Retail store goods reporting data processing method and system based on sales prediction model

Abstract

The invention provides a retail store goods reporting data processing method and system based on a sales volume prediction model, and relates to the technical field of computer data processing, and the method comprises the steps: 1, synchronizing store sales data, real-time weather data, holiday and festival information, promotion activity information and store real-time transaction data through a real-time data collection interface, and carrying out the real-time data collection of the store sales data, the real-time weather data, the holiday and festival information, the promotion activity information and the store real-time transaction data; constructing a multi-dimensional dynamic data set; 2, based on the multi-dimensional dynamic data set, identifying data distribution characteristics to establish an analysis reference so as to obtain two groups of data sequences representing different sales volume influence dimensions, calculating the correlation degree between the data sequences, and generating a sales volume change characteristic interval; according to the invention, the real-time inventory of the store, the supply chain period parameter and the sales prediction result are integrated, accurate generation of the goods reporting instruction and automatic circulation of the whole process of goods reporting are realized, and the efficiency and accuracy of goods reporting are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer data processing, in particular to a retail store reporting data processing method and system based on a sales prediction model. BACKGROUND

[0002] In the daily reporting of milk tea chain stores in summer, most of the traditional technology only uses past sales data as the basis for reporting, and rarely includes dynamic data such as external environment and store operation. For example, a milk tea store plans to report ice lemon tea for next Monday, and the traditional technology only refers to the average sales of this product on Monday in the past three weeks, with an average of 80 cups per day, to determine the reporting quantity of 80 cups. However, it does not synchronize the weather data for next Monday, which predicts that the temperature on the day will reach 38°C, the highest temperature in the past month. High temperatures in summer usually lead to an increase in the sales of ice drinks, resulting in the product being sold out before 12 noon on Monday next week, and potential sales are lost in the afternoon due to lack of stock.

[0003] In addition, under the traditional reporting technology, the process is mostly a linear mode of manual application, manual review and manual tracking. After the reporting instruction is generated, it is rarely compared with the actual sales and prediction results in real time. Even if deviations are found, they are mostly adjusted by manual review, lacking an automatic deviation warning and dynamic correction mechanism. It may be assumed that reporting is the end point for completing delivery, and subsequent deviations can be manually corrected in the next period. However, milk tea raw materials such as fresh milk and fresh fruit have a short shelf life, and if deviations are not discovered in time, it may lead to inventory loss or continuous lack of stock. SUMMARY

[0004] The technical problem to be solved by the present application is to provide a retail store reporting data processing method and system based on a sales prediction model, to reduce inventory accumulation and lack of stock, and to improve reporting accuracy and inventory turnover efficiency.

[0005] To solve the above technical problems, the technical solution of the present application is as follows: In a first aspect, a retail store reporting data processing method based on a sales prediction model, the method comprising: Step 1: Synchronize store sales data, real-time weather data, holiday information, promotion activity information and store real-time transaction data through a real-time data acquisition interface to build a multi-dimensional dynamic data set; Step 2: Based on the multi-dimensional dynamic data set, identify the data distribution characteristics to establish an analysis benchmark to obtain two groups of data sequences representing different sales influence dimensions, and calculate the correlation between the data sequences to generate a sales change characteristic interval; Step 3: Select a number of evaluation sample points within and outside the sales change characteristic interval, generate a change trajectory according to the evolution correlation of the sample points in the time dimension, and generate dynamic adjustment parameters according to the trajectory pattern; Step 4, input the dynamic adjustment parameter into the adaptive sales prediction model, real-time correct the model parameter, and dynamically adjust the importance of each external factor according to real-time weather, holidays and promotion events, to generate a sales prediction result; Step 5, integrate the sales prediction result with the real-time inventory data of the store and the supply chain cycle parameters to generate a goods ordering instruction; Step 6, according to the goods ordering instruction, drive the goods ordering task to automatically flow between the application, audit and distribution tracking links through the process engine, and synchronize the status to each terminal in real time; In the execution process, the deviation between the actual sales data and the prediction result is compared in real time, and if the deviation is greater than the preset threshold, a deviation analysis report is automatically triggered.

[0006] In the second aspect, a retail store goods ordering data processing system based on a sales prediction model includes: A data synchronization module is configured to synchronize store sales data, real-time weather data, holiday information, promotion activity information and store real-time transaction data through a real-time data acquisition interface, and construct a multi-dimensional dynamic data set; An identification generation module is configured to identify data distribution characteristics based on the multi-dimensional dynamic data set, establish an analysis benchmark, obtain two groups of data sequences representing different sales influence dimensions, calculate the correlation between the data sequences, and generate a sales change characteristic interval; An analysis generation module is configured to select a number of evaluation sample points within and outside the sales change characteristic interval, generate a change trajectory based on the evolution correlation of the sample points in the time dimension, and generate dynamic adjustment parameters based on the trajectory pattern; A prediction correction module is configured to input the dynamic adjustment parameters into the adaptive sales prediction model, real-time correct the model parameters, and dynamically adjust the importance of each external factor according to real-time weather, holidays and promotion events, to generate a sales prediction result; An instruction generation module is configured to integrate the sales prediction result with the real-time inventory data of the store and the supply chain cycle parameters to generate a goods ordering instruction; A process warning module is configured to automatically flow the goods ordering task between the application, audit and distribution tracking links through the process engine according to the goods ordering instruction, and synchronize the status to each terminal in real time; In the execution process, the deviation between the actual sales data and the prediction result is compared in real time, and if the deviation is greater than the preset threshold, a deviation analysis report is automatically triggered.

[0007] In the third aspect, a computing device includes: One or more processors; A storage device for storing one or more programs, when the one or more programs are executed by the one or more processors, so that the one or more processors implement the method.

[0008] In a fourth aspect, a computer-readable storage medium stores a program, which implements the method when executed by a processor.

[0009] The above solution of the present invention includes at least the following beneficial effects: By synchronizing store sales, real-time weather, holidays, promotional activities and real-time transaction data to build a dynamic data set, we can fully capture the combined impact of internal and external factors on sales, making the forecast results more in line with real consumption scenarios. Based on the time evolution trajectory of the sample points, dynamic adjustment parameters are generated, basic sales, trends and seasonal parameters are corrected in real time, and the weights of external factors are dynamically adjusted according to weather, holidays and promotional events to avoid forecast deviations. When generating delivery instructions, supply chain cycle parameters are fully integrated, and delivery time points and basic delivery quantities are planned to avoid inventory backlogs in transit due to early delivery or out-of-stock risks caused by late delivery. At the same time, combined with economic ordering The principles of cargo volume and supplier minimum order quantity reduce the hidden costs of warehousing leasing, logistics transportation and ordering; the process engine drives the automatic flow of reporting tasks in the application, review and delivery tracking links, without the need for manual transfer of documents and update of status. It not only reduces manual entry errors and repetitive work, but also shortens the reporting process cycle and improves the operational efficiency of stores and headquarters. The status of each link of the reporting task is synchronized in real time to the store management terminal, supplier collaboration platform and headquarters monitoring terminal, avoiding communication costs caused by information asymmetry. At the same time, it allows suppliers to plan production capacity in advance and logistics parties to accurately arrange transportation, thereby improving the overall collaborative efficiency of the supply chain. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 It is a flowchart of a method for processing retail store delivery data based on a sales forecast model provided by an embodiment of the present invention.

[0011] Figure 2 2 is a schematic diagram of a retail store reporting data processing system based on a sales forecasting model provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0012] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0013] like Figure 1 As shown, an embodiment of the present invention proposes a method for processing retail store delivery data based on a sales forecast model, the method comprising the following steps: Step 1, synchronizing store sales data, real-time weather data, holiday information, promotional activity information and store real-time transaction data through a real-time data acquisition interface, and constructing a multi-dimensional dynamic data set; Step 2, based on the multi-dimensional dynamic data set, identifying data distribution characteristics to establish an analysis benchmark, obtaining two groups of data sequences representing different sales impact dimensions, and calculating the correlation between the data sequences to generate a sales change characteristic interval; Step 3, selecting several evaluation sample points within and outside the sales change characteristic interval, generating a change trajectory based on the evolution correlation of the sample points in the time dimension, and generating dynamic adjustment parameters based on the trajectory pattern; Step 4, inputting the dynamic adjustment parameters into the adaptive sales prediction model to correct the model parameters in real time, and dynamically adjusting the importance of each external factor based on real-time weather, holidays and promotional events to generate a sales prediction result; Step 5, integrating the sales prediction result with store real-time inventory data and supply chain cycle parameters to generate a restocking instruction; Step 6, according to the restocking instruction, driving the restocking task to automatically flow through the application, review and distribution tracking links through the process engine, and synchronizing the status to each terminal in real time; in the execution process, the deviation between the actual sales data and the prediction result is compared in real time, and if the deviation is greater than the preset threshold, a deviation analysis report is automatically triggered.

[0014] In the embodiments of the present application, the multi-dimensional dynamic data set constructed by synchronizing store sales data, real-time weather data, holiday information, promotional activity information and store real-time transaction data through a real-time data acquisition interface can capture the internal and external key factors affecting sales, avoid analysis bias caused by partial data, identify data distribution characteristics based on multi-dimensional dynamic data set, establish an analysis benchmark, split data sequences of different sales impact dimensions and calculate correlation, and finally generate a sales change characteristic interval, which can clearly define the expected fluctuation range of sales under the action of complex factors, convert multi-dimensional data into sales change basis with clear reference value, help operators intuitively master the correlation logic of sales affected by internal operation and external environmental factors, reduce the subjectivity of sales change judgment, select evaluation sample points within and outside the sales change characteristic interval, generate a change trajectory based on the time dimension, and extract dynamic adjustment parameters, making the restocking adjustment more targeted and timely.

