A method for forecasting demand for multiple steel products based on smart supply chain

By using SARIMA and SARIMAX time series models to predict the demand for multiple varieties of steel in the smart supply chain, the problems of low prediction efficiency and insufficient accuracy in the existing technology are solved, and timely and accurate prediction of the demand for multiple varieties of steel is achieved, improving the improvement of the demand for multiple varieties of steel. Production efficiency and capacity resource utilization rate.

CN113780655BActive Publication Date: 2024-12-03欧冶云商股份有限公司
0 Cites 0 Cited by

Patent Information

Application Number
CN202111049815.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2021-09-08
Publication Date
2024-12-03
Estimated Expiration
2041-09-08

AI Technical Summary

Technical Problem

The existing multi-variety steel demand forecasting methods have problems such as time-consuming, poor systemicity, low integrity, low accuracy and slow response and output speed in steel production enterprises. It is difficult to improve production efficiency and capacity resource utilization, and cannot meet the wisdom. Intelligent linkage of the demand for multiple types of steel in the supply chain industrial product production.

Method used

Using a multi-variety steel demand forecasting method based on smart supply chain, by obtaining industrial product production data, using SARIMA and SARIMAX time series models for data cleaning, modeling and prediction, combining sliding averaging method and grid search optimization, the demand data time is constructed Sequence model to achieve accurate prediction.

Benefits of technology

It improves the accuracy and stability of demand forecasting, and can timely and accurately predict the demand for multiple steel varieties in a certain period of the future, helping steel mills to reasonably arrange production plans and improve production efficiency and capacity resource utilization.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN113780655B_ABST
    Figure CN113780655B_ABST
Patent Text Reader

Abstract

The present invention relates to a method for forecasting the demand for multiple varieties of steel based on a smart supply chain, comprising the following steps: 1) obtaining a time series of demand data for multiple varieties of steel based on industrial product production data of a smart supply chain system; 2) constructing a SARIMA time series model based on the time series of demand data for multiple varieties of steel; 3) inputting a training sample set to be tested into the SARIMA time series model to obtain a forecast result of the demand for multiple varieties of steel; 4) constructing a SARIMAX time series model based on the time series of demand data for multiple varieties of steel; 5) inputting a test sample set to be tested into the SARIMAX time series model to obtain a forecast result of the demand for multiple varieties of steel after algorithm correction; 6) obtaining a final forecast result based on the result output by the SARIMA model and the correction result output by the SARIMAX model. Compared with the prior art, the present invention has the advantages of being able to make the time series more stable and having a higher forecast accuracy.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of supply chain production management, and in particular to a method for forecasting demand for multiple varieties of steel products based on a smart supply chain. Background Art

[0002] With the development of the industrial Internet, the downstream supply chain of the steel industry has also ushered in transformation and upgrading. The production organization mode of its main raw material steel also needs to change accordingly. Steel production enterprises need to change from a rigid production organization mode based on production to a flexible production organization mode with balanced production and sales. It is necessary to utilize the information exchange of the manufacturing process of industrial products with downstream supply chain enterprises, combine effective technical means, explore the laws of industrial product production data, and predict the demand for multiple varieties of steel in the downstream supply chain of the steel industry in advance for a certain period in the future, improve the steel production efficiency and capacity resource utilization of steel production enterprises, and meet the steel consumption demand of the downstream supply chain more quickly and timely.

[0003] However, the current technical means for demand analysis, processing and decision-making in the downstream supply chain are relatively backward. The industrial product production data of downstream supply chain enterprises collected in the smart supply chain mainly rely on manual experience or simple mathematical formulas to predict the demand for corresponding steel varieties. Based on the current demand forecast, it is difficult for steel mills to reasonably and efficiently arrange production plans for different steel varieties, resulting in difficulty in improving production efficiency and capacity resource utilization. This is a technical problem that urgently needs to be improved and solved. In other fields, when forecasting demand through various big data forecasting technologies, more consideration is given to non-technical issues such as corporate profits and cost conditions. The current demand forecasting technical means and methods are time-consuming, poorly systematic, low in integrity, low in accuracy, and slow in response and output.

