Community group-buying grid demand prediction method

Abstract

Disclosed is a community group-buying grid demand prediction method, comprising the following steps: Step S1: acquiring real-time community group-buying data; Step S2: performing preprocessing on the acquired community group-buying data, to obtain a user feature set for community group-buying grid demand prediction; Step S3: inputting the acquired user feature set to be predicted into a trained community group-buying grid demand prediction model and performing community group-buying grid demand prediction, to obtain predicted values of community group-buying products; Step S4: performing product reservation based on the obtained predicted values of community group-buying products.

Description

Community group purchase grid demand prediction method Field of the invention

[0001] The present application relates to the field of commodity demand prediction, in particular to a community group purchase grid demand prediction method. BACKGROUND

[0002] In the post-epidemic era, the "money-bashing" mode of low-price dumping and expanding logistics scale is not a long-term business approach. It is imperative to optimize the logistics network layout as soon as possible in the low-profit space to reduce costs. On the one hand, each platform has to face the impact of physical stores. With the recovery of offline physical stores such as markets and convenience stores, the dependence of community residents on community group purchase has decreased significantly, and the average daily single quantity of community group purchase has begun to fall and gradually stabilized. On the other hand, while improving service levels, the dense warehouse network layout also results in high operating costs and management costs for enterprises, which embodies the "benefit back" principle in logistics.

[0003] Based on the above changes in user demand and e-commerce platform development trends, for self-operated retail e-commerce platforms, sales forecasting and procurement planning are of great significance. For procurement, the most important thing is to purchase appropriate categories and quantities of goods in a timely manner to meet subsequent sales demand. Based on this, the procurement link needs to reduce the deviation in sales forecasting. If there is a large deviation, it means that there may be a shortage of goods and opportunity cost losses, and it may also lead to inventory accumulation. The current community group purchase platform still faces the challenge of high supply chain investment costs in the early stage, as well as unstable user purchase behavior choices, which leads to fluctuations in user demand.

[0004] SUMMARY

[0005] The following presents a simplified summary of one or more aspects in order to provide a basic understanding of such aspects. This summary is not an extensive overview of all contemplated aspects, and is intended to neither identify key or critical elements of all aspects nor delineate the scope of any or all aspects. Its sole purpose is to present some concepts of one or more aspects in a simplified form as a prelude to the more detailed description that is presented later.

[0006] The purpose of the present application is to solve the above problems, and a community group purchase grid demand prediction method is provided, which comprehensively considers the key influencing factors of community resident user shopping behavior choices, analyzes user behavior according to the existing historical data of community group purchase, and constructs a prediction model to predict community group purchase grid demand.

[0007] The technical scheme of the present application is as follows:

[0008] The present application provides a community group purchase grid demand prediction method, comprising the following steps:

[0009] Step S1: obtaining real-time community group purchase data;

[0010] Step S2: preprocessing the obtained community group purchase data to obtain a user feature set for community group purchase grid demand prediction;

[0011] Step S3: inputting the obtained user feature set to be predicted into the trained community group purchase grid demand prediction model to perform community group purchase grid demand prediction, and obtaining a community group purchase commodity prediction value;

[0012] Step S4: reserving commodities according to the obtained community group purchase commodity prediction value.

[0013] According to an embodiment of the community group purchase grid demand prediction method, the community group purchase grid demand prediction method adopts a gating recurrent unit as a prediction model framework, and obtains a community group purchase grid demand prediction model for community group purchase grid demand prediction by model training on the gating recurrent unit. The community group purchase grid demand prediction method trains the community group purchase grid demand prediction model through the following steps:

[0014] Step C1: collecting historical community group purchase data, and preprocessing the collected community group purchase data to obtain a user feature set for model training;

[0015] Step C2: inputting the user feature set into the gating recurrent unit to be trained to perform model training, so as to obtain a community group purchase grid demand prediction model for community group purchase grid demand prediction;

[0016] Step C3: determining whether the model training termination condition is met under the current state; if yes, ending the model training and outputting the trained community group purchase grid demand prediction model; if no, repeating step C3 until the model training termination condition is met.

[0017] According to an embodiment of the community group purchase grid demand prediction method, the community group purchase data includes user basic data, user behavior data, commodity sales data, commodity click times, and commodity cart adding times. When the community group purchase grid demand prediction method preprocesses the community group purchase data, the obtained community group purchase data is labeled according to a preset user feature label, so as to obtain a user feature set.

[0018] According to an embodiment of the community group purchase grid demand prediction method, the user feature label includes a user label and an extended community grid label, and the community group purchase grid demand prediction method is used to distinguish and classify the obtained community group purchase data according to the preset user label and the extended community grid label, so as to obtain a corresponding user feature set; wherein the user label includes a user personal information label and a user behavior information label, and the extended community grid label includes a warehouse availability label, a user loss degree label, a product quality label, a community group purchase convenience label and a product real-time heat label.

[0019] According to an embodiment of the community group purchase grid demand prediction method, when the obtained community group purchase data is labeled, a corresponding label model is established for each preset extended community grid label, and a quantitative feature is obtained as a model input by quantifying the label model; wherein the quantitative feature includes user personal features, user behavior features, warehouse availability, user loss degree, product quality, community group purchase convenience and product real-time heat.

[0020] According to an embodiment of the community group purchase grid demand prediction method, a corresponding warehouse availability label model is established for the warehouse availability label, and the warehouse availability is quantified by using the delivery time of community group purchase from the distribution station to the self-service place; wherein the delivery time T of community group purchase from the distribution station to the self-service place is calculated according to the following formula:

[0021] Wherein T represents the delivery time of community group purchase from the distribution station to the self-service place,

[0022] v0 represents the driving speed of the delivery vehicle of community group purchase from the grid warehouse to the self-service place,

[0023] s ij represents the delivery distance from the grid warehouse to the self-service place,

[0024] represents the average waiting time of community group purchase from the grid warehouse to the self-service place j,

[0025] M j represents the product order quantity of the self-service place j,

[0026] τ represents the average demand time of product unloading.

