step S3: performing model calculation on the data processed by step S2; step S4: presenting the calculation result data to realize full-process business interaction. The present invention uses machine learning modeling to select a site for a forward warehouse, and adopts a data modeling method to improve the efficiency of site selection. Reduce model calculation overhead, improve system response speed, and optimize user perception.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a big data forward warehouse site selection method and device. Background Art

[0002] A forward warehouse is a chain store that sells fresh food and retail department stores in the form of warehousing. Customers do not need to go to the store to shop offline. They can directly place orders for goods through online apps or mini-programs, and store staff or professionals will complete the sorting and door-to-door delivery of goods. Depending on the content of the sale, the current forward warehouses mainly include fresh food, department stores, wine and beverages, etc. Common forward warehouses on the market include PuPu Life Supermarket, Daily Fresh, Meiyijia Youxuan, etc. Unlike traditional supermarkets and convenience stores, the forward warehouse has a relatively more prominent demand for distribution and customer sources. Each store of the forward warehouse is a small or medium-sized storage and distribution center, which allows the headquarters central warehouse to only supply goods to the store and cover the last mile. After the consumer places an order, the goods are shipped from a nearby retail store instead of from a warehouse far away in the suburbs. This allows it to be delivered within 3 kilometers of the store in 30 minutes.

[0003] According to iResearch, the scale of China's forward warehouse instant retail market has grown rapidly in recent years, and the number of forward warehouses has also continued to grow. Every time a new store is added, it involves the site selection of the store. In the life cycle of a chain store, "site selection" is a very important link. Coverage of user conditions, traffic convenience, competition, rent, etc. have relatively obvious specific significance for the site selection of the store.

[0004] However, both the traditional site selection method and the path-finding site selection model have their own shortcomings: the traditional site selection method is more inclined to the personal experience of the site selector, and the site selector is not adequately professionally trained. In the actual operation process, it relies more on personal experience and lacks objectivity. The applicable scenarios are relatively single, and the site selector's professional requirements are high. The site selection logic may vary greatly. After changing the scenario, the guiding value brought by the site selector's experience will be significantly reduced. There are few data collection and processing components, few analysis and operation tools, and a lack of standardized site selection evaluation tools.

[0005] Pathfinding algorithms such as the A-Star pathfinding algorithm are greatly affected by calculation accuracy in practical applications. The higher the road accuracy, the greater the amount of calculation. In application scenarios, site selection may need to consider path cost calculations within three or four kilometers. At the same time, combined with the fact that most deliveries are made by electric vehicles and motorcycles in real life, many county and township roads and various non-standard roads may need to be calculated during path calculation. This type of algorithm has a high cost when performing calculations.

[0006] Chinese invention patent CN 115034706 A discloses a method, device and computing equipment for determining the delivery range of a forward warehouse, wherein the method comprises: each sub-region includes a first-type sub-region and a second-type sub-region, and the delivery demand information of each sub-region is determined according to the historical order data corresponding to each sub-region; based on the delivery demand information of each sub-region and the delivery capacity of each forward warehouse in the area to be optimized, the unique forward warehouse corresponding to each sub-region is determined; for any forward warehouse, the delivery range of the forward warehouse is determined according to the sub-region corresponding to the forward warehouse. Summary of the invention

[0007] In response to the problems existing in the above-mentioned prior art, the present invention provides a big data forward warehouse site selection method and device. The technical problem to be solved by the present invention is to provide a big data forward warehouse site selection method and device with strong applicability that can reduce computing overhead and model calculation amount.

[0008] To achieve the above object, the present invention provides a method for selecting a big data forward warehouse site, the method comprising the following steps:

[0009] Step S1: Obtain basic data;

[0010] Step S2: converting the basic data to reduce the data capacity;

[0011] Step S3: performing model calculation on the data processed in step S2;

[0012] Step S4: Present the calculation result data to realize full-process business interaction.

[0013] Furthermore, the transformation of basic data includes constructing indicator features, fuzzifying boundaries, and determining whether data points are within the regional range.

