Supply chain demand prediction method based on fusion model

Abstract

The invention provides a supply chain demand prediction method based on a fusion model. The method comprises the steps of firstly, determining an interaction relationship between supply chains by collecting historical sales data, inventory conditions and other related supply chain participant operation data, constructing a supply chain relationship model, and then predicting the demand of each link in the supply chains by using prediction models such as ARIMA and the like. And fusing the supply chain relation model and the prediction model to obtain comprehensive prediction requirements of each link. And finally, comparing a prediction result with an actual condition so as to evaluate the performance of the model. The method can effectively improve the accuracy of supply chain demand prediction, thereby helping an enterprise to optimize supply chain management and improving the operation efficiency.

Description

Technical Field

[0001] The present invention belongs to the field of supply chain demand forecasting, and more specifically relates to a supply chain demand forecasting method based on a fusion model. Background Art

[0002] Traditional demand forecasting methods mainly include qualitative forecasting and quantitative forecasting. Qualitative forecasting mainly relies on the experience and intuition of experts, such as opinion surveys, expert evaluations, etc. They are useful in new product forecasting or when there is no historical data. However, qualitative methods are often subjective, the accuracy varies from person to person, and they cannot handle large amounts of data.

[0003] In contrast, quantitative forecasting methods can more accurately predict future demand, especially when there is a large amount of historical sales data. These methods include time series analysis, causal models, regression analysis, etc. They can analyze historical data, find patterns and trends hidden behind the data, and provide forecasts. Although these methods can handle large amounts of data, they require professional skills and may also be affected by data anomalies.

[0004] In recent years, with the rise of big data and artificial intelligence, new demand forecasting methods such as machine learning and deep learning have also begun to be applied to demand forecasting. They can process large and complex data, discover deep association rules and patterns, and make high-precision forecasts.

[0005] However, these models often ignore the relationships between participants in the supply chain and their interactions. This may ignore some important factors in demand forecasting, leading to forecasting errors. In this context, we propose a supply chain demand forecasting method based on a fusion model to improve the accuracy of forecasting by considering supply chain relationships. Summary of the invention

[0006] This invention mainly solves the problem that traditional demand forecasting methods ignore the interaction between supply chain participants and cannot accurately reflect demand changes and the nature of the supply chain system. By building a fusion of the supply chain relationship model and the forecasting model, the interaction relationship and interaction intensity between participants are taken into consideration, which improves the accuracy of demand forecasting and the efficiency of supply chain management.

[0007] In order to achieve the above object, the present invention is implemented by adopting the following technical scheme: the method comprises:

[0008] Data collection: Collect historical sales data, inventory data, and supply conditions;

[0009] Data processing: ensuring data quality and accuracy;

[0010] Build a supply chain relationship model: Identify all the key players in the supply chain, including suppliers, manufacturers, distributors, and end customers, and build a supply chain relationship model through their interactions;

[0011] Build a forecasting model: Build a forecasting model based on the supply chain relationship model;

[0012] Integration of supply chain relationship and forecasting models: Integrate forecasting models with supply chain relationship models;

[0013] Evaluate the fusion model: Use the fusion model to predict future demand and evaluate its performance.

[0014] In one solution, the data is collected and obtained from different information systems such as ERP system, WMS, or through a third-party data service provider.

[0015] In one embodiment, the data processing includes: data normalization; data partitioning, dividing the data into a training set, a validation set, and a test set.

[0016] In one embodiment, the construction of the supply chain relationship model includes:

[0017] (1) Identify the main players in the supply chain, including suppliers, manufacturers, wholesalers, retailers and end consumers;

[0018] (2) Clarify the relationships between participants: Understand the role of each participant in the supply chain and the relationship between them and other participants;

[0019] (3) Analyze the interaction mode: understand the interaction between participants, including: the flow of goods, the exchange of information, and quality control;

[0020] (4) Construct a supply chain relationship model using a graph model;

[0021] Each participant in the supply chain is considered as a node in the graph, and the interactions between participants are considered as directed edges. On this basis, weights are added to indicate the amount of flow or the intensity of interaction.

