Supply chain data processing method and device, equipment and storage medium

Abstract

The invention provides a supply chain data processing method, device and equipment and a storage medium, and the method comprises the steps: enabling a preset first agent, a preset second agent and a preset third agent to correspond to manufacturing, selling and recycling links respectively, and determining the dynamic processing logic of input data and output data, efficient data interaction and cooperation among links of the supply chain are formed, and the problem that the data utilization efficiency in a traditional supply chain is low is solved. Key data such as market demand prediction data, historical sales data and real-time sales data are transmitted to corresponding agents in real time, so that accurate calculation and dynamic adjustment of a production plan, sales prediction and a recovery plan are realized, and the phenomena of information delay and data islands are reduced; therefore, the overall operation efficiency and response capability of the supply chain are remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to a supply chain data processing method, apparatus, device, and storage medium. Background Art

[0002] In today's rapidly developing market environment, the life cycle management of home appliance products has become increasingly important. Home appliance products usually go through multiple links such as manufacturing, sales, recycling, disassembly, and reuse. Effective supply chain management can improve resource utilization efficiency, reduce costs, and promote sustainable development. However, traditional supply chain management methods often rely on manual judgment and experience, which lead to problems such as extended response time, information asymmetry, and insufficient decision-making accuracy. Therefore, there is an urgent need for an intelligent solution to improve the overall efficiency and coordination ability of the home appliance product supply chain.

[0003] Although traditional supply chain management has achieved certain results in some aspects, it has problems such as long response time, information asymmetry, and inaccurate decision-making. Due to relying on manual judgment, traditional supply chain management reacts slowly in the face of market changes. Information transmission between the manufacturing, sales, and recycling links often experiences delays, resulting in enterprises being unable to adjust production and sales strategies in a timely manner. Insufficient information sharing among the various links of the supply chain leads to differences in the understanding of information such as market demand, inventory status, and production capacity among different participants. This information asymmetry makes it difficult for all parties to form effective cooperation and increases the complexity of decision-making. Traditional management methods often make decisions based on historical data and experience, lacking the support of real-time data analysis. This affects decision-making and further affects the efficiency and effectiveness of the overall supply chain.

[0004] Therefore, in the scenario of supply chain management, there is a technical problem of low data utilization efficiency for the data of each link. Summary of the Invention

[0005] The purpose of the present invention is to overcome the above technical deficiencies and provide a supply chain data processing method, apparatus, device, and storage medium to solve the technical problem of low data utilization efficiency for the data of each link in the scenario of supply chain management.

[0006] To achieve the above technical objectives, the present invention adopts the following technical solutions:

[0007] In the first aspect, the present invention provides a supply chain data processing method, which at least includes:

[0008] A preset first intelligent agent receives first input data and outputs first output data; wherein, the first intelligent agent is an intelligent agent representing the manufacturing link in the supply chain, the first input data at least includes market demand prediction data, and the first output data at least includes production plan prediction data;

[0009] A preset second intelligent agent receives second input data and outputs second output data; wherein, the first intelligent agent is an intelligent agent representing the sales link in the supply chain, the second input data at least includes historical sales data, and the second output data at least includes the market demand prediction data;

[0010] A preset third intelligent agent receives third input data and outputs third output data; wherein, the third intelligent agent is an intelligent agent representing the recycling link in the supply chain, the third input data at least includes real-time sales data, and the third output data at least includes recycling plan data.

[0011] In a second aspect, the present invention provides a supply chain data processing device, which at least includes:

[0012] A first intelligent agent module, which is used to receive first input data and output first output data; wherein, the first intelligent agent is an intelligent agent representing the manufacturing link in the supply chain, the first input data at least includes market demand prediction data, and the first output data at least includes production plan prediction data;

[0013] A second intelligent agent module, which is used to receive second input data and output second output data; wherein, the first intelligent agent is an intelligent agent representing the sales link in the supply chain, the second input data at least includes historical sales data, and the second output data at least includes the market demand prediction data;

[0014] A third intelligent agent module, which is used to receive third input data and output third output data; wherein, the third intelligent agent is an intelligent agent representing the recycling link in the supply chain, the third input data at least includes real-time sales data, and the third output data at least includes recycling plan data.

[0015] In a third aspect, the present invention provides an electronic device, including: a memory, and one or more processors communicatively connected to the memory; instructions executable by the one or more processors are stored in the memory, and when the instructions are executed by the one or more processors, the one or more processors are caused to implement the above method.