[0027] According to an embodiment of the community group purchase grid demand prediction method of the present application, the community group purchase grid demand prediction method establishes a corresponding user loss degree label model for the user loss degree label, uses a data segmentation time window to segment user behavior data, and quantifies the user loss degree by judging the user behavior data in the data segmentation time window; wherein the data segmentation time window expression and the user loss degree expression are as follows:

[0028] Wherein, Δt represents the data segmentation time window used to segment the user behavior data.

[0029] According to an embodiment of the community group purchase grid demand prediction method of the present application, the community group purchase grid demand prediction method establishes a corresponding community group purchase product quality label model for the product quality label, and uses the freshness value of fresh products to quantify the community group purchase product quality; wherein the community group purchase product quality expression is as follows:

[0030] Wherein, Q j represents the community group purchase product quality at the pickup point j,

[0031] μ represents the product integrity rate parameter,

[0032] represents the product integrity rate function,

[0033] T represents the delivery time of the community group purchase from the distribution station to the pickup point,

[0034] v0 represents the driving speed of the delivery vehicle when the community group purchase is delivered from the grid warehouse to the pickup point,

[0035] s ij represents the delivery distance from the grid warehouse to the pickup point,

[0036] represents the average waiting time of the community group purchase delivery from the grid warehouse to the pickup point j,

[0037] M j represents the product order quantity at the pickup point j,

[0038] τ represents the average demand time for unloading goods.

[0039] According to an embodiment of the community group purchase grid demand prediction method of the present application, the community group purchase grid demand prediction method establishes a corresponding community group purchase convenience label model for the community group purchase convenience label, and uses the time spent by the user walking to the pickup point to quantify the community group purchase convenience; wherein the community group purchase convenience expression is as follows:

[0040] wherein S represents community group purchase convenience,

[0041] d represents the distance of the user to the self-service point,

[0042] v represents the self-service speed of the user.

[0043] According to an embodiment of the community group purchase grid demand prediction method, the community group purchase grid demand prediction method establishes a corresponding real-time heat model of a commodity for a real-time heat label of the commodity, and quantifies the real-time heat of the commodity according to a basic heat, a loss rate and a heat condition of the past three days; wherein the real-time heat value expression of the commodity is as follows:

[0044] H i = (1-l mi )D mi +D m3

[0045] wherein H i represents the real-time heat of the community group purchase commodity i,

[0046] l mi represents the loss rate of the community group purchase commodity i,

[0047] D mi represents the basic heat of the community group purchase commodity i,

[0048] D m3 represents the heat condition of the community group purchase commodity i in the past three days.

[0049] According to an embodiment of the community group purchase grid demand prediction method, after obtaining the quantified features as the model input, the community group purchase grid demand prediction method further extracts corresponding public opinion features according to social points and social public opinions, and then takes the extracted public opinion features and the quantified features as the model input.

[0050] According to an embodiment of the community group purchase grid demand prediction method, the group purchase demand prediction method further introduces a multiple attention layer for the extracted public opinion features, extracts deep features of the public opinion features through the multiple attention layer, and then takes the extracted deep features and the quantified features as the model input; wherein when the group purchase demand prediction method extracts the deep features of the public opinion features by using the multiple attention layer, the sequence of the public opinion features input into the multiple attention layer is divided into multiple different parts, each part corresponds to one attention, then the corresponding feature vector is calculated for each attention, and finally the feature vectors of each attention are spliced to obtain a complete output vector.

[0051] According to an embodiment of the community group purchase grid demand prediction method, when calculating each heavy attention feature vector, first, the key attention weight value between each opinion feature value in the divided opinion feature sequence and the overall value of the opinion feature sequence is calculated, and then the feature vector fused with the key attention weight value is calculated based on the calculated key attention weight value, and the expression is as follows:

[0052] a i = Softmax(Score(x i , (x k , v y ))

[0053] wherein a i represents the key attention weight value of the i-th opinion feature value,

[0054] x i represents the i-th opinion feature value,

[0055] x k represents the input vector in the multi-head attention mechanism,

[0056] v y represents the sum average operation on the opinion feature input,

[0057] k represents the size of the sliding window,

[0058] Y i represents the opinion feature value

[0059] h y represents the feature vector fused with the key attention.

[0060] According to an embodiment of the community group purchase grid demand prediction method, in step C2, when the community group purchase grid demand prediction model is trained, the community group purchase grid demand prediction method also uses the fish-eagle optimization algorithm to optimize the model parameters of the community group purchase grid demand prediction model, so as to obtain the optimal model parameters, including the following steps:

[0061] Step D1: setting fish-eagle optimization algorithm parameters and model parameters to be optimized, and initializing the population; wherein the fish-eagle optimization algorithm parameters include optimization dimension D, population size N and maximum iteration number T, the model parameters to be optimized include input layer neuron number numFeatures, hidden layer neuron number numHiddenUnits and learning rate irate;

[0062] Step D2: calculate the fitness value of each individual in the initialization population, and divide the population according to the fitness value of each individual, including fish group population and fish eagle population;

[0063] Step D3: update the position of each individual in the fish group population, and determine the fish to be captured by each fish eagle in the fish eagle population;

[0064] Step D4: determine whether to update the fish eagle position after the fish eagle completes the capture according to the fitness values of the fish eagle and the captured fish; if the fitness value of the captured fish is greater than the fitness value of the fish eagle, the fish eagle position is updated to the position of the captured fish; if the fitness value of the captured fish is less than or equal to the fitness value of the fish eagle, the fish eagle position is not updated;

[0065] Step D5: randomly generate a new capture position for each completed capture fish eagle, and determine whether to update the fish eagle position to the new capture position; if the fitness value of the new capture position is greater than the fitness value of the fish eagle, the fish eagle position is updated to the new capture position; if the fitness value of the new capture position is less than or equal to the fitness value of the fish eagle, the fish eagle position is not updated;

[0066] Step D6: determine whether each fish eagle in the fish eagle population has completed capture and position optimization update at the current iteration number t; if yes, execute the next step D7; if no, continue steps D4-D5 until each fish eagle in the fish eagle population has completed capture and position optimization update;

[0067] Step D7: determine whether the current iteration number t reaches the maximum iteration number T; if yes, end the fish eagle optimization algorithm and output the optimal model parameters; if no, repeat steps D3-D6 until the maximum iteration number T is reached to obtain the optimal model parameters.

