CN111539764A - Big data multiple access selection method based on submodular function - Google Patents

Big data multiple access selection method based on submodular function Download PDF

Info

Publication number
CN111539764A
CN111539764A CN202010305284.0A CN202010305284A CN111539764A CN 111539764 A CN111539764 A CN 111539764A CN 202010305284 A CN202010305284 A CN 202010305284A CN 111539764 A CN111539764 A CN 111539764A
Authority
CN
China
Prior art keywords
data
tree
model
multiple access
area
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202010305284.0A
Other languages
Chinese (zh)
Other versions
CN111539764B (en
Inventor
陈兴国
林洁
刘厚涛
吴多丰
朱洁
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Nanjing University of Posts and Telecommunications
Original Assignee
Nanjing University of Posts and Telecommunications
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Nanjing University of Posts and Telecommunications filed Critical Nanjing University of Posts and Telecommunications
Priority to CN202010305284.0A priority Critical patent/CN111539764B/en
Publication of CN111539764A publication Critical patent/CN111539764A/en
Application granted granted Critical
Publication of CN111539764B publication Critical patent/CN111539764B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/02—Marketing; Price estimation or determination; Fundraising
    • G06Q30/0201—Market modelling; Market analysis; Collecting market data
    • G06Q30/0204—Market segmentation
    • G06Q30/0205—Location or geographical consideration
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Strategic Management (AREA)
  • Finance (AREA)
  • Development Economics (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Accounting & Taxation (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Game Theory and Decision Science (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • Medical Informatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Electrically Operated Instructional Devices (AREA)

Abstract

The invention provides a big data multiple access selection method based on a submodular function, which comprises the following steps: step 1: collecting a plurality of characteristics of existing stores in an area to be addressed, wherein the characteristics can influence the quality of the address selection, and constructing a sample set; step 2: training evaluation models of a plurality of base learners by utilizing a sample set, and combining the base learners into a strong learner by utilizing a bagging integration method, namely a final address selection model; and step 3: respectively inputting the characteristics of a plurality of candidate stores in the area to be addressed into the addressing model, wherein the candidate store corresponding to the maximum output value is the target store; and 4, step 4: and (4) removing the selected target store and other candidate stores within the influence radius, and repeating the step 3 to obtain the next target store. The invention constructs the site selection model with wider coverage, and finally realizes the site selection strategy with scientificity, standardization and strong sustainability.

