CN114219169A - Script banner supply chain sales and inventory prediction algorithm model and application system - Google Patents

Abstract

Based on deep investigation and analysis of business requirements and pain points of sales forecast, procurement plan and execution, inventory replenishment and the like of enterprises on supply chain upstream and downstream, combined with years of actual combat experience in the fields of supply chain business management, purchase-sale-inventory management, industrial internet supply chain, electronic commerce and the like, the sales and inventory forecasting technologies such as neural network, weighted average, linear differential equation, solution space and the like are applied, on the basis of theoretical connection reality, a safety inventory forecasting model based on stock commodity SKU is developed, a sales scene dynamic self-adaption sales forecasting model and a time series regression sales and inventory forecasting algorithm model are developed, a corresponding gluozhu sales and inventory forecasting management application system is developed, and multiple rounds of operation, parameter optimization and result verification of actual production data are carried out on the algorithm model and the application system, the method plays an obvious promoting role in improving the efficiency and optimizing the benefit of a supply chain operation system of related industries.

Description

Script banner supply chain sales and inventory prediction algorithm model and application system

Technical Field

The invention relates to a supply chain sales and inventory prediction algorithm model and an application system, which apply neural network, weighted average, linear differential equation, solution space and other sales prediction technologies, and make corrections according to interference factors from a sales prediction basic stage, namely manual semi-automatic sales prediction and safe inventory prediction based on inventory commodity SKU to obtain purchase, sales and inventory plans of supply chain enterprises. And then, the method is transited to the advanced sales prediction step, so that iterative optimization of comprehensive prediction models such as digital full-automatic sales prediction, dynamic self-adaptive prediction of sales scenes, time series regression prediction and the like is achieved, the problems of randomness and blindness of the sales and inventory management of supply chain enterprises are solved, and the purchase, inventory and sales operation of the supply chain upstream and downstream enterprises are promoted to develop towards the lean direction.

Background

In the field of supply chain business management, inaccurate sales prediction is often encountered, which leads to untimely or beyond market expectation for upstream purchasing, processing and stock replenishment, which leads to insufficient supply, i.e. retail customer demand can not be met, or supply is over-demand, which causes waste of enterprises and supply chain as a whole, and finally leads to various defects of reduced customer satisfaction, low supply chain business management efficiency and the like.

Aiming at the demands and pain points of sales prediction, purchase planning and execution, inventory replenishment and the like of enterprises on the upstream and downstream of the supply chain, Shanghai mist flag technology companies develop supply chain sales prediction systems for small and medium-sized enterprises on the upstream and downstream of the supply chain, and accurately predict the production, supply and sales prediction capability and prediction data of each supply chain link from a dynamic view angle in a supply chain ecosystem, so that accurate decision support information is provided for enterprise operation management on each node of a supplier, the small and medium-sized enterprises on the upstream and downstream of the supply chain can better arrange purchasing, production processing and sales behaviors, and the efficiency and the benefit of the whole supply chain system are greatly improved. The method has wide market background and enterprise demand support and strong technical feasibility.

Disclosure of Invention

A supply chain sales and inventory prediction algorithm model and an application system are built through technical means, and the sales prediction technologies such as neural networks, weighted averages, linear differential equations, solution spaces and the like are applied, so that the purchase, sales and inventory plans of supply chain enterprises are obtained from the sales prediction basic stage, namely manual semi-automatic sales prediction, safety inventory prediction based on inventory commodity SKUs and correction according to interference factors. And then, the process is transited to the advanced sales prediction step, so that the iterative optimization of comprehensive prediction models such as digital full-automatic sales prediction, dynamic self-adaptive prediction of sales scenes, time series regression prediction and the like is achieved. Through calculation and correction of a mathematical model, sales and inventory data in a supplier flow are accurately predicted, the problems that operation decisions lack data support, subjective randomness and the like in enterprise supply chain management work are solved, and the method is the key value of the method.

The purpose of the invention is realized by four technical schemes:

1. developing a safety inventory prediction model based on inventory commodity SKUs;

2. developing a dynamic self-adaptive sales forecasting model of a sales scene;

3. developing a time series regression sales and inventory prediction model;

4. and creating a supply chain sales and inventory management plan according to the prediction result.