[0016] Fourthly, the present invention provides a computer-readable storage medium, in which a computer program is stored, and when the computer program is executed by a processor, the above-mentioned method is implemented.

[0017] Beneficial effects:

[0018] In the present invention, a preset first intelligent agent, second intelligent agent, and third intelligent agent respectively correspond to the manufacturing, sales, and recycling links, and by clarifying the dynamic processing logic of input data and output data, efficient data interaction and collaboration among all links of the supply chain are formed, solving the problem of low data utilization efficiency in traditional supply chains. By real-time transmitting key data such as market demand forecast data, historical sales data, and real-time sales data to the corresponding intelligent agents, accurate calculation and dynamic adjustment of production plans, sales forecasts, and recycling plans are achieved, reducing information delay and data island phenomena, thereby significantly improving the overall operation efficiency and response ability of the supply chain. Description of the drawings

[0019] Figure 1 is a schematic flowchart of a supply chain data processing method provided by an embodiment of the present invention;

[0020] Figure 2 is a schematic flowchart of a supply chain data processing method provided by an embodiment of the present invention;

[0021] Figure 3 is a schematic flowchart of a supply chain data processing method provided by an embodiment of the present invention;

[0022] Figure 4 is a supply volume process relationship diagram provided by an embodiment of the present invention;

[0023] Figure 5 is a schematic flowchart of the workflow between AI agents provided by an embodiment of the present invention;

[0024] Figure 6 is a block diagram of a supply chain data processing device adopted by an embodiment of the present invention;

[0025] Figure 7 is a block diagram of an electronic device adopted by an embodiment of the present invention. Detailed implementation manners

[0026] In order to enable those skilled in the art to better understand the solutions of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0027] In the related art, in modern manufacturing and retail environments, the supply chain management of household electrical appliance products faces rapidly changing market demands and diverse consumer preferences. With the development of technological progress and globalization, enterprises need to respond more agilely to market changes to ensure that products can reach consumers in a timely and effective manner. At the same time, consumers' expectations for product quality and service are also constantly increasing, which requires enterprises to maintain high standards of product quality and customer experience while achieving rapid delivery. The complex market environment makes the traditional supply chain management model inefficient and urgently needs transformation and upgrading.

[0028] However, traditional supply chain management methods are already difficult to cope with current challenges. The rapid changes in market dynamics have led to difficulties in resource allocation and process coordination for enterprises, and it is often impossible to achieve the best operational efficiency. In the manufacturing, sales, recycling, disassembly, and reuse of household electrical appliance products, any lag in any link will affect the operation of the entire supply chain. This not only increases operating costs but may also lead to a decline in customer satisfaction. When faced with a large amount of data, enterprises are difficult to analyze effectively and lack the ability to extract valuable information from the data, further affecting the accuracy and timeliness of decision-making.

[0029] Therefore, this embodiment proposes a supply chain data processing method, aiming to achieve efficient resource management and decision support by intelligently analyzing and matching each link in the supply chain. This method uses machine learning and data analysis technologies to monitor the data of links such as manufacturing, sales, recycling, disassembly, and reuse in real time, and automatically identify and predict market demands. By constructing multi-level AI agents, the information of each link can be seamlessly connected to ensure the coordinated operation of the supply chain, improve the flexibility and response ability of enterprises when market demands change, and achieve the goal of sustainable development.

[0030] As Figure 1 shown, this embodiment provides a supply chain data processing method, and the method may include:

[0031] Step S12: A preset first intelligent agent receives first input data and outputs first output data; wherein, the first intelligent agent is an intelligent agent representing the manufacturing link in the supply chain link, the first input data at least includes market demand prediction data, and the first output data at least includes production plan prediction data.

[0032] In this embodiment, the first intelligent agent may be configured with a first AI model. The first AI model may be a prediction model based on machine learning or a prediction model based on deep learning. Specifically, the first AI model may be an LSTM network combined with time series analysis or a reinforcement learning model.

[0033] In this embodiment, the first input data may include market demand forecast data. This market demand forecast data may further include: product models, forecast demand quantities, delivery times, and so on.

[0034] In this embodiment, the first input data may also include inventory status data. For example, the current inventory levels of raw materials and semi-finished products.

[0035] In this embodiment, the first input data may also include reusable data. For example, the quantity and quality assessment results of reusable components.