[0014] Furthermore, the indicator characteristics include: basic characteristic data, regional characteristics and road information.

[0015] Furthermore, the indicator feature is constructed based on the basic requirement of "people and vehicles can arrive".

[0016] Furthermore, the indicator characteristics use the area that non-motor vehicles can reach as the data statistical range, and calculate the levels of various data indicators in the area.

[0017] Furthermore, a convex hull algorithm is used to optimize the region boundary and determine whether the data point is within the region.

[0018] Furthermore, in step S2, optimizing the region boundary using the convex hull algorithm includes the following steps:

[0019] (1) Sort the original points in the basic data from small to large longitude. If the longitudes are the same, sort them from small to large latitude.

[0020] (2) Traverse all original points, compare them with the previous points, and remove duplicate points;

[0021] (3) Record the convex hull boundary points;

[0022] (4) Determine whether the current point is an estimated convex hull point. If it is determined to be an estimated point, make judgments on the previous points in turn and delete non-convex hull data;

[0023] (5) Output the convex hull set st composed of convex hull points.

[0024] Furthermore, in step S2, using a convex hull algorithm to determine whether a data point is within the region includes the following steps:

[0025] (1) Traverse and calculate the distance between point p and the points in the convex hull set st. If all distances are greater than the set value, the judgment process ends and the output result is: outside the coverage range;

[0026] (2) The calculated point p and the convex hull set st are used as the original data point set, and the convex hull algorithm is run to optimize the region boundary. The new convex hull set st_new is output. After the calculation is completed, it is determined whether the calculated point p is in the new convex hull set st_new.

[0027] Furthermore, the basic data includes: web crawler data, public service data, operator data, forward warehouse merchant's own data and road data.

[0028] Furthermore, in step S3, model calculation includes data preparation, data set division, model training, hyperparameter adjustment, model generalization and model evaluation; the classification model is used for calculation, and combined with the softmax output score, to represent the comprehensive evaluation of the site selection point.

[0029] Furthermore, in step S4, the calculation result data is displayed on the map through the public map service interface and front-end and back-end network development technology, completing the full-process business interaction.

[0030] On the other hand, the present application provides a big data forward warehouse site selection device, which includes: a data acquisition module, a data conversion module, a model calculation module, a web front-end and back-end service management module and an external data access module. The device uses the above method to select a site for the forward warehouse.

[0031] Furthermore, in the data acquisition module, the existing forward warehouse stores are crawled as training data through scrapy and selenium technical tools, and the commercial activity information data in different areas are obtained through the public poi service; the operator information service interface is used to obtain the personnel and passenger flow indicator information in each area.

[0032] Furthermore, the data conversion module processes the basic data through basic alignment, and then further reduces the data capacity through the convex hull algorithm to simplify the model calculation.

[0033] Furthermore, a common classification model is used in the model calculation module, and combined with the softmax output score to represent the comprehensive evaluation of the site selection point.

[0034] Furthermore, in the web front-end and back-end service management module, data is displayed on the map through the public map service interface and front-end and back-end network development technology, completing the whole process business interaction.

[0035] Through a big data forward warehouse site selection method and device with the above characteristics, the impact of specific data on the model is reduced by using the "area that non-motorized vehicles can reach" to divide the data; the convex hull algorithm idea is used to reduce the model calculation amount and improve the model performance; the remaining points in the data set are wrapped inside to optimize the boundary and make the received data more comprehensive. Moreover, after optimizing the boundary through convex hull calculation, the number of points that make up the boundary range will be greatly reduced, which makes the calculation of determining whether the specified point is within the coverage area simple and efficient.