[0015] The dynamic adjustment parameter is input into the adaptive sales prediction model for real-time correction, and the importance of external factors is dynamically adjusted according to real-time weather, holidays and promotion events, so that the accuracy of the sales prediction result is improved. The real-time correction of the parameter can timely correct the past prediction deviation, and the dynamic adjustment of the weight of the external factor can make the prediction model more suitable for the actual market situation at the moment, so as to ensure that the prediction result can accurately reflect the future sales trend and provide a reliable basis for the goods ordering decision; the goods ordering instruction is generated by integrating the sales prediction result, the real-time inventory data of the store and the supply chain cycle parameter, so that the feasibility of the goods ordering scheme is ensured. The integration of real-time inventory data can avoid the problem of inventory accumulation or shortage caused by blind ordering, and the consideration of the supply chain cycle parameter can accurately plan the goods ordering time point and the goods ordering quantity, so as to avoid the problem of goods arriving at the store too late or too early caused by not considering the supply chain time efficiency. The goods ordering task is automatically transferred in the application, audit and distribution tracking links driven by the process engine, and the state is synchronized to each terminal in real time, so that the operation efficiency of the goods ordering process is improved.

[0016] In a preferred embodiment of the present application, step 1 is performed by synchronizing store sales data, real-time weather data, holiday information, promotion activity information and store real-time transaction data through a real-time data acquisition interface to build a multi-dimensional dynamic data set, which can include: In the embodiment of the present application, interfaces are established with corresponding data source parties for store sales data, real-time weather data, holiday information, promotion activity information and store real-time transaction data. For internal data, sales data and real-time transaction data, the real-time data output interface of the store POS system and the cash register system is connected. For external data, the open data interface of the third-party meteorological platform and the official holiday release platform is connected for real-time weather data and holiday information. For promotion activity information, the activity data interface of the enterprise headquarters marketing management system is connected. At the same time, a unified data acquisition rule is set to determine the acquisition fields of each type of data, such as the need for store sales data to include product code, sales quantity, sales amount, sales timestamp; real-time weather data needs to include temperature, precipitation probability, wind force level of the area where the store is located; holiday information needs to include holiday name, start time and end time; promotion activity information needs to include activity product range, discount strength, activity start and end time; and store real-time transaction data needs to include transaction serial number, transaction time, payment method and associated product details. The acquisition frequency of each data is determined, such as the acquisition of store sales data and real-time transaction data at a minute level, the acquisition of real-time weather data at an hour level, the advance acquisition and update of holiday information at a quarterly level, and the real-time synchronization of promotion activity information after the creation of the activity.

[0017] The interface regularly extracts the sales records of goods within a specified time period, such as the last minute, from the store POS system and sales management system. According to the preset product coding rules, it filters out the valid sales data within the store's business scope, and eliminates the sales records corresponding to test transactions and invalid refunds to ensure that the data only contains real product sales behaviors. The interface obtains the current weather data of the corresponding area from the third-party meteorological platform based on the store's registered geographical location information, such as province, city, district / county, and compares the timestamp in the data with the collection time to filter out real-time weather data with a time difference of less than 10 minutes, and excludes outdated weather information; the interface extracts statutory holidays, weekends and company-defined holiday information in the next 12 months from the official holiday platform or the company's preset holiday database, and uses date format rules such as YYYY-MM-DD to sort holidays. Holiday times are standardized, and holiday types are marked, such as statutory holidays, weekends, and special event days, to filter out expired holiday information; the interface extracts promotional activity data that is currently in the preparation and implementation stages from the headquarters marketing management system, and filters out promotional information related to the store based on the activity coverage, such as promotions for national stores and regional stores. At the same time, the product list, discount ratio, and activity start / end time corresponding to the activity are extracted, and promotional activity data that has ended or been canceled is eliminated; the interface extracts the transaction flow data of each transaction within the last 1 minute from the store cash register system and payment system, and filters out valid transactions based on the transaction status, such as paid or completed, and associates the transaction serial number with the corresponding product details, such as matching the purchased product code and quantity through the transaction serial number, and eliminates invalid transaction data that is pending payment or has been canceled.

[0018] The sales time stamp, transaction time in the store sales data and real-time transaction data are uniformly converted into the time format of YYYY-MM-DD HH:MM:SS; the start / end time of holiday information and the start / end time of promotion activities are also standardized in the same format to ensure that the time dimensions of all data can be aligned. The product codes in the store sales data, real-time transaction data and promotion activity information are checked and corrected according to the enterprise unified product code rules, such as category code and single product code, to ensure that the codes corresponding to the same product in different data types are consistent and avoid association failure due to code differences. The sales quantity and sales amount in the store sales data are unified in units according to the enterprise financial accounting standards, and the units of temperature, wind power and other units in the real-time weather data are fixed to ensure that the units of numerical data can be directly compared and calculated. After standardization, the various types of data are integrated through the association field, with time as the core association dimension, and the store sales data, real-time transaction data and real-time weather data in the same time interval are matched, such as the sales data and transaction data of 2025-08-25 10:00:00-10:01:00 are associated with the weather data in this time period. The holiday information and promotion activity information are associated with the sales data and transaction data with date and product code as the association dimension, such as the holiday information of 2025-10-01 (National Day) is associated with the sales data of all products in the store on that day. The product codes corresponding to a certain promotion activity are associated with the sales data of the product during the activity period to form an integrated data set including time, product, sales, weather, holiday, promotion and transaction multi-dimensional information.

[0019] A dynamic updating mechanism is established to repeat the above data synchronization, screening, standardization and integration process according to the set collection frequency, and real-time append the newly collected effective data to the existing integrated data set to realize dynamic expansion of the data set, and ensure that the data set can reflect the latest store operation and external environment. At the same time, after each data update, integrity check is performed to count the actual collection quantity and the expected collection quantity of various types of data, such as the sales data volume and transaction data volume estimated according to the number of store POS machines and transaction frequency. If the actual collection quantity of a certain type of data is less than 95% of the expected collection quantity, the data collection mechanism is triggered to re-call the corresponding interface to obtain the data of the missing time period. The matching rate of the association field is checked, such as the matching rate of the time dimension and the matching rate of the product code. If the matching rate is less than 98%, the format problem in the data standardization process or the field missing problem in the interface synchronization is checked and corrected, and then the association integration is performed again to finally form a multi-dimensional dynamic data set with complete data, unified format and real-time update.

[0020] In a preferred embodiment of the present application, the step 2, based on the multi-dimensional dynamic data set, identifies the data distribution characteristics to establish an analysis benchmark, to obtain two groups of data sequences representing different sales impact dimensions, and calculates the correlation degree between the data sequences to generate the sales change characteristic interval, which can include: In the embodiment of the present application, step 220, data distribution characteristics of multi-dimensional dynamic data set are identified, and data analysis benchmark is established by analyzing statistical characteristics, specifically including: collecting multi-dimensional dynamic data recorded by week in the past 8 years, including internal operation indicators such as product production batches per week, store salesperson number, online promotion investment, customer return and exchange number, warehouse turnover rate, and external environment indicators such as regional total population flow per week, per capita consumer spending, competitor promotion frequency, logistics transportation cost, and weather abnormal day number, a total of 416 weekly data, cleaning the data, eliminating duplicate records such as repeated data entry in the same week, marking missing values such as unrecorded logistics transportation cost in a week, and labeling to be supplemented, ensuring that the effective data amount of each indicator is not less than 380; taking the weekly online promotion investment as an example, adding the 380 valid data to get a total of 1.9 million yuan, and dividing by 380 to get the mean value of 5000 yuan; sort the data from small to large, such as 3000 yuan, 3200 yuan, …, 8000 yuan, take the average of the 190th and 191st data, 4900 yuan and 5100 yuan, and get the median value of 5000 yuan; count the number of the most frequently occurring values, 5000 yuan appears 45 times, and the mode is determined as 5000 yuan; taking the competitor promotion frequency as an example, find the maximum value of 8 times / week and the minimum value of 1 time / week, and the range is 7 times; after calculating the mean value of 4 times / week, calculate the difference between each data and the mean value, such as 1-4=-3, 2-4=-2, etc., square the difference and sum, such as 9+4+…=2660, divide by 380 to get the variance 7; take the square root of 7 to get the standard deviation of about 2.65 times / week; draw a histogram of warehouse turnover rate, with the turnover rate interval on the horizontal axis, such as 3-4 times / week, 4-5 times / week, etc., and the number of weeks in the corresponding interval on the vertical axis, if the right tail of the histogram is significantly longer, the number of weeks in the high turnover rate interval is more, it is determined as right-skewed distribution, the peak interval is recorded, 4-5 times / week and the skewness direction, and the extreme high value interval of more than 7 times / week is determined through data distribution observation.

[0021] For normally distributed indicators such as per capita consumer spending, the mean is 3000 yuan / week and the standard deviation is 400 yuan. The normal fluctuation range is 3000±3×400 yuan, i.e. 1800-4200 yuan. Data outside this range is marked for review. For skewed distribution indicators such as customer returns, the upper quartile is 6 pieces for the 285th data and the lower quartile is 2 pieces for the 95th data. The interquartile range is 4 pieces. The reasonable interval is 4±4 pieces, i.e. 0-8 pieces. Data exceeding 8 pieces is marked as abnormally high. The baseline parameters of all indicators are summarized to form a baseline table including indicator name, data type, central tendency value, dispersion value, normal interval, and abnormal marking rules, serving as a unified standard for subsequent data processing.