[0036] In this embodiment, the first agent may perform data cleaning and normalization on the received first input data. For example, handling missing values, outliers, and standardizing data from different sources, and so on.

[0037] In this embodiment, the input of the first AI model may be market demand forecast data, or it may be market demand forecast data, inventory status data, and reusable data. The output of the first AI model may be production plan forecast data. This production plan forecast data may be production batches (the production quantity and priority of each product), production times (the specific schedule of production tasks), and production times (the specific schedule of production tasks).

[0038] In this embodiment, the first agent may also be configured with a first policy engine. The first policy engine may generate corresponding production strategies based on the market demand forecast data, inventory status data, and reusable data. Therefore, the first output data may also include production strategy data. This first policy engine can provide a flexible set of production strategies in addition to the production plan forecast data, ensuring an efficient and responsive manufacturing process while coping with potential resource constraints, market changes, and emergencies.

[0039] Specifically, this production strategy may be a production priority strategy. That is, according to the market demand forecast data, priorities are assigned to each product. For example, products with high demand or high profit margins are given priority.

[0040] Specifically, this production strategy may also be a resource allocation strategy. For example, it is possible to determine the allocation method of raw materials and reusable components in different production tasks, ensuring the priority use of inventory resources and maximizing the benefits of reusable components.

[0041] More specifically, multiple policy rules can be preset in the first policy engine for quickly executing known business logics. For example, "Policy Rule 1: When the inventory status is lower than the threshold, preferentially use reusable components for replacement." Another example is, "Policy Rule 2: When the demand forecast shows a sharp increase in the demand for a certain product, dynamically increase the production priority of this product." Correspondingly, an optimization algorithm can also be preset in the first policy engine for generating dynamic and complex production strategies. For example, the dynamic programming algorithm.

[0042] The first policy engine first receives the market demand forecast data, inventory status data, and reusable data. Then it executes the policy rules to trigger the policy rules that match the current input data (such as priority adjustment, resource scheduling). Next, it calls the optimization algorithm to generate specific production strategy data based on the current market demand, inventory status, and production capacity. Finally, it outputs the production strategy data.

[0043] Step S14: The preset second agent receives the second input data and outputs the second output data; wherein, the first agent is an agent representing the sales link in the supply chain, the second input data at least includes historical sales data, and the second output data at least includes the market demand forecast data.

[0044] In this embodiment, the second agent can be configured with a second AI model. The second AI model can be a prediction model based on machine learning or a prediction model based on deep learning. Specifically, the second AI model can be a bidirectional LSTM model, which has the ability to learn from both past and future contexts, enabling the model to more comprehensively understand time series data. The bidirectional LSTM model sets the number of hidden layer units to 12 and the number of layers to 3 to balance the complexity of the model and the training efficiency. The mean squared error (MSE) is used as the loss function to effectively measure the difference between the predicted value and the actual value. At the same time, the Adam optimizer is selected to update the model parameters to improve the training speed and performance.

[0045] In this embodiment, the second AI model can also be a random forest regression model, a reinforcement learning model, or a prediction model based on a graph neural network (GNN), etc.

[0046] In this embodiment, the second input data can also include real-time sales data.

[0047] In this embodiment, the second output data can also include real-time sales analysis data.

[0048] In this embodiment, the second input data may further include consumer behavior data, market competition data, and so on. The second output data may further include market competition analysis data, and so on.

[0049] In this embodiment, the second agent may extract historical sales data from the supply chain database, and the historical sales data may include sales records of different product models, seasonal variations, regional distributions, and so on. The second agent may obtain the latest real-time sales data from the front-end sales system. For example, the order quantity, sales amount, and customer behavior feedback. After receiving the input data, the second agent may perform data cleaning. For example, cleaning missing values and outliers to ensure data quality. Then, the historical sales data may be used as training data to train the second AI model. For example, the historical sales data may be split into a training set and a validation set for training a bidirectional LSTM model. Then, using the trained bidirectional LSTM model and inputting the real-time sales data, market demand prediction data may be generated. The market demand prediction data may be data representing the predicted demand for different product models in a future period of time. For example, it may include information such as the prediction time range, regional distribution, and so on.

[0050] In this embodiment, the second agent may be configured with a second analysis engine. Specifically, the second analysis engine may perform dynamic analysis on the latest real-time sales data, extract key indicators, and form real-time sales analysis data. The second analysis engine may compare the real-time sales data with the historical sales data, analyze abnormal fluctuations or growth points, and form real-time sales analysis data. The second analysis engine may also extract core sales features such as the sales volume change rate, sales trend, and regional distribution from the real-time sales data to provide support for the second AI model.

