Cross-regional supply chain management method and system based on artificial intelligence

Abstract

The invention relates to the technical field of artificial intelligence and the like, and provides a cross-regional supply chain management method and system based on artificial intelligence, and the method comprises the steps: carrying out the fusion processing of data related to product purchase, obtaining multi-dimensional purchase related data, inputting the data into an LSTM model, carrying out the purchase prediction, obtaining the purchase demand prediction data in a prediction period, and carrying out the prediction of the purchase demand. The method comprises the steps of automatically matching a preset purchase plan according to purchase demand prediction data, continuously comparing actual product sales data in a prediction period with the purchase demand prediction data through a reinforcement learning algorithm, evaluating a prediction deviation of purchase demands in real time, and automatically adjusting parameters of an LSTM model when demand sudden increase or market fluctuation occurs. And the LSTM model outputs updated purchase demand prediction data after parameter adjustment, automatically matches a preset purchase plan according to the updated purchase demand prediction data until the end of the prediction period, and obtains a final reference purchase plan in the current prediction period, thereby realizing reduction of purchase prediction deviation and flexible response to demand change.

Description

Technical Field

[0001] The present invention relates to the technical fields of artificial intelligence, Internet services, and in particular to a cross-regional supply chain management method and system based on artificial intelligence. Background Art

[0002] Cross-regional supply chain management usually refers to the coordination and optimization of supply chain activities between different geographical regions. It mainly includes cross-regional collaboration in procurement, supplier management, production, warehousing, transportation and distribution, aiming to improve the efficiency and responsiveness of the supply chain. In the procurement stage, the current supply chain management system can analyze historical sales data, market trends and seasonal fluctuations to a certain extent, predict product demand in the future, and thus generate corresponding procurement plans in advance. In the supplier management stage, the current supply chain management system will also count the delivery speed, product quality, cost and other data of different suppliers to improve supply efficiency. In the warehousing stage, the current supply chain management system can basically analyze the changes in sales demand in different regions, adjust the inventory levels in each region, and transfer goods between warehouses to prevent inventory shortages or backlogs. However, when the current supply chain management faces sudden demand or abnormal market fluctuations in the procurement stage, the forecast is often biased, which can easily lead to insufficient or excessive procurement. Once the market environment changes rapidly, it is difficult to adjust the procurement plan in time and cannot flexibly respond to changes in demand. In the supplier management stage, the current supply chain management is difficult to comprehensively process the supplier's historical data, market environment and real-time data, resulting in one-sided supplier evaluation. Moreover, when suppliers fail to deliver on time or product quality is unstable, the current supply chain management lacks an early warning mechanism, making it difficult to make timely adjustments. In the warehousing stage, the current supply chain management mainly adjusts inventory based on historical data, which is difficult to cope with rapidly changing sales demand. The transfer of goods between warehouses is limited by traditional routes and manual decision-making, making it difficult to optimize in real time.

[0003] To sum up, the existing cross-regional supply chain management technology has technical problems such as large deviations in procurement forecasts and inability to flexibly respond to changes in demand when faced with sudden demands or abnormal market fluctuations; one-sided supplier evaluation; lack of early warning mechanisms and difficulty in making timely adjustments when suppliers fail to deliver on time or product quality is unstable; warehouse management mainly adjusts inventory based on historical data, making it difficult to cope with rapidly changing sales demands; and the transfer of goods between warehouses is limited by traditional paths and manual decisions, making it difficult to optimize in real time. Summary of the invention

[0004] In view of the shortcomings of the above-mentioned prior art, the present invention provides a cross-regional supply chain management method and system based on artificial intelligence, so as to reduce procurement forecast deviations, flexibly respond to changes in demand, comprehensively evaluate suppliers, and issue abnormal warnings for suppliers, so as to realize automatic adaptation of warehouse management to rapidly changing sales demand and intelligent allocation of goods between warehouses.

[0005] In a first aspect, the present invention provides a cross-regional supply chain management method based on artificial intelligence, comprising:

[0006] The historical sales data, market trend data, seasonal fluctuation data and external environment data related to product procurement are integrated and processed to obtain multi-dimensional procurement-related data;

[0007] Input the multi-dimensional procurement-related data into the LSTM model for procurement forecasting, obtain procurement demand forecast data within the forecast period, and automatically match the preset procurement plan according to the procurement demand forecast data;

[0008] By using a reinforcement learning algorithm, the actual product sales data within the forecast period is continuously compared with the procurement demand forecast data, and the forecast deviation of procurement demand is evaluated in real time. When there is a sudden increase in demand or market fluctuations, the parameters of the LSTM model are automatically adjusted. After the parameters are adjusted, the LSTM model outputs updated procurement demand forecast data;

[0009] The preset procurement plan is automatically matched according to the updated procurement demand forecast data until the end of the forecast period, so as to obtain the final reference procurement plan within the current forecast period.

[0010] Furthermore, after obtaining the final reference procurement plan within this forecast period, supplier evaluation is performed through the supplier's comprehensive evaluation model to obtain multiple candidate suppliers that match the reference procurement plan and provide reference for the procurement department.