[0022] Step 221, based on the data analysis baseline, the multi-dimensional dynamic data set is processed by dimension deconstruction and sequence generation to obtain two groups of data sequences representing the influence of internal operation factors and external environmental factors, specifically including: product production batches, store salespersons, online promotion investment, customer returns, and warehouse turnover rate are clearly classified as internal operation factor indicators; regional total population flow, per capita consumer spending, competitor promotion frequency, logistics transportation cost, and weather abnormal days are classified as external environmental factor indicators, forming a classification list and labeling the business meaning of each indicator. Check the logistics transportation cost data against the baseline table and find that the cost for a certain week is 12000 yuan, which exceeds the normal interval of 5000-10000 yuan. After verification, it is an additional cost caused by temporary heavy rain. The average value of the previous 3 weeks is 9000 yuan, 9000 yuan, 8500 yuan, and 9500 yuan. Corrected; another week's cost is 9800 yuan, close to the upper limit but within reasonable fluctuations, keep the original value and mark it as close to the upper limit. For missing values in weather abnormal days, fill in the weekly average abnormal days for the month, such as 2 days / week in July.

[0023] In chronological order, arrange the processed internal indicator data from week 1 to week 416, each time point corresponds to a complete set of data, for example, week 1 has 15 batches of production, 30 salespersons, 4800 yuan of promotion investment, 3 returned goods, and a turnover rate of 4.2 times; week 2 has 16 batches of production, 32 salespersons, 5200 yuan of promotion investment, 2 returned goods, and a turnover rate of 4.5 times… Form a sequence of internal data at 416 time points, saved as a table that increases by week; Similarly, arrange the processed external indicator data in chronological order, each time point corresponds to a set of data, for example, week 1 has a population flow of 50,000 person-times, an average consumption of 2900 yuan per person, 3 competitive product promotions, a logistics cost of 8000 yuan, and 1 abnormal day; Week 2 has a population flow of 52,000 person-times, an average consumption of 2950 yuan per person, 2 competitive product promotions, a logistics cost of 7800 yuan, and 0 abnormal days… Form a sequence of external data at 416 time points, check the time tags of the two sequences week by week to ensure that weeks 1 to 416 correspond completely without time displacement; Check the data format of each indicator, such as numerical units and integer / decimal types, and remove format error data, such as changing 5 times to 5.

[0024] Step 222, using the two data sequences, calculate the dynamic correlation degree between each other to get the analysis result of the dynamic correlation degree, specifically including: combining the weekly data characteristics, set the time window to 12 weeks, i.e. 3 months, the sliding step is 2 weeks, calculate the correlation degree of the new window every 2 weeks, ensure that the window covers the complete sales cycle and avoid too much data overlap; The first window, weeks 1-12 are calculated, extract all data of 12 weeks in the internal sequence, such as 12 production batch values, 12 salesperson values, etc. and the data of 12 weeks in the external sequence, such as 12 population flow values, 12 average consumption values, etc. Treat the two sets of data as whole sequences respectively, and calculate the correlation degree: First, calculate the comprehensive mean of the internal sequence, add all internal indicator data and divide by the total number of data to get the mean A and the comprehensive mean of the external sequence, similarly get the mean B; Second, calculate the difference between each week's internal data and A, and the difference between each week's external data and B, multiply the two difference values of the same week and accumulate to get the total sum C, such as the product of the first week's difference + the product of the second week's difference +…+ the product of the twelfth week's difference = 1200; Third, calculate the sum of squares of the internal difference, total sum D = 800 and the sum of squares of the external difference, total sum E = 1800; Fourth, divide C by the square root of (D x E) to get the correlation degree of this window as 1.0 (this is an example, actual calculation needs to be based on real data).

[0025] Slide to the 3-14th week by steps, repeat the above four steps to calculate, get the correlation degree 0.85; Continue to slide until all windows are calculated (a total of (416-12) ÷2+1=203 windows), get 203 correlation degree values, arrange 203 correlation degree values in window order, such as 1.0, 0.85, 0.92, …, 0.78, the total sum is 162.4, the average is 0.8 (162.4 ÷ 203); The maximum value is 1.0 (1-12 weeks), the minimum value is 0.55 (380-391 weeks); Calculate the change rate of adjacent windows, such as the change rate of the 2nd window (0.85) and the 1st window (1.0) is (0.85-1.0) ÷1.0=-15%, indicating that the correlation degree decreases by 15%, by drawing the correlation degree sequence line graph, it is observed that the correlation degree of the first 50 windows is between 0.8-1.0 (high correlation), the 51-150 windows are between 0.7-0.9 (medium-high correlation), and the 151-203 windows are between 0.55-0.8 (medium correlation), it is summarized that the correlation degree decreases in a stepwise manner over time, and the interval with the largest decrease is marked, which is the 140-150th window, with a decrease of 12%.

[0026] Step 223, based on the analysis result of dynamic correlation degree, generate a sales change characteristic interval for describing the expected fluctuation range of sales under the action of composite factors, specifically including: according to the distribution characteristics of 203 correlation degree values, divide into three intervals, high correlation (0.8-1.0), medium correlation (0.6-0.8), and low correlation (0.5-0.6), the number of windows contained in each interval is 60, 120, and 23 respectively, extract the actual weekly sales data of each window, such as the weekly sales of the 1-12th week is 1200 pieces, the 3-14th week is 1150 pieces, etc.

[0027] High correlation interval: the total sum of the weekly sales of 60 windows is 72000 pieces, the average value is 1200 pieces / week; Calculate the sales fluctuation range (maximum value-minimum value) of each window, the total sum is 30000 pieces, the average fluctuation range is 500 pieces / week; Medium correlation interval: the total sum of the weekly sales of 120 windows is 120000 pieces, the average value is 1000 pieces / week; The average fluctuation range is 400 pieces / week; Low correlation interval: the total sum of the weekly sales of 23 windows is 18400 pieces, the average value is 800 pieces / week; The average fluctuation range is 300 pieces / week.

[0028] Sales change characteristic interval generation: High correlation interval: take 1200 pieces as the center, calculate the upper and lower limits as 1200±250 pieces, i.e. 950-1450 pieces / week, according to the average fluctuation range 500 pieces; Medium correlation interval: 1000±200, i.e. 800-1200 pieces / week; Low correlation interval: 800±150, i.e. 650-950 pieces / week.

[0029] An attribute interval table is formed, and the corresponding sales fluctuation range of each correlation degree interval is clearly marked, such as high correlation (0.8-1.0), sales 950-1450 pieces / week, and the window quantity proportion of each interval is noted (high correlation 29.6%, medium correlation 59.1%, and low correlation 11.3%).

[0030] Through the concentration trend, dispersion degree calculation and distribution form analysis of each index, the data analysis benchmark established can accurately locate abnormal data and avoid the interference of abnormal values on analysis; at the same time, the benchmark table provides a unified evaluation standard for different indexes, ensures the objectivity and consistency of data processing, lays a reliable data foundation for the entire analysis process, dimension decomposition classifies complex data according to influencing factors, clearly separates the influence paths of internal operation and external environment, avoids analysis confusion caused by mixed data of different dimensions, and the generated time sequence completely retains the time attribute of the data, can directly show the change law of the two types of factors with time, and improves the pertinence and accuracy of analysis.

[0031] In a preferred embodiment of the present application, the step 3, selecting a plurality of evaluation sample points within and outside the sales change attribute interval, generating a change trajectory according to the evolution correlation of the sample points in the time dimension, and generating a dynamic adjustment parameter according to the trajectory form, can include: In the embodiment of the present application, step 330, taking the boundary of the sales change attribute interval as a reference, combining the fluctuation rate and statistical distribution characteristics of the sales data, a plurality of representative evaluation sample points are selected in the internal and external neighborhood of the interval, specifically including: collecting daily or weekly sales data in the past period, such as the past 12 months, first calculating the average value of all sales in this period, then calculating the difference between each sales data and the average value, squaring these differences and then averaging them to obtain a value called variance, and the square root of the variance is the standard deviation, and the lower boundary of the sales change attribute interval is obtained by subtracting 2 times the standard deviation from the average value, and the upper boundary is obtained by adding 2 times the standard deviation to the average value, and the 2 times the standard deviation can be adjusted according to the requirement of the business on the data accuracy, such as 1.5 times or 3 times; using the sales data used to determine the interval boundary, first find the difference between the adjacent two sales data, such as the second sales minus the first sales, the third sales minus the second sales, and so on until the last sales minus the second last sales, then add the absolute values of these differences to obtain a total fluctuation, and then divide the total fluctuation by the product of (data number-1) and sales average value, and the result is the sales data fluctuation rate, which can reflect the fluctuation size of the sales relative to the average sales in unit time. Based on the sales data, in addition to the average value and standard deviation that has been calculated, the median is also calculated, and all sales data is sorted in ascending order, if the number of data is odd, the middle number is the median; if it is even, the average of the two middle numbers is the median. At the same time, the maximum sales value and the minimum sales value in the data set are found, and through these values, the distribution of sales data can be judged, such as whether the data is symmetrically distributed around the average value, if so, it is close to normal distribution, or whether it is skewed to one side, such as most data is small and a few data is large, which is right-skewed distribution; otherwise, it is left-skewed distribution. According to the interval boundary, volatility and statistical distribution characteristics calculated before, sample points are selected. Within the interval, if the sales data is close to normal distribution and the volatility is small, the midpoint of the interval and the lower boundary plus the lower boundary divided by 2 are taken as the center, and sample points are selected at equal intervals, such as dividing the interval into 5 segments, and the endpoints of each segment are taken as sample points, plus the average value point and the median point of sales within the interval to ensure that different sales levels within the interval are covered. In the adjacent area outside the interval, the range of this area should refer to the volatility. If the volatility is large, the upper boundary is expanded outward (the upper boundary minus the lower boundary) multiplied by the volatility to 2 times (the upper boundary minus the lower boundary) multiplied by the volatility, the lower boundary is expanded outward 2 times (the upper boundary minus the lower boundary) multiplied by the volatility to (the upper boundary minus the lower boundary) multiplied by the volatility. If the volatility is small, the expansion range is reduced by about half. In this external area, points are also selected in combination with the data distribution, such as right-skewed data, which selects several points outside the upper boundary; left-skewed data, which selects several points outside the lower boundary. Finally, a total of 8 to 12 evaluation sample points that can represent different situations are selected within and outside the interval.