[0022] Suppose graph G = (V, E), where V represents a set of nodes, which represent the participants in the supply chain, suppliers, manufacturers, distributors and retailers; and E represents a set of directed edges, representing the interactive relationships between participants.

[0023] Assume V = {v1, v2, v3, ..., vn}, where vi (i = 1, 2, ..., n) represents different participants in the supply chain;

[0024] E is the set of edges, defined as E = {(vi, vj, w)}|vi, vj belongs to V, w represents the weight, and the weight indicates the quantity of goods flow between suppliers and manufacturers, manufacturers and distributors, and distributors and retailers;

[0025] Finally, the adjacency matrix A (an n×n matrix, where n represents the number of nodes) is used to represent this graph model; if there is an edge from vi to vj, then the corresponding matrix element aij is the weight w, otherwise aij=0.

[0026] In one solution, the construction of the forecasting model includes: using and ARIMA model to forecast demand;

[0027] Specifically including: ARIMA model construction and prediction.

[0028] The ARIMA model consists of three parts: AR model (autoregressive model), I model (differential integration model) and MA model (moving average model), which correspond to the three parameters p, d, q of the ARIMA model respectively;

[0029] In the ARIMA model, the p-order model of the AR part is expressed as:

[0030]

[0031] X t is the current value, c is a constant, is the autoregressive coefficient, ε t It is white noise;

[0032] The q-order model of the MA part is expressed as:

[0033] X t =μ+ε t +∑(from i=1 to q)θ i *ε (t-i)

[0034] Here μ is the mean of the sequence, θ i is the moving average coefficient;

[0035] The difference part includes one or more order differences of the original time series to make the time series stationary, which is expressed by the following formula:

[0036] Y t =X t -X (t-1)

[0037] Integrating these three aspects together, the ARIMA (p, d, q) model is expressed as follows:

[0038]

[0039] Among them, B is the backoff operator, B k *X t =X (t-k) , B to the power of d is the d-order difference. The left part corresponds to AR and I, and the right part corresponds to MA.

[0040] In one embodiment, the integrated supply chain relationship and forecasting model includes:

[0041] (1) Use the forecasting model ARIMA (p, d, q) to forecast demand at each node; that is, use the ARIMA model to predict future demand at the participants (nodes) such as suppliers, manufacturers, distributors, and retailers; assume that the demand forecast value of node vi is ARIMA vi (p, d, q), expressed as:

[0042]

[0043] Among them, t represents the prediction time, Y vit represents the predicted demand of node vi at time t;

[0044] (2) Apply the supply chain relationship model to adjust the weights on each interaction relationship (edge); assuming that the interaction weight from node vi to node vj is wij, then the predicted demand from node vi to vj will be obtained by adjusting the predicted demand value of node vi after weight adjustment. It can be expressed as:

[0045] Y′ vi_vi_t =wij*ARIMA vi (p, d, q)

[0046] Among them, Y′ vi_vj_t represents the predicted demand from node vi to node vj at time t.

[0047] (3) Summarize the forecast demand on each interaction relationship to obtain the final forecast demand; N is all the interaction relationships in the supply chain, then the final forecast demand Y_t is expressed as:

[0048]

[0049] In one embodiment, the evaluation fusion model includes:

[0050] Collect actual demand values ​​to compare with our forecast; the actual demand value is A t , predicted demand value Y t , calculate the prediction error:

[0051] E t =A t -Yt

[0052] Use statistical indicators of error to comprehensively evaluate the prediction performance.

[0053] The mean error ME represents the average level of prediction deviation and is calculated by the following formula:

[0054] ME=∑E t / T

[0055] Where T is the total prediction time.