1. Developing a stock keeping model based on stock keeping commodity SKU

The safety stock prediction model based on the stock commodity SKU is a safety stock prediction model obtained according to the evolution of a supply chain demand prediction model, so that the defect of static, independent and same-distribution demand prediction adopted by most safety stock prediction models is overcome. According to the analysis of the actual operation condition of the script banner supply chain service application system, the predicted value of the safety stock can be greatly changed under the conditions of different industries, different commodities, short and prosperous sale seasons and the like.

The supply chain involves many interested parties, and the interested parties pull one another to move the whole body. The safety stock therein is an important node. The safety stock precision in the supply chain system is guaranteed, and the method has important significance and influence on efficient and economic operation of the supply chain and higher supply chain participant customer satisfaction.

Current widespread safety inventory forecasting is based on statistical principles of safety inventory determination and safety inventory determination under bulk orders. Mainly solves the problem of the safety stock cost under the condition of mass production.

Conventional inventory forecasting methods typically estimate forecasts from historical inventory data for a good, assuming a distribution of demand for the good. A parametric model for inventory forecasting based on a single forecasting time series and a single metric forecasting error is used as a basic inventory forecasting framework.

In reality, there are a plurality of factors affecting inventory prediction in a real-world business scene, and the factors are not limited to inventory historical data, such as macroscopic economic trends, epidemic situation influence, commodity price indexes, supply chain topology layout, seasonal factors and the like.

The invention adopts two safety stock prediction models, namely a linear regression prediction model and a target library function prediction model, to deal with the safety stock prediction under different service scenes and demand conditions.

And predicting the safety stock by using the linear regression model, and setting a safety stock value by using the basic data of the linear regression model and adding estimation error correction.

And predicting the safety stock by using the target library function, and determining safety stock data by using a library function extrapolation variable least square calculation value under different target and service level constraints.

The influencing factors of the accuracy of the safety stock management prediction model used by the invention comprise:

<1> correlation of non-linear performance function level;

<2> predicted safety inventory control variables for each supply chain node;

<3> interactions and dependencies between different supply chain nodes.

The method takes the performance evaluation function and the solving control variable as two basic points to derive a safety stock prediction calculation framework.

And (3) performing performance function evaluation by using off-line calculation of a discrete event simulation model, and establishing a safety inventory prediction management model taking linear programming as a characteristic.

Safety stock service levels at the production and logistics stages are important influencing factors and decision variables for safety stock forecasting. The interdependence between the non-linear performance function of the safety inventory forecast and the independent variables is fully embodied in a safety inventory forecast model featuring linear programming.

During the operation of the safety inventory prediction model, the invention particularly pays attention to the demand variance estimation and the prediction error analysis. The supply period demand variance predicted by the safety inventory can be obtained by calculating the functional relation between the inventory commodity demand variance and the supply period in each supply chain period.

According to the invention, through analysis of model data calculation results, a prediction algorithm combining moving average and exponential smoothing in a safety stock prediction model is found, the correction precision of a demand variance expression under a batch ordering strategy can be obviously improved, and the operation efficiency of a batch ordering supply chain business process is consolidated.

2. Research and development of dynamic self-adaptive sales prediction model of sales scene

The sales scene dynamic self-adaptive sales prediction model can dynamically and automatically determine to adopt different sales prediction models according to different sales scenes.

In the actual supply chain business, many sales scenes exist, such as supply ratio commodity display sales prediction, micro-merchant marketing influence path prediction, correlation analysis commodity promotion sales prediction, sub-customer group sales prediction and the like.

According to the method, through the analysis of the sales scenes and the research of the sales prediction models, the adaptability of different sales prediction models to different sales scenes is obtained.

The association rule prediction model is suitable for the forecast scene of the display and sales of the goods in supply ratio.

And for the forecasting scene of the marketing impact path of the micro-businessman, a social network forecasting model is suitable to be adopted.

And for the correlation analysis commodity sales promotion and sales prediction scene, a classification regression prediction model is suitable for being adopted.

For the sales prediction scene of the sub-client group, a clustering analysis prediction model is suitable.

Different sales scenes correspond to a sales prediction model diagram, as shown in the attached figure 1 of the specification.