[0032] In step 331, the evaluation sample points are sorted according to the corresponding time stamps, and adjacent sample points are connected according to the time sequence to calculate the weighted average value between adjacent points for smooth fitting to form a change trajectory to represent the evolution process and correlation of sales in the time dimension. Specifically, the time records corresponding to each evaluation sample point are first collected, and the unit of the time record should be consistent with the recording unit of the sales data, such as recording sales once a day, and the time record is accurate to the day; recording sales once a week, and the time record is accurate to the week. Then, all evaluation sample points are sorted in chronological order to form a sample point sequence arranged in chronological order, and each sample point contains corresponding time and sales data. For the ordered sample point sequence, the two adjacent sample points are connected in turn, and the weighted average of the two adjacent points is calculated. The weight is determined according to the time interval between the two points. The smaller the time interval, the greater the weight of the sample point, because the sample points with small time intervals can better reflect the change of sales in a short period of time. For example, if two sample points are separated by 2 days, the time interval is 2, the weight of the first sample point is (2+1) / (2*2+1), and the weight of the second sample point is 2*(2*2+1). Then, the sales of the first sample point multiplied by its weight is added to the sales of the second sample point multiplied by its weight, and the result is the weighted average of the two adjacent points. According to this method, the weighted average of all adjacent sample points is calculated, and each weighted average corresponds to a time point, which is the average of the time of the two adjacent sample points. The ordered original sample points and the calculated weighted average points are connected in time sequence to form a preliminary broken line. In order to make the line smoother, a smoothing process is performed again. The possible original sample points and weighted average points are found on the preliminary broken line, and the time point and sales prediction value between the two points are calculated. The intermediate time point is the average of the time of the two points, and the intermediate sales prediction value is the average of the sales of the two points. These intermediate points are also added to the point sequence. Finally, all points, original sample points, weighted average points, and intermediate points are connected in time sequence to form a smoother curve, which is the change trajectory that can reflect the change process and mutual relationship of sales over time.

[0033] In step 332, the morphological characteristics of the change trajectory are analyzed, the trajectory inflection points, trend direction and fluctuation amplitude are identified, and the adjustment parameters for dynamically adjusting the order quantity and order frequency are generated. Specifically, it includes: finding three consecutive points on the change trajectory, first calculating the slope from the first point to the second point, subtracting the sales of the first point from the sales of the second point, and then dividing by the time difference between the second point and the first point; then calculating the slope from the second point to the third point, subtracting the sales of the second point from the sales of the third point, and then dividing by the time difference between the third point and the second point; first set a slope difference threshold, which is determined according to the average size of the slope difference of all adjacent points on the trajectory, for example, take 1.5 times the average value, if the first slope is positive and the second slope is negative, and the absolute value of the difference between the two slopes exceeds the set threshold, the middle point is the inflection point where the sales change from rising to falling; if the first slope is negative and the second slope is positive, and the absolute value of the difference exceeds the threshold, the middle point is the inflection point where the sales change from falling to rising. According to this method, all inflection points on the trajectory are found out. The changing trajectory is divided into several time periods according to time, such as calculating one time period for every 3 adjacent points, and the specific number is adjusted according to the length of the trajectory. The average value of the slope of all adjacent points in each time period is calculated. If the average value is positive, it means that the sales in this time period is in an upward trend; if it is negative, it is in a downward trend; if the absolute value of the average value is particularly small, such as less than 0.01, the specific value is determined according to the business trend sensitivity, that is, it is a stable trend. Combining the trend situation of all time periods, the overall trend of the entire trajectory can be known, such as first rising, then stable, then falling, and so on. Find a complete fluctuation period on the changing trajectory, from a sales peak to the next peak, or from a sales trough to the next trough. The fluctuation amplitude of this period is obtained by subtracting the sales of the lowest point from the sales of the highest point in this period. If the trajectory has multiple fluctuation periods, the average value of the fluctuation amplitudes of all periods is calculated as the average fluctuation amplitude of the entire trajectory. At the same time, the maximum and minimum fluctuation amplitudes are found to fully understand the fluctuation of the trajectory. According to the sales trend and fluctuation amplitude, the order quantity adjustment parameter is determined. First, set a fluctuation amplitude threshold. If the overall trend is rising and the average fluctuation amplitude is less than the set threshold, set the order quantity adjustment parameter to increase. The increase ratio is determined according to the trend slope. First, find the maximum slope in the data. Divide the average slope of the current time period by the maximum slope, and multiply a basic adjustment ratio, such as 10%, to get the specific increase ratio. For example, if the maximum slope is 0.2 and the current average slope is 0.05, the increase ratio is (0.05 ÷ 0.2) × 10%, which is 2.5%. The order quantity is 2.5% more than the last time. If the trend is downward and the average fluctuation amplitude is greater than the threshold, set the order quantity adjustment parameter to decrease. The decrease ratio is calculated in a similar way, but the basic adjustment ratio can be increased appropriately, such as 15%. In addition, if there is an inflection point of rising to falling in the near future, such as in the next month, the adjustment ratio is reduced by a certain amount, such as multiplied by 0.8. If there is an inflection point of falling to rising, the adjustment ratio is increased by a certain amount, such as multiplied by 1.2.

[0034] According to the fluctuation amplitude and the inflection point position, the reporting frequency adjustment parameter is determined, the largest fluctuation amplitude is found first, a basic adjustment interval is set, such as 3 days, and the original reporting frequency is adjusted, if the average fluctuation amplitude is large, it means that the sales change is unstable, and more frequent reporting is needed, so the reporting interval is shortened, the current average fluctuation amplitude is divided by the maximum fluctuation amplitude, and then multiplied by the basic adjustment interval, to get the shortened days, and the original reporting interval is reduced by this day, which is the new reporting interval, for example, the original 7-day reporting is once, the maximum fluctuation amplitude is 100, and the current average fluctuation amplitude is 50, so the shortened day is (50 ÷ 100) x 3, that is, 1.5 days, and the new interval is about 5 days or 6 days, if the average fluctuation amplitude is small, the sales are stable, and the reporting interval is extended, the calculation method is similar, but the basic adjustment interval can be smaller, such as 2 days, if there is an inflection point recently, and the trend changes obviously after the inflection point, the change range of the adjustment interval is enlarged a little more, such as multiplied by 1.3, to adapt to the sales change faster.

[0035] When selecting sample points, interval boundaries, volatility and statistical distribution characteristics are combined to avoid the problem of sample not being representative caused by random selection. More sample points are set in places with high volatility to accurately capture situations with dramatic changes in sales. Combining data distribution to select points can cover different distribution intervals, especially when data is skewed, it can supplement sample points in key areas. Not only are points selected within the interval of normal sales fluctuations, but also points are selected in adjacent areas outside the interval to consider extreme cases where sales exceed the normal range. This makes the adjustment parameter more comprehensive and improves the adaptability of the reporting decision. When generating the trajectory, the weighted average value and quadratic smoothing are calculated to filter out random interference in the sales data, avoiding irregular fluctuations in the original data line, and clearly presenting the real sales trend. The sample points are sorted by time and connected to form a trajectory, which intuitively shows the change process of sales over time and clearly shows the relationship between sales at different time points.

[0036] In a preferred embodiment of the present application, the above step 4 inputs the dynamic adjustment parameter into the adaptive sales prediction model to real-time correct the model parameters, and dynamically adjusts the importance of each external factor according to real-time weather, holidays and promotion events to generate sales prediction results, which can include: In the embodiment of the present application, in step 440, the dynamic adjustment parameter is input into the adaptive sales prediction model trained based on the past sales records, an error feedback mechanism is used to correct the basic sales parameter, the trend parameter and the seasonal parameter in the prediction model in real time to obtain the prediction model after updating the parameters, and specifically, in the dynamic adjustment parameter input link, first, the past sales records are collected, the records contain sales data in different time periods, such as daily, weekly and monthly, and sales-related influencing factor data, such as weather data, holiday arrangements, promotion activity information and the like, are also collected, based on the past sales records, the data is preprocessed by statistical analysis methods, such as calculating the mean, variance and the like, to remove abnormal values, such as sales values far beyond the normal range due to data entry errors, then the preprocessed past sales records are used as training set data, the training set data is input into the initial model in time sequence, the error between the predicted sales output by the initial model and the actual sales in each time period, such as 1 month, is calculated, if the error exceeds the preset threshold, such as 10%, the initial model parameters are adjusted, for example, when the predicted sales are continuously lower than the actual sales, if it is determined that the trend parameter is too low, the trend parameter is increased by 0.5%-1% on the basis of the original trend parameter, the specific amplitude is adjusted according to the error size, the larger the error, the larger the amplitude; if it is determined that the seasonal parameter is deviated, such as the actual summer sales proportion is 5% higher than the predicted value, the summer seasonal parameter is increased by 3%-4%, each time the parameter is adjusted, the training set data in the next time period is used to recheck the error, the iteration process is repeated until the errors of 3-5 consecutive time periods are stabilized within the preset threshold, such as 5%-8%, finally the trained prediction model is verified by the verification set data, if the verification set error is also within the threshold, the prediction model training is completed; if the verification set error exceeds the threshold, the training set needs to be returned to adjust the parameters and iterate until the prediction model reaches a stable error level on both the training set and the verification set, and the adaptive sales prediction model trained based on the past sales records is obtained.