[0056] The variance of the error, VE, indicates the stability of the prediction deviation and is calculated by the following formula:

[0057]

[0058] Beneficial effects of the present invention:

[0059] The present invention effectively proposes a new demand forecasting strategy by integrating the supply chain relationship model and the forecasting model, taking into account the interaction relationship and interaction intensity between participants, thereby significantly improving the accuracy of the forecast results. In addition, this new forecasting method can effectively give supply chain management more predictability and pertinence, making management tasks such as production scheduling and inventory strategies more efficient, thereby greatly optimizing the operating cost and response speed of the entire supply chain and improving the economic benefits of the enterprise. At the same time, since the accuracy of the forecast results is significantly improved, customer satisfaction and the competitiveness of the enterprise are also greatly enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 is a flow chart of the method of the present invention;

[0061] Figure 2 Construct a supply chain relationship model flow chart;

[0062] Figure 3 Flowchart for integrating supply chain relationships and forecasting models. DETAILED DESCRIPTION

[0063] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Typical embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described in the present invention. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0064] Unless otherwise defined, all technical and scientific terms used in the present invention have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used in the present invention in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Typical embodiments of the present invention are shown in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described in the present invention. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0065] like Figure 1 As shown, the supply chain demand forecasting method based on the fusion model includes:

[0066] S1. Data collection: Collect historical sales data, inventory data, and supply status.

[0067] Determine the type of data that needs to be collected. At this stage, historical sales data, inventory data, supply status, and other data related to demand forecasting are needed.

[0068] Determine the data sources from which this data can be obtained. Data sources include internal systems such as ERP, WMS, etc., as well as external data sources such as supplier data, third-party market reports, etc.

[0069] Once the data is obtained, it needs to be verified for accuracy and completeness. This may include checking for missing values ​​or errors in the data, and comparing it with known information or standards.

[0070] Store the collected data where further analysis can be performed, such as in a database or data warehouse. At the same time, consider data security and privacy requirements and ensure compliance with relevant laws and industry regulations.

[0071] S2. Data processing: ensure the quality and accuracy of data.

[0072] Data normalization: Adjust the data to a standard scale, such as between 0 and 1 (normalization) or to a standard normal distribution (standardization). This ensures that in the subsequent modeling process, data of different scales will not have a significant impact on the final result.

[0073] The data is divided into training set, validation set and test set, so that the generalization ability of the model can be evaluated to prevent overfitting or underfitting. The ratio of training set, validation set and test set is 6:2:2.

[0074] S3. Build a supply chain relationship model: Identify all the main participants in the supply chain, including suppliers, manufacturers, distributors and end customers, and build a supply chain relationship model through their interactions.

[0075] like Figure 2 As shown, the construction of the supply chain relationship model includes:

[0076] S301. First, identify the main participants in the supply chain, including suppliers, manufacturers, wholesalers, retailers and end consumers.

[0077] S302. Clarify the relationship between participants: Understand the role of each participant in the supply chain and the relationship between them and other participants. For example, the relationship with suppliers may be cooperative, competitive or dependent.

[0078] S303. Analyze interaction methods: Have a clear understanding of possible interaction methods between participants, such as the flow of goods, exchange of information, quality control, etc.

[0079] S304, construct a supply chain relationship model, construct a graph model, and use mathematical formulas in graph theory to describe the interaction relationship. The relationship between participants in the supply chain is represented as a mathematical model.

[0080] Each participant in the supply chain is considered as a node in the graph, and the interactions between participants are considered as directed edges. On this basis, weights are added to indicate the amount of flow or the intensity of interaction.

[0081] Suppose graph G = (V, E), where V represents a set of nodes, which represent the participants in the supply chain, suppliers, manufacturers, distributors and retailers; and E represents a set of directed edges, representing the interactive relationships between participants.

[0082] Assume V = {v1, v2, v3, ..., vn}, where vi (i = 1, 2, ..., n) represents different participants in the supply chain;

[0083] E is the set of edges, defined as E = {(vi, vj, w)}vi, vj belongs to V, w represents the weight, and the weight indicates the quantity of goods flow between suppliers and manufacturers, manufacturers and distributors, and distributors and retailers;

[0084] Finally, the adjacency matrix A (nxn matrix, n represents the number of nodes) is used to represent this graph model; if there is an edge from vi to vj, then the corresponding matrix element aij is the weight w, otherwise aij = 0.