Description

Big data multiple access selection method based on submodular function
Technical Field
The invention relates to the field of machine learning and big data, in particular to a big data multiple access selection method based on a submodular function.
Background
With the development of social economy, the linkage industry is in the spotlight. Chain enterprises are becoming active in the catering, retail and hospitality industries. The site selection is a very important thing for a chain of enterprises, and may even be an important link for determining the success or failure of enterprise operation, because many other factors influencing the enterprise operation condition, such as marketing, personnel service, price, etc., can be changed faster according to the current condition so as to better adapt to the current development condition, and once the site selection is finished, the position cannot be changed easily, and meanwhile, the customer group covered by the shop itself is also mostly determined. Therefore, when a new store is to be started, the initial site selection work is most important, because it often determines the life and death of the store.
In small and medium-sized chain enterprises, site selection is judged to be good or bad by depending on the site selection experience of the enterprises for many years, and scientificity and standardization are lacked. The method has the advantages that firstly, scientific basis is not provided, and success or failure has certain contingency and great risk; secondly, even if one or a plurality of chain stores succeed, the possibility of quick re-copying is not provided. In addition, a relatively scientific and standardized site selection mode can be established by a small part of other enterprises, and the enterprises can manually collect some relevant data, such as business circles around the candidate stores, people flow, surrounding traffic road conditions and other factors related to success or failure of site selection results. The data is then collated and analyzed to finally select a relatively suitable address. This solution is certainly much better than the previous one, which is purely addressed empirically, but has some inherent drawbacks, namely that the data is collected manually, the period is long, and the comprehensiveness and accuracy of the data are not very high, and it is difficult to quickly copy the data to other sites of stores. The second is the inherent problem of site selection of chain stores, the problem of mutual interference caused by a plurality of stores, when the number of stores is large, the influence ranges of the stores are overlapped, and the income of a single store is negatively influenced.
Some people have optimized the traditional site selection method, especially under the condition of rapid development of current big data and artificial intelligence. In an addressing method and apparatus (application publication No. 108984561A, application No. 201710405595.2), the system generates frequent trajectories of a user by obtaining and analyzing spatiotemporal data of the user to determine alternative addresses. The strategy highlights the influence of the human flow on site selection, and the influence of other conditions on the site selection cannot be analyzed and measured. In another site selection model construction and site selection method, device and equipment (application publication number: 110009379A, application number: 201811428103.2), the system ranks the business conditions of merchants as labels, and trains a model by using position information and transaction characteristics as characteristic values. This method is only applicable to companies that offer payment methods such as payers, and is not applicable to individuals and small enterprises that want site-selection analysis because countries and enterprises do not publish the business situation of a single store. In a GIS-based bus station site selection method (application publication No. 108388970A, application No. 201810238719.7), a self-defined distance estimation value is used as output, and a linear regression method is used for site selection after relevant influence factors are collected. The method uses a linear model for analysis, ignores the mutual influence among characteristics and the nonlinear causal relationship, is easily influenced by data with larger deviation, and causes the real situation that the model can not well reflect. In a business site selection method and system (application publication No. 107909105A, application No. 201711117740.3), it is judged whether a place is suitable as a site selection place by analyzing characteristics of satellite data and city data. The method uses a logistic regression mode to judge whether the region is suitable for certain type of shop opening, all the regions with the type are set as 1, the current situation that shops in many industries bloom in a passing way is ignored, and the result of the logistic regression is difficult to select a better place in a plurality of suitable preselected regions.
After the existing site selection system establishes a model, only a single site can be selected as a preselected area through the model each time. In order to set up a plurality of stores, the model must be operated for a plurality of times or areas with previous effects must be selected as site selection sites, and the influence of mutual competition with the existing stores is not considered. Therefore, such a site selection method cannot meet the requirement that the coverage range of opening stores pursued by chain enterprises is as large as possible. The problem of trying to cover a larger area in a limited number of open stores is an NP-hard problem. The advantage of using the method of the submodular function to solve the problem of maximum coverage of the store chain is that a best possible solution can be selected in a relatively short time.
Disclosure of Invention
In order to overcome the defects of the prior art, the invention provides a big data multiple access selection method based on a submodular function. The method takes the online evaluation quantity of stores as a standard for evaluating the site selection of the stores, and in addition, data is collected by means of big data, including taxi GPS data, the quantity of shared single cars, pedestrian heat and the like as influence factors.
In order to achieve the above effects, the technical scheme of the invention is designed as follows:
the method comprises the following steps: and selecting proper candidate cities, wherein the cities are developed economically, the quantity and the types of chain stores are many, and various types of related data of the cities can be obtained relatively easily. The method has the advantages that enough data volume can be ensured when the site selection model is established later, so that a more objective and effective model is established;
step two: in some embodiments, wherein: the target area is divided into grids, and a set of all grids is established, so that the target problem becomes how to select a certain grid area from the set to make the addressing result best. The method has the advantages that the problem that the user can select at any place is changed into the problem that the user can select in a limited number, the problem solving difficulty is reduced, and meanwhile, the accuracy can be ensured;