[0036] The present invention introduces big data and machine learning modeling, integrates multiple data sources, and forms a relatively objective and effective forward warehouse site selection evaluation model; with the help of the model's "one-time training, multiple uses" method, it reduces the computational overhead, and improves its applicability through the optimization processing of "model generalization" in the model algorithm; through a series of optimization technical means, it further reduces the model calculation amount and improves the model calculation efficiency; by combining model parameters, model scoring and other information, it forms a site selection evaluation system including data acquisition, data conversion, model prediction, and comprehensive evaluation. It uses a more generalized modeling idea, which is more versatile and extensible. Users can create new feature indicators according to business rules or introduce data from other business fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A process of a big data forward warehouse site selection method provided by an embodiment of the present invention Figure 1 ;

[0038] Figure 2A schematic diagram of a big data forward warehouse site selection device provided in an embodiment of the present invention Figure 1 ;

[0039] Figure 3 A process of a big data forward warehouse site selection method provided by an embodiment of the present invention Figure 2 ;

[0040] Figure 4 A schematic diagram of a 3km reachable range boundary provided by an embodiment of the present invention;

[0041] Figure 5 A schematic diagram of comparing a boundary after introducing a convex hull with an original boundary range provided by an embodiment of the present invention;

[0042] Figure 6 A schematic diagram of a big data forward warehouse site selection device provided in an embodiment of the present invention Figure 2 ; DETAILED DESCRIPTION

[0043] The preferred implementation modes of the present invention will be described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are not all embodiments of the present invention, and all other embodiments obtained by ordinary technicians in the field without making creative work are within the protection scope of the present invention.

[0044] In order to better understand the purpose, structure and function of the present invention, a big data forward warehouse site selection method of the present invention is further described in detail below in conjunction with the accompanying drawings.

[0045] Embodiment 1:

[0046] Figure 1 The following is an exemplary flowchart of a method for selecting a location for a big data forward warehouse provided by an embodiment of the present invention. The method includes:

[0047] Step S1: Obtain basic data;

[0048] Step S2: converting the basic data to reduce the data capacity;

[0049] Step S3: performing model calculation on the data processed in step S2;

[0050] Step S4: Present the calculation result data to realize full-process business interaction.

[0051] Embodiment 2:

[0052] Figure 1 The following is an exemplary flowchart of a method for selecting a location for a big data forward warehouse provided by an embodiment of the present invention. The method includes:

[0053] Step S1: Obtain basic data;

[0054] Step S2: converting the basic data to reduce the data capacity;

[0055] Step S3: performing model calculation on the data processed in step S2;

[0056] Step S4: Present the calculation result data to realize full-process business interaction.

[0057] The difference between this embodiment and embodiment 1 is that:

[0058] like Figure 3 and Figure 6 As shown, the basic data includes: web crawler data, public service data, operator data, forward warehouse merchant's own data and road data; the conversion of basic data includes constructing indicator features, fuzzy processing of boundaries and judging whether data points are within the regional scope; the indicator features include: basic feature data, regional features and road information; the indicator features are constructed with "people and vehicles can reach" as the basic requirement; the indicator features take the area that non-motorized vehicles can reach as the data statistical scope, and calculate the levels of various data indicators in the area.

[0059] Embodiment 3:

[0060] Figure 1 The following is a schematic diagram of a method for selecting a location for a big data forward warehouse provided by an embodiment of the present invention. The method includes:

[0061] Step S1: Obtain basic data;

[0062] Step S2: converting the basic data to reduce the data capacity;

[0063] Step S3: performing model calculation on the data processed in step S2;

[0064] Step S4: Present the calculation result data to realize full-process business interaction.

[0065] The difference between this embodiment and embodiment 2 is that Figure 4 and Figure 5 As shown, the convex hull algorithm is used to optimize the region boundary and determine whether the data point is within the region:

[0066] (i) In step S2, optimizing the region boundary using the convex hull algorithm includes the following steps:

[0067] (1) Sort the original points in the basic data from small to large longitude. If the longitudes are the same, sort them from small to large latitude.

[0068] (2) Traverse all original points, compare them with the previous points, and remove duplicate points;

[0069] (3) Record the convex hull boundary points;

[0070] (4) Determine whether the current point is an estimated convex hull point. If it is determined to be an estimated point, make judgments on the previous points in turn and delete non-convex hull data;

[0071] (5) Output the convex hull set st composed of convex hull points.