[0068] According to an embodiment of the community group purchase grid demand prediction method of the application, when the fish eagle optimization algorithm is used to optimize the model parameters of the community group purchase grid demand prediction model, the population position is initialized by the following formula:

[0069] wherein, represents the new position of the jth dimension of the ith fish eagle in the first stage,

[0070] represents the population mean,

[0071] lb j represents the lower boundary of the optimization region,

[0072] ub i represents the upper boundary of the optimization region,

[0073] rand represents random selection.

[0074] According to an embodiment of the community group purchase grid demand prediction method, when the fish eagle optimization algorithm is used to optimize the model parameters of the community group purchase grid demand prediction model, a dynamic position movement control parameter a is added to expand the optimization area of the fish eagle, and the expression is as follows:

[0075] Wherein, j represents the dimension,

[0076] lb j represents the lower boundary of the optimization area,

[0077] ub j represents the upper boundary of the optimization area,

[0078] t represents the current iteration number,

[0079] T represents the maximum iteration number.

[0080] According to an embodiment of the community group purchase grid demand prediction method, the community group purchase grid demand prediction method further presets an error threshold for evaluating the accuracy of the community group purchase grid demand prediction model prediction; wherein,

[0081] If the error value between the actual sales data of the commodity and the community group purchase commodity prediction value is higher than the preset error threshold, the community group purchase grid demand prediction model is updated online based on the actual commodity sales data;

[0082] If the error value between the actual sales data of the commodity and the community group purchase commodity prediction value is less than the preset error threshold, it is proved that the current community group purchase grid demand prediction model is accurate and does not need to be updated.

[0083] The application also provides a computer readable medium storing computer program code, which, when executed by a processor, implements the method as described above.

[0084] The application also provides a community group purchase grid demand prediction device, comprising:

[0085] a memory for storing instructions executable by the processor; and

[0086] a processor for executing the instructions to implement the method as described above.

[0087] The present application has the following beneficial effects compared with the prior art: the present application is aimed at the popular community group purchase mode at present, is centered on user behavior, analyzes user behavior data according to existing historical data of community group purchase, obtains a user feature set, then combines opinion features to train a model, and obtains a community group purchase grid demand prediction model, and the trained model is used to predict the demand of the community group purchase grid. Compared with the prior art, the present application can reserve goods in advance according to the community group purchase commodity prediction value, not only can meet the user group purchase demand, but also is beneficial to the optimization of the commodity reserve scheme, reduces the commodity storage cost, and further greatly reduces the input cost of the commodity supply chain. BRIEF DESCRIPTION OF DRAWINGS

[0088] The above features and advantages of the present application can be better understood by reading the following detailed description of embodiments of the present application in conjunction with the accompanying drawings. In the drawings, components are not necessarily drawn to scale, and components of similar or identical structure or function can have identical or similar reference numbers.

[0089] Fig. 1 is a step flow chart showing an embodiment of the community group purchase grid demand prediction method of the present application.

[0090] Fig. 2 is a step flow chart showing an embodiment of training the community group purchase grid demand prediction model of the present application.

[0091] Fig. 3 is a step flow chart showing an embodiment of using the fish eagle optimization algorithm to optimize the model parameters of the community group purchase grid demand prediction model of the present application.

[0092] Detailed description of the application

[0093] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are only some examples or embodiments of the present application, and for those skilled in the art, the present application can also be applied to other similar scenarios without creative labor. Unless the context clearly indicates otherwise or otherwise stated, the same reference numbers in the drawings represent the same structure or operation.

[0094] As shown in the present application and claims, unless the context clearly indicates otherwise or otherwise stated, “one”, “a”, “an” and / or “the” do not refer to the singular, but also include the plural. Generally speaking, the terms “include” and “contain” only indicate the inclusion of the steps and elements explicitly identified, and these steps and elements do not constitute an exclusive list, and the method or device can also contain other steps or elements.

[0095] The foregoing is considered as illustrative only of the principles of the application. Further, since numerous modifications and changes will readily occur to those skilled in the art, it is not desired to limit the application to the exact construction and practice described. Accordingly, all suitable modifications and equivalents can be resorted to falling within the scope of the application. Unless otherwise indicated herein, the procedures recited in the examples and experimental data set forth herein are conducted at ambient temperature and pressure, unless otherwise specified. The relative arrangement of components and steps, the numerical expressions, and numerical values set forth in the examples herein are not meant to limit the scope of the present application, unless otherwise specifically stated. It is to be understood that the various parts shown in the drawings are not necessarily drawn to scale and that, for purposes of convenience and clarity, not all components and steps can be shown in a given figure. Techniques, methods, and apparatus known to those of ordinary skill can not be discussed in detail, but are to be considered as part of the present disclosure, where appropriate. In all examples shown and discussed herein, any specific values should be interpreted as merely illustrative and not as a limitation of the scope of the exemplary embodiments. Thus, other examples of the exemplary embodiments can have different values. It is to be noted that like reference numerals and letters refer to like items in the drawings, and, once an item is defined in one drawing, it need not be further discussed in subsequent drawings.

[0096] In describing and claiming the present application, the following terminology will be used in accordance with the definitions set out below. It is also to be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting, unless otherwise specifically stated. Furthermore, it is contemplated that any embodiment described herein can be implemented in software, hardware, firmware, or any combination thereof.

[0097] In the description of the present application, it is to be understood that the orientation or positional relationships indicated by orientation words such as "front, back, upper, lower, left, right", "horizontal, vertical, perpendicular, horizontal" and "top, bottom" and the like are generally based on the orientation or positional relationships shown in the drawings, and are only for the convenience of description and simplification of the description, and do not indicate and imply that the devices or elements referred to must have a particular orientation or be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the scope of protection of the present application; the orientation words "inner, outer" refer to the inner and outer of the contour of each component itself.