step three: the category of stores to be addressed is selected, and the chain of businesses with the most stores from the stores of the category in the city is selected as the standard for the address selection of the stores. The method has the beneficial effect that the size of the data volume has great influence on the accuracy of model establishment. Meanwhile, the weight of the influence factors of different industry types for site selection is different. Therefore, the enterprise with the largest number of stores in the same industry in the current area is selected to establish the site selection model of the industry.
Step four: in some embodiments, wherein: the evaluation quantity of the same chain store on related websites in different regions is collected as the basis for grading the site selection, and the influence of other factors except the region on the store is avoided, so that the result is more objective and accurate. The restaurant evaluation method has the advantages that the evaluation standard of a restaurant can have a plurality of different angles, such as turnover, flow rate, evaluation number and the like. Unfortunately, some data are difficult to obtain and belong to business secrets, so the evaluation number on the network is selected as an index for evaluating the position selection, and the data are relatively easy to obtain and can reflect the position selection of the store;
step five: in some embodiments, wherein: by applying big data and crawler technology, data including resident data, pedestrian data, public transportation data, cost and the like are obtained from websites such as popular comment, Gaode map, 58 city of the same city, Hello bicycle and the like. The crawler-based data acquisition system has the advantages that through a crawler means, enough and comprehensive data can be collected, so that the influence of various factors on site selection can be better shown;
step six: in some embodiments, wherein: processing the original data collected in the last step, namely processing data which can show problems in the original file, such as cleaning missing data and abnormal values which may exist; secondly, various characteristics belonging to different areas are summarized and normalized or standardized, so that troubles caused by data of different scales are avoided. The method has the advantages that the collected data cannot be directly used, the data needs to be converted into data meeting requirements according to the requirements, and the standardization and normalization are convenient for the next step, so that the model is established more quickly and accurately;
step seven: in some embodiments, wherein: and modeling the address by using a machine learning method, wherein the bagging machine learning method is used. The basis learner selects a linear regression model, a decision tree and an XGboost respectively, and then synthesizes the models to generate a final model. The method has the advantages that the final model effect obtained by the bagging method is generally higher than that of a single learner, certain noises in the base learner can be overcome, when the single learner has problems, the final result cannot be influenced, and the method is strong in robustness. Meanwhile, the base learners have no strong dependence relationship, and the accuracy of a single learner can be obtained. Here we use these three different basis learners: the linear regression model is a model which is applied more at present and is used in the problem to see the accuracy rate; the XGboost method is a relatively popular machine learning method recently, and is often the best method in various machine learning competitions, and the Xgboost and related methods are not used as precedent in the problem of site selection before the Xgboost method is used; the reason for selecting the decision tree model is that the decision tree is also a classical machine learning method, and the basis of the integrated method like XGboost is the decision tree, so that the decision method can be compared with the XGboost, and how the improvement effect of the XGboost is compared with that of the decision tree is checked.
Step eight: in some embodiments, wherein: and deleting the grids where the stores are located and the grids which can be covered from the set, selecting the grids with the highest scores from the rest grids as areas for next address selection by using an address selection model, and repeating the process until the number of the stores which need to be selected is reached. The method has the advantages that the problem of maximum coverage of site selection is solved by using an optimization algorithm of a submodular function, so that site selection of a plurality of stores is more systematic and comprehensive.
Drawings
FIG. 1 is a flow chart of an address selection entity;
FIG. 2 is a flow chart of machine learning model building;
FIG. 3 is a flow chart of address selection of a submodular function.
Detailed Description
The invention provides a big data multiple access selection method based on a submodular function, a flow chart of which is shown in figure 1 and comprises the following steps:
selecting a proper city as a target area of site selection according to the developed degree of economy, culture and related industries;
furthermore, the network grid is used for dividing the preselected area, the purpose of selection is changed, and the best site selection point is selected from all points on the map into the best site selection area selected from a certain number of intervals. The size of the grid of the network may be a fixed value, for example, 80% of the customers in a particular store are within a distance of 5km, and the side length of the selected grid is 5 km; or setting the side length of the grid as a variable, establishing a model for each grid within the range of 1km to 10km and the interval of 200m, then carrying out error analysis on the test set, and selecting the side length of the grid with the minimum error.
Further, the number of preselected intervals may be reduced. It goes without saying that from the viewpoint of the terrain, not every area is functionally suitable as an area for addressing, and the number of addressing can be reduced. Finally, numbering the preset areas and putting the preset areas into a set;
and step two, determining the industry category of the store to be addressed, collecting several chain store brands with a large number and a widest coverage in the current industry in the current city, putting the stores of the chain store brands into a set for training as samples, and searching and sequencing the chain store brands from each large navigation or life software in a specific searching mode. And then, collecting factors influencing the site selection and the operation condition of the store, including resident data, pedestrian data, public transportation data, cost and website evaluation quantity information for evaluating the site selection of the store.First collecting the prices of the surrounding residences may approximately reflect the consumer's ability to consume. The number of pedestrians in the coming and going department directly influences whether a store can have a certain pedestrian flow, and here, taxis, data of shared bicycles and thermodynamic diagrams of the pedestrians are collected to serve as evaluation indexes of the conditions of the pedestrians. Secondly, the convenience of public transportation is a very important factor for measuring whether stores have potential markets. The number of surrounding bus stops, subway exits and parking lots is collected through map software to serve as factors for judging whether public transportation is convenient or not. Again, another very important consideration in store location is cost, which primarily includes store rentals. And finally, the number of other shops with the same function in the same area is the number of other brands, and the commenting number of the other shops on the network is taken as a measure for the competitiveness of the shops. Wherein x is1The number of times of stopping in the region in taxi GPS data; x is the number of2Is the bicycle data in the area, x3Is the pedestrian heat condition in the area; x is the number of4Is a subway station within a region, x5Is the number of bus stops in the area; x is the number of6Is the number of parking lots; x is the number of7Means the number of rentals, x, of the store8The evaluation quantity of competitors in the same area is referred to; x is the number of9Which refers to the rent of the surrounding residential areas.