[0085] S4. Build a prediction model: Build a prediction model based on the supply chain relationship model.

[0086] The construction of the forecasting model includes: using the ARIMA model to forecast demand; specifically including: constructing and forecasting the ARIMA model. The full name of the ARIMA model is the Autoregressive Integrated Moving Average Model. This is a forecasting model for time series data based on a combination of autoregressive process and moving average process.

[0087] In the ARIMA model, we perform differential processing on the time series data to convert it into a stationary series. Then we predict the stationary series and reverse-differentiate the predicted results to obtain the predicted results of the original series.

[0088] The ARIMA model consists of three parts: AR model (autoregressive model), I model (differential integration model) and MA model (moving average model), which correspond to the three parameters p, d, q of the ARIMA model respectively;

[0089] In the ARIMA model, the p-order model of the AR part is expressed as:

[0090]

[0091] X t is the current value, c is a constant, is the autoregressive coefficient, ε t It is white noise;

[0092] The q-order model of the MA part is expressed as:

[0093] X t =μ+ε t +∑(from i=1 to q)θ i *ε (t-i)

[0094] Here μ is the mean of the sequence, θ i is the moving average coefficient;

[0095] The difference part includes one or more order differences of the original time series to make the time series stationary, which is expressed by the following formula:

[0096] Y t =X t -X (t-1)

[0097] Integrating these three aspects together, the ARIMA (p, d, q) model is expressed as follows:

[0098]

[0099] Among them, B is the backoff operator, B k *X t =X (t-k) , B to the power of d is the d-order difference. The left part corresponds to AR and I, and the right part corresponds to MA.

[0100] like Figure 3 As shown, S5, integrating supply chain relationship and forecasting model: integrating the forecasting model and the supply chain relationship model.

[0101] The integrated supply chain relationship and forecasting model includes:

[0102] S501. Use the forecasting model ARIMA (p, d, q) to forecast demand at each node; that is, use the ARIMA model to forecast future demand at the participants (nodes) such as suppliers, manufacturers, distributors, and retailers; assume that the demand forecast value of node vi is ARIMA vi (p, d, q), expressed as:

[0103] Y vit =ARIMA vi (p, d, q)

[0104] Among them, t represents the prediction time, Y vit represents the predicted demand of node vi at time t;

[0105] S502, use the supply chain relationship model to adjust the weights on each interaction relationship (edge); assuming that the interaction weight from node vi to node vj is wij, then the predicted demand from node vi to vj will be obtained by adjusting the predicted demand value of node vi after weight adjustment. It can be expressed as:

[0106] Y′ vi_vi_t =wij*ARIMA vi (p, d, q)

[0107] Among them, Y′ vi_vj_t represents the predicted demand from node vi to node vj at time t.

[0108] S503, summarizing the forecasted demand on each interaction relationship to obtain the final forecasted demand; N is all the interaction relationships in the supply chain, then the final forecasted demand Y_t is expressed as:

[0109]

[0110] S6. Evaluate the fusion model: Use the fusion model to predict future demand and evaluate its performance.

[0111] The evaluation fusion model includes:

[0112] Collect actual demand values ​​to compare with our forecast; the actual demand value is A t , predicted demand value Y t , calculate the prediction error:

[0113] E t =A t -Y t

[0114] Use statistical indicators of error to comprehensively evaluate the prediction performance.

[0115] The prediction error can reflect the deviation between the predicted value and the actual value, and is an intuitive indicator for evaluating the prediction performance. However, it should be noted that a single error value cannot fully reflect the prediction performance, because the predicted value may be affected by random errors. Therefore, the statistical indicators of the error are used to comprehensively evaluate the prediction performance.

[0116] The mean error ME represents the average level of prediction deviation and is calculated by the following formula:

[0117] ME=∑E t / T

[0118] Where T is the total prediction time.