[0072] The specific implementation steps are as follows:

[0073] Input: A two-dimensional array poly(n-rows, 2-cols) consisting of the original boundary longitude and latitude points, where 2 means that each point consists of two data points, longitude and latitude, and n means there are a total of n longitude and latitude point sets.

[0074] Output: Data set st consisting of convex hull longitude and latitude points. The coverage formed by these convex hull points can cover all data points in the original boundary points. The convex hull points are saved in the same way as the original boundary points, that is, the longitude and latitude information of a single convex hull point is saved in an array, and all convex hull points are saved in a two-dimensional array.

[0075] (Note: In the algorithm calculation, a stack structure is needed to save the current convex hull set, and the abbreviation of the stack is used as the output result identifier)

[0076] initialization:

[0077] [1] A one-dimensional array visit(n) is used to identify whether a data point has been added to the convex hull set. The initial state of each point is false.

[0078] [2] is used to identify the next data point index ind to be accessed, initialized to 2.

[0079] [3] The stack data structure st used to record the current convex hull point set is initialized to empty.

[0080] Algorithm calculation steps:

[0081] (1) Sort the original points poly from small to large longitude. If the longitudes are the same, sort them from small to large latitude. (2) Traverse the points in the poly and compare them with the previous point. If they are the same, remove them to remove duplicate points in the data. (3) The leftmost point and the rightmost point must be the boundary points of the convex hull, recorded as l and r respectively.

[0082] (4) Push the two points with indexes 0 and 1 into the stack, and define visit[0] = true and visit[1] = true

[0083] (5) Traverse in counterclockwise order and make position judgment. Take the two top points a (the second element from the top of the stack) and b (the top element of the stack), take the next untraversed element poly[ind] from the original data set poly, and calculate the vector cross product between the vector AB composed of a and b and the vector APOLY composed of a and poly[ind]:

[0084] [5.1] AB x APOLY < 0, then the APOLY vector is on the right side of the AB vector, and the poly[ind] point is on the right side of the extension line of the line connecting the points a and b, that is, the convex hull formed by the line connecting a and poly[ind] can enclose the point b. At this time, pop the top element b from the stack and define visit[b] = false.

[0085] [5.2]AB x APOLY>=0. At this time, it is impossible to determine whether the current point is a convex hull point. Push it into the stack and temporarily regard it as a potential convex hull data. Define visit[poly[ind]]=true.

[0086] (6) Loop to calculate the next ind position

[0087] [6.1] ind == 0, indicating that all points have been traversed, ending the traversal in (5)

[0088] [6.2a]visit[r]==true, indicating that the rightmost point has been added to the convex hull, the convex hull below the line connecting l and r has been calculated, and the convex hull data will be calculated. At this time, ind=ind-1

[0089] [6.2b]visit[r]==true, indicating that the rightmost point has not been visited yet and the lower convex hull calculation is still in progress.

[0090] =ind+1

[0091] [6.3a]visit[ind]==true, indicating that the point ind has been added to the convex hull set and no further judgment is required. Return to step 6.1

[0092] [6.3b]visit[ind] == false. At this time, a new point that has not been added to the convex hull is found. Jump out of the loop in (6) and return to (5) to determine whether it is a convex hull point.

[0093] The traversal is completed and the st array is returned.

[0094] The purpose of the above method is to use the smallest data point to wrap the rest of the points in the data set, thereby optimizing the data boundary.