[0098] For the convenience of description, spatial relationship words such as "under", "below", "lower", "under", "above", "upper" and the like can be used herein to describe the relationship of one element or feature to other elements or features shown in the drawings. It will be understood that these spatial relationship words are intended to include other directions of the device in use or operation in addition to the directions depicted in the drawings. For example, if the device in the drawings is turned over, the orientation of the element described as "below" or "under" or "under" the other element or feature will be changed to "above" the other element or feature. Thus, the exemplary words "below" and "under" can include both the upper and lower directions. The device can also have other orientations (rotated 90 degrees or in other directions), so the spatial relationship description words used herein should be interpreted accordingly. Furthermore, it will also be understood that when a layer is referred to as "between" two layers, it can be the only layer between the two layers, or there can be one or more intervening layers.

[0099] In the context of this application, structures described as being "on" "connected to" "coupled with" or "in contact with" one another should be understood in a broad sense to potentially encompass structures that are directly in contact with one another, as well as structures that are not in direct contact with one another. As examples of the latter, one component can be on, connected to, coupled with or in contact with another component even though one or multiple intervening components are present. In contrast, when one component is described as being "directly on" "directly connected to" "directly coupled with" or "directly in contact with" another component, there are no intervening components present. Similarly, when a first component is described as being "electrically in contact with" or "electrically coupled with" a second component, there is an electrical path between the first component and the second component that allows current to flow. The electrical path can include capacitors, coupled inductors and / or other components that allow current to flow, even though there is no direct contact between electrically conductive components.

[0100] It will be understood that when a component is referred to as being "on" "connected to" "coupled with" or "in contact with" another component, it can be directly on, connected to, coupled with, or in contact with the other component, or one or more intervening components can also be present. In contrast, when a component is referred to as being "directly on" "directly connected to" "directly coupled with" or "directly in contact with" another component, there are no intervening components present. Similarly, when a first component is referred to as being "electrically in contact with" or "electrically coupled with" a second component, there is an electrical path between the first component and the second component that allows current to flow. The electrical path can include capacitors, coupled inductors and / or other components that allow current to flow, even though there is no direct contact between electrically conductive components.

[0101] In addition, it should be pointed out that the use of the terms "first", "second" and the like in connection with various elements is merely intended for identification purposes and does not have a special meaning other than to distinguish one element from another, unless otherwise specifically indicated. In addition, although the terms used in the present application are selected from publicly known and used terms, some of the terms mentioned in the description of the present application can be created by the applicant in his or her own judgment, and the detailed meanings thereof are disclosed in relevant parts of the description herein. Furthermore, the present application is not intended to be limited only to the actual terms used but is intended to cover each and every term derived from the meanings of the actual terms used.

[0102] An embodiment of a community group purchase grid demand prediction method is disclosed herein. FIG. 1 is a flow chart illustrating the steps of an embodiment of the community group purchase grid demand prediction method of the present application. Referring to FIG. 1, the following is a detailed description of each step of the community group purchase grid demand prediction method.

[0103] Step S1: Obtain real-time community group purchase data.

[0104] Step S2: Preprocess the obtained community group purchase data to obtain a user feature set for community group purchase grid demand prediction.

[0105] In this embodiment, in order to meet the future community group purchase demand, it is necessary to obtain real-time community data, including various user behavior data of users, user basic data, and commodity sales data, commodity click times and commodity cart adding times, etc. Then, the data is preprocessed to obtain a user feature set.

[0106] Specifically, in this embodiment, when preprocessing the community group purchase data, the obtained community group purchase data needs to be labeled according to the preset user feature label to obtain a user feature set. The user feature label includes a user label and an extended community grid label. The community group grid demand prediction method discriminates and classifies the obtained community group purchase historical data according to the preset user label and the extended community grid label, thereby obtaining the corresponding user feature set. The user label includes a user personal information label and a user behavior information label. The extended community grid label includes a warehouse availability label, a user loss degree label, a commodity quality label, a community group purchase convenience label, and a commodity real-time hotness label.

[0107] In addition, in this embodiment, since the extended community grid label is not original data that can be directly obtained, when the obtained community group purchase data is labeled, a corresponding label model is also established for the preset extended community grid label, and the quantified features as the model input are obtained by quantifying the label model. The quantified features include user personal features, user behavior features, warehouse availability, user loss degree, commodity quality, community group purchase convenience, and commodity real-time hotness.

[0108] Specifically, in this embodiment, a corresponding warehouse availability label model is established for the warehouse availability label. The distance from the community group grid warehouse to the self-service place, the availability, speed, and distance of the warehouse distribution vehicle, the average waiting time of the community group distribution from the grid warehouse to the self-service place, and the loading and unloading time caused by different order quantities are considered. Therefore, the distribution time T of the community group from the distribution station to the self-service place is used to quantify the warehouse possibility model, thereby realizing the quantification of the warehouse availability. The calculation formula of the distribution time T of the community group from the distribution station to the self-service place is as follows:

[0109] Wherein, T represents the distribution time of the community group from the distribution station to the self-service place, v0 represents the driving speed of the community group distribution vehicle from the grid warehouse to the self-service place, s ij represents the distribution distance from the grid warehouse to the self-service place, represents the average waiting time of the community group distribution from the grid warehouse to the self-service place j, M j represents the order quantity of the goods at the self-service place j, and τ represents the average demand time for unloading goods.

[0110] For the community group purchase convenience label, a corresponding community group purchase convenience label model is established. When analyzing the user behavior data of each user, a data segmentation time window is set to segment the user behavior data. That is, first, the user behavior data is segmented according to the set fixed data segmentation time window from the user registration time, and then it is analyzed whether the user has login, search, purchase or add purchase behavior in the time period. If there is no relevant behavior, it is considered that the user has lost in the time period. The data segmentation time window expression is as follows:

[0111] Where Δt represents the data segmentation time window for segmenting user behavior data.