[0004] In summary, the existing method for forecasting demand for multiple varieties of steel cannot improve production efficiency and capacity resource utilization, and cannot meet the intelligent linkage of industrial product production in the smart supply chain to the demand reserve and scheduling of multiple varieties of steel in steel production enterprises. Summary of the invention

[0005] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a method for forecasting the demand for multiple varieties of steel based on a smart supply chain. The method utilizes the rules of industrial product production data in the smart supply chain and targeted data processing methods to make timely, accurate and stable forecasts on the demand for multiple varieties of steel in a certain period of time in the future.

[0006] The purpose of the present invention can be achieved by the following technical solutions:

[0007] A method for forecasting demand for multiple varieties of steel products based on a smart supply chain, the method comprising the following steps:

[0008] S1: Based on the industrial product production data of the smart supply chain system, obtain the time series of demand data for multiple varieties of steel. The industrial product production data of the smart supply chain system includes production BOM data, production data of the main machine manufacturer, and production data of the parts manufacturer. The specific steps for obtaining the time series of demand data for multiple varieties of steel include:

[0009] 11) The production data of industrial products in the smart supply chain system is used as the original data time series;

[0010] 12) Based on the SARIMA model technical framework, the original data time series is cleaned, the data date format is adjusted, the missing data is filled using the sliding average method, the threshold number of months is set, and the data above the threshold number of consecutive missing months is filled with zeros, and the top 90% of the data in a certain period are selected as available data sources;

[0011] 13) According to the production BOM data, the production data of the host manufacturer and the production data of the parts manufacturer, establish the functional relationship between production data and demand data W t =g(P 1 ,P 2 ,...,P n ), where W t is the demand value of different types of steel in different months, P 1 ,P 2 ,...,P n For the production values ​​of different industrial products, the production data of smart supply chain industrial products are solved and converted into the corresponding time series of demand data of multiple varieties of steel products using functional relationships;

[0012] 14) Perform sliding average processing on the demand data time series of multiple varieties of steel obtained in step 13) to obtain the sample data length used for rolling forecasting of the demand forecasting model and the initial stable demand data time series.

[0013] S2: Based on the time series of demand data of multiple varieties of steel, a SARIMA time series model is constructed.

[0014] The specific steps include:

[0015] 21) Based on seasonality, construct a SARIMA time series model:

[0016]

[0017] Where W t 、w t-n 、w t-sn are the demand values ​​of different types of steel in months t, tn, and t-sn, μ is a constant term, εt , ε t-n , ε t-sn The error of the demand value of different types of steel in months t, tn, and t-sn, p is the number of trend autoregressive items, q is the number of trend sliding average items, P is the number of seasonal autoregressive items, and Q is the number of seasonal sliding average items. n is the trend autoregressive coefficient, θ n is the trend moving average coefficient, φ n is the seasonal autoregressive coefficient, η n is the seasonal moving average coefficient.

[0018] 22) According to the results of the three characteristic sub-items obtained by model decomposition, the standard deviation between the seasonal characteristic component and the random characteristic component is calculated, and the abnormal values ​​of all demand data time series are adjusted within the set interval; the calculation formula of the standard deviation σ between the seasonal characteristic component and the random characteristic component is:

[0019]

[0020] In the formula, Cs and Cr are the seasonal characteristic component and random characteristic component automatically decomposed from the demand data time series;

[0021] The upper and lower limits of the demand data time series values ​​before each prediction are obtained using the trend characteristic component Ct±standard deviation, and all abnormal values ​​of the demand data time series are adjusted within the program setting range, that is, [Ct-σ, Ct+σ].