[0095] For better understanding, the specific code examples are as follows:

[0096] #Calculate the cross product of vectors p1p2 and p1p3

[0097] def__direction(self,p1,p2,p3)->bool:

[0098] #Judge whether vector p1p2 is on the left of p1p3

[0099] #Less than 0 is on the right, greater than 0 is on the left, and equal to 0 is on the same line

[0100] assert isinstance(p1,list)and isinstance(p2,list)and isinstance(p3,list)

[0101] assert len(p1)==2and len(p2)==2and len(p3)==2

[0102] [x1,y1]=[p2[0]-p1[0],p2[1]-p1[1]]

[0103] [x2,y2]=[p3[0]-p1[0],p3[1]-p1[1]]

[0104] onTheLeft=x1*y2-x2*y1>0

[0105] return onTheLeft

[0106] #Calculate the convex hull of the specified area

[0107] def convex(self,_poly)->list:

[0108] poly=_poly.copy()

[0109] poly.sort()

[0110] polyND=[]

[0111] for vec in poly:

[0112] if len(polyND)>0and vec[0]==polyND[-1][0]and vec[1]==polyND[-1][1]:

[0113] continue else:

[0114] polyND.append(vec)

[0115] polyND.sort()#horizontal sorting

[0116] l0 = polyND[0] # The leftmost endpoint must be a convex hull vertex

[0117] r0 = polyND[len(polyND)-1] #The rightmost endpoint must be the convex hull vertex

[0118] st=[]

[0119] vis={}

[0120] st.append(polyND[0])

[0121] st.append(polyND[1])

[0122] vis[str(polyND[0])]=0

[0123] vis[str(polyND[1])]=1

[0124] ind=2

[0125] while(st[len(st)-1]!=l0):

[0126] #Judge whether the current viewing point is on the left or right of the vector composed of the two elements at the top of the stack while(len(st)>1and not self.__direction(st[len(st)-2],st[len(st)-1],polyND[ind])):

[0127] vis.pop(str(st[len(st)-1]))

[0128] st.pop()

[0129] st.append(polyND[ind])

[0130] vis[str(polyND[ind])]=ind

[0131] #Find the next ind

[0132] while(vis.get(str(polyND[ind]))):

[0133] ind=ind-1if vis.get(str(r0))else ind+1

[0134] return st

[0135] (ii) Figure 4 and Figure 5 As shown, in step S2, using the convex hull algorithm to determine whether the data point is within the region includes the following steps:

[0136] (1) Traverse and calculate the distance between point p and the points in the convex hull set st. If all distances are greater than the set value, the judgment process ends and the output result is: outside the coverage range;

[0137] (2) The calculated point p and the convex hull set st are used as the original data point set, and the convex hull algorithm is run to optimize the region boundary. The new convex hull set st_new is output. After the calculation is completed, it is determined whether the calculated point p is in the new convex hull set st_new.

[0138] The specific implementation steps are as follows:

[0139] Input: A one-dimensional array of points p(2), where 2 means that the array contains only two data points: longitude and latitude.

[0140] Output: Is p within or outside the coverage range?

[0141] Calculation process:

[0142] (1) Traverse and calculate the distances between point p and the points in the convex hull set st. If all distances are greater than 3 km, the judgment process ends and the output result is: outside the coverage range.

[0143] (2) Take the point p and the icon set st as the original data point set, run the convex hull boundary algorithm in Example 2, and output the new convex hull set st_new. After the calculation is completed, determine whether p is in the new convex hull set st_new:

[0144] [a]st_new.find(p) == true: After introducing point p, a new convex hull boundary can be formed by point p, that is, p is outside the original convex hull boundary

[0145] [b]st_new.find(p) == false: After the point p is introduced, a new convex hull boundary cannot be formed from the point p, that is, p is within the original convex hull boundary

[0146] The above two steps (1) and (2) are used to determine whether a point is within the boundary range. The first step traverses the distance between all points and the current point, and the algorithm complexity is O(n). The second step is the convex hull algorithm process, and the algorithm complexity is O(nlog(n)). The total algorithm complexity is O(nlog(n)).

[0147] In addition, in practical applications, the above algorithm is mainly applicable to the scenario of judging various types of POIs or other element information (such as the number of hospitals, the number of schools, etc.) within the convex hull range of 3km with the estimated site selection point p as the center. Assuming that the total number is m, the algorithm complexity is O(mnlog(n)). Considering that the order of magnitude of the convex hull point data volume n is basically below 1000, the data range of POI, i.e. the data volume of other element information m, is basically within 10000, the overall complexity can be controlled within 1e8, and the calculation can be completed within 1 second using a regular server, which meets the real-time requirements.