[0112] Let the user registration time be t0 and the last behavior data time be t e , then first look at whether the user has lost in the first sliding window t0-t0+Δt, Since there is user behavior data in this time period, the user has not lost. Then judge whether the next sliding window t0+Δt-t0+2Δt has lost. Finally, when the window slides to t0+nΔt-t e , at this time, since the user has no behavior data, the corresponding output variable is that the user has lost.

[0113] For the product quality label, a corresponding community group purchase product quality label model is established. Considering that when purchasing fresh products in the community group purchase mode, there is a transportation time and a delivery time, the freshness value of the fresh products will decrease with the change of time, resulting in a decrease in product quality when the fresh products are delivered to the terminal. Product quality is a characteristic variable of the community group purchase platform. Therefore, when quantifying the community group purchase product quality, the community group purchase product quality is defined by referring to the fresh product spoilage equation. The expression of the community group purchase product quality is as follows:

[0114] Where Q j represents the community group purchase product quality at the self-pickup place j, μ represents the product integrity rate parameter, represents the product integrity rate function, T represents the delivery time of the community group purchase from the distribution station to the self-pickup place, v0 represents the driving speed of the delivery vehicle when the community group purchase is delivered from the grid warehouse to the self-pickup place, s ij represents the delivery distance from the grid warehouse to the self-pickup place, represents the average waiting time of the community group purchase when delivered from the grid warehouse to the self-pickup place j, M j represents the product order quantity at the self-pickup place j, and τ represents the average demand time for unloading goods.

[0115] For the community group purchase convenience label, a corresponding community group purchase convenience label model is established. In community group purchase grid demand prediction, the pickup time is an important indicator for measuring convenience. Therefore, by calculating the time spent by the user walking to the pickup location, the convenience of community group purchase can be quantified. The community group purchase convenience expression is as follows:

[0116] Wherein, S represents the community group purchase convenience, d represents the distance from the user to the pickup point, and v represents the user pickup speed.

[0117] Step S3: input the obtained user feature set to be predicted into the trained community group purchase grid demand prediction model to perform community group purchase grid demand prediction, and obtain a community group purchase commodity prediction value.

[0118] In this embodiment, after obtaining the user feature data for community group purchase grid demand prediction through the above steps, a trained community group purchase grid demand prediction model is used to perform community group purchase grid demand prediction, thereby obtaining a community group purchase commodity prediction value.

[0119] Wherein, when constructing the community group purchase grid demand prediction model, a gated recurrent unit (GRU) is used as the prediction model framework, and the gated recurrent unit is trained to obtain a community group purchase grid demand prediction model for community group purchase grid demand prediction. FIG. 2 is a step flow chart illustrating an embodiment of training a community group purchase grid demand prediction model of the present application. Please refer to FIG. 2, and the following is a detailed description of each step of training the community group purchase grid demand prediction model.

[0120] Step C1: collect historical community group purchase data, and pre-process the collected community group purchase data to obtain a user feature set for model training.

[0121] In this embodiment, when training the community group purchase grid demand prediction model, first, historical community group purchase data is collected, including user basic data, user behavior data, commodity sales data, commodity click times, and commodity cart addition times, etc. These data are used to analyze user behavior and predict future user group purchase grid demand.

[0122] Wherein, for the obtained original historical community group purchase data, the historical community group purchase data is pre-processed by referring to the method of labelizing the obtained community group purchase data according to the preset user feature label used in step S2 above, and the community grid label is further quantified, thereby obtaining quantized features for model training, which will not be described here.

[0123] Step C2: input the user feature set into the gated recurrent unit to be trained for model training, so as to obtain a community group grid demand prediction model for community group grid demand prediction.

[0124] In this embodiment, when the community group grid demand prediction model is trained, the fish-eagle optimization algorithm (GOA) is also used to optimize the model parameters of the community group grid demand prediction model, so as to obtain the optimal model parameters. FIG. 3 is a flow chart showing an embodiment of the fish-eagle optimization algorithm for optimizing the model parameters of the community group grid demand prediction model according to the present application. Please refer to FIG. 3, and the following is a detailed description of each step of the fish-eagle optimization algorithm for optimizing the model parameters of the community group grid demand prediction model.

[0125] Step D1: set the fish-eagle optimization algorithm parameters and the model parameters to be optimized, and initialize the population; wherein the fish-eagle optimization algorithm parameters include the optimization dimension D, the population size N, and the maximum number of iterations T, and the model parameters to be optimized include the input layer neuron number numFeatures, the hidden layer neuron number numHiddenUnits, and the learning rate irate.

[0126] In this embodiment, when setting the parameters of the community group grid demand prediction model, in addition to setting the model parameters to be optimized (input layer neuron number numFeatures, hidden layer neuron number numHiddenUnits, and learning rate irate), some other community group grid demand prediction model parameters also need to be set, and their values are set according to the following table, thereby completing the initial construction of the community group grid demand prediction model:

[0127] In addition, in this embodiment, a new initialization species updating method is used for the initialization of the population, which is used to improve the problems of reduced population diversity, low population quality, and reduced algorithm convergence speed caused by uneven distribution of the generated population when randomly generating the initial population. Among them, the population position initialization expression is as follows:

[0128] wherein, represents the new position of the i-th fish-eagle in the first stage, represents the population mean. lb j represents the lower boundary of the optimization region, ub j represents the upper boundary of the optimization region, and rand represents a random selection.

[0129] Step D2: calculate the fitness value of each individual in the initialized population, and divide the population according to the fitness value of each individual, including the fish group population and the fish-eagle population.

[0130] In this embodiment, each individual in the population is a set of model parameter schemes to be optimized, and the model parameter schemes to be optimized are optimized by optimizing the positions of the fish eagles in the fish eagle population.

[0131] Step D3: updating the position of each individual in the fish group population and determining the fish to be captured by each fish eagle in the fish eagle population.

[0132] Step D4: judging whether to update the position of the fish eagle after the fish eagle completes the capture according to the fitness values of the fish eagle and the captured fish; if the fitness value of the captured fish is greater than the fitness value of the fish eagle, the position of the fish eagle is updated to the position of the captured fish; if the fitness value of the captured fish is less than or equal to the fitness value of the fish eagle, the position of the fish eagle is not updated.