The association rule prediction model for supply ratio commodity display sales prediction is an algorithm for mass data mining, and aims to mine frequent items and corresponding association rules from a series of transactions. The association rule prediction model is widely applied to industries such as commercial supermarkets and the like, assists the supermarkets to find some goods shopping behaviors with associations, and can improve the shopping experience of customers and the sales volume of the supermarkets by properly adjusting the goods. Therefore, the association rule prediction model is suitable for the supply ratio commodity display sales prediction scene.

The association rule prediction model includes two subtasks of frequent pattern discovery and association rule generation.

Frequent pattern discovery: also called frequent pattern mining, frequent item mining, etc., refers to selecting frequent parts from a series of candidate items, and the degree of frequent measurement may be the frequency of occurrence of each item, and when a certain threshold is exceeded, the item is frequently tasked.

And (3) generating an association rule: in the most frequent set of items that have been found, association rules are found whose confidence level is not less than the minimum consistency given by the user.

The association rule prediction model inputs a hidden output three-layer architecture diagram, as shown in the figure 2 of the specification.

The social network prediction model for predicting the marketing influence path of the WeChat dealer is used for acquiring attribute data, behavior data and content data of members of a target customer group of the WeChat dealer, and analyzing and deducing the attribute data, the behavior data and the content data through a certain social network behavior algorithm to obtain a logic rule and a data parameter with higher marketing influence path weight.

By collecting customer attribute data in the social network, such as identity, gender, age, academic, occupation, party, community and the like, and customer behavior data, such as web pages browsed by customers, movie and television works watched, mobile APP is used, and customer content data, such as WeChat friend circles published by customers, microblogs, uploaded photos and videos and the like, and then adopting algorithms such as data mining business intelligence and machine self-learning to predict and refine those elements with business promotion values, such as brand preference, purchasing habit, transaction characteristics, commodity attributes, evaluation expressions and word patterns, which are correlated, interacted and influenced with the customer attributes, behaviors and contents, and finally packaging and processing the information, submitting the information to merchants and advertisers engaged in the WeChat business, banks, insurance companies, financial institutions and professional data brokers, thereby obtaining economic value.

Social network behavior algorithms include various types of classifiers, vector machines, and deep learning and clustering classifiers.

The method comprises the steps of comparing and analyzing a social network prediction result of a target client on a WeChat marketing influence path with an actual WeChat marketing effect, finding out an influence factor influencing the difference of hit rates, describing the influence factor into a high-dimensional vector, and repeatedly learning and training by using a set classifier, so that the precision and hit rate of the social network prediction result on the forecasting of the WeChat marketing influence path can be continuously improved. And more commercial values and economic benefits can be obtained through targeted iterative optimization of the model prediction result and the actual marketing operation effect.

The social network prediction model test group prediction fitting graph is shown in the specification and attached to the figure 3.

The classification regression prediction model for correlation analysis of sales prediction of commodity sales promotion is a machine learning task with a supervised learning algorithm as a typical characteristic, and comprises two models of classification prediction and regression prediction.

The classification prediction model divides the commodity samples into categories suitable for promotion according to the correlation characteristics of the promotion commodities, so that a certain commodity promotion function is achieved. The specific operation flow is that training samples related to the sales promotion commodities are utilized to carry out sales promotion prediction training, on the basis, the relational mapping from the characteristics of the sales promotion commodity samples to the sample labels is obtained, then the mapping relation is utilized to calculate the sample label value of the sales promotion commodities, and finally the purpose of classifying the commodity samples to be suitable for the types of the sales promotion commodities is achieved.

The general classification prediction model adopts a binary classification problem mode, namely, the classification of an object or an event is judged through the existing characteristic attributes, and the result is only two results of '0' and '1', and the object or the event belongs to the classification or does not belong to the classification.

The relevance analysis commodity promotion sales classification prediction model used by the invention adopts a multivariate classification problem mode, namely, the promotion categories of related commodities are judged according to the existing characteristic attributes of commodity promotion, the generated classification results have five different categories, namely, the classification results are not very suitable for promotion, are not suitable for promotion, are neutral in promotion, are suitable for promotion and are very suitable for promotion, and the promotion commodity sample label values from 0 to 4 are respectively used for representing the five different commodity promotion categories.

The relevance analysis commodity sales promotion classification prediction mode used by the invention respectively adopts a linear classification model, a decision tree classification model and a naive Bayes classification model.