And thirdly, putting the collected data of each characteristic into an area in the set according to the longitude and latitude information, and counting the quantity of each characteristic in different areas to be used as each characteristic index of the area.
And step four, establishing a final addressing model by using a machine learning method, wherein the flow of the model is as shown in figure 2. First we take a random sample from the set for each base learner that has been replaced, the number of times being the same as the number of elements in the set. Secondly, three different machine learning methods are adopted to respectively establish a model, a base learner is respectively trained by using linear regression, a decision tree and an XGboost algorithm, and then the mean value of the model is calculated to be used as a final result.
Wherein the linear regression model is:
y＝β+α1x1+α2x2+α3x3+α4x4+α5x5+α6x6+α7x7+α8x8+α9x9
in the formula, y is the prediction output of the linear regression model and is used for representing the difference between the prediction result of the model and the actual value, (a)1,a2,a3,a4,a5,a6,a7,a8,a9) Is a parameter of various characteristics, parameter (a)1～a9) The method is obtained by learning a linear regression model, and the specific learning method can be obtained by using a gradient descent method to indicate the influence degree of various characteristics on the result, xiAre numerical values for each of the various characteristics,
further, because of the regression model, the decision tree we choose is the classification and regression tree cart (classification and regression tree). The core step is to select the cut point of the bifurcation feature. The formula is as follows:
Figure BDA0002455553240000061
in the formula, j and s respectively represent segmentation attribute and segmentation value, x and y respectively represent feature set and output of current data, i represents serial number of current data, and c1And c2The values of the two diverging nodes representing the current node of the decision tree, respectively.
And further establishing a model by adopting an XGboost method. XGBoost is a very popular and very effective learning method in data science competitions recently. XGBoost is actually a combinatorial optimization method of multiple CART (classified in a regression tree) with the result that all trees are linearly summed. Meanwhile, a progressive relation exists between the multiple classifications and the regression tree, and input data of the nth classification and the regression tree is a difference value between an added value and a true value of a fitting result of the previous n-1 trees. Each classification and regression tree is a binary tree, the condition of splitting the left and right nodes downwards each time is each value of each feature, the condition with the minimum loss is selected to obtain two leaf nodes, and then traversal is continued in the same mode. The splitting is finished under the condition of adding a threshold, when the gain is larger than the set threshold, the node is continuously split, otherwise, the splitting is finished. The specific process is as follows:
1: the building of the first tree is completed. Splitting from the root node to the bottom, and selecting a classification point with the best characteristic to maximize Gain, wherein the Gain is calculated by the following method,
Figure BDA0002455553240000062
wherein G islAnd GRRespectively, the first derivative, H, of the right child node after splittinglAnd HRThe second derivatives of the left and right subnodes after classification, respectively, and gamma is the threshold value customized to control whether splitting occurs.
2: when the tree is no longer split to a predetermined depth, the leaf node values can be calculated. The calculation method of the leaf node is as follows:
Figure BDA0002455553240000063
wherein g isjAnd hiHeterodynia and second derivative, respectively, and λ is the regularization value.
3: the splitting method and the generation of leaf nodes for the latter tree are the same as for the first tree, the only difference being that the latter tree needs to be updated with y on the basis of the results of the former tree, in other words to fit the residual of the former tree.
4: when all trees are completed, predict values
Figure BDA0002455553240000064
ft(x) The T-th tree is shown, where T is the number of trees.
Finally, we take the value obtained by arithmetically averaging the obtained results of the 3 basis learners as the result of the final model.
Step five, after the site selection model is completed, another unique problem of site selection of a plurality of stores needs to be solved, namely how to cover a larger range by using a smaller number of stores, so that the advantage of expanding the chain stores in a healthy posture instead of random blind copying is realized, and meanwhile, mutual influence and shunting among the stores are avoided, and the profit of each store is ensured. Meanwhile, a chain store with a larger coverage area, and the larger coverage area means more propaganda and plays a role in popularizing brands. The maximum coverage problem is an NP-hard problem, where we use a submodular approach to solve the maximum coverage problem. The submodular function means that the formula is satisfied:
f(x+i)-f(x)≥f(y+i)-f(y)if
Figure BDA0002455553240000071
we first divide the map of the whole area into 5km by 5km grids, here we set the influence range of the store to k with the expert's advice we set all grids as a complete set, delete the grid with the store in the grid and the grids within the circle with radius k that it can influence from the set, then use the trained model to select the grid with the best evaluation as the location of the next store in the remaining grids, and delete the grids that the new store can influence in the remaining grids, and then continue to loop in this way until the number of additional stores needed is reached.
Compared with the existing site selection scheme, the invention has the advantages that: the site selection model is constructed by combining a big data technology and an artificial intelligence algorithm, and a scientific, standardized and highly continuous solution is provided for site selection of a plurality of stores by utilizing an optimization method of a submodular function.