[0011] Furthermore, when evaluating suppliers through the supplier's comprehensive evaluation model, it includes: inputting supplier evaluation data into the supplier's comprehensive evaluation model, the supplier evaluation data including the supplier's delivery speed, product quality, cost and stability of previous cooperation; the supplier's comprehensive evaluation model conducts a comprehensive evaluation of the supplier based on the supplier evaluation data, and divides the suppliers into high-quality suppliers and risky suppliers, the high-quality suppliers being candidate suppliers matching the reference procurement plan.

[0012] Furthermore, after the supplier delivers goods, the supplier's supply status is monitored in real time. When an abnormal supply status occurs, the supplier evaluation data is modified according to the cause of the abnormal supply status, and the modified supplier evaluation data is input into the supplier's comprehensive evaluation model to re-evaluate the supplier comprehensively.

[0013] Furthermore, the cross-regional supply chain management method based on artificial intelligence is characterized in that it also includes: obtaining final procurement demand forecast data updated by the final deadline within the forecast period, final actual product sales data of the final deadline within the forecast period, and inventory data of the final deadline within the forecast period; dynamically adjusting the inventory level of each warehouse based on the inventory data, the final procurement demand forecast data, and the final actual product sales data to achieve on-demand allocation of inventory in each warehouse.

[0014] In a second aspect, the present invention provides an artificial intelligence-based cross-regional supply chain management system, which applies the above-mentioned artificial intelligence-based cross-regional supply chain management method.

[0015] Compared with the prior art, the present invention has the following beneficial effects:

[0016] The present invention provides a cross-regional supply chain management method and system based on artificial intelligence. The method and system obtain multi-dimensional procurement-related data by fusing historical sales data, market trend data, seasonal fluctuation data and external environment data related to product procurement, input the multi-dimensional procurement-related data into an LSTM model for procurement forecasting, obtain procurement demand forecast data within a forecast period, automatically match a preset procurement plan according to the procurement demand forecast data, continuously compare actual product sales data within the forecast period with the procurement demand forecast data through a reinforcement learning algorithm, evaluate the forecast deviation of procurement demand in real time, automatically adjust the parameters of the LSTM model when a sudden increase in demand or market fluctuation occurs, and output updated procurement demand forecast data after the parameters are adjusted. The preset procurement plan is automatically matched according to the updated procurement demand forecast data until the end of the forecast period, and the final reference procurement plan within the current forecast period is obtained, thereby reducing the procurement forecast deviation and flexibly responding to demand changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of the present invention. The exemplary embodiments of the present invention and their description are used to explain the present invention and do not constitute an improper limitation of the present invention. Some specific embodiments of the present invention will be described in detail in an exemplary and non-restrictive manner with reference to the drawings. The same reference numerals in the drawings indicate the same or similar components or parts. It should be understood by those skilled in the art that these drawings are not necessarily drawn to scale. In the drawings:

[0018] Figure 1 This is a flow chart of a cross-regional supply chain management method based on artificial intelligence according to an embodiment of the present invention;

[0019] Figure 2 It is a flow chart of the input data source of the LSTM model in the embodiment of the present invention;

[0020] Figure 3 It is a schematic diagram of the architecture of an artificial intelligence-based cross-regional supply chain management system according to an embodiment of the present invention. DETAILED DESCRIPTION

[0021] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only an embodiment of a part of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work should fall within the scope of protection of the present invention.

[0022] Embodiment 1

[0023] See also Figure 1-Figure 3 This embodiment provides a cross-regional supply chain management method based on artificial intelligence, comprising the following steps:

[0024] S101, fusing historical sales data, market trend data, seasonal fluctuation data, and external environment data related to product procurement to obtain multi-dimensional procurement-related data;

[0025] S102, inputting the multi-dimensional procurement-related data into the LSTM model for procurement forecasting, obtaining procurement demand forecast data within the forecast period, and automatically matching a preset procurement plan according to the procurement demand forecast data;

[0026] S103, using a reinforcement learning algorithm, continuously comparing the actual product sales data within the forecast period with the procurement demand forecast data, evaluating the forecast deviation of procurement demand in real time, and automatically adjusting the parameters of the LSTM model when there is a sudden increase in demand or market fluctuations, and the LSTM model outputs updated procurement demand forecast data after the parameters are adjusted;

[0027] S104. Automatically match the preset procurement plan according to the updated procurement demand forecast data until the end of the forecast period to obtain the final reference procurement plan within the current forecast period.