[0148] For better understanding, the specific code examples are as follows:

[0149] #Calculate the distance between two points

[0150] def geoDist(self,lon1,lat1,lon2,lat2):

[0151] #Calculate the straight-line distance between two points Unit: m

[0152] [lon1,lat1,lon2,lat2]=map(math.radians,[float(lon1),float(lat1),float(lon2),float(lat2)])

[0153] lat=abs(lat1-lat2)

[0154] lon=abs(lon1-lon2)

[0155] val=pow(math.sin(lat / 2),2)+math.cos(lat1)*math.cos(lat2)*pow(math.sin(lon / 2),2)

[0156] dist=2*6378.137*math.asin(pow(val,0.5))

[0157] return 1000*dist

[0158] #Calculate whether the point is in the convex hull plane

[0159] def withinPolygon(self,point,_poly):

[0160] poly=_poly.copy()

[0161] poly.append(point)

[0162] conv = self.convex(poly)

[0163] for [lng, lat] in conv:

[0164] if lng==point[0]and lat==point[1]:

[0165] return False

[0166] return True

[0167] Embodiment 4:

[0168] Figure 1 The following is an exemplary flowchart of a method for selecting a location for a big data forward warehouse provided by an embodiment of the present invention. The method includes:

[0169] Step S1: Obtain basic data;

[0170] Step S2: converting the basic data to reduce the data capacity;

[0171] Step S3: performing model calculation on the data processed in step S2;

[0172] Step S4: Present the calculation result data to realize full-process business interaction.

[0173] The present embodiment is different from the above embodiments in that:

[0174] like Figure 3 As shown, in step S3, model calculation includes data preparation, data set division, model training, hyperparameter adjustment, model generalization and model evaluation; the classification model is used for calculation, and combined with the softmax output score, to represent the comprehensive evaluation of the site selection point.

[0175] Data preparation mainly refers to integrating and aggregating data from various channels and storing them in Excel or other standard data files; data set division is used in the model training process, by dividing the collected and sorted data into training set, validation set and test set, and performing model training, model adjustment and model effect evaluation in the modeling process; model training refers to the process of learning the features in the data in a progressive parameter learning manner to form a model. Most common models can be represented by a multivariate equation, such as y=ax^3+by^2+cz, where abc is the model parameter, that is, training The parameters to be learned during training; hyperparameters specifically refer to other parameters other than model parameters, such as the number of training iterations, training step size, etc. In model training, when the amount of data, data features, and model complexity do not match, the model will learn all the features in this batch of data sets. At this time, for a new data, it may cause the model's judgment to be abnormal, that is, overfitting; model generalization refers to the use of regular processing or the addition of penalty coefficients to reduce the learning of specific data in the data set and make the model more general; after the model training is completed, the model's capabilities need to be evaluated to generate a model with certain confidence guidance capabilities. In the classification model, a classification result for this data will be output. In many scenarios, we not only need the classification result, but also its confidence score, for example, we need to know how confident we are in judging it as a certain category. The common processing method is to use the softmax function to output the probability of each category and use it as the corresponding scoring scenario.

[0176] In step S4, the calculation result data is displayed on the map through the public map service interface and front-end and back-end network development technology, completing the full-process business interaction.

[0177] On the other hand, in order to better understand the purpose, structure and function of the present invention, a big data forward warehouse site selection device of the present invention is further described in detail below in conjunction with the accompanying drawings.

[0178] Embodiment 5:

[0179] like Figure 2 As shown, a big data forward warehouse site selection device includes a data acquisition module, a data conversion module, a model calculation module, a web front-end and back-end service management module and an external data access module. The device adopts the above method to select the site for the forward warehouse.

[0180] like Figure 3As shown, in the data acquisition module, the existing forward warehouse stores are crawled as training data through scrapy and selenium technical tools, and the commercial activity information data in different regions are obtained through the public poi service; the operator information service interface is used to obtain the personnel and passenger flow indicator information in each region.