[0133] Steps D3 and D4 are in the positioning stage, in which any fish eagle i in the fish eagle population randomly detects the position of a fish in the fish group population and attacks and captures it. If the new position of the captured fish is better, the position of the fish eagle i is replaced by the position of the captured fish, i.e., the position of the fish eagle i is updated to X i .

[0134] Step D5: randomly generating a new capture position for each fish eagle that completes the capture and judging whether to update the position of the fish eagle to the new capture position; if the fitness value of the new capture position is greater than the fitness value of the fish eagle, the position of the fish eagle is updated to the new capture position; if the fitness value of the new capture position is less than or equal to the fitness value of the fish eagle, the position of the fish eagle is not updated.

[0135] This step belongs to the exploration stage, in which the fish eagle i selects a safe position after hunting and randomly generates a new capture position If the objective function value is improved at the new capture position, the original position of the fish eagle i is replaced by the new capture position, and the position of the fish eagle i is updated to X i .

[0136] In addition, in the positioning and exploration stages, a dynamic position movement control parameter a is added in this embodiment, which expands the optimization area of the fish eagle through the dynamic position movement control parameter to increase the local development capability. The expression is as follows:

[0137] wherein j represents the dimension, lb j represents the lower boundary of the optimization area, ub j represents the upper boundary of the optimization area, t represents the current iteration number, and T represents the maximum iteration number.

[0138] Step D6: Determine whether each fish eagle in the fish eagle population has completed capture and position optimization update at the current iteration number t; if yes, execute the next step D7; if no, continue steps D4-D5 until each fish eagle in the fish eagle population has completed capture and position optimization update.

[0139] Step D7: Determine whether the current iteration number t reaches the maximum iteration number T; if yes, end the fish eagle optimization algorithm and output the optimal model parameters; if no, repeat steps D3-D6 until the maximum iteration number T is reached to obtain the optimal model parameters.

[0140] In this embodiment, after obtaining the optimal solution of the model parameters to be optimized by the fish eagle algorithm, the obtained optimal solution and other community group grid demand prediction model parameters are output together, and the optimal model parameters are obtained.

[0141] Step C3: Determine whether the current state meets the model training termination condition; if yes, end the model training and output the trained community group grid demand prediction model; if no, repeat step C3 until the model training termination condition is met.

[0142] In this embodiment, after obtaining the optimal model parameters by steps D1-D7, the community group grid demand prediction model configured with the optimal model parameters is used for model training until the trained community group grid demand prediction model is obtained, and then the user feature set to be predicted is input into the trained community group grid demand prediction model for community group grid demand prediction, thereby obtaining the community group commodity prediction value.

[0143] During model training, in addition to using quantitative data for model training, the influence of social points and social public opinion is also considered, and the corresponding public opinion features are extracted according to the social points and social public opinion, and then the extracted public opinion features and quantitative features are input into the model for model training.

[0144] Specifically, in this embodiment, a multiple attention layer is introduced for the public opinion features to be extracted, and the deep features of the public opinion features are extracted through the multiple attention layer, and then the extracted deep features and quantitative features are input into the model. Wherein, when extracting the deep features of the public opinion features, the multiple attention layer first divides the public opinion feature sequence input into the multiple attention layer into multiple different parts, each part corresponds to a heavy attention, then calculates the corresponding feature vector for each heavy attention, and finally splices the feature vectors of each heavy attention to obtain a complete output vector.

[0145] Wherein, in the calculation of each heavy attention feature vector, first, the key attention weight value between each opinion feature value in the divided opinion feature sequence and the overall value of the opinion feature sequence is calculated, and then the feature vector of the fused key attention weight value is calculated based on the calculated key attention weight value, and the expression is as follows:

[0146] a i = Softmax(Score(x i , (x k , v y ))

[0147] Wherein, a i represents the key attention value, x i represents the input quantization feature value, x k represents the input vector in the multi-head attention mechanism, v y represents the sum average operation on the opinion feature input, k represents the sliding window size, Y i represents the opinion feature value, h y represents the feature vector of the fused key attention.

[0148] Step S4: According to the obtained community group purchase commodity prediction value, the commodity is reserved.

[0149] In this embodiment, after obtaining the future community group purchase commodity prediction value through the above step S3, the community grid is reserved according to the community group purchase commodity prediction value, so as to meet the user demand and reduce the storage cost.

[0150] In addition, in this embodiment, the actual commodity sales data is also obtained regularly, and the actual sales data is compared with the community group purchase commodity prediction value to judge whether the error value is higher than the preset error threshold. Wherein, if the error value is higher than the error threshold, the community group grid demand prediction model is updated online based on the actual commodity sales data. If the error value between the actual sales data of the commodity and the community group purchase commodity prediction value is less than the preset error threshold, it proves that the current community group grid demand prediction model is accurate and does not need to be updated.

[0151] In this specification, a computer readable medium storing computer program code is also provided, which, when executed by a processor, implements the community group grid demand prediction method as described above.

[0152] In this specification, a community group grid demand prediction device is also provided, which includes a storage that stores instructions executable by a processor, and a processor for executing the instructions in the instruction storage to implement the community group grid demand prediction method as described above.

[0153] The previous description of the disclosure is provided to enable any person skilled in the art to make or use the disclosure. Various modifications to the disclosure will be readily apparent to those skilled in the art, and the generic principles defined herein can be applied to other variations without departing from the spirit or scope of the disclosure. Thus, the disclosure is not intended to be limited to the examples described herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0154] Those of skill would further appreciate that the various illustrative logical blocks, modules, circuits, and algorithm steps described in connection with the embodiments disclosed herein can be implemented as electronic hardware, computer software, or combinations of both. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps have been described above generally in terms of their functionality, without limitation. Such

[0155] The various illustrative logical blocks, modules, and circuits described in connection with the embodiments disclosed herein can be implemented or performed with a general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general purpose processor can be a microprocessor, but in the alternative, the processor can be any conventional processor, controller, microcontroller, or state machine. A processor can also be implemented as a combination of computing devices, e.g., a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration.