[0028] It should be noted that supply chain management needs to consider factors such as historical sales data, market trends, seasonal fluctuations, and external environment (such as economic policies, logistics conditions, etc.), which will have a complex impact on procurement demand. In step S101, multidimensional data fusion processing can more comprehensively capture the various factors affecting procurement and provide a reliable data basis for subsequent demand forecasting. For example, in the cross-regional supply chain management of an electronic product, historical sales data can show past sales trends; market trend data can reflect the market potential of new products; seasonal data can help predict holiday demand peaks; external environment data can consider changes in import and export policies, etc. The integration of this information can provide more accurate data input for the next LSTM model. The LSTM model (i.e., long short-term memory model) is an ideal model for processing time series data, suitable for capturing long-term and short-term trend changes in data. In step S102, by predicting procurement demand through the LSTM model, accurate prediction of future demand can be achieved based on multidimensional data, thereby reducing the risk of out-of-stock or overstock. Taking electronic products as an example, after the input multidimensional data is processed by LSTM, the product procurement demand in the next quarter can be predicted. Assuming that the forecast shows that demand will peak in December, the supply chain team can prepare in advance to ensure adequate supply during the peak period. In actual applications, changes in market demand are often random and unpredictable. In step S103, the actual sales data and LSTM forecast data can be compared in real time through the reinforcement learning algorithm to identify forecast deviations. When there is a surge in demand or market fluctuations, the parameters of the LSTM model are automatically adjusted to improve the adaptability of the model and the accuracy of the forecast. For example, assuming that in actual sales, a certain electronic product has a significant increase in sales due to a sudden increase in market publicity. The reinforcement learning algorithm can detect this change in time and dynamically adjust the parameters of the LSTM model to make the model more adaptable to new market trends and output updated forecast data. It can be understood that in step S103, the actual sales data is continuously compared with the procurement demand forecast of the LSTM model through the reinforcement learning algorithm to evaluate the forecast deviation in real time. The core goal of reinforcement learning is to adjust the LSTM model according to the size of the forecast error and optimize the forecast accuracy of the model. The input of the LSTM model comes from the current state space, which contains multiple factors that affect procurement demand, such as historical sales data, market trend data, and seasonal fluctuation data. These data together form the prediction basis of the LSTM model, and the LSTM model generates forecast data for procurement demand based on these inputs. In order to guide reinforcement learning to optimize the LSTM model, a reward function can be set. The reward function is related to the size of the prediction error, specifically the negative value of the prediction error. Among them, the prediction error refers to the difference between the actual sales data and the data predicted by the LSTM model. The smaller the error, the greater the reward, and vice versa, the larger the error, the smaller the reward.In this way, reinforcement learning forces the LSTM model to gradually reduce the forecast error, thereby improving the forecast accuracy. Demand surges and market fluctuations are important factors that reinforcement learning needs to consider when adjusting the LSTM model. If the actual sales volume exceeds a preset percentage (such as 30%) or more of the forecast demand in a certain period, it is considered a demand surge. If the sales fluctuation in a certain month exceeds a preset multiple (such as twice) of the standard deviation of that month, it is considered a demand surge. External environment (such as weather, macroeconomic changes, policy adjustments, etc.) may cause large fluctuations in market demand. When there is a large and persistent deviation between actual sales data and forecast data, especially when there are changes in the external environment, it can be considered a signal of market fluctuations. When there is a sudden increase in demand or market fluctuations, it usually leads to greater forecast deviations because the LSTM model may not be able to capture these sudden changes in time. Reinforcement learning ensures that the LSTM model can adapt to these dynamic changes by monitoring these deviations and making feedback adjustments. Under the feedback mechanism of reinforcement learning, the parameters of the LSTM model are dynamically adjusted according to real-time feedback. Specifically, the parameters that can be adjusted include: learning rate, number of hidden layer units, regularization parameter, batch size, etc. The learning rate determines the magnitude of each update of the model during the training process. By adjusting the learning rate, reinforcement learning can speed up or slow down the model training process to adapt to different types of data changes. The number of hidden layer units affects the expressiveness of the model and determines whether the model can capture long-term dependencies in the data. According to the complexity of the data, adjusting the size of the hidden layer can help the model better predict demand changes. Regularization is used to control the complexity of the LSTM model to avoid overfitting. Reinforcement learning optimizes the generalization ability of the model by dynamically adjusting the regularization parameter, so that it can maintain good prediction accuracy under various conditions. The batch size affects the efficiency and stability of gradient descent during training. Reinforcement learning adjusts the batch size according to the changes in the current data, thereby improving the efficiency and stability of the training process. During each training, the reinforcement learning algorithm adjusts the parameters of the LSTM model (such as learning rate, number of hidden layer units, regularization parameters, etc.) according to the current prediction error. These adjustments can help the LSTM model better adapt to sudden or fluctuating demand changes when predicting demand. When the LSTM model parameters are adjusted, it will be retrained to generate updated procurement demand forecast data. These updated forecast data will be used again to evaluate the forecast error and provide a basis for the next round of parameter adjustments. In step S104, the procurement plan is matched based on the updated forecast data to optimize the response speed and flexibility of the entire supply chain, which can reduce the lag of each link in the supply chain and avoid waste of resources or insufficient inventory due to inaccurate forecasts.For example, suppose the forecast is updated to show that demand will continue to rise in the coming months. At this time, the procurement plan is automatically matched and adjusted to make procurement more closely aligned with demand until the end of the forecast period, thereby ultimately providing an optimized procurement plan, reducing procurement forecast deviations, and flexibly responding to changes in demand.