[0022] 23) Grid search training and optimization for model selection are used to obtain different combinations of SARIMA(p,d,q)(P,D,Q)s parameters, where d is the number of trend differences made to make it a final stationary series, D is the number of seasonal differences made to make it a final stationary series, s is the number of time steps during a single season, p is the number of trend autoregressive terms, q is the number of trend sliding average terms, P is the number of seasonal autoregressive terms, and Q is the number of seasonal sliding average terms; the AIC values ​​of different combinations are calculated, and the parameter combination with the smallest AIC value is selected as the optimal parameter combination to complete the final conditions for establishing the SARIMA time series model.

[0023] S3: Input the training sample set to be tested into the constructed SARIMA time series model to obtain the demand forecast results of multiple varieties of steel.

[0024] S4: Based on the time series of demand data for multiple types of steel in S1, a SARIMAX time series model is constructed.

[0025] The specific steps include:

[0026] 41) Based on the time series of demand data of multiple varieties of steel corresponding to the production data of the main engine manufacturer and the production data of the parts manufacturer obtained in S1, normalize them to obtain two groups of data mapped in the range of (0, 1);

[0027] 42) The correlation coefficient of the two groups of mapped data is calculated by correlation analysis, and the data corresponding to the correlation coefficient ≥ 0.7 are selected as the leading data. The demand data of multiple varieties of steel corresponding to the production data of the parts manufacturer before normalization are used as the external variable X of the SARIMAX model, then:

[0028]

[0029] In the formula, is the demand value of multiple varieties of steel corresponding to the production data of the parts manufacturer, β n For external variables The regression coefficient of , r is the number of regression items of the external variable;

[0030] 43) Based on the demand data time series obtained after S2 processing, the grid search of model selection is used to train and optimize again to obtain different combinations of SARIMAX(p,d,q)(P,D,Q)s parameters, calculate the AIC values ​​of different combinations, and select the parameter combination with the smallest AIC value as the optimal parameter combination to complete the final conditions for establishing the SARIMAX time series model.

[0031] S5: Input the training sample set to be tested into the constructed SARIMAX time series model to obtain the demand forecast results of multiple varieties of steel after algorithm correction.

[0032] S6: Obtain the final forecast result based on the demand forecast result output by the SARIMA time model obtained in S3 and the revised demand forecast result output by the SARIMAX time series model obtained in S5.

[0033] The specific contents are:

[0034] Demand forecast results based on the SARIMA time model output obtained from S3 The revised demand forecast results obtained from the SARIMAX time series model output by S5 Get the final prediction results in The demand time series is the output of the SARIMA time series model, excluding the output of the available SARIMAX time series model.

[0035] Furthermore, in step 12), based on the rule that twelve months is a complete cycle, the top 90% of the data in a complete cycle are selected as available data sources.

[0036] Further, in step 42), the correlation coefficient is ≥ 0.7 and W b W in the middle a The prior data, W a , W b The data mapped within the two groups (0, 1) obtained in step 41).

[0037] The method for forecasting demand for multiple varieties of steel products based on a smart supply chain provided by the present invention has at least the following beneficial effects compared to the prior art:

[0038] 1) The present invention can utilize the regularity of industrial product production data in the smart supply chain and a targeted data processing method to obtain the sample data length and the initial stable demand data time series used in the rolling forecast of the demand forecast model. The demand data time series is more stable, which helps to improve the accuracy of the forecast;

[0039] 2) SARIMA and SARIMAX time series models are established to predict the time series operation data of steel demand data of multiple varieties, that is, non-stationary time series. Not only can the non-stationary time series be converted into a stationary time series, but also the influence of seasonal factors is taken into account. Moreover, by mining the natural laws between the production data of different industrial products, effective features for building models are formed, and the algorithm efficiency and complexity of data sourcing and correlation factor analysis are reduced, so that prediction results with high accuracy and engineering applicability can be obtained based on the above time series;

[0040] 3) Through the method for forecasting the demand for multiple varieties of steel based on a smart supply chain of the present invention, timely, accurate and stable forecasts can be made for the demand for multiple varieties of steel in a certain period of time in the future, thereby solving the technical problem that steel mills are unable to reasonably and efficiently arrange production plans for different varieties of steel, making it difficult to improve production efficiency and capacity resource utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is a flowchart of a method for forecasting demand for multiple varieties of steel based on a smart supply chain in an embodiment. DETAILED DESCRIPTION

[0042] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work should belong to the scope of protection of the present invention.