The regression prediction model and the classification prediction model are machine learning tasks of a supervised learning algorithm. The difference between the two is that in the classification prediction algorithm, the sample labels are discrete values, and each label represents a class. In the regression prediction algorithm, the sample label is some continuous value. The regression prediction algorithm is to obtain the relational mapping from the sample characteristics of the sales promotion commodities to the sample labels through the machine learning of the training samples of the sales promotion commodities.

The common regression prediction model generally calculates the relationship mapping from the sample characteristics to the sample labels according to experimental empirical values, so as to calculate the quantitative result of the regression prediction.

The regression prediction model used by the invention adopts an algorithm for solving a regression equation, predicts the continuous numerical commodity promotion target value through derivation, deduction and calculation of a commodity promotion regression equation, receives a series of continuous relevant commodity promotion sales data, and solves a coefficient value most suitable for the commodity promotion regression equation, thereby predicting the relevant commodity promotion sales result.

The relevance analysis commodity sales promotion regression prediction mode used by the invention respectively adopts a linear regression model, a gradient regression model and a random forest regression model.

Correlation analysis goods promotion sales regression prediction model goods prediction situation graph, as shown in the specification and figure 4.

The clustering analysis model for the sales prediction of the sub-client group is used for determining the affinity and the sparseness of the samples by using Q-type clustering and R-type clustering algorithms according to the sales attributes and the characteristics of the samples of the client group, and objectively typing and classifying the samples according to the affinity and the sparseness of the samples of the client group and the sales prediction result so as to obtain a measurable clustering system suitable for the sales prediction.

The Q-type clustering client group sales prediction algorithm is used for carrying out sales prediction clustering on client group samples, and gathering client group samples with similar sales prediction characteristics together to separate client group samples with large sales prediction differences.

The R-type clustering customer group sales forecasting algorithm is used for clustering the sales forecasting variables in a customer group mode, the sales forecasting variables with large differences are separated, and the sales forecasting variables with customer group similarity are gathered together.

The R-type clustering can select a few variable attributes with customer group representativeness from sales predictive variables to participate in analysis, and achieves the dual purposes of reducing the number of the sales predictive variables and reducing the dimension of the variables.

In an actual sales scenario, different customer groups have different sales prediction reference frames and prediction results. There are different sales strategies, sales processes and sales performances for different types of customer groups. All customer groups have corresponding sales prediction models.

The method clusters the sales prediction modes of all types of customer groups to obtain the large categories of the sales prediction modes of various types of customer groups, then finds out the category which is most likely to be similar to the sales prediction mode of a certain specific customer group by using a clustering analysis judgment method such as an expert clustering analysis method, a hierarchical clustering analysis method and the like, and uses the category as a reference system. Modeling analysis is carried out on the selected mode of the customer group sales prediction reference system, and after the sales prediction rule is found, the sales prediction rule can be applied to the related customer group sales prediction mode.

The method comprehensively utilizes Q-type clustering and R-type clustering algorithms of clustering analysis, takes Euclidean distance (Euclidean square) as a measure standard and minimum centroid distance as a clustering criterion, eliminates the difference between absolute log magnitudes of sales prediction data of all sub client groups, normalizes the sales prediction data into 0-1 interval numerical values, and performs clustering analysis on sales predictions of 9 client group types to obtain a clustering analysis result.

Through multiple rounds of operation, the 9 types of branch client group sales prediction models are clustered into 3 types, and the 3 types of branch client group sales prediction algorithms have good high-cohesion and low-coupling clustering characteristics.

The method is characterized in that a flow chart is applied to an accessory forecast replenishment dispatching business scene by a client group sales forecast clustering analysis model, and is shown in the attached figure 5 in the specification.

Based on 3 sub-client group sales prediction algorithms, modeling analysis is performed by adopting actual sales data of clients, regression fitting is performed on each type of client group according to the continuity and shape characteristic values of the sales prediction data, and residual analysis is performed. Through comparison and analysis of actual sales performance data of an actual client group and a sales prediction algorithm calculation result, the method is suitable for performing sales prediction on different client groups.

3. Developing time series regression sales and inventory forecasting models

According to the time series operation conditions and characteristics related to sales and inventory forecasting work, the time series regression sales and inventory forecasting mode developed by the invention uses long-term trend forecasting, seasonal trend forecasting, popularity trend forecasting and randomness trend forecasting, and has four time series analysis forecasting methods in total.