Claims (7)

1. A big data multiple access selection method based on a submodular function is characterized by comprising the following steps:
step 1: collecting a plurality of characteristics of existing stores in an area to be addressed, wherein the characteristics can influence the quality of the address selection, and constructing a sample set;
step 2: training evaluation models of a plurality of base learners by utilizing a sample set, and combining the base learners into a strong learner by utilizing a bagging integration method, namely a final address selection model;
and step 3: respectively inputting the characteristics of a plurality of candidate stores in the area to be addressed into the addressing model, wherein the candidate store corresponding to the maximum output value is the target store;
and 4, step 4: and (4) removing the selected target store and other candidate stores within the influence radius, and repeating the step 3 to obtain the next target store.
2. The big data multiple access selection method based on the submodular function according to claim 1, wherein the characteristics in step 1 comprise resident data, pedestrian data, public transportation data, cost data and website evaluation quantity information for evaluating the shop site.
3. The big data multiple access selection method based on the submodular function according to claim 2, wherein the residential data is rent of surrounding residential areas; the pedestrian data comprises the number of times of stopping in the region in taxi GPS data, the single-car data in the region and the pedestrian heat condition in the region; the public transportation data comprises subway stations in the region, the number of bus stations in the region and the number of parking lots; the cost data is the number of rentals of the store.
4. The big data multiple access selection method based on the submodular function of claim 3, wherein the specific steps of step 2 comprise:
step 2.1: training a first base learner by using a linear regression model;
step 2.2: training a second base learner by using the classification and regression tree model;
step 2.3: training a third base learner by using XGboost;
step 2.4: and combining the three base learners into a strong learner by using a bagging integration method, namely a final address selection model.
5. The big data multiple access selection method based on the submodular function of claim 4, wherein the linear regression model is:
y＝β+α1x1+α2x2+α3x3+α4x4+α5x5+α6x6+α7x7+α8x8+α9x9
y is the predicted output of the linear regression model, representing the difference between the predicted result and the actual value of the model, α1，a2，α3，α4，a5，a6，a7，a8，a9Are parameters of various characteristics; x is the number of1The number of times of stopping in the area in taxi GPS data, x2Is the bicycle data in the area, x3Is the pedestrian heat condition in the area; x is the number of4Is a subway station within a region, x5Is the number of bus stops in the area; x is the number of6Is the number of parking lots; x is the number of7Means the number of rentals, x, of the store8The evaluation quantity of competitors in the same area is referred to; x is the number of9Which refers to the rent of the surrounding residential areas.
6. The big data multiple access selection method based on the submodular function of claim 4, wherein the classification and regression tree model is:
Figure FDA0002455553230000021
in the formula, j and s respectively represent segmentation attribute and segmentation value, x and y respectively represent feature set and output of current data, i represents serial number of current data, and c1And c2The values of the two diverging nodes representing the current node of the decision tree, respectively.
7. The big data multiple access selection method based on the submodular function of claim 4, wherein the specific steps of step 2.3 comprise:
step 2.3.1: completing the establishment of the first tree; splitting from the root node to the bottom, selecting a classification point with the best characteristic to maximize Gain, wherein the Gain is calculated by the following method,
Figure FDA0002455553230000022
wherein G islAnd GRRespectively, the first derivative, H, of the right child node after splittinglAnd HRSecond derivatives of the classified left and right child nodes, respectively, and gamma is a threshold value used for self-defining to control whether splitting is performed or not;
step 2.3.2: calculating the values of the leaf nodes when the tree is not split any more to a preset depth; the calculation method of the leaf node is as follows:
Figure FDA0002455553230000023
wherein g isjAnd hiHeterodyning and second derivative, respectively, λ is the regularization value;
step 2.3.3: the splitting method of the later tree and the generation of the leaf nodes are the same as those of the first tree, and the later tree needs to update y on the basis of the result of the former tree, namely, the residual error of the former tree is fitted;
step 2.3.4: when all trees are completed, predict values
Figure FDA0002455553230000024
ft(x) The T-th tree is shown, where T is the number of trees.
CN202010305284.0A 2020-04-17 2020-04-17 Big data multiple access selection method based on submodular function Active CN111539764B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010305284.0A CN111539764B (en) 2020-04-17 2020-04-17 Big data multiple access selection method based on submodular function