[0051] In this embodiment, the second analysis engine may include preset statistical analysis algorithms, clustering analysis algorithms, or anomaly detection algorithms, and so on. For example, it may include algorithms for calculating the mean, variance, year-on-year growth rate, and grouping and clustering algorithms for regional sales data, and so on.

[0052] In this embodiment, the output data may be transmitted to other agents (such as manufacturing agents and recycling agents) through a message queue or API for dynamic adjustment and optimization of the supply chain.

[0053] Step S16: A preset third agent receives third input data and outputs third output data; wherein, the third agent is an agent representing the recycling link in the supply chain, the third input data includes at least real-time sales data, and the third output data includes at least: recycling plan data.

[0054] In this embodiment, the third input data may further include: historical sales data, historical recycling record data, market characteristic data, and so on.

[0055] In this embodiment, the third output data may further include: recycling priority data, recycling cost analysis data, recycling trend analysis data, and so on.

[0056] In this embodiment, the third agent may be configured with a third AI model. The third AI model may be a linear regression model to predict the recycling quantity.

[0057] The input of the third AI model may be real-time sales data. It may also be real-time sales data, historical sales data, historical recycling record data, market characteristic data, and so on. Its output is recycling plan data. The third AI model can analyze the relationship between the input data and the recycling quantity, and predict the recycling quantity of different products in different regions. The prediction results can be converted into specific recycling plan data, including recycling targets, time arrangements, and regional distributions.

[0058] The third AI model can process various types of data inputs to ensure the comprehensiveness and accuracy of the prediction. Its output is recycling plan data, which may include recycling targets, recycling time arrangements, and recycling regions. The recycling target is expressed as the product type or model to be recycled, and the recycling target quantity for each model. The recycling time arrangement can be expressed as the start time, end time, and key nodes of the recycling, and so on. The recycling region can be expressed as the recycling regions to be covered and the recycling target quantity for each region.

[0059] In this embodiment, the third AI model can use real-time sales data combined with historical data and market characteristic data to dynamically generate a recycling plan.

[0060] In this embodiment, the third AI model may also be a random forest regression model, a support vector machine regression model, an LSTM model, and so on.

[0061] In this embodiment, the preset first agent, second agent, and third agent respectively correspond to the manufacturing, sales, and recycling links. By clarifying the dynamic processing logic of the input data and output data, an efficient data interaction and collaboration among the various links of the supply chain are formed, solving the problem of low data utilization efficiency in the traditional supply chain. By real-time transmitting key data such as market demand prediction data, historical sales data, and real-time sales data to the corresponding agents, accurate calculation and dynamic adjustment of production plans, sales forecasts, and recycling plans are realized, reducing information delay and data island phenomena, thereby significantly improving the overall operation efficiency and response ability of the supply chain.

[0062] Such as Figure 2As shown, in some embodiments, the method further includes:

[0063] Step S18: A preset fourth agent receives fourth input data and outputs fourth output data; wherein, the fourth input data at least includes the recycling plan data, and the fourth output data at least includes the disassembly plan data.

[0064] In this embodiment, the disassembly plan data may include: product disassembly targets, disassembly sequences, and disassembly resource allocations, etc. The product disassembly target represents the specific product type or model to be disassembled. The disassembly sequence represents the disassembly priorities of different products and the step sequence of disassembly operations. The disassembly resource allocation represents the allocation plan for disassembly equipment, tools, and human resources. The disassembly plan data may also include disassembly quantity prediction data. For example, the predicted disassembly quantity of each product model. For another example, among 1000 units recycled in a certain area, 600 units need to be disassembled. The disassembly plan data may also include recycling target data. For example, the expected types and quantities of different materials obtained from disassembly.

[0065] In this embodiment, the fourth agent may be configured with a fourth AI model. The fourth AI model may be a random forest model to predict the disassembly quantity.

[0066] In this embodiment, the fourth input data may also include product historical disassembly data. For example, the historical disassembly records of different model products, as well as the disassembly efficiency, material recovery rate, and the proportion of reusable parts.

[0067] In this embodiment, the fourth input data may further include product material composition data. For example, the material composition ratios of each product model.

[0068] In this embodiment, the input of the fourth AI model may be the recycling plan data, product historical disassembly data, and product material composition data. The fourth AI model may be a random forest model, and this random forest model can improve the accuracy and robustness of the model through the comprehensive prediction of multiple decision trees. Its output is the disassembly plan data.

