CN-122022689-A - Intelligent automatic replenishment demand prediction method for convenience store based on big data analysis

CN122022689ACN 122022689 ACN122022689 ACN 122022689ACN-122022689-A

Abstract

The invention discloses a convenience store intelligent automatic replenishment demand prediction method based on big data analysis, which comprises the steps of obtaining a business original data set, generating a business event input structure, constructing a potential demand event stream and an inventory modulation rule based on the business event input structure, constructing an improved neural Hox process model and a multi-event type input sequence, generating a basic demand intensity sequence and a basic replenishment intensity sequence, constructing a bidirectional self-excitation modulation mechanism, executing bidirectional coupling on an intensity update path, generating a coupling demand intensity sequence, constructing a hierarchical intensity generation structure, generating a hierarchical coupling demand intensity sequence, executing a shortage/deletion sensing inference based on the inventory modulation rule, generating a correction demand intensity prediction result, determining replenishment cycle parameters based on the correction demand intensity prediction result, and generating a replenishment demand prediction result. The invention improves the accuracy of demand prediction and the stability of replenishment decision.

Inventors

Han Chengxuan
YIN JINGYI

Assignees

兔悠网络科技(沈阳)有限公司

Dates

Publication Date: 20260512
Application Date: 20260203

Claims (8)