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010305284.0A CN111539764B (en) 2020-04-17 2020-04-17 Big data multiple access selection method based on submodular function

Publications (2)

Publication Number Publication Date
CN111539764A true CN111539764A (en) 2020-08-14
CN111539764B CN111539764B (en) 2022-08-16

Family

ID=71979971

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010305284.0A Active CN111539764B (en) 2020-04-17 2020-04-17 Big data multiple access selection method based on submodular function

Country Status (1)

Country Link
CN (1) CN111539764B (en)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111985576A (en) * 2020-09-02 2020-11-24 南宁师范大学 Shop address selection method based on decision tree
CN112418445A (en) * 2020-11-09 2021-02-26 深圳市洪堡智慧餐饮科技有限公司 Intelligent site selection fusion method based on machine learning
CN112884224A (en) * 2021-02-20 2021-06-01 杭州比智科技有限公司 Method and device for selecting address of entity object, computing equipment and computer storage medium
CN112929215A (en) * 2021-02-04 2021-06-08 博瑞得科技有限公司 Network flow prediction method, system, computer equipment and storage medium
CN118365377A (en) * 2024-03-26 2024-07-19 连云港市芳欣科技有限公司 Industry investigation analysis system

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103440589A (en) * 2013-09-17 2013-12-11 上海商学院 Store site selection system and method
CN106548367A (en) * 2016-10-12 2017-03-29 电子科技大学 The site selection model and its applied research of multi-source data
CN110135913A (en) * 2019-05-20 2019-08-16 智慧足迹数据科技有限公司 Training method, shop site selecting method and the device of shop site selection model
CN110991914A (en) * 2019-12-09 2020-04-10 朱递 Facility site selection method based on graph convolution neural network

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103440589A (en) * 2013-09-17 2013-12-11 上海商学院 Store site selection system and method
CN106548367A (en) * 2016-10-12 2017-03-29 电子科技大学 The site selection model and its applied research of multi-source data
CN110135913A (en) * 2019-05-20 2019-08-16 智慧足迹数据科技有限公司 Training method, shop site selecting method and the device of shop site selection model
CN110991914A (en) * 2019-12-09 2020-04-10 朱递 Facility site selection method based on graph convolution neural network