The sales and inventory forecasting work is divided into long-term trends, seasonal trends, popularity trends and randomness trends according to four time series characteristics for analysis and forecasting, then forecasting results of the factors are integrated, and finally sales and inventory forecasting result values and correction values are obtained.

<1> Long-term Trend prediction

Influenced by factors such as customer group characteristics, product attributes, warehouse geographical layout and the like, the commodity sales and inventory data show relatively stable and long-term change trends according to time change, and stably and permanently rise and fall according to certain rules and logics.

For example, the application of polyethylene plastic products in agricultural mulching film and easily degradable packaging industries shows stable and long-term commodity consumption trend and the change characteristics of the quantity of stored commodities. By using a moving weighting method, a trend curve fitting method, and an exponential tangent smoothing method for the sales and inventory history data of the above industries, an algorithmic prediction is made of the long-term trends of the sales and inventory changes of goods in the field.

For the long-term trend prediction transfer and transfer description of inventory commodities among the summary central library, the distribution central library and the node library, a long-term trend prediction inventory accessory transfer and transfer flow chart is shown in the attached figure 6 of the description.

<2> seasonal trend prediction

The commodity sales and inventory change of some industries show obvious seasonal trends, are greatly influenced by action factors of seasonal alternation, show strong regular change which is characterized by seasonal periods on time series characteristics, and are called seasonal index factors.

For example, in the garment design, production and sale industry, seasonal trends have a strong influence on the sales and inventory data performance of down jackets, T-shirts and other garment products. Seasonal trend prediction results and optimization conclusions of sales and inventory data of the commodities in the clothing industry can be obtained by using a seasonal index factor usage time series analysis method for off-season peak season sales and inventory accumulation data of the commodities.

Seasonal trend prediction daily weekly monthly prediction and replenishment flow chart, as shown in the attached figure 7 of the specification.

<3> prediction of prevalence trends

The popularity trend prediction is between the deterministic change and the stochastic change predicted by time series analysis, and is characterized by the relative certainty of popularity definition and the relative randomness of popularity occurrence. We define this feature as a prevalence fluctuation parameter.

For example, the use of network popular terms in the commodity advertising industry shows a distinct popularity trend characteristic. The creation and application scenes of popular network terms such as recent hotspots, annual vocabularies, hot search and ranking and the like can promote the updating and the use of the internet advertising language. According to a fluctuation rate equation and a trend fitting algorithm of time series trend analysis, marketing schemes and audience analysis with obvious popularity characteristics can be analyzed and pre-judged with certain accuracy.

The popularity trend prediction demand acquisition flow chart is shown in the specification and attached with figure 8.

<4> stochastic trend prediction

Stochastic trend prediction pertains to time series analysis changes caused by many uncertainty factors, exhibiting stochastic, uncertainty, and perturbative characteristics over the changes in trend. We define this stochastic trend characteristic as a stochastic perturbation factor.

For example, in the process of issuing, underwriting and underwriting of equity investment products of securities investment funds, a certain random trend is presented, which is not completely determined by factors such as macroscopic economic situation, issuer condition, underwriter capability and the like. And (3) establishing a time series analysis prediction model according to the randomness disturbance factor of the time series trend analysis, and performing randomness trend analysis prediction on the issuing, underwriting and subscription processes of the ticket investment fund equity investment products by adopting algorithms such as a high-order regression equation, a least square method, a white noise nonlinear fitting curve and the like.

The random trend prediction demand management flow chart is shown in the figure 9 in the specification.

4. Creating supply chain sales and inventory management plans based on forecasted results

And inputting the prediction results of the sales and inventory according to the safety inventory prediction model, the sales scene dynamic self-adaptive sales prediction model and the time series regression sales and inventory prediction model of the commodity SKU into a supply chain purchasing, sales and inventory application system, and generating a corresponding supply chain sales and inventory management plan based on the sales and inventory analysis prediction model.

The supply chain sales and inventory management plan is divided into three functional modules, sales and inventory forecasting result input, sales and inventory business management flow description and analysis, and sales and inventory management plan output.

<1> sales and inventory forecast result input module

And the sales and inventory prediction result data are combed and regulated, and the requirements of a data array and a normal form of a sales and inventory business management flow description and analysis module are met.