[0029] In some further embodiments, when the historical sales data, market trend data, seasonal fluctuation data and external environment data related to product procurement are fused, it includes: collecting the historical sales data, market trend data, seasonal fluctuation data and external environment data related to product procurement, cleaning the collected data and aligning the time series; performing feature engineering on the data after cleaning and time series alignment, extracting the features of each data source, and splicing and fusing the features of each data source in time series to obtain the multi-dimensional procurement-related data. It should be noted that supply chain management needs to analyze and predict procurement needs from multiple perspectives, and these data are key factors affecting procurement needs. Historical sales data can reflect the sales trend of products, market trend data reveals market attention, seasonal fluctuation data can capture consumption changes in a specific period, and external environment data (such as economic and policy factors) affects the overall demand environment. The data structure, format and frequency of each data source may be inconsistent. For example, historical sales data may be recorded on a daily basis, while external environment data may be released on a monthly basis. Data cleaning helps to eliminate noise, outliers, and fill in missing data, while time series alignment ensures that the time dimension of different data sources is unified, making subsequent feature extraction more reliable. Extracting key features through feature engineering helps to transform the complex information of each data source into input variables that are easy for the model to understand. For example, in historical sales data, growth rate, sales volatility, etc. can be used as features, while in market trend data, the growth rate of discussion heat can be used as a feature. Extracting these features can condense information and make the predictions of subsequent models more representative. By splicing the features of each data source according to the time series, an input matrix containing time and multi-dimensional information can be formed, providing all-round information for subsequent prediction models. In this way, the model can simultaneously receive comprehensive information such as product sales history, market trends, and environmental factors, and make more accurate predictions.

[0030] In some further embodiments, in the multidimensional procurement-related data, different procurement forecast weights are assigned to the historical sales data, the market trend data, the seasonal fluctuation data, and the external environment data, and the procurement forecast weights of the historical sales data, the market trend data, the seasonal fluctuation data, and the external environment data are successively reduced. It should be noted that historical sales data usually reflects the long-term sales trend and short-term fluctuations of products most directly, which are important foundations for supply chain forecasting. It provides detailed information about past demand changes and can reveal the performance patterns of specific products under different conditions. Since the advantage of the LSTM model is to process time series data, historical sales data can help the model capture the cyclical characteristics and overall growth trend of sales, and is the most reliable data source for predicting future demand. Giving the highest weight to historical sales data can ensure that the LSTM model obtains a relatively stable forecast result. Market trend data reflects the focus of the current market and consumers and the dynamics of demand changes to a certain extent. Although market trends are relatively short-term, their impact on demand is usually more direct. For example, the discussion heat on social media or other network platforms can often reflect consumers' willingness to buy, which may cause demand changes in the short term. Giving the second highest weight to market trend data can help the LSTM model respond quickly to these short-term changes. Seasonal fluctuation data can reveal the cyclical changes in demand (such as quarterly, monthly or holiday demand), and this type of information can improve the accuracy of the forecast. Among them, seasonal fluctuation data is very helpful for capturing cyclical demand patterns, but it is a more regular and stable factor, so it does not require a higher weight, but as an auxiliary feature, it ensures that the LSTM model can identify regular demand fluctuations. External environmental data (such as economic conditions, policy changes, etc.) are indirect factors affecting demand. External environmental data usually have a relatively macro impact. Although they will indirectly change consumers' purchasing power or preferences, their impact on specific product demand is longer and the impact is more difficult to quantify. Therefore, the lowest weight should be given in the purchase forecast, which is mainly used to supplement and adjust the forecast in special situations, and is not used as the main reference factor in conventional forecasts. In this embodiment, by giving historical sales data, market trend data, seasonal fluctuation data and external environmental data successively reduced purchase forecast weights, the LSTM model can give priority to the most reliable data source (such as historical sales data), supplemented by trend, seasonality and external environmental data, and ultimately improve the accuracy and adaptability of purchase forecasts. Under this configuration, the LSTM model is able to more accurately capture long-term trends, short-term fluctuations, and cyclical changes in demand, while retaining appropriate responsiveness to unpredictable changes in the external environment.

[0031] In some preferred embodiments, after completing a procurement process according to the final reference procurement plan, the actual procurement data is fed back to the LSTM model, and the LSTM model adjusts the prediction error according to the actual procurement data to gradually optimize the prediction accuracy of the model. It should be noted that the actual procurement data is the most real and direct reflection of demand. By feeding the actual data back to the model, LSTM can identify the deviation between the prediction result and the actual situation, and then optimize the prediction error. Market demand and supply chain conditions are dynamically changing, and many external and internal factors will affect the actual procurement demand. For example, changes in consumer preferences, emergencies, or adjustments to market policies. By continuously feeding back the actual procurement data to the model, LSTM can learn these dynamic changes. The feedback mechanism ensures that the model gradually adapts to new market trends and demand patterns, and has stronger adaptability and prediction accuracy.