[0181] The data conversion module processes the basic data through basic alignment, and then further reduces the data capacity through the convex hull algorithm to simplify the model calculation.

[0182] A common classification model is used in the model calculation module, and the softmax output score is combined to represent the comprehensive evaluation of the site selection point.

[0183] In the web front-end and back-end service management module, data is displayed on the map through the public map service interface and front-end and back-end network development technology, completing the full-process business interaction.

[0184] It can be understood that various different implementation modes can be further derived by combining or modifying different implementation modes and various technical features and steps in different ways.

[0185] The above describes the big data forward warehouse site selection method and device of the preferred embodiment of the present invention in combination with the specific embodiment. It can be understood that the above description is only exemplary and not restrictive, and other variations and modifications can be implemented by those skilled in the art without departing from the spirit and scope of the present invention. All these variations and modifications fall within the scope of the present invention.

Claims

1. A big data forward warehouse site selection method, characterized in that: The following steps are involved: Step S1: Obtain basic data; Step S2: converting the basic data to reduce the data capacity; the conversion of the basic data includes constructing index features, fuzzifying the boundaries and judging whether the data points are within the regional range; the index features take the area that non-motor vehicles can reach as the data statistical range, and calculate the levels of various data indicators in the area; using the convex hull algorithm to optimize the regional boundaries and judge whether the data points are within the area; Among them, the optimization of the region boundary using the convex hull algorithm includes the following steps: (1) Sort the original points in the basic data from small to large longitude. If the longitudes are the same, sort them from small to large latitude. (2) Traverse all original points, compare them with the previous points, and remove duplicate points; (3) Record the convex hull boundary points; (4) Determine whether the current point is an estimated convex hull point. If it is determined to be an estimated point, make judgments on the previous points in turn and delete non-convex hull data; (5) Output the convex hull set st composed of convex hull points; Using the convex hull algorithm to determine whether a data point is within the region includes the following steps: (1) Traverse and calculate the distance between point p and the points in the convex hull set st. If all distances are greater than the set value, the judgment process ends and the output result is: outside the coverage range; (2) Take the calculated point p and the convex hull set st as the original data point set, run the convex hull algorithm to optimize the region boundary, output the new convex hull set st_new, and after the calculation is completed, determine whether the calculated point p is in the new convex hull set st_new; Step S3: performing model calculation on the data processed in step S2; Step S4: Present the calculation result data to realize full-process business interaction.

2. The big data forward warehouse site selection method according to claim 1 is characterized in that ,The index characteristics include: basic characteristic data, regional ,characteristics and road information.

3. The big data forward warehouse site selection method according to claim 1 is characterized in that: The basic data include: web crawler data, public service data, operator data, forward warehouse merchant’s own data and road data.

4. The big data forward warehouse site selection method according to claim 1 is characterized in that: In step S3, model calculation includes data preparation, data set division, model training, hyperparameter adjustment, model generalization and model evaluation; the classification model is used for calculation, and combined with the softmax output score, to represent the comprehensive evaluation of the site selection point.

5. The big data forward warehouse site selection method according to claim 1 is characterized in that: In step S4, the calculation result data is displayed on the map through the public map service interface and front-end and back-end network development technology, completing the full-process business interaction.

6. A big data forward warehouse site selection device, characterized in that: The device includes a data acquisition module, a data conversion module, a model calculation module, a web front-end and back-end service management module and an external data access module. The device selects a location for a forward warehouse using the method described in any one of claims 1 to 2.

Citation Information

Patent Citations

  • Method and device for determining distribution range of front warehouse and computing equipment

    CN115034706A

  • Warehousing and distribution network optimization method and device, storage medium and program product

    CN114676875A

  • Fresh food e-commerce preposed warehouse site selection-path planning method

    CN114897217A

  • Distribution task distribution method and device

    CN115511302A

  • Tobacco industry distribution line optimization method, system, equipment and medium

    CN116862091A