And then preprocessing the sorted and structured sales and inventory forecast data such as linear correlation, key value pair matching, curve fitting and the like.

And finally writing the corresponding database table according to the storage requirements of the relational database and the NoSql database.

Sales and inventory forecast result data combing and cleaning method step diagram, as shown in the attached figure 10 of the specification.

<2> sales and inventory business management flow description and analysis module

And inputting the data after the sales and inventory prediction result input module is carded and normalized into a safety inventory prediction model based on stock commodity SKU, a sales scene dynamic self-adaptive sales prediction model and a time series regression sales and inventory prediction model, calculating and adjusting and optimizing through an algorithm model to obtain the process description and stage analysis results of the sales and inventory business management process, and laying and preparing in the aspects of functions and data for finally determining the sales and inventory management plan.

A sales and inventory business management scenario description and flow analysis diagram, as described in the specification with reference to fig. 11.

<3> sales and inventory management plan output module

On the basis of the process description and stage analysis result calculated by the sales and inventory business management flow description and analysis module, specific execution plans under different sales and inventory management scenes are output and obtained according to the formulation rule and logic of the sales and inventory management plans, so that the algorithm implementation and application ground of sales and inventory prediction is formed, and finally the theoretical and empirical closed-loop result of the sales and inventory prediction invention patent is achieved.

The sales and inventory management plan outputs a map of the optimal selection method, as described in the specification with reference to fig. 12.

The innovation and the positive progress effects of the invention are represented as follows:

in the actual business operation process of supply chain management of different industries, the sales and inventory expected results of commodities generally lack scientific and effective prediction, and the supply chain management decision of enterprises is mostly based on personal experience and subjective assumption. Thus, commodity flow, fund flow and document flow in and among enterprises on the upstream and downstream of the supply chain are caused, efficient and consistent matching and interaction are difficult to form, the unfavorable situation that supply is over-demand or under-demand occurs in the supply chain, and loss and waste in the aspects of manpower, funds and other resources are generated.

Shanghai Yingzhao flag technology Limited company deeply ploughs in the field of supply chain science and technology finance for many years, and business and technical backbone personnel thereof have clear and profound understanding on pain points and blockage points of supply chain full-flow closed-loop management. A gluoze flag company carries out deep and detailed business analysis and research on enterprise clients upstream and downstream of a supply chain of service, applies a supply chain sales and inventory prediction algorithm model and a computing technology to an actual business scene and a process, determines to adopt a safe inventory prediction model based on stock commodity SKU, a sales scene dynamic self-adaptive sales prediction model and a time series regression sales and inventory prediction model as three basic algorithm models of a sales and inventory prediction application system by combining years of actual combat experiences of the company in the fields of a supply chain business management system, a purchase, sales and inventory application system, an industrial internet supply chain, an electronic commerce platform and the like on the basis of theoretical connection actual, and then supplements 5500 supply chain sales and inventory basic business data of the enterprise upstream and downstream of the supply chain, which are acquired and accumulated by the application systems such as the supply chain management system, the purchase, the sales and inventory management system, the supply chain, the electronic commerce platform and the like which are developed and operated on line by the company, and carrying out multiple rounds of operation, parameter optimization and result verification on actual production data of the sales and inventory prediction algorithm model and the application system.

The sales and inventory forecast algorithm and various business and technical indexes of the application system can reach and meet the target and condition of supply chain sales and inventory forecast management of enterprises on the upstream and downstream of the supply chain.

According to the method, members in a vast supply chain ecosystem are cultured, and the prediction capability and prediction data of production, supply and sales of each supply chain link are accurately predicted from the dynamic view of a supply chain operation system, so that accurate decision support information is provided for enterprise operation management of each node of a supplier, and small and medium enterprises on the upstream and downstream of the supply chain can better arrange purchasing, production processing and sales behaviors, thereby driving the efficiency and the benefit of the whole supply chain system to be greatly improved, and establishing good industry demonstration and typical cases.

Practice proves that the sales and inventory prediction algorithm and the application system developed by Shanghai imperial banner technology Limited have wide market development prospect and customer demand background, and have firm and steady product and technical feasibility.