[0043] Example

[0044] like Figure 1 As shown, the present invention relates to a method for forecasting demand for multiple varieties of steel products based on a smart supply chain, the method comprising the following steps:

[0045] Step 1: Obtain the time series of demand data for multiple varieties of steel.

[0046] Obtain production data of industrial products from the smart supply chain system, including production BOM (bill of materials) data, production data of host manufacturers, production data of parts manufacturers and other raw data time series; perform data preprocessing such as cleaning, screening, conversion, and sliding average on the raw data time series based on the SARIMA model technical framework to obtain the sample data length of multiple steel varieties used in the rolling forecast of the demand forecast model and the required stable demand data time series. Specifically:

[0047] Step 101: Obtain the production data of industrial products in the smart supply chain system, including the production BOM (bill of materials) data, the production data of the host manufacturer, the production data of the parts manufacturer and other original data time series;

[0048] Step 102: Clean the original data time series based on the SARIMA model technical framework, use the program to adjust the data date format, use the sliding average method to fill in the missing data, set the threshold to 6 months, and fill in the missing data for more than 6 consecutive months with 0. In order to improve the stability of the time series, combined with the rule that the production data of industrial products is a complete cycle of 12 months, select the top 90% of the data in a complete cycle as the available data source;

[0049] Step 103: According to the production BOM (Bill of Materials) data, the production data of the host manufacturer, and the production data of the parts manufacturer, a functional relationship between the production data and the demand data is established. t =g(P 1 ,P 2 ,...,P n ), solve the production data of industrial products in the smart supply chain and convert it into the corresponding time series of demand data of multiple varieties of steel, where W t is the demand value of different types of steel in different months, P 1 ,P 2 ,...,P n is the production value of different industrial products;

[0050] Step 104: Perform a 3-month moving average process on the demand data time series of multiple steel varieties obtained in Step 103 to obtain the sample data length used for the rolling forecast of the demand forecasting model and the initial stable demand data time series.

[0051] Step 2: Establish a SARIMA time series model based on the demand data time series obtained in step 1.

[0052] The SARIMA algorithm model is continuously adjusted and optimized to automatically decompose the trend characteristic component, seasonal characteristic component and random characteristic component from the demand data time series; according to the results of the three characteristic sub-items obtained by the model decomposition, the program is used to calculate the standard deviation between the seasonal characteristic component and the random characteristic component, and the upper and lower limits of the demand data time series value are obtained by using the trend characteristic component ± standard deviation, and all abnormal values ​​of the demand data time series are adjusted within the program setting range to avoid the influence of extreme demand data in individual months on the overall data law; different combinations of SARIMA (p, d, q) × (P, D, Q) parameters are obtained by grid search training and optimization of model selection, and the AIC (Akaike Information Criterion) values ​​of different combinations are calculated by the program, and the parameter combination with the smallest AIC value is selected as the optimal parameter combination to complete the final conditions for establishing the SARIMA time series model. The specific steps include:

[0053] Step 201: Since the production data of smart supply chain industrial products related to the consumption of steel varieties has a relatively obvious seasonal pattern, that is, the production volume in summer and winter is greater than that in spring and autumn, the demand data of multiple varieties of steel will also have seasonal changes, and a SARIMA time series model can be constructed:

[0054]

[0055] Where W t 、w t-n 、w t-sn are the demand values ​​of different types of steel in months t, tn, and t-sn, μ is a constant term, ε t , ε t-n , ε t-sn The error of the demand value of different types of steel in months t, tn, and t-sn, p is the number of trend autoregressive items, q is the number of trend sliding average items, P is the number of seasonal autoregressive items, and Q is the number of seasonal sliding average items. n is the trend autoregressive coefficient, θ n is the trend moving average coefficient, φ n is the seasonal autoregressive coefficient, η n is the seasonal moving average coefficient.