[0037] First, determine the error calculation standard of the predicted value of the prediction model and the actual sales value. Usually, a certain period of time such as a day or a week is taken as a calculation period. In each calculation period, the predicted sales value corresponding to the basic sales parameter output by the adaptive sales prediction model in the period, the predicted sales trend value corresponding to the trend parameter, and the predicted sales seasonal fluctuation value corresponding to the seasonal parameter are obtained. At the same time, the actual sales data in the period is collected. The basic sales parameter refers to the normal sales benchmark value of the product without the influence of special external factors. The error between the predicted sales value of the basic sales parameter and the actual sales value in the period is calculated. If the error is within the allowed range, such as the error rate is less than 5%, the basic sales parameter remains unchanged. If the error exceeds the allowed range, the error reason needs to be analyzed. If it is caused by short-term market fluctuations not covered in the data, the basic sales parameter is adjusted in proportion according to the difference between the actual sales and the predicted sales. For example, the actual sales are 10% higher than the predicted sales, and it is determined that it is caused by short-term market benefits. The basic sales parameter is increased by 8%. After multiple error comparisons and adjustments, the basic sales parameter is corrected in real time. The trend parameter reflects the long-term change direction and change amplitude of product sales, such as whether the sales are in an upward, downward, or stable trend, and the speed of upward or downward. First, according to the predicted sales trend value corresponding to the trend parameter in the period, the cumulative predicted sales under the trend is calculated. Then, it is compared with the actual cumulative sales in the period and the previous continuous multiple periods, such as 4 continuous periods, which can be adjusted according to the sales cycle characteristics of the product. The trend prediction error is calculated. If the error is small, such as the trend deviation rate is less than 3%, the trend parameter does not need to be adjusted. If the error is large, analyze whether the trend direction is wrong or the change amplitude is not accurate. If the trend direction is wrong, such as predicted to rise but actually fell, first reverse the trend direction, and then adjust the change amplitude value of the trend parameter according to the difference between the actual sales change amplitude and the predicted change amplitude. For example, the actual monthly sales decreased by 3% while the predicted monthly sales increased by 5%. First, adjust the trend direction to downward. Then, according to the total difference of 8% between the actual downward amplitude and the predicted upward amplitude, adjust the change amplitude of the trend parameter to 4% downward per month. The seasonal parameter reflects the law of product sales changing with seasons, such as the sales of some products are high in summer and low in winter. At least 3 complete sales years of seasonal sales data are collected, the average sales of each period in each season is calculated as the standard seasonal sales reference value of the season, the seasonal parameter prediction sales seasonal fluctuation value of the season in which the current period is located is compared with the standard seasonal sales reference value of the season and the actual sales of the current period, the seasonal prediction error is calculated, if the error is within a reasonable range, such as the error rate is less than 4%, the seasonal parameter remains unchanged; if the error is out of range, the fluctuation coefficient of the corresponding season in the seasonal parameter is adjusted according to the actual sales fluctuation of different seasons, for example, the average sales of a product in summer in previous years is 20% higher than that in spring, the actual sales of the first month of summer this year is 28% higher than the average sales in spring, then the seasonal fluctuation coefficient of summer is adjusted from 20% to 26%, and the real-time correction of the seasonal parameter is completed in this way, and finally the prediction model after updating the parameter is obtained.

[0038] Step 441, based on the prediction model after updating the parameter, an external influence factor evaluation system is established, the degree coefficients of the influence of real-time weather conditions, holiday types and promotion activity intensity on sales are determined by calculating the correlation between each external factor and sales data, and the importance weight of each external factor in the prediction model is dynamically adjusted according to the degree coefficients, which specifically includes: based on the prediction model after updating the parameter, the specific categories of external influence factors are determined, i.e. real-time weather conditions, holiday types and promotion activity intensity, specific evaluation indexes are set for each category of external factors, such as different level indexes of real-time weather conditions, such as sunny, cloudy, rainy and snowy days; holiday types can be set as legal holidays, short holidays, ordinary holidays and weekends; promotion activity intensity can be set as large-scale promotion, medium-scale promotion and small-scale promotion. At the same time, the sales influence range reference value corresponding to each evaluation index is determined, which is based on data statistics, such as data showing that the sales of a product in rainy days is 15% lower than that in sunny days, then the influence range reference value of rainy days on sales is set to about-15%. Select at least 12 months of data, the weather conditions and the day of the day of the sale data one by one, for each weather grade, the average sales of the weather grade appears in the number of days, and then calculate the overall average of the average sales of all weather grades, then, the average sales of each weather grade and the difference between the overall average, divided by the overall average, the degree of association of the weather grade and the sales data, for example, in the past 12 months, sunny days 180 days, the average daily sales of 100 pieces; all weather grade average daily sales of the overall average of 95 pieces, then the sunny day and the degree of association of the sales data is (100-95) / 95≈5.26%, indicates that the sunny day sales compared to the average level of about 5.26%; Collect the sales data during the holiday and non-holiday in the past 3 years, classify the holiday according to the set type index, for each holiday type, calculate the average daily sales during the holiday of this type, at the same time, calculate the average daily sales during the non-holiday in the same period, use the average daily sales during each holiday type and the difference between the average daily sales during the non-holiday in the same period, divided by the average daily sales during the non-holiday in the same period, get the degree of association of the holiday type and the sales data, for example, in the past 3 years, during the Spring Festival, a total of 15 days, the average daily sales of 300 pieces; the average daily sales during the non-holiday in the same period is 120 pieces, then the degree of association of the Spring Festival holiday type and the sales data is (300-120) / 120=150%, it is explained that the sales during the Spring Festival is 150% higher than that during the non-holiday. The sales data during the promotion activities and non-promotion periods in the past 2 years are sorted according to the set promotion activity intensity indicators. For each promotion activity intensity, the average daily sales during the intensity promotion activity and the average daily sales during the same period non-promotion period are calculated. The difference between the average daily sales during each promotion activity intensity and the average daily sales during the same period non-promotion period is divided by the average daily sales during the same period non-promotion period to obtain the correlation degree of the promotion activity intensity and the sales data. For example, during the large-scale promotion activity in the past 2 years, the average daily sales of 20 days is 250 pieces; the average daily sales during the same period non-promotion period is 100 pieces, and the correlation degree of the large-scale promotion activity intensity and the sales data is (250-100) / 100=150%. According to the correlation degree of each external factor and the sales data obtained by calculation, combined with the actual situation of each external factor, the influence degree coefficient is determined. The greater the absolute value of the correlation degree, the greater the influence of the external factor on the sales, and the greater the absolute value of the corresponding degree coefficient. At the same time, if the correlation degree is positive, the degree coefficient is positive, and if the correlation degree is negative, the degree coefficient is negative. For example, the current weather is rainy, and the correlation degree of rain and sales data is calculated to be-15%, so the degree coefficient of real-time weather condition is set to-0.15. Currently, it is National Day, and the correlation degree of statutory holiday and sales data is calculated to be 120%, so the degree coefficient of holiday type is set to 1.2. Currently, a medium-scale promotion activity is being carried out, and the correlation degree of medium-scale promotion activity and sales data is calculated to be 80%, so the degree coefficient of promotion activity intensity is set to 0.8. First, set the initial importance weight of the three external factors of real-time weather condition, holiday type, and promotion activity intensity. The initial weight can be set according to the average degree of influence of each factor on sales in the data, such as setting the initial weight to 0.33, and the total is 1. Then, adjust the initial weight according to the degree coefficient of each factor. The adjustment formula can be referred to, the adjusted weight of a factor = the initial weight of the factor x (1 + the absolute value of the degree coefficient of the factor / the sum of the absolute values of all factor degree coefficients). The specific adjustment method needs to be adjusted flexibly according to the actual situation to ensure that the adjusted weight can reasonably reflect the current importance of each factor. For example, the real-time weather condition degree coefficient is -0.15, the holiday type is 1.2, and the promotion activity intensity is 0.8. The sum of the absolute values of all factor degree coefficients is 0.15+1.2+0.8=2.15. Then, the adjusted weight of the real-time weather condition = 0.33x(1+(-0.15) / 2.15)≈0.33x(1-0.0697)≈0.33x0.9303≈0.307; the adjusted weight of the holiday type = 0.33x(1+1.2 / 2.15)≈0.33x(1+0.5581)≈0.33x1.5581≈0.514; the adjusted weight of the promotion activity intensity = 0.33x(1+0.8 / 2.15)≈0.33x(1+0.3721)≈0.33x1.3721≈0.453. After adjustment, the weights need to be normalized so that the sum of the weights of the three is still 1. The total of the above calculation results is about 0.307+0.514+0.453=1.274. After normalization, the weight of the real-time weather condition ≈0.307 / 1.274≈0.241, the weight of the holiday type ≈0.514 / 1.274≈0.403, and the weight of the promotion activity intensity ≈0.453 / 1.274≈0.356. Through such calculation, the dynamic adjustment of the importance weight of each external factor in the prediction model is completed.