While specific embodiments of the present invention have been described above, it will be appreciated by those skilled in the art that these are merely illustrative of a script banner marketing and inventory forecasting algorithm and application, and that the scope of the present invention is defined by the appended claims. Various changes and modifications to these embodiments may be made by those skilled in the art, which fall within the scope and spirit of the invention.

The present invention also has certain drawbacks and weaknesses that need to be improved and perfected in future further theoretical studies and practices.

Drawings

FIG. 1 is a diagram of sales forecast models corresponding to different sales scenarios

FIG. 2 is a three-layer architecture diagram of the input hidden output of the association rule prediction model

FIG. 3 is a test set prediction fit graph of a social network prediction model

FIG. 4 is a diagram of the prediction of goods by the regression prediction model for sales promotion of goods by correlation analysis

FIG. 5 is a flowchart of the application of the cluster analysis model for forecasting the sales of the customer groups in the scenario of the parts forecasting replenishment dispatching business

FIG. 6 is a flow chart of long term trend predictive inventory parts allocation and transfer

FIG. 7 is a flow chart of seasonal trend prediction daily weekly monthly forecast and replenishment

FIG. 8 is a flow chart of the popularity trend prediction demand acquisition

FIG. 9 is a flow chart of stochastic trend prediction demand management

FIG. 10 is a diagram of the steps of a sales and inventory forecast result data combing and cleaning method

FIG. 11 is a sales and inventory business management scenario description and flow analysis diagram

FIG. 12 is a diagram of a sales and inventory management plan output optimization selection method.

Claims (8)

1. Claim 1

A linear regression safety stock prediction model based on stock keeping good SKUs comprising the steps of:

firstly, calculating to obtain a safety stock basic data result of a single supply chain node based on stock commodity SKU through a linear regression model;

secondly, solving error correction values of single supply chain nodes by using the safety inventory statistical historical data;

thirdly, comprehensively calculating to obtain a safety stock predicted value of a single supply chain node according to a calculation result value of the linear regression model and an error correction value of the statistical historical data;

fourthly, circularly traversing and calculating the linear regression safety stock prediction value of each supply chain node by using the method from the first step to the third step;

fifthly, the linear regression safety stock prediction calculation values of the supply chain nodes are adjusted and optimized integrally according to linear and nonlinear correlation functions among different nodes of the supply chain.

2. Claim 2

A target library function secure inventory prediction model based on inventory commodity SKUs, comprising the steps of:

firstly, determining a target library function of a single supply chain node under the constraint conditions of different targets, service level agreements and the like;

secondly, calculating safety stock predicted values of different target library functions by using a least square extrapolation method;

thirdly, solving the weighted average value of the function safety stock prediction calculation values of different target libraries of a single supply chain node;

fourthly, circularly traversing and calculating the predicted value of the target library function safety stock of each supply chain node by using the method from the first step to the third step;

fifthly, the target library function safety stock prediction calculation value of the supply chain node is adjusted and optimized integrally according to linear and nonlinear correlation functions among different nodes of the supply chain.

3. Claim 3

An adaptive prediction model for a sales scenario of association rules, comprising the following steps:

firstly, preparing basic data and business rules of a supply ratio commodity display sales forecasting scene;

secondly, executing frequent mode discovery tasks from a group of related sales prediction scenes, and determining frequently-occurring task items;

thirdly, in the determined task item subset with the maximum frequency, searching a minimum consistency association rule with the confidence coefficient being more than or equal to a given threshold value;

fourthly, applying the screened association rule to a goods supply ratio commodity display sales forecasting scene, and calculating to obtain a sales forecasting value;

fifthly, performing exponential iteration and logarithmic tuning on the sales predicted value obtained by the association rule by adopting a functional regression method.

4. Claim 4

A social network sales scenario adaptive prediction model is characterized by comprising the following steps:

firstly, collecting customer attribute data, customer behavior data and customer content data in a forecasting scene of a marketing influence path of a micro dealer;

secondly, predicting and refining attribute element topics which are correlated with the attributes of the customers by adopting a classifier algorithm;

thirdly, predicting and refining a behavior element theme which is interacted with the customer behavior by adopting a vector machine algorithm;

fourthly, predicting and refining content element themes which are correlated with the customer content by adopting a deep learning algorithm;

fifthly, applying the calculated social network attribute, behavior and content element theme to a forecasting scene of the marketing influence path of the WeChat to find out a difference influence factor influencing the forecasting hit rate;

and sixthly, describing the influence factors into high-dimensional vectors, and repeatedly learning and training by using a set integrated classifier to improve the accuracy and hit rate of forecasting the marketing influence path of the micro-businessman.