1. The intelligent automatic replenishment demand prediction method for the convenience store based on big data analysis is characterized by comprising the following steps of: acquiring a business original data set in a convenience store transaction system, executing event analysis and time unified processing, and generating a business event input structure; Based on the business event input structure, constructing a potential demand event stream, executing observable projection to generate a sales volume observation event stream, and constructing an inventory modulation rule; Constructing an improved neural Hox process model, simultaneously constructing a multi-event type input sequence by combining a potential demand event stream, and driving the improved neural Hox process model to execute intensity updating to generate a basic demand intensity sequence and a basic replenishment intensity sequence; In an improved neural Hox process model, a bidirectional self-excitation modulation mechanism is constructed, and a basic replenishment intensity sequence and a basic demand intensity sequence are combined to execute bidirectional coupling on an intensity update path of the model to generate a coupling demand intensity sequence; Constructing a hierarchical strength generation structure, generating global strength parameters, store strength parameters and commodity strength parameters by combining the business event input structure, injecting a strength update path corresponding to a bidirectional self-excitation modulation mechanism, performing hierarchical update on the coupling demand strength sequence, and generating a hierarchical coupling demand strength sequence; Performing out-of-stock deletion perception deduction on the hierarchical coupling demand intensity sequence based on inventory modulation rules to generate a corrected demand intensity prediction result; And determining a replenishment period parameter based on the corrected demand intensity prediction result and the business original data set, and aggregating the corrected demand intensity prediction result into a replenishment demand prediction result.
2. The method for predicting intelligent automatic restocking needs of convenience stores based on big data analysis of claim 1, wherein the generating of the business event input structure comprises: reading sales record data, inventory change record data, replenishment execution record data, time identification data, store identification field and commodity identification field from a convenience store transaction system, and correlating to form a business original data set; Executing event analysis on sales record data in the business original data set, generating a demand related event, and constructing a demand related event sequence; executing event analysis on the inventory change record data to generate an inventory state sequence; executing event analysis on the replenishment execution record data to generate a replenishment related event sequence; performing time unification processing on the demand related event sequence, the inventory state sequence and the replenishment related event sequence based on the time identification data, and mapping the time unification processing to discrete time scales under a unified time axis; after the time unification processing is completed, the demand related event sequence, the replenishment related event sequence, the inventory state sequence, the store identification field and the commodity identification field are subjected to structured packaging, and a business event input structure is generated.
3. The method for intelligent automatic restocking demand prediction of convenience stores based on big data analysis according to claim 1, wherein the generation of the potential demand event stream, sales observation event stream and inventory modulation rule comprises: Based on a business event input structure, reading a demand related event sequence and an inventory state sequence, establishing a corresponding time index relation for each demand related event, and mapping the demand related event sequence into a potential demand event stream; Performing observable projection processing on the potential demand event stream based on the inventory state sequence, marking the event in the potential demand event stream as an observable event in a time interval with the inventory quantity larger than zero in the inventory state sequence, and generating a sales observed event stream; constructing an inventory modulation rule based on the inventory state sequence, judging the inventory state sequence, and determining a continuous time interval with zero inventory quantity and duration exceeding a preset time threshold in the inventory state sequence as a stock out deletion interval; based on inventory modulation rules, marking a demand event corresponding to a potential demand event stream as an event which can not be directly observed in a stock-out deleting interval, and prohibiting judging the state of an event which does not occur in a sales observed event stream as zero; In the non-backout deleting interval, the sales quantity observation event stream is started to be used as a direct observation result of the requirement related event sequence, and in the backout deleting interval, the sales quantity observation event stream does not participate in the judgment of zero requirement, so that a requirement observation mechanism based on an inventory modulation rule is formed.
4. The method for predicting intelligent automatic restocking demands of convenience stores based on big data analysis according to claim 1, wherein the generating of the basic demand intensity sequence and the basic restocking intensity sequence comprises: based on the potential demand event stream and the replenishment related event sequence, an improved neural Hox process model is constructed, and an event type set is defined, wherein the event type set comprises potential demand event types and replenishment event types; executing event type coding processing on the events in the potential demand event stream and the replenishment related event sequence, generating a corresponding event type identifier according to the event type to which the event belongs, and generating an event time identifier based on the event occurrence time; calculating event time intervals for adjacent events in the potential demand event stream and the replenishment related event sequence based on the event time identification as time features; The potential demand event stream and the replenishment related event sequence after the event type encoding and the event time interval encoding are completed are uniformly organized into a multi-event type input sequence, and the historical evolution process of the demand event and the replenishment event under a uniform time axis is represented; inputting the multi-event type input sequence into an improved neural Hox process model, and performing recursive updating on the intensity functions corresponding to the potential demand event types and the replenishment event types based on event types and event time intervals of historical events in the multi-event type input sequence; And after finishing the intensity recurrence updating, outputting a basic demand intensity sequence corresponding to the potential demand event type and a basic replenishment intensity sequence corresponding to the replenishment event type.
5. The method for predicting intelligent automatic restocking needs of convenience stores based on big data analysis according to claim 1, wherein the generating of the coupling demand intensity sequence comprises: In an improved neural Hox process model, a bidirectional self-excitation modulation mechanism is constructed, a demand side modulation path and a replenishment side modulation path are established based on a basic demand intensity sequence and a basic replenishment intensity sequence, and a coupling structure which is mutually related is formed; In the demand side modulation path, a basic replenishment intensity sequence is used as demand intensity modulation input to be introduced into a demand intensity updating path, and intensity change corresponding to a replenishment event is introduced into a recursion updating process of demand intensity to be used as a modulation factor of demand intensity updating; In the replenishment side modulation path, a basic demand intensity sequence is used as replenishment intensity modulation input to be introduced into a replenishment intensity updating path, and intensity change corresponding to a demand event is introduced into a recurrence updating process of replenishment intensity to be used as a modulation factor of replenishment intensity updating; Based on the demand side modulation path and the replenishment side modulation path, performing bidirectional coupling processing on an intensity updating path of the improved neural Hox process model, and performing coupling processing of mutually restricting the demand intensity updating process and the replenishment intensity updating process in the same time evolution process; and after the bidirectional coupling processing is completed, summarizing the output results of the required strength updating path to generate a coupling required strength sequence.
6. The method for predicting intelligent automatic restocking needs of convenience stores based on big data analysis of claim 1, wherein the generating of the hierarchical coupling demand intensity sequence comprises: In an improved neural Hox process model, constructing a hierarchical strength generation structure, wherein the hierarchical strength generation structure consists of a global strength generation unit, a store strength modulation unit and a commodity strength modulation unit; Driving a global intensity generating unit based on a business event input structure, and performing aggregation processing on statistical characteristics of a demand related event sequence and a replenishment related event sequence in an overall time range to generate a global intensity parameter; a store intensity modulating unit is driven based on the store identification field, and store-level parameter generating processing is executed on a business event input structure corresponding to the same store identification field to generate store intensity parameters; Driving a commodity intensity modulation unit based on the commodity identification field, and executing commodity-level parameter generation processing on a business event input structure corresponding to the same commodity identification field to generate commodity intensity parameters; the global strength parameter, the store strength parameter and the commodity strength parameter are respectively injected into a required strength updating path corresponding to the bidirectional self-excitation modulation mechanism, and hierarchical modulation processing is carried out on the coupling required strength sequence in the strength recursive updating process; and after the hierarchical modulation processing is completed, summarizing the output result of the demand intensity updating path to generate a hierarchical coupling demand intensity sequence.
7. The method for predicting intelligent automatic restocking demand in convenience stores based on big data analysis according to claim 1, wherein the generating of the corrected demand intensity prediction result comprises: Performing time interval judgment processing on the hierarchical coupling demand intensity sequence based on an inventory modulation rule, and aligning the time range of the backout and deletion interval with the hierarchical coupling demand intensity sequence on a unified time axis to obtain a demand intensity time sequence marked with the backout and deletion interval; in the backout deleting interval, deleting perception constraint processing is executed on the hierarchical coupling demand intensity sequence, the time state of an event which does not occur in the sales quantity observation event stream is forbidden to be judged as zero, and the potential demand event stream is taken as the constraint of the existence of the demand and is estimated according to the participation demand intensity; Based on event arrival time distribution of potential demand event streams in the backorder deletion interval, executing adjustment processing on demand intensity estimation results of corresponding time intervals in the hierarchical coupling demand intensity sequence, introducing inventory removal constraint processing on the demand intensity estimation results in the backorder deletion interval, and avoiding direct inhibition of inventory unavailable states on the demand intensity estimation results; In a non-backorder deleting interval, consistency correction processing is carried out on the hierarchical coupling demand intensity sequence, sales volume observation event streams are used as direct observation basis of demands, and the demand intensity estimation result is aligned with the actual sales volume event arrival condition; And based on the deletion-sensing constraint processing result in the backout-of-stock deletion interval and the consistency correction processing result in the non-backout-of-stock deletion interval, performing unified correction on the hierarchical coupling demand intensity sequence within a complete time range, and generating a correction demand intensity prediction result.
8. The method for predicting intelligent automatic restocking demand in convenience stores based on big data analysis according to claim 1, wherein the generating of the restocking demand prediction result comprises: Based on the correction demand intensity prediction result, reading an order period field and an arrival completion time stamp field in the replenishment execution record data, and performing time difference analysis on order triggering time and arrival completion time corresponding to the historical replenishment execution record to form replenishment period parameters; Performing time window division processing on the corrected demand intensity prediction result in a time interval corresponding to the replenishment cycle parameter, and mapping the corrected demand intensity prediction result into a demand intensity sequence in a discrete replenishment cycle window; in each replenishment cycle window, performing accumulation aggregation processing on the demand intensity sequence to obtain a cycle demand quantity prediction result, and forming a replenishment demand prediction result; and carrying out structural organization on the replenishment demand prediction result according to the store identification field and the commodity identification field, and outputting the replenishment demand prediction result for intelligent automatic replenishment of the convenience store.