[0032] In some preferred embodiments, the historical sales data and the seasonal fluctuation data are obtained from the data center of the local server, and the market trend data and the external environment data are captured from the Internet platform through a Web crawler program. The historical sales data include sales quantity, sales time, sales region, sales channel and customer category; the seasonal fluctuation data include demand fluctuation data affected by holidays, demand fluctuation data affected by seasonal changes and demand fluctuation data affected by specific months; the market trend data include industry growth rate data, market competition data, consumption preference change data and consumption capacity change data; the external environment data includes macroeconomic data, policy and regulatory data, natural disaster data and weather change data. It should be noted that historical sales data and seasonal fluctuation data are important data within the enterprise, which are usually highly confidential and sensitive. Therefore, saving these data on the local server of the enterprise can better control data access rights and ensure data security. At the same time, these data are stable data accumulated over a long period of time, and local storage can ensure their efficient access and call. Historical sales data (such as sales quantity, time, region, channel, customer category) is the data actually recorded by the enterprise in the past, so its accuracy is relatively high. These data directly reflect the sales of products in different periods and different markets, and are the core data source for procurement forecasts. Seasonal fluctuation data (such as demand fluctuations during holidays, seasons, and specific months) are also obtained based on years of historical data analysis, and are suitable for storage and management in local data centers, which are convenient for regular updates and maintenance. For example, an enterprise has accumulated sales data for multiple holidays in the past few years. These data are stored in local servers and can be called at any time to provide direct data support for predicting demand during specific holidays. Market trend data and external environment data (such as industry growth rate, market competition, consumer preferences, macroeconomics, policies and regulations, etc.) are highly real-time and dynamic, and often need to be updated frequently. By crawling these data from the Internet platform through Web crawlers, the latest market dynamics and external environment information can be obtained. The real-time nature of external data plays a key role in capturing changes in market trends and emergencies in the macro environment (such as policy changes or natural disasters). Market trend data and external environment data come from a wide range of sources, which may involve multiple platforms such as industry reports, news, and social media. Through Web crawlers, enterprises can obtain the latest and extensive data from these external platforms to form a comprehensive understanding of the market environment, thereby enhancing the flexibility and adaptability of procurement forecasts. For example, by obtaining the latest industry growth rate data through web crawlers, the model can identify the overall upward trend of the industry, thereby increasing the demand expectations for the product when making forecasts; and obtaining weather change data from weather stations can predict the potential impact of severe weather on logistics and supply chains.

[0033] It should also be noted that historical sales data includes data such as sales volume, sales time, sales region, sales channel, and customer category. These data directly reflect past sales characteristics and help the model understand the performance of products in different markets, channels, and customer groups. For example, the peak demand period in specific regions and channels can be used as an important reference for procurement plans. Seasonal fluctuation data includes data such as demand fluctuations in holidays, seasons, and specific months. Through these data, periodic patterns can be discovered and the prediction accuracy of the prediction model in different seasons and specific months can be improved. Among them, specific months include the month when students start school. Market trend data includes data such as industry growth rate, market competition, changes in consumer preferences, and changes in consumer capacity. These data can reveal the current market demand direction and competition, and help companies make procurement decisions in sync with the industry. For example, when the industry growth rate increases or consumer preferences shift to a certain type of product, the model can adjust the procurement demand forecast accordingly. External environment data includes data such as macroeconomic data, policies and regulations, natural disasters, and weather changes. These data can reflect a wider range of economic and environmental impacts, especially in special circumstances, when changes in macroeconomics and policies will directly affect demand. For example, an economic recession or the promulgation of new policies may lead to a sudden drop in consumer demand, while natural disasters and weather changes may affect the stability of logistics and supply chains. It should be noted that in this embodiment, through the combination of local data and external data, the LSTM model is able to find a balance between data reliability and real-time performance. Local data provides a stable basic support, while external data captured by web crawlers provides the model with the latest market information and environmental impact data. The combination of the two makes the forecasting model both historically stable and dynamically adaptable, thereby improving the accuracy and flexibility of procurement forecasts.

[0034] In some further preferred embodiments, in the historical sales data, different purchase forecast weights are assigned to the sales quantity, the sales time, the sales region, the sales channel and the customer category, and the purchase forecast weights of the sales quantity, the sales time, the sales region, the sales channel and the customer category are successively reduced. It should be noted that the sales quantity is the most critical data that directly reflects the demand for the product. In the purchase forecast, the sales quantity represents the actual sales volume in a certain period of time in the past, and is the most direct and important reference indicator for determining the demand. By analyzing the trend of the sales quantity, the model can more accurately predict the future purchase demand. Therefore, the sales quantity is set to the highest weight. The sales time is an important time dimension information that can reveal the sales periodicity and trend of the product. By analyzing the sales time, the model can identify seasonal or cyclical demand patterns (such as monthly, quarterly, and annual cycles), and take into account the impact of time factors on demand when forecasting. Therefore, the sales time is set to the second highest weight. The demand in different regions varies greatly, and regional factors play a role in positioning and market segmentation in purchase forecasting. The information of the sales region can help the model identify the impact of geographical factors on demand, such as consumption levels and cultural preferences in different regions. Therefore, the sales region is set to the third highest weight. Sales channels (such as online, offline, specific retailers, etc.) affect how products are sold, and channel information reflects customer preferences and demand in different channels to a certain extent. Some channels may drive sales growth for specific products (such as online promotions), but their impact is relatively indirect. Therefore, sales channels are set to the fourth highest weight. Customer category (such as age group, occupation, purchase frequency, etc.) has an indirect impact on purchasing demand. It reflects the characteristics of the product's audience, but usually does not directly determine the overall demand. Customer category information can help the model make more personalized predictions in the market segment, but it has little impact on the overall purchasing forecast. Therefore, customer category is set to the lowest weight.