5. Claim 5

A classification regression sale scene self-adaptive prediction model is characterized by comprising the following steps:

firstly, selecting a correlation analysis commodity sales promotion and sales prediction scene, implanting a classification regression acquisition factor, and collecting basic data of classification and regression prediction;

secondly, carrying out sales promotion prediction training by using training samples related to the sales promotion commodities to obtain the relation mapping from the characteristics of the sales promotion commodities to sample labels;

thirdly, calculating a label value of the sales promotion commodity sample by using the mapping relation, and dividing the commodity sample into classification categories of the sales promotion commodities;

fourthly, solving the coefficient value of the promotion regression equation by adopting a regression equation algorithm according to the five classified commodity sample data which are not suitable for promotion, are neutral in promotion, are suitable for promotion and are very suitable for promotion;

fifthly, according to the coefficient value calculation result of the commodity sample promotion regression equation, the related commodity promotion sales result is predicted and iteratively adjusted.

6. Claim 6

A cluster analysis sales scenario adaptive prediction model is characterized by comprising the following steps:

firstly, determining sales attributes and characteristics of a client group sample of a client group sales prediction scene;

secondly, carrying out sales prediction clustering on the customer group samples by using a Q-type clustering algorithm, clustering the customer group samples with similar sales prediction characteristics together, and separating the customer group samples with large sales prediction differences;

thirdly, performing sub-client cluster clustering on the sales forecast variables by using an R-type clustering algorithm, separating the sales forecast variables with large differences, and clustering the sales forecast variables with client cluster similarity;

fourthly, finding out a class with the largest sales prediction similarity of a certain specific client group as a reference system for the large class of the sales prediction modes of the client groups obtained by the Q-type and R-type clustering algorithms, and carrying out modeling analysis on the prediction of the sub-client groups;

and fifthly, after modeling analysis of the sales prediction of the sub-client group is performed for multiple turns, gradually optimizing and approximating to obtain a high-cohesion and low-coupling sales prediction clustering prediction result.

7. Claim 7

A time series regression sales and inventory forecasting model, comprising the steps of:

firstly, collecting time series operation conditions and characteristic data related to sales and inventory forecasting work, and summarizing actual service scenes according with time series regression sales and inventory forecasting;

secondly, calculating the long-term trend sales and inventory prediction results and optimizing of the polyethylene plastic product industry by using time series regression algorithms such as a mobile weighting method, a trend curve fitting method, an exponential tangent smoothing method and the like;

thirdly, calculating the prediction results and optimization of seasonal trend sales and inventory of the clothing design, production and sales industry by using a time series regression prediction algorithm of the seasonal index factors;

fourthly, calculating a popularity trend marketing scheme in the network popularity expression field and a prediction result and tuning of audience influence by using a time series regression prediction algorithm of the popularity fluctuation parameters;

fifthly, calculating the prediction results and the tuning of the random trend product distribution, the reimbursement and the subscription of the equity investment products of the securities investment fund by using a time series regression prediction algorithm of the random disturbance factors;

sixthly, performing exponential iterative optimization and evolution on the time series regression algorithm calculation results of the long-term trend, the seasonal trend, the popularity trend and the randomness trend by adopting a functional analysis partial differential equation.

8. Claim 8

A method of creating a supply chain sales and inventory management plan, comprising the steps of:

firstly, sorting and normalizing the prediction result data of a safety stock prediction model, a sales scene dynamic self-adaptive sales prediction model and a time series regression sales and stock prediction model of a commodity SKU;

secondly, preprocessing the sorted and structured sales and inventory prediction data such as linear correlation, key value pair matching, curve fitting and the like, and writing the preprocessed data into a database table;

thirdly, sales and inventory forecast data in the database are subjected to sales and inventory business management flow description and stage result analysis;

fourthly, on the basis of sales and inventory forecast business process description and stage analysis, specific execution plans under different sales and inventory management scenes are obtained according to the formulation rules and logic of sales and inventory management plans;

fifthly, according to actual results of sales and inventory service operation under different scenes, iterative optimization is carried out on corresponding sales and inventory management plans.

Images