[0039] Step 442, input the weight-adjusted external factors into the parameter-updated prediction model, generate the sales prediction result of the future time period through the multidimensional calculation processing of the prediction model, which specifically includes: arranging the real-time weather condition, holiday type, and promotion activity intensity data of the three external factors after completing the dynamic adjustment of the importance weight in step 441 according to the data format required by the parameter-updated prediction model, ensuring that the data format matches the input interface of the prediction model, and then inputting the arranged external factor data into the parameter-updated prediction model. Prediction model multidimensional calculation processing: The first dimension is a basic sales calculation. The corrected basic sales parameter is called first, and the length of the current time period is combined. For example, if the future sales of 7 days are predicted, the length is 7 days. The basic prediction sales is calculated without considering the influence of external factors, i.e., basic prediction sales = corrected basic sales parameter x length of future time period. The second dimension is a trend influence calculation. The corrected trend parameter is called. According to the change direction and change amplitude of the sales reflected by the trend parameter, the influence amount of the trend factor on the basic prediction sales in the future time period is calculated. If the trend parameter is 5% increase in monthly sales, and the future time period is 1 month, then the trend influence amount = basic prediction sales x 5%. If the trend parameter is 3% decrease in monthly sales, then the trend influence amount = basic prediction sales x (-3%). The basic prediction sales and the trend influence amount are added to obtain the prediction sales considering the trend factor. The third dimension is a seasonal influence calculation. The corrected seasonal parameter is called. According to the season in which the future time period is located, the seasonal fluctuation coefficient corresponding to the season is determined, and the influence amount of the seasonal factor on the prediction sales considering the trend factor is calculated, i.e., seasonal influence amount = prediction sales considering the trend factor x seasonal fluctuation coefficient. The prediction sales considering the trend factor and the seasonal influence amount are added to obtain the prediction sales considering the trend and seasonal factors. The fourth dimension is an external factor influence calculation. For each input external factor adjusted by the weight, the influence amount of the factor on the sales is calculated. Taking the real-time weather condition as an example, first, the sales influence coefficient corresponding to the current real-time weather condition is determined based on data statistics, such as the sales influence coefficient corresponding to rainy days being -0.1. The influence amount of the real-time weather condition on the sales is calculated based on the importance weight, i.e., real-time weather condition influence amount = prediction sales considering the trend and seasonal factors x real-time weather condition sales influence coefficient x real-time weather condition importance weight. Similarly, the influence amounts of the holiday type and the promotion activity intensity on the sales are calculated, i.e., holiday type influence amount = prediction sales considering the trend and seasonal factors x holiday type sales influence coefficient x holiday type importance weight, and promotion activity intensity influence amount = prediction sales considering the trend and seasonal factors x promotion activity intensity sales influence coefficient x promotion activity intensity importance weight. Then, the influence amounts of the three external factors are added to obtain the total influence amount of the external factors. The predicted sales volume considering the trend and seasonal factors is added to the total influence of external factors to obtain a preliminary sales volume prediction result of the future time period, and then the preliminary prediction result is reasonably checked, and the sales data under the influence of similar external factors in the same period are referred to, if the preliminary prediction result is within a reasonable range, such as the sales difference rate of similar situations in the same period is less than 10%, the preliminary prediction result is determined as the final sales volume prediction result of the future time period; if it is out of the reasonable range, it is necessary to recheck the calculation process of each dimension and correct the possible calculation errors, such as rechecking the parameter values, influence coefficients, weights and the like, until a reasonable sales volume prediction result is obtained.

[0040] The real-time correction of the basic sales volume parameters, the trend parameters and the seasonal parameters through the error feedback mechanism can timely find the deviation of the parameters from the actual sales data, and adjust according to the deviation, avoid the problem of inaccurate prediction when facing market changes due to fixed parameters, make the parameters always have high fitting degree with the current market actual situation, and the input and real-time correction mechanism of dynamic adjustment of parameters can quickly respond to short-term fluctuations, long-term trend changes and seasonal regulation adjustment in the market. In a preferred embodiment of the present application, step 5, integrating the sales volume prediction result with the real-time inventory data of the store and the supply chain cycle parameters to generate a goods ordering instruction, can include: In the embodiment of the present application, step 550, based on the sales volume prediction result, the real-time inventory data of the store, the in-transit goods information and the preset safety inventory threshold are obtained to generate inventory prediction comprehensive analysis data, specifically including: extracting the predicted sales quantity of each commodity in the future sales cycle from the sales volume prediction result, classifying and counting according to the commodity category and SKU (inventory unit), obtaining the real-time inventory data of the store through the inventory management system interface, including the current actual inventory quantity, inventory location and inventory state of each commodity, screening out the normal saleable inventory quantity, calling the in-transit goods information, including the quantity of each commodity that has been sent out but has not yet arrived at the store, the expected arrival time and the transportation state, sorting according to the expected arrival time, counting the total quantity of in-transit goods that can be reached in the future sales cycle, extracting the preset safety inventory threshold of each commodity in the system, which is set according to the commodity sales fluctuation range, replenishment cycle and emergency degree, such as 3 days of sales for daily stable sales commodity safety inventory and 5 days of sales for seasonal commodity, adding the real-time saleable inventory quantity to the in-transit commodity quantity that can be reached in the future sales cycle to obtain the current disposable inventory total quantity of each commodity, integrating the predicted sales quantity, the current disposable inventory total quantity and the safety inventory threshold of each commodity to form a corresponding relationship table including commodity SKU-predicted sales-disposable inventory-safety inventory.

[0041] Step 551, using inventory forecast comprehensive analysis data, determine the net demand in the future sales cycle through demand calculation, specifically including: for each commodity, calculate the inventory consumption forecast in the future sales cycle, subtract the forecasted sales amount in the cycle from the current total available inventory to obtain the inventory remaining forecast, if the result is negative, it means that the inventory is not enough to meet the forecast demand, compare the inventory remaining forecast with the safety stock threshold, if the inventory remaining forecast is greater than or equal to the safety stock threshold, it means that the current inventory is sufficient, and the net demand is temporarily set to 0; if the inventory remaining forecast is less than the safety stock threshold, the difference (safety stock threshold - inventory remaining forecast) is calculated, and the above difference is taken as the base value, if the base value is positive, the net demand is equal to the base value; if the base value is negative, that is, the inventory remaining forecast is lower than the safety stock and cannot meet the forecast sales, the net demand is the difference between the forecasted sales amount and the current total available inventory, which ensures to meet the sales demand and replenish to the safety stock level, and the net demand of all commodities is summarized to form a net demand list sorted by commodity category, and the commodity with a net demand of 0 is marked as not replenished temporarily.

[0042] Step 552, according to the net demand, combined with the supply chain cycle parameters, including the supplier delivery cycle, logistics transportation time and acceptance and shelving time, calculate and determine the time point and basic order quantity, specifically including: obtaining the corresponding supplier delivery cycle of each commodity from the supplier information library, such as A supplier delivery cycle is 3 days, B supplier is 5 days, obtaining the average logistics transportation time of the corresponding transportation route from the logistics system, such as the transportation time from the warehouse to the store is 1 day, obtaining the acceptance and shelving time from the store operation specification, which is fixed at 0.5 days, adding the supplier delivery cycle, logistics transportation time and acceptance and shelving time to obtain the total replenishment cycle of each commodity from ordering to completing shelving, such as A commodity total replenishment cycle = 3 + 1 + 0.5 = 4.5 days, rounding up to 5 days; according to the start date of the future sales cycle and the total replenishment cycle, if it is planned to ensure the shelving of the commodity on the first day in the future, the ordering time point is the current date plus (total replenishment cycle - 1) days, such as the total replenishment cycle is 5 days, then the ordering needs to be initiated 4 days in advance to ensure the completion of shelving on the 5th day, taking the net demand determined in step 551 as the basis, if the net demand is positive, the basic order quantity is equal to the net demand; if there are multiple suppliers providing the same commodity, the basic order quantity is split according to the proportion of each supplier's quota, such as supplier C supplies 60% and supplier D supplies 40%, if there is a promotion activity in the future sales cycle, increase the promotion increment on the basic order quantity, according to the promotion strength estimation, such as the full price reduction activity increases by 20%, the buy one get one free activity increases by 50%, to obtain the adjusted basic order quantity.

[0043] Step 553, based on the basic order quantity, comprehensively consider the supplier minimum order quantity, distribution batch requirement and economic order quantity principle, generate the final order instruction, specifically including: from the system to obtain the minimum order quantity of each supplier, such as the minimum order quantity of supplier E is 10 pieces, and the minimum order quantity of supplier F is 20 pieces, and the distribution batch requirement, such as it must be ordered in multiples of 5, if the basic order quantity is greater than or equal to the minimum order quantity, the quantity is directly reserved; if it is less than the minimum order quantity, the order quantity is adjusted to the minimum order quantity, such as the basic order quantity is 8 pieces, and the minimum order quantity is 10 pieces, then it is adjusted to 10 pieces, the adjusted quantity is rounded according to the distribution batch, such as the batch requirement is 5 times, 12 pieces are adjusted to 15 pieces, and 17 pieces are adjusted to 20 pieces, according to the data statistics, the unit storage cost of the commodity is, such as 0.5 yuan per piece per day, and the order cost is, such as the fixed procedure fee is 50 yuan, by comparing the total cost (storage cost + order cost) under different order batch, the order batch range with the lowest cost is determined, such as the economic batch is between 20-30 pieces, under the premise of meeting the minimum order quantity and distribution batch, the quantity in the economic order quantity range is preferentially selected, such as the adjusted quantity is 15 pieces, the economic batch is 20-30 pieces, then the final quantity is determined as 20 pieces, the final order quantity of each commodity, the corresponding supplier, the order time point, the required arrival time and the commodity specification are integrated, the order instruction sheet is generated in a unified format, and the inventory prediction analysis data required for approval is attached as a basis.