[0056] By utilizing the composition of the SARIMA algorithm model, automatic adjustment and optimization are performed to automatically decompose the trend characteristic component, seasonal characteristic component and random characteristic component from the demand data time series, which are recorded as Ct, Cs and Cr respectively.

[0057] Step 202: Based on the results of the three characteristic sub-items obtained by model decomposition, the standard deviation between the seasonal characteristic component and the random characteristic component is calculated using the program. The calculation formula is as follows:

[0058]

[0059] In order to avoid the influence of extreme demand data in individual months on the overall data pattern, the upper and lower limits of the demand data time series value before each prediction are obtained by using the trend characteristic component ± standard deviation, and all abnormal values ​​of the demand data time series are adjusted within the program setting range, that is, [Ct-σ, Ct+σ].

[0060] Step 203: Use the grid search training and optimization of model selection to obtain different combinations of SARIMA (p, d, q) (P, D, Q) s parameters, d is the number of trend differences made to make it a final stationary series, D is the number of seasonal differences made to make it a final stationary series, and s is the number of time steps during a single season. The program is used to calculate the AIC (Akaike Information Criterion) values ​​of different combinations, and the parameter combination with the smallest AIC value is selected as the optimal parameter combination to complete the final conditions for establishing the SARIMA time series model.

[0061] Step 3. Based on the SARIMA time series model obtained in step 2 and the final stable demand data time series obtained in steps 1 and 2, select the data with a time period length of [T-24, T] as the training sample set, input the training sample set to be tested, and make a rolling forecast for the demand of multiple varieties of steel in T+2 months (the next 30 to 60 days) to obtain the demand forecast results of multiple varieties of steel.

[0062] Step 4: Since the production data of the main engine manufacturer and the parts manufacturer have obvious natural laws, the production data of the parts manufacturer always precedes the changes in the production data of the main engine manufacturer. Based on this law, the time series of demand data of multiple varieties of steel corresponding to the production data of the main engine manufacturer and the parts manufacturer are obtained based on step 1, and normalized to obtain two groups of data mapped within the range of (0, 1); the correlation coefficient of the two groups of mapped data is calculated using correlation analysis, and the external variable X that can be used as the SARIMAX model is screened out. Based on the SARIMA algorithm model of steps 2 and 3, the grid search training and optimization of the model selection are used to obtain different combinations of SARIMAX (p, d, q) (P, D, Q)s parameters, and the AIC (Akaike Information Criterion) values ​​of different combinations are calculated using the program. The parameter combination with the smallest AIC value is selected as the optimal parameter combination to complete the final conditions for establishing the SARIMAX time series model. The specific steps include:

[0063] Step 401: Based on the production data of the main machine manufacturer and the production data of the parts manufacturer obtained in step 1, the demand data time series of multiple varieties of steel materials corresponding to the production data are normalized to obtain two groups of data mapped in the range of (0, 1), which are respectively denoted as W a , W b ;

[0064] Step 402: Use correlation analysis to calculate the correlation coefficient of the two groups of mapped data, and select the data with a correlation coefficient ≥ 0.7 and W b W in the middle a The prior data will be used to normalize the demand data of multiple steel varieties corresponding to the production data of the parts manufacturer before normalization, as the external variable X of the SARIMAX model, then:

[0065]

[0066] in, is the demand value of multiple varieties of steel corresponding to the production data of the parts manufacturer, β n For external variables The regression coefficient of , r is the number of regression items of the external variable;

[0067] Step 403: Based on the demand data time series obtained after processing in step 2, in order to avoid the influence of adding external variables X on the combination selection of SARIMAX(p,d,q)(P,D,Q)s parameters, the grid search selected by the model is used to train and optimize again to obtain different combinations of SARIMAX(p,d,q)(P,D,Q)s parameters. The AIC (Akaike Information Criterion) values ​​of different combinations are calculated using the program, and the parameter combination with the smallest AIC value is selected as the optimal parameter combination to complete the final conditions for establishing the SARIMAX time series model.