[0119] The variance of the error, VE, indicates the stability of the prediction deviation and is calculated by the following formula:

[0120]

[0121] Example:

[0122] Take a manufacturing enterprise as an example. The supply chain of the enterprise includes suppliers, factories, distributors and retailers, which constitute the four nodes in the supply chain. In the ERP and WMS systems within the manufacturing enterprise, we collected sales data, inventory data, material consumption data, etc. for the past two years. We also collected the supply of relevant raw materials from suppliers, and collected data such as consumer consumption habits from portal websites and industry reports.

[0123] In the S2 stage, the collected data are cleaned and processed to remove outliers and missing values, and the non-numeric data are digitized and encoded to finally obtain the cleaned data.

[0124] In the S3 stage, set

[0125] V = {v1 (Supplier), v2 (Factory), v3 (Distributor), v4 (Retailer)} is used as a node in the supply chain relationship model, and the interaction between the four is analyzed to define weights, such as raw material supply, product manufacturing and sales volume, etc., and edges from supplier to factory v1-v2, from factory to distributor v2-v3, and from distributor to retailer v3-v4 are constructed to construct the supply chain relationship model.

[0126] In the S4 stage, the ARIMA model is used to independently predict the demand of each node by setting the model parameters p, d, q to 1, 1, 1 respectively, combined with the above-collected historical data. Assume that the forecast results are: v1 (Supplier) forecast demand: 1000 pieces of raw materials / v2 (Factory) forecast demand: 900 pieces of products / v3 (Distributor) forecast demand: 800 pieces of products / v4 (Retailer) forecast demand: 700 pieces of products.

[0127] In the S5 stage, the results of S3 and S4 are integrated. By setting the edge weight w_ij and predicting the demand of each node based on the above, it can be concluded that after the weight adjustment, the demand forecast value from v1 to v2 is 0.91000=900 products, the demand forecast value from v2 to v3 is 0.89900=801 products, and the demand forecast value from v3 to v4 is 0.87*800=696 products.

[0128] In the S6 stage, the performance of the model is evaluated by comparing it with the actual data. The evaluation criteria are the accuracy of the predicted results and the actual values. Assuming that the actual v4 (Retailer) demand is 695 products, which is within the prediction error range, it means that the model prediction accuracy is high.

[0129] The demand forecast results we expect should be able to guide companies to make more reasonable production decisions, inventory management decisions and supply chain collaboration decisions in actual operations, and continuously optimize the model based on actual results to improve forecast accuracy. In this way, it can not only reduce unnecessary waste and inventory, optimize supply chain operations, but also improve customer satisfaction and enhance the economic benefits of enterprises.

[0130] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the program can be stored in a computer-readable storage medium, and when the program is executed, it can include the processes of the embodiments of the above-mentioned methods. The storage medium can be a disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM).

[0131] It should be understood that the detailed description of the technical solutions of the present invention by means of the preferred embodiments is illustrative rather than restrictive. A person skilled in the art may modify the technical solutions described in the embodiments, or replace some of the technical features by equivalents, based on reading the specification of the present invention; and these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A supply chain demand forecasting method based on a fusion model, characterized by: The method includes: Data collection: Collect historical sales data, inventory data, and supply conditions; Data processing: ensuring data quality and accuracy; Build a supply chain relationship model: Identify all the key players in the supply chain, including suppliers, manufacturers, distributors, and end customers, and build a supply chain relationship model through their interactions; Build a forecasting model: Build a forecasting model based on the supply chain relationship model; Integration of supply chain relationship and forecasting models: Integrate forecasting models with supply chain relationship models; Evaluate the fusion model: Use the fusion model to predict future demand and evaluate its performance.

2. The supply chain demand forecasting method based on the fusion model according to claim 1 is characterized by: The data collection is obtained from different information systems such as ERP system, WMS, or through a third-party data service provider.

3. The supply chain demand forecasting method based on the fusion model according to claim 1 is characterized by: The data processing includes: data normalization; data partitioning, dividing the data into a training set, a validation set and a test set.