[0035] In some preferred embodiments, after obtaining the final reference procurement plan within the current forecast period, the supplier evaluation is performed through the supplier's comprehensive evaluation model to obtain multiple candidate suppliers matching the reference procurement plan, providing a reference for the procurement department. Further, when evaluating the supplier through the supplier's comprehensive evaluation model, it includes: inputting supplier evaluation data into the supplier's comprehensive evaluation model, the supplier evaluation data includes the supplier's delivery speed, product quality, cost and stability of previous cooperation; the supplier's comprehensive evaluation model conducts a comprehensive evaluation of the supplier based on the supplier evaluation data, and divides the supplier into high-quality suppliers and risk suppliers, and the high-quality suppliers are candidate suppliers matching the reference procurement plan. It should be noted that the supplier's comprehensive evaluation model aims to evaluate the capabilities of different suppliers in a systematic way, and conduct quantitative analysis from multiple dimensions to help enterprises make the best choice. The supplier's comprehensive evaluation model can objectively reflect the comprehensive strength of the supplier, especially when choosing between multiple suppliers, it can reduce subjective judgment and human bias, and improve the scientificity and accuracy of decision-making. In addition, the supplier's comprehensive evaluation model can ensure that all suppliers participating in the supply chain have undergone unified evaluation standards and can truly reflect their performance in different key aspects. In this way, the purchasing department can quickly screen out the supplier that best meets the needs based on the various capabilities of the supplier, avoiding wrong decisions caused by single-dimensional judgment. In addition, in this embodiment, by using multiple key evaluation indicators of the supplier (such as delivery speed, product quality, cost, and cooperation stability), the comprehensive evaluation model can more comprehensively measure the reliability, quality, and cooperation potential of the supplier. Each indicator plays an important role in purchasing decisions, ensuring that the evaluation does not only focus on price or delivery speed, but comprehensively considers the sustainability of long-term cooperation. Delivery speed is one of the key factors for the smooth implementation of the procurement plan. The supplier's delivery capability directly affects the company's production plan, inventory management, and product launch speed. In the supply chain, delayed delivery may lead to production line shutdowns, customer loss, and other problems, so delivery speed is very important as a core evaluation indicator. Product quality is a crucial factor in purchasing decisions, which directly affects the final quality of the product, customer satisfaction, and the brand reputation of the company. If the supplier's product quality cannot be stably guaranteed, it will lead to an increase in the company's return rate, an increase in customer complaints, and even affect the company's long-term development. Cost is one of the core factors of corporate procurement decisions. Although high-quality suppliers usually offer higher prices, if the product quality and delivery speed they provide are more in line with demand, they may bring greater value. From a cost perspective, evaluating the rationality of suppliers' quotations, cost-effectiveness, and price fluctuations in long-term cooperation can ensure that the budget of the procurement plan is controlled within a reasonable range. The supplier's historical performance and cooperation stability directly affect the quality and risk of future cooperation.A stable supplier usually has a stronger ability to fulfill contracts, can better cope with market fluctuations and changes in demand, and reduce uncertainties in the supply chain. Suppliers with a good history of cooperation can better maintain long-term cooperative relationships with enterprises and reduce the probability of risky suppliers. Therefore, the stability of past cooperation reflects the credibility and reliability of suppliers, which can help the procurement department avoid choosing suppliers that have had performance problems or frequent changes in the past, and ensure the smoothness and continuity of the supply chain. In this embodiment, by dividing suppliers into high-quality suppliers and risky suppliers, it is helpful to quickly screen alternative suppliers that meet the reference procurement plan. Through this classification, the procurement department can clearly identify which suppliers are reliable partners for long-term cooperation in actual operations, and which suppliers may bring risks and need to be carefully selected in specific circumstances. For high-quality suppliers, the procurement department can give more orders and cooperation opportunities based on their past performance, while for risky suppliers, the number of orders may be limited or alternative suppliers may be found.

[0036] In some further preferred embodiments, in the comprehensive evaluation model of the supplier, the product quality, the delivery speed, the cost and the stability of the previous cooperation are given evaluation weights that decrease in sequence. It should be noted that product quality is one of the most core indicators in supplier evaluation. Regardless of the delivery speed, cost or cooperation stability, product quality is always the first priority in supplier selection. If the product quality is not up to standard, it may eventually lead to serious consequences such as returns, rework, customer loss, and even affect the company's brand image and customer satisfaction. Therefore, in the supplier evaluation model, giving the product quality the highest weight can ensure that the product can meet the quality standards and avoid a series of risks caused by quality problems. Delivery speed directly affects production progress, inventory management, customer delivery time, etc., and is a key factor in ensuring a smooth supply chain. If a supplier fails to deliver on time, it may hinder the entire production plan, thereby affecting the company's operations and customer satisfaction. Therefore, delivery speed is given the second highest weight to ensure that the procurement plan can be implemented smoothly. Cost is a very important consideration for enterprises in procurement decisions. However, when selecting suppliers, cost is often considered on the premise of meeting product quality and delivery requirements. Although low cost is the goal of enterprises to optimize procurement budgets, excessive focus on cost may cause enterprises to ignore product quality and delivery reliability, resulting in higher indirect costs. Therefore, the weight of cost ranks third. Although the stability of previous cooperation can reflect the supplier's fulfillment ability and historical performance, its influence is usually lower than product quality, delivery speed and cost. Therefore, the weight of stability of previous cooperation ranks lowest.