[0068] Step 5. Based on the SARIMAX time series model obtained in step 4 and the final steady demand data time series obtained in steps 1 and 2, select data with a time period length of [T-24, T] as the training sample set, input the training sample set to be tested, and predict the demand for multiple varieties of steel in T+2 months (the next 60 days) to obtain the demand forecast results for multiple varieties of steel after algorithm correction.

[0069] Step 6: Demand forecast results based on the SARIMA model output obtained in step 3 And the revised demand forecast results based on the SARIMAX model output obtained in step 5 The final prediction result in Demand time series using the SARIMA model output, excluding the available SARIMAX model output results.

[0070] The present invention can realize the use of the laws of industrial product production data in the smart supply chain and targeted data processing methods to obtain the sample data length used for rolling forecasting of the demand forecasting model and the initial stable demand data time series. The demand data time series is more stable, which helps to improve the accuracy of the forecast. SARIMA and SARIMAX time series models are established to predict the demand data time series operation data of multiple varieties of steel, that is, non-stationary time series. Not only can the non-stationary time series be converted into a stationary time series, but also the influence of seasonal factors is considered. Moreover, by mining the natural laws between the production data of different industrial products, effective features for building models are formed, and the algorithm efficiency and complexity of data sourcing and correlation factor analysis dimensions are reduced, so that based on the above time series, a high-accuracy and engineering-applicable forecast result can be obtained.