[0044] By comprehensively collecting and integrating sales forecast, real-time inventory, in-transit goods and safety stock data, the dynamic change of future inventory can be clearly mastered, the possible shortage situation can be found in advance, and the net demand can be accurately calculated, which not only ensures that there is enough goods to meet the demand in the future sales cycle, but also maintains reasonable safety stock, avoids inventory accumulation and capital occupation caused by excessive ordering, and also prevents loss caused by insufficient ordering. Combined with the supply chain cycle parameters to determine the order time point and the basic order quantity, the goods can be ensured to arrive in time when needed, avoiding the accumulation of inventory caused by ordering too early, and preventing the shortage situation caused by ordering too late. Considering the requirements of suppliers and logistics and the economic order quantity principle, the final order instruction generated not only meets the actual operation conditions, but also reduces the procurement and inventory costs.

[0045] In a preferred embodiment of the present application, step 6, according to the order instruction, the order task is automatically transferred between the application, review and distribution tracking links through the process engine, and the state is synchronized to each terminal in real time. In the execution process, the deviation between the actual sales data and the prediction result is compared in real time, if the deviation is greater than the preset threshold, a deviation analysis report is automatically triggered, which can include: In an embodiment of the present invention, step 660, based on the delivery instruction, the process engine analyzes the instruction content, automatically creates a delivery task including product details, delivery quantity and required arrival time, and pushes the delivery task to the application processing link. Specifically, the process engine reads key information in the delivery instruction, including product code, product name, delivery quantity and required arrival time of each product, verifies the product details in the delivery instruction, and checks whether the product code exists in the product database registered in the system. If so, extracts basic information such as specifications and units of the product; if not, marks it as an abnormal product and prompts for additional information, summarizes the detailed information, delivery quantity and required arrival time of all valid products, and automatically generates a delivery task number using the rule of year + month + date + serial number, such as 20250826001, where the serial number starts from 001 and increases in the order of daily delivery tasks; stores the generated delivery task information in the task management database, and sends the task to the work queue of the application processing link through the system push mechanism.

[0046] Step 661: After the application processing phase is completed, the delivery task status is automatically updated to pending review. Based on the preset review rules and authority settings, the delivery task is assigned to the corresponding reviewer terminal. Specifically, after the operator in the application processing phase completes the preliminary information confirmation of the delivery task, including verifying whether the product is a commonly used category in the store and whether the reported quantity meets the regular replenishment range, and submits it, the system automatically updates the delivery task status from pending to pending review. The system calls a preset review rule library, which contains review authority levels corresponding to different product categories. For example, low-value daily necessities are reviewed by regional managers, while high-value electrical appliances are reviewed by headquarters purchasing managers. The system matches the product category in the delivery task with the review authority level, determines the corresponding reviewer range, and, based on the reviewer's current task load, counts the number of uncompleted review tasks for each reviewer. Prioritizes tasks assigned to those with the least number of tasks, selects a specific reviewer from the determined reviewer range, pushes the delivery task to the to-do list on their terminal, and sends a reminder notification.

[0047] Step 662, after the review link is passed, the delivery task status is automatically updated to reviewed, and a delivery task order is generated at the same time, and the task is transferred to the delivery tracking link, specifically including: after the reviewer completes the review operation at the terminal and clicks pass, the system automatically updates the delivery task status from pending review to reviewed. Based on the reviewed delivery task information, the system automatically generates a delivery task order, which includes the receiving store address, contact person, contact number, product details and corresponding delivery quantity, required arrival time and other information. According to the distribution area division rules, the delivery task order is assigned to the corresponding distribution center according to the administrative area or logistics distribution zone to which the store address belongs, and the delivery task is transferred to the delivery tracking link.

[0048] Step 663, in the distribution tracking link, real-time collection of goods out-of-warehouse status, transportation trajectory and estimated arrival time, through the state monitoring mechanism to update the goods reporting task progress, including: the distribution center operator when the goods are out of warehouse, through the code gun to scan the bar code on the commodity packaging, the system real-time collection of the number of goods out-of-warehouse, out-of-warehouse time, and update the status of the corresponding goods in the goods reporting task to out-of-warehouse, the GPS positioning device installed on the transport vehicle uploads the position information to the system every 5 minutes, the system marks these position information on the electronic map, forming the transportation trajectory, combined with the transportation trajectory, the current road section traffic time and the remaining distance, the estimated arrival time is calculated, if the delay occurs during transportation, such as the estimated arrival time exceeds the required arrival time, the system automatically marks the goods reporting task progress as delayed, and records the delay reasons, such as traffic congestion, vehicle failure, etc.

[0049] Step 664, real-time synchronization of the processing status and progress data of each link to the store management terminal, supplier collaboration platform and headquarters monitoring terminal, including: the system triggers the data synchronization mechanism at each state change node of the goods reporting task, such as from pending to pending, from audited to distribution, extracts the state information of the node, including state name, change time, operator and progress data, such as the percentage of completed links to total links, encapsulates the data in the preset data format, such as JSON format, and sends the encapsulated data to the store management terminal, the supplier collaboration platform and the headquarters monitoring terminal respectively, displays the goods reporting progress of the store, the distribution of the goods to be supplied, and the overall goods reporting progress of all stores, ensures that each terminal receives the updated data within 10 seconds.

[0050] Step 665, in the process of goods reporting task execution, real-time acquisition of actual sales data of the store through the data acquisition interface, comparison and analysis of actual sales data and sales forecast results, calculation of sales deviation data, including: the system acquires actual sales data from the POS system of the store through the data acquisition interface every hour, including the sales quantity, sales amount, sales time of each commodity, extracts the sales forecast results of the goods reporting task corresponding commodities in the same time period, for each commodity, calculates the difference between actual sales quantity and predicted sales quantity (actual sales quantity-predicted sales quantity), and then divides the difference by the predicted sales quantity to obtain the sales deviation rate (deviation rate=(actual sales quantity-predicted sales quantity) / predicted sales quantity x 100%), and the sales deviation rate of all commodities is summarized to form a sales deviation data set corresponding to the goods reporting task.

[0051] Step 666, based on the sales deviation data, through the preset deviation threshold judgment mechanism, when the deviation value is greater than the set threshold, an early warning signal including deviation details and impact assessment is automatically generated. Specifically, the system calls the preset deviation threshold judgment mechanism, which sets corresponding deviation thresholds for different product categories, such as a deviation threshold of ±15% for fast-moving consumer goods and a deviation threshold of ±20% for durable goods. The system compares the sales deviation rate of each product calculated in step 665 with the corresponding deviation threshold of the product category. If the absolute value of the sales deviation rate of any product is greater than its corresponding deviation threshold, the system automatically determines that a warning is needed and generates a warning signal. The warning signal includes deviation details such as product name, actual sales, predicted sales, deviation rate, and impact assessment such as impact on inventory. If the actual sales are much higher than the prediction, it may lead to inventory shortage; if the actual sales are much lower than the prediction, it may lead to inventory accumulation.

[0052] Step 667, according to the early warning signal, start the deviation analysis processing flow, through analyzing the sales law, market factors and supply chain conditions, generate the deviation analysis report including inventory adjustment scheme, replenishment plan revision suggestion and prediction parameter adjustment scheme, specifically including: after receiving the early warning signal, start the deviation analysis processing flow, first call the sales data of the product in the past 3 months, analyze its sales law, such as whether there is a weekend sales peak, whether it is affected by holidays, etc., collect the market factor data in the same period, such as competitors' promotion activities, changes in population flow in the region, related policy adjustments, etc., analyze the influence degree of these factors on sales deviation, estimate the influence proportion by comparing the sales difference with and without market factors; check the supply chain conditions, such as whether the suppliers have delayed delivery recently, whether the logistics transportation has appeared abnormally, etc., judge whether it leads to sales deviation, and generate the inventory adjustment scheme according to the comprehensive analysis results, such as for the product with sales much higher than the prediction, suggest to increase inventory immediately; for the product with sales much lower than the prediction, suggest to suspend subsequent replenishment, replenishment plan revision suggestion such as adjusting the replenishment cycle, changing the replenishment quantity and prediction parameter adjustment scheme such as adjusting the weight value of holiday factor in the sales prediction model, finally form a complete deviation analysis report.

[0053] Through the flow engine, the report task automatically flows through each link, reducing manual intervention, avoiding the delay and errors of traditional manual document transfer, shortening the processing time of the report task from application to distribution, improving the response speed of the overall supply chain, and synchronizing the processing status and progress data of each link to each terminal in real time, so that the store, supplier and headquarters can timely grasp the report dynamic, facilitate all parties to prepare in advance, and improve the collaboration efficiency. By comparing the actual sales data with the prediction results in real time, the deviation exceeding the threshold can be quickly identified, the early warning can be triggered in time, and the inventory accumulation or shortage problem caused by the sales anomaly not being found can be avoided, reducing the operating risk.

[0054] As Figure 2 shown, the embodiment of the present application also provides a retail store reporting data processing system based on a sales prediction model, comprising: A data synchronization module is configured to synchronize store sales data, real-time weather data, holiday information, promotional event information and store real-time transaction data through a real-time data acquisition interface, and construct a multi-dimensional dynamic data set; An identification generation module is configured to identify data distribution characteristics based on the multi-dimensional dynamic data set, establish an analysis benchmark, obtain two groups of data sequences representing different sales influence dimensions, calculate the correlation between the data sequences, and generate a sales change characteristic interval; An analysis generation module is configured to select a number of evaluation sample points within and outside the sales change characteristic interval, generate a change trajectory according to the evolution correlation of the sample points in the time dimension, and generate a dynamic adjustment parameter according to the trajectory form; A prediction correction module is configured to input the dynamic adjustment parameter into the adaptive sales prediction model, correct the model parameters in real time, and dynamically adjust the importance of each external factor according to real-time weather, holidays and promotional events, and generate a sales prediction result; An instruction generation module is configured to integrate the sales prediction result with store real-time inventory data and supply chain cycle parameters to generate a reporting instruction; A process warning module is configured to automatically transfer the reporting task among the application, review and distribution tracking links through a process engine according to the reporting instruction, and synchronizes the status to each terminal in real time; during the execution process, the actual sales data is compared with the deviation of the prediction result in real time, and if the deviation is greater than a preset threshold, a deviation analysis report is automatically triggered.