Description

Intelligent automatic replenishment demand prediction method for convenience store based on big data analysis Technical Field The invention relates to the field of retail supply chain management, in particular to a convenience store intelligent automatic replenishment demand prediction method based on big data analysis. Background The convenience store operation has the characteristics of more commodity types, high sales frequency and short replenishment period, and the replenishment decision directly influences inventory turnover efficiency and the shortage risk. The conventional convenience store replenishment management is based on historical sales statistics, manual experience rules or a time sequence-based prediction method, and calculates future demands according to sales data of a past period of time, and accordingly determines the number of orders and replenishment cycles. However, in actual business processes, convenience store sales data is often directly affected by inventory status. When a commodity is out of stock, sales data does not truly reflect potential demands, and observation deletion caused by the out-of-stock exists in a large number in the history sales. In the prior art, the sales volume is usually zero and is directly regarded as the demand is zero, or a simple missing value filling mode is adopted to process the missing interval, so that the real demand change is difficult to accurately describe, and the underestimation of the demand or the late replenishment is easy to cause. Meanwhile, the existing replenishment prediction method generally treats replenishment behaviors as exogenous conditions, does not describe dynamic interaction relations between replenishment events and subsequent demand changes, and is difficult to reflect the influence of replenishment rhythms on demand release and sales fluctuation. In addition, aiming at the difference between different stores and different commodities, the prior art mostly adopts independent modeling or simple layering processing modes, lacks the capability of collaborative depiction in a unified model structure, and has limited stability and generalization capability of a prediction result. Therefore, how to provide a method for predicting intelligent automatic replenishment demands of convenience stores based on big data analysis is a problem to be solved by those skilled in the art. Disclosure of Invention The invention aims to provide an intelligent automatic replenishment demand prediction method for convenience stores based on big data analysis, which is characterized in that a business event input structure is obtained and constructed, a demand related event sequence is mapped into a potential demand event stream, a sales observed event stream is formed by combining an inventory state sequence, and the potential demand event stream and the replenishment related event sequence are jointly modeled in an improved neural Hox process model. According to the invention, a bidirectional self-excitation modulation mechanism and a hierarchical strength generation structure are introduced into a model, a hierarchical coupling demand strength sequence is generated, and a stock modulation rule is based on which a stock shortage perception deduction is executed to obtain a corrected demand strength prediction result, and the corrected demand strength prediction result is further aggregated to form a replenishment demand prediction result. The invention can depict the dynamic coupling relation between the demand and the replenishment, reduce the influence of the shortage and the deletion on the demand forecast, and improve the stability and the usability of the replenishment demand forecast result. According to the embodiment of the invention, the intelligent automatic replenishment demand prediction method for the convenience store based on big data analysis comprises the following steps: acquiring a business original data set in a convenience store transaction system, executing event analysis and time unified processing, and generating a business event input structure; Based on the business event input structure, constructing a potential demand event stream, executing observable projection to generate a sales volume observation event stream, and constructing an inventory modulation rule; Constructing an improved neural Hox process model, simultaneously constructing a multi-event type input sequence by combining a potential demand event stream, and driving the improved neural Hox process model to execute intensity updating to generate a basic demand intensity sequence and a basic replenishment intensity sequence; In an improved neural Hox process model, a bidirectional self-excitation modulation mechanism is constructed, and a basic replenishment intensity sequence and a basic demand intensity sequence are combined to execute bidirectional coupling on an intensity update path of the model to generate a coupling demand intensity sequence; Constr