[0156] The steps of a method or algorithm described in connection with the embodiments disclosed herein can be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. A software module can reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor such that the processor can read information from, and write information to, the storage medium. In the alternative, the storage medium can be integral to the processor. The processor and the storage medium can reside in an ASIC. The ASIC can reside in a user terminal. In the alternative, the processor and the storage medium can reside as discrete components in a user terminal.

[0157] In one or more exemplary embodiments, the functions described can be implemented in hardware, software, firmware, or any combination thereof. If implemented in software as a computer program product, the functions can be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media includes both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. Storage media can be any available media that can be accessed by a computer. By way of example, and not limitation, such computer-readable media can comprise RAM, ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to carry or store desired program code in the form of instructions or data structures and that can be accessed by a computer. Also, any connection is properly termed a computer-readable medium. For example, if the software is transmitted from a website, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwave, then the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwave are included in the definition of medium. Disk and disc, as used herein, includes compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk and blu-ray disc where disks usually reproduce data magnetically, while discs reproduce data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.

Claims

1. A community group-buying grid demand forecasting method, characterized in that, Includes the following steps: Step S1: Obtain real-time community group buying data; Step S2: Preprocess the acquired community group buying data to obtain a set of user features for predicting community group buying grid demand; Step S3: Input the obtained set of user features to be predicted into the trained community group buying grid demand prediction model to predict the community group buying grid demand and obtain the predicted value of community group buying products; Step S4: Stock up on goods based on the obtained community group-buying product forecast.

2. The community group-buying grid demand forecasting method according to claim 1, characterized in that, The community group-buying grid demand forecasting method uses a gated recurrent unit as the prediction model framework. By training the gated recurrent unit, a community group-buying grid demand forecasting model is obtained. The community group-buying grid demand forecasting method trains the community group-buying grid demand forecasting model through the following steps: Step C1: Collect historical community group buying data and preprocess the collected community group buying data to obtain a set of user features for model training; Step C2: Input the user feature set into the gated recurrent unit to be trained to train the model, so as to obtain the community group buying grid demand prediction model for community group buying grid. Step C3: Determine whether the model training termination condition is met in the current state; if yes, end the model training and output the trained community group buying grid demand prediction model; if no, repeat step C3 until the model training termination condition is met.

3. The community group-buying grid demand forecasting method according to claim 1 or 2, characterized in that, The community group buying data includes basic user data, user behavior data, product sales data, product click counts, and product addition counts. The community group buying grid demand prediction method preprocesses the community group buying data by labeling the acquired data according to preset user feature tags, thereby obtaining a set of user features.

4. The community group-buying grid demand forecasting method according to claim 3, characterized in that, User feature tags include user tags and extended community grid tags. The community group buying grid demand prediction method judges and classifies the acquired community group buying data based on preset user tags and extended community grid tags, thereby obtaining the corresponding user feature set. Among them, user tags include user personal information tags and user behavior information tags, and extended community grid tags include warehouse availability tags, user churn tags, product quality tags, community group buying convenience tags, and product real-time popularity tags.

5. The community group-buying grid demand forecasting method according to claim 4, characterized in that, The community group-buying grid demand prediction method, when processing the acquired community group-buying data by labeling, establishes a corresponding label model for each preset extended community grid label, and obtains quantitative features as input to the model by quantifying the label model; wherein, the quantitative features include user personal characteristics, user behavior characteristics, warehouse availability, user churn rate, product quality, community group-buying convenience, and real-time popularity of products.

6. The community group-buying grid demand forecasting method according to claim 5, characterized in that, The community group-buying grid demand forecasting method establishes a corresponding warehouse availability label model for each warehouse availability label, and quantifies warehouse availability using the delivery time from the delivery station to the self-pickup point. The formula for calculating the delivery time T from the delivery station to the self-pickup point is as follows: Where T represents the delivery time from the distribution station to the self-pickup point in the community group buying service. v0 represents the speed at which delivery vehicles travel from the community group-buying warehouse to the self-pickup point. s ij This indicates the delivery distance from the grid warehouse to the self-pickup point. This represents the average waiting time from the community group-buying delivery point, from the grid warehouse to the self-pickup location. M j This indicates the quantity of goods ordered at self-pickup point j. τ represents the average time required to unload the goods.

7. The community group-buying grid demand forecasting method according to claim 5, characterized in that, The community group-buying grid demand prediction method establishes a corresponding user churn label model for user churn labels, and uses a data segmentation time window to segment user behavior data. User churn is quantified by judging user behavior data within the data segmentation time window. The data segmentation time window expression and the user churn expression are as follows: Where Δt represents the time window used to segment user behavior data.

8. The community group-buying grid demand forecasting method according to claim 5, characterized in that, The community group-buying grid demand forecasting method establishes a corresponding community group-buying product quality label model for each product quality label, using the freshness value of fresh produce to quantify the quality of community group-buying products; the expression for community group-buying product quality is as follows: Among them, Q j The quality of community group-buying goods represented by self-pickup point j μ represents the product integrity rate parameter. The product integrity rate function, T represents the delivery time from the distribution station to the self-pickup point for community group buying. v0 represents the speed at which the delivery vehicle travels from the grid warehouse to the self-pickup point during community group buying deliveries. s ij This indicates the delivery distance from the grid warehouse to the self-pickup point. This indicates the average waiting time for community group-buying deliveries from the grid warehouse to the self-pickup point. M j This indicates the quantity of goods ordered at self-pickup point j. τ represents the average time required for unloading cargo.

9. The community group-buying grid demand forecasting method according to claim 5, characterized in that, The community group-buying grid demand forecasting method establishes a corresponding community group-buying convenience label model for each community group-buying convenience label, and quantifies the convenience of community group-buying by using the time spent by users walking to the self-pickup point; wherein, the community group-buying convenience expression is as follows: Where S represents the convenience of community group buying, d represents the distance from the user to the self-pickup point. v represents the user's self-pickup speed.