4. The supply chain demand forecasting method based on the fusion model according to claim 1 is characterized by: The construction of the supply chain relationship model includes: (1) Identify the main players in the supply chain, including suppliers, manufacturers, wholesalers, retailers and end consumers; (2) Clarify the relationships between participants: Understand the role of each participant in the supply chain and the relationship between them and other participants; (3) Analyze the interaction mode: understand the interaction between participants, including: the flow of goods, the exchange of information, and quality control; (4) Construct a supply chain relationship model using a graph model; Each participant in the supply chain is considered as a node in the graph, and the interactions between participants are considered as directed edges. On this basis, weights are added to indicate the amount of flow or the intensity of interaction. Assume that graph G = (V, E), where V represents a set of nodes, which represent the participants in the supply chain, suppliers, manufacturers, distributors, and retailers; and E represents a set of directed edges, which represent the interactive relationships between participants; Assume V = {v1, v2, v3, ..., vn}, where vi (i = 1, 2, ..., n) represents different participants in the supply chain; E is the set of edges, defined as E = {(vi, vj, w)}vi, vj belongs to V, w represents the weight, and the weight indicates the quantity of goods flow between suppliers and manufacturers, manufacturers and distributors, and distributors and retailers; Finally, the adjacency matrix A (an n×n matrix, where n represents the number of nodes) is used to represent this graph model; if there is an edge from vi to vj, then the corresponding matrix element aij is the weight w, otherwise aij=0.

5. The supply chain demand forecasting method based on the fusion model according to claim 1 is characterized by: The construction of the forecasting model includes: using the ARIMA model to forecast demand; Specifically include: ARIMA model construction and prediction; The ARIMA model consists of three parts: AR model (autoregressive model), I model (differential integration model) and MA model (moving average model), which correspond to the three parameters p, d, q of the ARIMA model respectively; In the ARIMA model, the p-order model of the AR part is expressed as: X t is the current value, c is a constant, is the autoregressive coefficient, ε t It is white noise; The q-order model of the MA part is expressed as: X t =μ+e t +∑(fromi=1to q)θ i *e (t-i) Here μ is the mean of the sequence, θ i is the moving average coefficient; The difference part includes one or more order differences of the original time series to make the time series stationary, which is expressed by the following formula: Y t =X t -X (t-1) Integrating these three aspects together, the ARIMA (p, d, q) model is expressed as follows: Among them, B is the backoff operator, B k *X t =X (t-k) , B to the power of d is the d-order difference. The left part corresponds to AR and I, and the right part corresponds to MA.

6. The supply chain demand forecasting method based on the fusion model according to claim 1 is characterized by: The integrated supply chain relationship and forecasting model includes: (1) Use the forecasting model ARIMA(p, d, q) to forecast demand at each node; that is, use the ARIMA model to predict future demand at the supplier, manufacturer, distributor, and retailer participants (nodes); assume that the demand forecast value of node vi is ARIMA vi (p, d, q), expressed as: Y vit =ARIMA vi (p,d,q) Among them, t represents the prediction time, Y vit represents the predicted demand of node vi at time t; (2) Apply the supply chain relationship model to adjust the weights of each interaction relationship (edge); assuming that the interaction weight from node vi to node vj is wij, then the predicted demand from node vi to vj will be obtained by adjusting the predicted demand value of node vi after weight adjustment; expressed as: Y′ vi_vj_t =we*ARIMA vi (p,d,q) Among them, Y′ vi_vj_t represents the predicted demand from node vi to node vj at time t; (3) Summarize the forecast demand on each interaction relationship to obtain the final forecast demand; N is all the interaction relationships in the supply chain, then the final forecast demand Y_t is expressed as:

7. The supply chain demand forecasting method based on the fusion model according to claim 1 is characterized by: The evaluation fusion model includes: Collect actual demand values ​​to compare with our forecast; the actual demand value is A t , predicted demand value Y t , calculate the prediction error: AND t =A t -AND t Use statistical indicators of error to comprehensively evaluate the prediction performance. The mean error ME represents the average level of prediction deviation and is calculated by the following formula: ME=∑Et / T Where T is the total time of prediction; The variance of the error, VE, indicates the stability of the prediction deviation and is calculated by the following formula:

Images