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111985576A (en) * 2020-09-02 2020-11-24 南宁师范大学 Shop address selection method based on decision tree
CN111985576B (en) * 2020-09-02 2023-11-03 南宁师范大学 Shop site selection method based on decision tree
CN112418445A (en) * 2020-11-09 2021-02-26 深圳市洪堡智慧餐饮科技有限公司 Intelligent site selection fusion method based on machine learning
CN112929215A (en) * 2021-02-04 2021-06-08 博瑞得科技有限公司 Network flow prediction method, system, computer equipment and storage medium
CN112929215B (en) * 2021-02-04 2022-10-21 博瑞得科技有限公司 Network flow prediction method, system, computer equipment and storage medium
CN112884224A (en) * 2021-02-20 2021-06-01 杭州比智科技有限公司 Method and device for selecting address of entity object, computing equipment and computer storage medium
CN118365377A (en) * 2024-03-26 2024-07-19 连云港市芳欣科技有限公司 Industry investigation analysis system

Also Published As

Publication number Publication date
CN111539764B (en) 2022-08-16

Similar Documents

Publication Publication Date Title
CN111539764B (en) Big data multiple access selection method based on submodular function
Lall et al. Industrial location and spatial inequality: Theory and evidence from India
Macharis et al. Reviewing the use of Multi-Criteria Decision Analysis for the evaluation of transport projects: Time for a multi-actor approach
González-Feliu et al. New trends on urban goods movement: modelling and simulation of e-commerce distribution
Menou et al. Decision support for centralizing cargo at a Moroccan airport hub using stochastic multicriteria acceptability analysis
Zhao et al. Identification of land-use characteristics using bicycle sharing data: A deep learning approach
Wang et al. A service demand forecasting model for one-way electric car-sharing systems combining long short-term memory networks with Granger causality test
Gan et al. The identification of truck-related greenhouse gas emissions and critical impact factors in an urban logistics network
US20220215749A1 (en) Method for predicting at least one profile of the speed of a vehicle on a road network
Wyatt The development of a property information system for valuation using a geographical information system (GIS)
CN112668803B (en) Automobile service chain enterprise shop-opening and site-selecting method based on LightGBM model
De Carvalho et al. Evaluation of the city logistics performance through structural equations model
Ge et al. Intelligent site selection for bricks-and-mortar stores
Nilsen et al. Where and why do firms choose to move? Empirical evidence from Norway
Wang et al. Risk analysis and route optimization of dangerous goods transportation based on the empirical path set
Sattayaprasert et al. Creating a risk-based network for hazmat logistics by route prioritization with AHP: case study: gasoline logistics in Rayong, Thailand
CN109409563A (en) A kind of analysis method, system and the storage medium of the real-time number of bus operation vehicle
CN110309953B (en) Urban security monitoring layout system and method adopting target mobility distribution prediction
Narayanan et al. Expansion of a small-scale car-sharing service: A multi-method framework for demand characterization and derivation of policy insights
Gonzalez Feliu et al. New trends on urban goods movement modelling: proximity delivery versus shopping trips
CN115170183A (en) Off-line store site selection evaluation method and device and electronic equipment
CN111461163B (en) Urban interior PM2.5 concentration simulation and population exposure evaluation method and device
Velumani et al. A mass appraisal assessment study of land values using spatial analysis and multiple regression analysis model (MRA)
Fridrikhson et al. Methodology for justification of the location of railcar repair complex enterprises
Ghaemi et al. Modeling construction freight using OLS, robust inference and FGLS

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
CB02 Change of applicant information
CB02 Change of applicant information

Address after: 210009, 66 new model street, Gulou District, Jiangsu, Nanjing

Applicant after: NANJING University OF POSTS AND TELECOMMUNICATIONS

Address before: Yuen Road Ya Dong Qixia District of Nanjing City, Jiangsu province 210023 New District No. 9

Applicant before: NANJING University OF POSTS AND TELECOMMUNICATIONS

GR01 Patent grant
GR01 Patent grant