[0069] In this embodiment, the fourth intelligent agent may be configured with a fourth policy engine. The fourth policy engine is capable of generating a material procurement strategy based on the disassembly plan data (e.g., quantity prediction data). For example, the fourth policy engine derives the types and quantities of recyclable materials according to the predicted disassembly quantity, and dynamically formulates a material procurement plan in combination with market demand and inventory status. Therefore, the fourth input data of the fourth intelligent agent may also include inventory status data (e.g., the inventory quantity, quality grade, and storage location of current materials and components). Accordingly, the fourth output data of the fourth intelligent agent may also include a material procurement plan (e.g., the types and quantities of materials to be procured).

[0070] In this embodiment, the fourth policy engine may also be preset with multiple policy rules for quickly executing known business logics. For example, "Policy Rule 1: When the inventory is below the lower limit, trigger a replenishment strategy", "Policy Rule 2: Priority is given to the procurement of scarce materials and they are quickly put into storage". The fourth policy engine may also be preset with an optimization algorithm to generate corresponding strategies, such as a material procurement strategy.

[0071] Step S110: A preset fifth intelligent agent receives fifth input data and outputs fifth output data; wherein, the fifth input data at least includes the disassembly plan data, and the fifth output data at least includes reusable data.

[0072] In this embodiment, the fifth intelligent agent may be configured with a fifth AI model. The fifth AI model may be a random forest model, a support vector machine regression model, or a deep learning model, etc.

[0073] In this embodiment, the fifth input data may also include historical reuse data. For example, the reuse records of different components and materials.

[0074] In this embodiment, the fifth input data may also include market demand prediction data.

[0075] In this embodiment, the fifth input data may further include inventory status data.

[0076] In this embodiment, the reusable data may include the names of materials or components, the reuse quantity, and the reuse scenarios, etc.

[0077] In this embodiment, the fifth output data may also include reuse priority data, reuse cost analysis data, etc.

[0078] In this embodiment, the fifth intelligent agent may be configured with a fifth policy engine. The fifth policy engine is capable of generating a reuse strategy (such as a reuse priority strategy) based on the output result of the fifth AI model.

[0079] In this embodiment, by setting up the fourth intelligent agent and the fifth intelligent agent, the intelligent collaborative operation between the recycling link and the subsequent disassembly and reuse links is realized. The fourth intelligent agent generates accurate disassembly plan data based on the recycling plan data to ensure the efficiency of the disassembly process and the maximization of resource utilization; the fifth intelligent agent generates reusable data based on the disassembly plan data to provide high-quality reusable parts for the manufacturing link. This phased and modular intelligent agent design significantly improves the overall efficiency of the supply chain, reduces resource waste and disassembly costs.

[0080] As Figure 3 shown, in some embodiments, the first input data further includes: reusable data; the step of the preset first intelligent agent receiving the first input data and outputting the first output data includes:

[0081] Step S122: Through a preset message queue, the first intelligent agent receives the market demand forecast data from the second intelligent agent and receives the reusable data from the fifth intelligent agent.

[0082] In this embodiment, the message queue can provide an asynchronous communication mechanism to decouple the direct dependencies between intelligent agents. Specifically, in the message queue, separate queues can be set for different data types, and a unified JSON or other standard format can be adopted, with fields including data type identifiers and data contents. The storage time of the queue can also be set to ensure that the first intelligent agent can obtain the latest data in a timely manner and avoid excessive historical data accumulation in the queue.

[0083] In this embodiment, the first intelligent agent can subscribe to the queue to obtain the market demand forecast data and the reusable data. Then, by decoding the JSON or other message formats, the key fields can be extracted. Finally, the market demand forecast data and the reusable data are integrated to form a complete input data set.

[0084] In this embodiment, the message queue can be a Kafka-based message queue.

[0085] Step S124: The first intelligent agent generates the production plan forecast data through a preset first AI model, the market demand forecast data, and the reusable data.

[0086] In this embodiment, the first AI model can be a prediction model based on machine learning or a prediction model based on deep learning. Specifically, the first AI model can be an LSTM network combined with time series analysis or a reinforcement learning model.

[0087] The first AI model can also be a hybrid model of LSTM network + reinforcement learning. For example, the LSTM network can predict the market demand for a period of time in the future in advance. The reinforcement learning model can adjust the production plan according to the prediction results and the current resource constraints.