10. The community group-buying grid demand forecasting method according to claim 5, characterized in that, The community group-buying grid demand forecasting method establishes a corresponding real-time product popularity model for each product's real-time popularity tag, quantifying the product's real-time popularity based on basic popularity, loss rate, and popularity over the past three days; the expression for the product's real-time popularity value is as follows: H i =(1-l mi )D mi +D m3 Among them, H i Represents the real-time popularity of community group-buying products. l mi Represents the loss rate of community group-buying products. D mi This represents the basic popularity of community group-buying products. D m3 This represents the popularity of community group-buying products over the past three days.

11. The community group-buying grid demand forecasting method according to claim 5, characterized in that, After obtaining the quantitative features used as model input, the community group-buying grid demand prediction method also extracts corresponding public opinion features based on social settings and public opinion. Then, the extracted public opinion features and quantitative features are used as model input.

12. The community group-buying grid demand forecasting method according to claim 11, characterized in that, The group-buying demand prediction method introduces a multi-attention layer to extract the deep features of public opinion features. The deep features and quantitative features are then used as model inputs. When extracting the deep features of public opinion features using the multi-attention layer, the method divides the sequence of public opinion features input to the multi-attention layer into multiple different parts, each part corresponding to a single attention layer. The corresponding feature vector is then calculated for each attention layer, and finally, the feature vectors of each attention layer are concatenated to obtain a complete output vector.

13. The community group-buying grid demand forecasting method according to claim 12, characterized in that, The community group-buying grid demand prediction method, when calculating the feature vector of each level of attention, first calculates the key attention weight between each public opinion feature value in the divided public opinion feature sequence and the overall value of the public opinion feature sequence. Then, based on the calculated key attention weights, it calculates the feature vector that fuses the key attention weights, as shown in the following expression: a i =Softmax(Score(x) i , (x k v y ))) Among them, a i This represents the key attention weight for the i-th public opinion feature value. x i This represents the i-th public opinion feature value. x k This represents the input vector in the multi-head attention mechanism. v y This indicates a summation and averaging operation on the input of public opinion characteristics. k represents the sliding window size. Y i Indicates the characteristic value of public opinion. h y This represents the feature vector that incorporates key attention.

14. The community group-buying grid demand forecasting method according to claim 2, characterized in that, In step C2, the community group-buying grid demand prediction method, when training the community group-buying grid demand prediction model, also uses the Osprey optimization algorithm to optimize the model parameters of the community group-buying grid demand prediction model, thereby obtaining the optimal model parameters, including the following steps: Step D1: Set the parameters of the Osprey optimization algorithm and the parameters of the model to be optimized, and initialize the population; where the Osprey optimization algorithm parameters include the optimization dimension D, the population size N and the maximum number of iterations T, and the parameters of the model to be optimized include the number of input layer neurons numFeatures, the number of hidden layer neurons numHiddenUnits and the learning rate irate. Step D2: Calculate the fitness value of each individual in the initial population, and divide the population according to the fitness value of each individual, including the fish population and the osprey population; Step D3: Update the position of each individual in the fish population and determine the fish that each osprey in the osprey population should catch; Step D4: Determine whether to update the osprey's position after it completes the capture based on the fitness values ​​of the osprey and the captured fish; if the fitness value of the captured fish is greater than the fitness value of the osprey, then update the osprey's position. The new location is the position of the captured fish; if the fitness value of the captured fish is less than or equal to the fitness value of the osprey, the osprey's position is not updated. Step D5: For each captured osprey, randomly generate a new capture location and determine whether to update the osprey's location to the new capture location; if the fitness value of the new capture location is greater than the fitness value of the osprey, then update the osprey's location to the new capture location; if the fitness value of the new capture location is less than or equal to the fitness value of the osprey, then do not update the osprey's location. Step D6: Determine whether every osprey in the osprey population has completed capture and position optimization update at the current iteration number t; if yes, proceed to the next step D7; if no, continue with steps D4 to D5 until every osprey in the osprey population has completed capture and position optimization update. Step D7: Determine whether the current iteration number t has reached the maximum iteration number T; if yes, end the Osprey optimization algorithm and output the optimal model parameters; if no, repeat steps D3 to D6 until the maximum iteration number T is reached to obtain the optimal model parameters.

15. The community group-buying grid demand forecasting method according to claim 14, characterized in that, When the community group-buying grid demand forecasting method uses the Osprey optimization algorithm to optimize the model parameters of the community group-buying grid demand forecasting model, the population location is initialized using the following formula: in, This represents the new position of the i-th osprey in the j-th dimension during the first stage. Represents the population mean. lb j This represents the lower boundary of the optimization region. ub j Represents the upper boundary of the optimization region. rand means to select randomly.

16. The community group-buying grid demand forecasting method according to claim 14, characterized in that, When the community group-buying grid demand forecasting method uses the Osprey optimization algorithm to optimize the model parameters of the community group-buying grid demand forecasting model, it also adds a dynamic position movement control parameter α. The optimization area of ​​the Osprey is expanded by the dynamic position movement control parameter, and the expression is as follows: Where j represents the dimension, lb j This represents the lower boundary of the optimization region. ub j Represents the upper boundary of the optimization region. t represents the current iteration number. T represents the maximum number of iterations.

17. The community group-buying grid demand forecasting method according to claim 14, characterized in that, The community group-buying grid demand forecasting method also presets an error threshold to evaluate the accuracy of the community group-buying grid demand forecasting model; wherein, If the error between the actual sales data of a product and the predicted value of the product in the community group buying is higher than the preset error threshold, the community group buying grid demand prediction model will be updated online based on the actual sales data of the product. If the error between the actual sales data of the goods and the predicted value of the goods in the community group buying is less than the preset error threshold, it proves that the current community group buying grid demand prediction model is accurate and does not need to be updated.

18. A computer-readable medium storing computer program code, characterized in that, The computer program code, when executed by a processor, implements the method as described in any one of claims 1-17.

19. A community group-buying grid demand forecasting device, characterized in that, include: Memory is used to store instructions that can be executed by the processor; as well as A processor for executing the instructions to implement the method as described in any one of claims 1-17.

Images