[0055] It should be noted that the system is a system corresponding to the above method, and all implementation manners in the above method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0056] The embodiment of the present application also provides a computing device, comprising a processor and a memory storing a computer program, wherein the computer program is executed by the processor to perform the method as described above. All implementation manners in the above method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0057] The embodiment of the present application also provides a computer readable storage medium storing instructions, which, when executed on a computer, cause the computer to perform the method as described above. All implementation manners in the above method embodiment are applicable to this embodiment and can achieve the same technical effects.

[0058] The above is the preferred embodiment of the present application, it should be noted that for those skilled in the art, without departing from the principles described in the present application, can also be made several improvements and refinements, these improvements and refinements should also be considered the scope of protection of the present application.

Claims

1. A method for processing retail store delivery data based on a sales forecast model, characterized in that: The method comprises: Step 1: Synchronize store sales data, real-time weather data, holiday information, promotional activity information, and real-time store transaction data through the real-time data collection interface to build a multi-dimensional dynamic data set; Step 2: Based on the multi-dimensional dynamic data set, identify data distribution characteristics and establish an analysis benchmark to obtain two sets of data series representing different sales impact dimensions. Calculate the correlation between the data series and generate sales change characteristic intervals. Step 3: Select several evaluation sample points within and outside the sales volume change characteristic interval, generate a change trajectory based on the evolution correlation of the sample points in the time dimension, and generate dynamic adjustment parameters based on the trajectory shape; Step 4: Input the dynamically adjusted parameters into the adaptive sales forecast model, perform real-time corrections on the model parameters, and dynamically adjust the importance of each external factor based on real-time weather, holidays, and promotional events to generate sales forecast results. Step 5: Integrate the sales forecast results with the store's real-time inventory data and supply chain cycle parameters to generate a delivery order; In step 6, based on the delivery instructions, the process engine drives the delivery task to automatically flow through the application, review and delivery tracking links, and synchronizes the status to each terminal in real time. During the execution process, the deviation between the actual sales data and the forecast results is compared in real time. If the deviation is greater than the preset threshold, an early warning is automatically triggered to generate a deviation analysis report.

2. The method for processing retail store delivery data based on a sales forecast model according to claim 1, characterized in that: Based on a multi-dimensional dynamic data set, we identify data distribution characteristics and establish an analysis benchmark to obtain two sets of data series representing different sales impact dimensions. We then calculate the correlation between the data series and generate sales change characteristic intervals, including: Identify data distribution characteristics of multi-dimensional dynamic data sets and establish data analysis benchmarks by analyzing statistical characteristics; Based on the data analysis benchmark, the multi-dimensional dynamic data set is subjected to dimensional deconstruction and sequence generation processing, resulting in two sets of data sequences representing the impact of internal operational factors and external environmental factors respectively; Using two sets of data series, the dynamic correlation between them is calculated to obtain the analysis results of the dynamic correlation; Based on the analysis results of dynamic correlation, a sales volume change characteristic interval is generated to describe the expected fluctuation range of sales volume under the influence of complex factors.

3. The method for processing retail store report data based on the sales forecast model according to claim 2, characterized in that: Select several evaluation sample points within and outside the sales volume change characteristic interval. Generate a change trajectory based on the evolutionary correlation of the sample points in the time dimension, and generate dynamic adjustment parameters based on the trajectory shape, including: Based on the boundaries of the sales volume change characteristic interval, combined with the volatility and statistical distribution characteristics of the sales data, multiple representative evaluation sample points are selected within and outside the interval; The evaluation sample points are sorted by their corresponding timestamps, and adjacent sample points are connected in chronological order. The weighted average between adjacent points is calculated and smoothed to form a change trajectory to represent the evolution process and correlation of sales volume in the time dimension. Analyze the morphological characteristics of the changing trajectory, identify the trajectory inflection points, trend direction and fluctuation amplitude, and generate adjustment parameters for dynamically adjusting the reporting quantity and frequency.

4. The method for processing retail store report data based on the sales forecast model according to claim 3 is characterized in that: Dynamically adjust parameters are input into the adaptive sales forecast model, and the model parameters are corrected in real time. The importance of each external factor is dynamically adjusted based on real-time weather, holidays, and promotional events to generate sales forecast results, including: The dynamically adjusted parameters are input into the adaptive sales forecasting model trained based on past sales records. The basic sales parameters, trend parameters, and seasonal parameters in the forecasting model are corrected in real time using an error feedback mechanism to obtain a forecasting model with updated parameters. Based on the updated forecast model, an external influencing factor evaluation system is established. By calculating the correlation between each external factor and sales data, the degree of influence of real-time weather conditions, holiday types, and promotion intensity on sales is determined. The importance weight of each external factor in the forecast model is dynamically adjusted based on the degree of influence. The weighted external factors are input into the forecasting model after parameter update, and the sales forecast results for the future time period are generated through multi-dimensional calculation processing of the forecasting model.

5. The method for processing retail store delivery data based on a sales forecast model according to claim 4, characterized in that: Integrate sales forecast results with real-time store inventory data and supply chain cycle parameters to generate order instructions, including: Based on sales forecast results, obtain real-time store inventory data, goods-in-transit information, and preset safety stock thresholds to generate comprehensive inventory forecast analysis data; Utilize inventory forecasting to analyze data and determine the net demand in the future sales cycle through demand calculation; Based on the net demand, combined with supply chain cycle parameters, including supplier delivery cycle, logistics transportation time, and acceptance and shelf time, calculate and determine the delivery time point and basic delivery quantity; Based on the basic reported quantity, the final report instruction is generated by comprehensively considering the supplier's minimum order quantity, delivery batch requirements and economic order quantity principle.

6. The method for processing retail store report data based on the sales forecast model according to claim 5, characterized in that: Based on the delivery instructions, the process engine drives the delivery task to automatically flow through the application, review and delivery tracking links, and synchronizes the status to each terminal in real time, including: Based on the delivery order, the process engine analyzes the order content and automatically creates a delivery task including product details, delivery quantity, and required delivery time, and pushes the delivery task to the application processing link; After the application processing is completed, the status of the delivery task will be automatically updated to pending review, and the delivery task will be assigned to the corresponding reviewer terminal according to the preset review rules and authority settings; After the review phase is passed, the delivery task status will be automatically updated to reviewed, a delivery task order will be generated, and the task will be transferred to the delivery tracking phase; In the delivery tracking phase, the company collects the goods’ outbound status, transportation trajectory, and estimated arrival time in real time, and updates the progress of the delivery task through the status monitoring mechanism; The processing status and progress data of each link are synchronized in real time to the store management terminal, supplier collaboration platform and headquarters monitoring terminal.

7. The method for processing retail store delivery data based on a sales forecast model according to claim 6, characterized in that: During the execution process, the deviation between actual sales data and forecast results is compared in real time. If the deviation is greater than the preset threshold, an early warning is automatically triggered and a deviation analysis report is generated, including: During the execution of the delivery task, the actual store sales data is obtained in real time through the data collection interface, and the actual sales data is compared and analyzed with the sales forecast results to calculate the sales deviation data; Based on sales deviation data, through the preset deviation threshold judgment mechanism, when the monitored deviation value exceeds the set threshold, an early warning signal including deviation details and impact assessment is automatically generated; Based on the early warning signal, the deviation analysis processing process is initiated. By analyzing sales patterns, market factors and supply chain conditions, a deviation analysis report is generated, including inventory adjustment plans, replenishment plan revision suggestions and forecast parameter adjustment plans.

8. A retail store report data processing system based on a sales forecast model, the system implementing the method according to any one of claims 1 to 7, characterized in that: include: The data synchronization module is used to synchronize store sales data, real-time weather data, holiday information, promotional information, and real-time store transaction data through the real-time data collection interface to build a multi-dimensional dynamic data set; The identification and generation module is used to identify data distribution characteristics based on a multi-dimensional dynamic data set to establish an analysis benchmark, thereby obtaining two sets of data series representing different sales impact dimensions, calculating the correlation between the data series, and generating sales change characteristic intervals; The analysis and generation module is used to select several evaluation sample points within and outside the sales volume change characteristic interval, generate change trajectories based on the evolution correlation of the sample points in the time dimension, and generate dynamic adjustment parameters based on the trajectory shape; The forecast correction module is used to input dynamic adjustment parameters into the adaptive sales forecast model, perform real-time correction on the model parameters, and dynamically adjust the importance of various external factors based on real-time weather, holidays, and promotional events to generate sales forecast results; The instruction generation module is used to integrate sales forecast results with real-time store inventory data and supply chain cycle parameters to generate delivery instructions; The process warning module is used to drive the automatic flow of the delivery task through the application, review and delivery tracking links according to the delivery instruction through the process engine, and synchronize the status to each terminal in real time; During the execution process, the deviation between actual sales data and forecast results is compared in real time. If the deviation is greater than the preset threshold, an early warning is automatically triggered to generate a deviation analysis report.

9. A computing device, characterized in that include: one or more processors; A storage device for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a program, which, when executed by a processor, implements the method according to any one of claims 1 to 7.

Images