[0037] In some further preferred embodiments, in the comprehensive evaluation model of the supplier, the qualified line standards are preset for the product quality, the delivery speed, the cost and the stability of the previous cooperation. When the comprehensive evaluation model of the supplier is evaluated, the qualified lines of the product quality, the delivery speed, the cost and the stability of the previous cooperation are evaluated first according to the evaluation weights of the product quality, the delivery speed, the cost and the stability of the previous cooperation from high to low. When the evaluation data of the supplier with high evaluation weight is unqualified, the evaluation data of the supplier with low weight is stopped, and the supplier exclusion prompt is output. It should be noted that when evaluating multiple suppliers, directly evaluating all dimensions may consume a lot of computing resources and time. By first screening according to high-weight indicators (such as product quality and delivery speed), once it is found that the supplier is unqualified in the key indicators of high priority, the supplier can be directly excluded to avoid unnecessary subsequent evaluations (such as cost and historical stability) on it, thereby reducing redundant operations in the calculation process and improving calculation efficiency. In addition, the evaluation of suppliers is usually a complex process involving scoring and weighting of multiple dimensions. By evaluating in descending order of evaluation weights and checking the most critical evaluation dimensions first, suppliers that clearly do not meet the standards can be quickly screened out, thereby improving the efficiency of the entire evaluation process. In actual supply chain management, ensuring that suppliers perform well in key areas is more important than their performance in other secondary areas. If a supplier has problems in high-priority areas (such as quality and delivery), it cannot be considered a suitable partner even if it performs well in other dimensions. Therefore, evaluating high-priority indicators first and promptly excluding suppliers that do not meet the standards will help reduce the risk of selecting inappropriate suppliers.

[0038] In some preferred embodiments, after the supplier supplies, the supplier's supply status is monitored in real time. When an abnormal supply status occurs, the supplier evaluation data is modified according to the cause of the abnormal supply status, and the modified supplier evaluation data is input into the supplier's comprehensive evaluation model to re-evaluate the supplier comprehensively. It should be noted that suppliers may perform differently in different supply cycles, and some sudden abnormal supply status may reveal the supplier's potential risks or instability in its operations. Therefore, real-time monitoring of the supplier's supply status and modifying the supplier evaluation data according to the abnormal cause can help the system dynamically adjust the supplier evaluation and ensure that the evaluation results reflect the current true supplier performance. Among them, supply anomalies may include quality problems, delivery delays, quantity errors, etc. These abnormal events may be only temporary or may reflect the supplier's long-term risks. If these abnormal events are not adjusted in a timely manner, the supplier's evaluation results may be distorted, leading to wrong procurement decisions.

[0039] In some preferred embodiments, the cross-regional supply chain management method based on artificial intelligence also includes: obtaining the final purchase demand forecast data updated by the final deadline during the forecast period, the final actual product sales data of the final deadline during the forecast period, and the inventory data of the final deadline during the forecast period; dynamically adjusting the inventory level of each warehouse according to the inventory data, the final purchase demand forecast data, and the final actual product sales data to achieve on-demand inventory allocation of each warehouse. Further, the final purchase demand forecast data is obtained from the LSTM model, the final actual product sales data is obtained from the sales management system of the local server data center, and the inventory data is obtained from the warehouse management system of the local server data center. It should be noted that inventory management is one of the core links in supply chain management, and accurate inventory management can help enterprises avoid overstocking or out-of-stock, thereby achieving cost control and supply chain efficiency optimization. By dynamically adjusting the inventory level of each warehouse, timely adjustments can be made according to actual demand and sales conditions to avoid resource waste or supply delays caused by static inventory management strategies. In this embodiment, the LSTM model can provide more accurate demand forecasts by learning multidimensional data such as historical sales data and market trends. Combining actual sales data and inventory data, the inventory level can be further accurately adjusted to ensure that the inventory of each warehouse can be allocated on demand. The final actual product sales data and inventory data during the forecast period are all from the local server data center, which can reflect the actual operation status of the current supply chain. Therefore, combining these real-time data to dynamically adjust the inventory level can ensure that inventory management is more in line with actual needs and avoid relying on the errors of a single forecasting model (such as LSTM). Allocation on demand is one of the key goals in cross-regional supply chain management. By predicting demand and real-time sales data, the inventory of each warehouse is dynamically adjusted to ensure that the inventory level can meet the actual needs of different regions and markets without causing unnecessary waste or shortages.