[0071] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person familiar with the technical field can easily think of various equivalent modifications or substitutions within the technical scope disclosed by the present invention, and these modifications or substitutions should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for forecasting demand for multiple varieties of steel based on a smart supply chain, characterized in that: The following steps are involved: 1) Based on the industrial product production data of the smart supply chain system, obtain the time series of demand data for multiple varieties of steel; 2) Based on the time series of demand data of multiple varieties of steel, a SARIMA time series model is constructed; 3) Input the training sample set to be tested into the constructed SARIMA time series model to obtain the demand forecast results of multiple varieties of steel; 4) Based on the time series of demand data of multiple varieties of steel in step 1), a SARIMAX time series model is constructed; 5) Input the training sample set to be tested into the constructed SARIMAX time series model to obtain the demand forecast results of multiple varieties of steel; 6) Obtain the final forecast result based on the demand forecast result output by the SARIMA time series model obtained in step 3) and the demand forecast result output by the SARIMAX time series model obtained in step 5); The industrial product production data of the smart supply chain system includes production BOM data, host manufacturer production data, and component manufacturer production data; The specific steps to obtain the time series of demand data for multiple varieties of steel include: 11) The production data of industrial products in the smart supply chain system is used as the original data time series; 12) Based on the SARIMA model technical framework, the original data time series is cleaned, the data date format is adjusted, the missing data is filled using the sliding average method, the threshold number of months is set, and the data with consecutive missing months exceeding the threshold number are filled with zeros, and the top 90% of the data in a certain period are selected as available data sources; 13) Establish a functional relationship between production data and demand data based on production BOM data, production data of the host manufacturer and production data of the parts manufacturer ,in is the demand value of different types of steel in different months, For the production values ​​of different industrial products, the production data of smart supply chain industrial products are solved and converted into the corresponding time series of demand data of multiple varieties of steel products using functional relationships; 14) Performing sliding average processing on the demand data time series of multiple steel varieties obtained in step 13) to obtain the sample data length used for rolling forecast of the demand forecasting model and the initial stable demand data time series; The specific steps of step 2) include: 21) Construct SARIMA time series model based on seasonality; 22) Based on the results of the three characteristic sub-items obtained by model decomposition, calculate the standard deviation between the seasonal characteristic component and the random characteristic component, and adjust the abnormal values ​​of all demand data time series within the set range; wherein, the standard deviation between the seasonal characteristic component and the random characteristic component The calculation formula is: , In the formula, Cs and Cr are the seasonal characteristic component and random characteristic component automatically decomposed from the demand data time series; The upper and lower limits of the demand data time series before each prediction are obtained by using the trend characteristic component Ct±standard deviation, and all abnormal values ​​of the demand data time series are adjusted within the program setting range, that is, ; 23) Grid search training and optimization of model selection are used to obtain different combinations of SARIMA (p, d, q) (P, D, Q)s parameters, where d is the number of trend differences made to make it a final stationary series, D is the number of seasonal differences made to make it a final stationary series, and s is the number of time steps during a single season. p is the number of trend autoregressive terms, q is the trend sliding average item, P is the number of seasonal autoregressive terms, Q is the number of seasonal sliding average items; calculate the AIC values ​​of different combinations, select the parameter combination with the smallest AIC value as the optimal parameter combination, and complete the final conditions for establishing the SARIMA time series model; In step 21), the expression of the constructed SARIMA time series model is: , In the formula, , , Different types of steel t, tn, t-sn Monthly demand value, is a constant term, , , Different types of steel t, tn, t-sn The error in the demand value for the month, p is the number of trend autoregressive terms, q is the trend sliding average item, P is the number of seasonal autoregressive terms, Q is the seasonal moving average term, is the trend autoregressive coefficient, is the trend moving average coefficient, is the seasonal autoregressive coefficient, is the seasonal moving average coefficient; The specific steps of step 4) include: 41) Based on the time series of demand data of multiple varieties of steel corresponding to the production data of the host manufacturer and the production data of the parts manufacturer obtained in step 1), normalize them to obtain two groups of data mapped in the range of (0, 1); 42) Use correlation analysis to calculate the correlation coefficient of the two groups of mapped data, select the data corresponding to the correlation coefficient ≥ 0.7 as the leading data, and use the demand data of multiple varieties of steel corresponding to the production data of the parts manufacturer before normalization as the external variables of the SARIMAX model, then: , In the formula, The demand value of multiple types of steel corresponding to the production data of the parts manufacturer, For external variables The regression coefficient of r is the number of regression terms of external variables; 43) Based on the demand data time series obtained after processing in step 1), the grid search selected by the model is used to train and optimize again to obtain different combinations of SARIMAX (p, d, q) (P, D, Q)s parameters, calculate the AIC values ​​of different combinations, and select the parameter combination with the smallest AIC value as the optimal parameter combination to complete the final conditions for establishing the SARIMAX time series model.

2. The method for forecasting demand for multiple varieties of steel products based on a smart supply chain according to claim 1 is characterized in that: The specific contents of step 6) are: Demand forecast results based on the SARIMA time series model output obtained in step 3) The demand forecast results output by the SARIMAX time series model obtained in step 5) , get the final prediction result = + ,in The demand time series is the output of the SARIMA time series model, excluding the output of the available SARIMAX time series model.

3. The method for forecasting demand for multiple varieties of steel products based on a smart supply chain according to claim 1 is characterized in that: In step 12), based on the rule that a complete cycle is twelve months, the top 90% of the data in a complete cycle are selected as available data sources.

4. The method for forecasting demand for multiple varieties of steel products based on a smart supply chain according to claim 1 is characterized in that: In step 42), select the correlation coefficient ≥ 0.7 and Zhongwei The leading data, , The two sets of data mapped within the range of (0, 1) obtained in step 41).