[0040] Embodiment 2

[0041] See also Figure 1-Figure 2The present embodiment provides an artificial intelligence-based cross-regional supply chain management system, which runs on a server. The artificial intelligence-based cross-regional supply chain management system adopts the artificial intelligence-based cross-regional supply chain management method described in any of the above embodiments, and obtains multi-dimensional procurement-related data by fusing historical sales data, market trend data, seasonal fluctuation data, and external environment data related to product procurement. The multi-dimensional procurement-related data is input into the LSTM model for procurement forecasting to obtain procurement demand forecast data within the forecast period, and a preset procurement plan is automatically matched according to the procurement demand forecast data. Through a reinforcement learning algorithm, the actual product sales data within the forecast period is continuously compared with the procurement demand forecast data, and the forecast deviation of procurement demand is evaluated in real time. When there is a sudden increase in demand or market fluctuations, the parameters of the LSTM model are automatically adjusted. After the parameters are adjusted, the LSTM model outputs updated procurement demand forecast data, and the preset procurement plan is automatically matched according to the updated procurement demand forecast data until the end of the forecast period, so as to obtain the final reference procurement plan within the current forecast period, thereby reducing the procurement forecast deviation and flexibly responding to demand changes. The historical sales data and the seasonal fluctuation data are obtained from the data center of the local server, and the market trend data and the external environment data are captured from the Internet platform through a Web crawler program.

[0042] It should be pointed out that the above embodiments are only preferred specific implementation modes of the present invention, and the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention. The protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A cross-regional supply chain management method based on artificial intelligence, characterized in that: include: The historical sales data, market trend data, seasonal fluctuation data and external environment data related to product procurement are integrated and processed to obtain multi-dimensional procurement-related data; Input the multi-dimensional procurement-related data into the LSTM model for procurement forecasting, obtain procurement demand forecast data within the forecast period, and automatically match the preset procurement plan according to the procurement demand forecast data; By using a reinforcement learning algorithm, the actual product sales data within the forecast period is continuously compared with the procurement demand forecast data, and the forecast deviation of procurement demand is evaluated in real time. When there is a sudden increase in demand or market fluctuations, the parameters of the LSTM model are automatically adjusted. After the parameters are adjusted, the LSTM model outputs updated procurement demand forecast data; The preset procurement plan is automatically matched according to the updated procurement demand forecast data until the end of the forecast period, so as to obtain the final reference procurement plan within the current forecast period.

2. The cross-regional supply chain management method based on artificial intelligence as claimed in claim 1, characterized in that: After completing a procurement process according to the final reference procurement plan, the actual procurement data is fed back to the LSTM model, and the LSTM model adjusts the prediction error according to the actual procurement data to gradually optimize the prediction accuracy of the model.

3. The cross-regional supply chain management method based on artificial intelligence as claimed in claim 1, characterized in that: The historical sales data and the seasonal fluctuation data are obtained from the data center of the local server, and the market trend data and the external environment data are captured from the Internet platform through a Web crawler program.

4. The cross-regional supply chain management method based on artificial intelligence as claimed in claim 1, characterized in that: The historical sales data includes sales quantity, sales time, sales region, sales channel and customer category; the seasonal fluctuation data includes demand fluctuation data affected by holidays, demand fluctuation data affected by changes of four seasons and demand fluctuation data affected by specific months; the market trend data includes industry growth rate data, market competition data, consumption preference change data and consumption capacity change data; the external environment data includes macroeconomic data, policy and regulatory data, natural disaster data and weather change data.

5. The cross-regional supply chain management method based on artificial intelligence as claimed in claim 1, characterized in that: After obtaining the final reference procurement plan during this forecast period, the supplier evaluation is carried out through the supplier's comprehensive evaluation model to obtain multiple candidate suppliers that match the reference procurement plan and provide reference for the procurement department.

6. The cross-regional supply chain management method based on artificial intelligence as claimed in claim 5, characterized in that: When evaluating suppliers through the supplier's comprehensive evaluation model, it includes: inputting supplier evaluation data into the supplier's comprehensive evaluation model, the supplier evaluation data including the supplier's delivery speed, product quality, cost and stability of previous cooperation; the supplier's comprehensive evaluation model conducts a comprehensive evaluation of the supplier based on the supplier evaluation data, and divides the suppliers into high-quality suppliers and risky suppliers, the high-quality suppliers being candidate suppliers matching the reference procurement plan.

7. The cross-regional supply chain management method based on artificial intelligence as claimed in claim 6, characterized in that: After the supplier delivers goods, the supplier's supply status is monitored in real time. When an abnormal supply status occurs, the supplier evaluation data is modified according to the cause of the abnormal supply status, and the modified supplier evaluation data is input into the supplier's comprehensive evaluation model to re-evaluate the supplier comprehensively.

8. The cross-regional supply chain management method based on artificial intelligence as claimed in claim 1, characterized in that: Also includes: Obtain the final purchase demand forecast data updated by the final deadline within the forecast period, the final actual product sales data by the final deadline within the forecast period, and the inventory data by the final deadline within the forecast period; The inventory level of each warehouse is dynamically adjusted according to the inventory data, the final purchase demand forecast data, and the final actual product sales data to achieve on-demand allocation of inventory in each warehouse.

9. The cross-regional supply chain management method based on artificial intelligence as claimed in claim 8, characterized in that: The final purchase demand forecast data is obtained from the LSTM model, the final actual product sales data is obtained from the sales management system of the local server data center, and the inventory data is obtained from the warehouse management system of the local server data center.

10. A cross-regional supply chain management system based on artificial intelligence, characterized in that: The cross-regional supply chain management system based on artificial intelligence applies the cross-regional supply chain management method based on artificial intelligence as described in any one of claims 1-9.

Images