[0088] In this embodiment, the first agent can perform data cleaning and normalization on the market demand prediction data and the reusable data. For example, the input data can be formatted and standardized, and missing values and outliers can be processed. Time series data (such as market demand prediction data) can also be segmented at a unified time granularity (such as days or weeks).

[0089] In this embodiment, the first AI model can be a model pre-trained with training data. The training data can be historical data. For example, historical market demand data (historical sales data, seasonal changes, regional demand fluctuations), historical reusable data (historical records of recycled parts, reuse rates, applicable scenarios, etc.). The training objective can be to minimize the error between the prediction results (production plan prediction data) and the actual data. For example, the mean squared error (MSE) can be used as the loss function. During the training process, the historical data can be divided into an 80% training set and a 20% validation set. The model parameters can be optimized through multiple trainings to reduce the error.

[0090] This embodiment makes the production plan prediction more accurate and efficient by integrating reusable data into the input data of the manufacturing agent. The addition of reusable data can help the manufacturing agent better understand the recycling and reuse of waste products, so that the recycling and utilization of materials can be better considered in the process of producing new products, reducing production costs and improving resource utilization efficiency. Therefore, this embodiment has a positive effect on improving supply chain management efficiency, reducing production costs, improving resource utilization efficiency, and promoting environmental protection.

[0091] In some embodiments, the step in which the preset second agent receives the second input data and outputs the second output data includes:

[0092] Step S142: The preset second agent obtains the historical sales data from the storage device and collects real-time sales data.

[0093] In this embodiment, the storage device can be a sales management system or a data warehouse.

[0094] In this embodiment, the second agent can extract the sales records of a specific time period and product from the data warehouse through predefined SQL statements. It can also connect to the sales system through an API to synchronize data regularly.

[0095] In this embodiment, the second agent can obtain orders and sales data (real-time sales data) from the sales system in real time through real-time API calls. It can also subscribe to the real-time sales data stream through tools such as Kafka or RabbitMQ.

[0096] In this embodiment, after the second agent obtains the historical sales data and real-time sales data, it can perform data cleaning and data normalization to make their features within the same numerical range, which is convenient for subsequent model processing.

[0097] Step S144: The second agent generates the market demand prediction data through a preset second AI model and the historical sales data and real-time sales data.

[0098] In this embodiment, the second AI model can be a prediction model based on machine learning or a prediction model based on deep learning. Specifically, the second AI model can be a bidirectional LSTM model, which has the ability to learn from the contexts of both the past and the future, enabling the model to more comprehensively understand time series data. The bidirectional LSTM model sets the number of hidden layer units to 12 and the number of layers to 3 to balance the complexity of the model and the training efficiency. The mean squared error (MSE) is used as the loss function to effectively measure the difference between the predicted value and the actual value. At the same time, the Adam optimizer is selected to update the model parameters to improve the training speed and performance.

[0099] In this embodiment, the input of the second AI model can be the historical sales data and real-time sales data. Its output can be the market demand prediction data.

[0100] In this embodiment, the second AI model can be a model pre-trained with historical data. For example, it can be historical sales data and historical market demand data.

[0101] This embodiment realizes the accurate prediction of market demand by the second agent integrating historical sales data and real-time sales data and combining a preset second AI model (bidirectional LSTM or other prediction models). This method uses historical data to capture long-term trends and at the same time reflects the current market dynamics through real-time data, making the prediction results more comprehensive and real-time. In addition, through the in-depth analysis and dynamic adjustment of time series data by the AI model, the prediction accuracy and the supply chain response ability can be effectively improved, thereby helping enterprises optimize production plans, improve resource utilization efficiency, and quickly adapt to market changes.

[0102] In some embodiments, the step in which the preset third agent receives the third input data and outputs the third output data includes:

[0103] Step S162: The third intelligent agent collects the real-time sales data.

[0104] In this embodiment, the third intelligent agent can obtain orders and sales data (real-time sales data) in real time from the sales system through real-time API calls. It can also subscribe to the real-time sales data stream through tools such as Kafka or RabbitMQ.

[0105] Step S164: The third intelligent agent generates recycled product data through a preset third AI model and the real-time sales data;

[0106] In this embodiment, the third AI model can be a linear regression model to predict the recycling volume. Therefore, the input of this linear regression model can be real-time sales data (sales volume, return volume, etc.), and its output is the predicted number of recycled products. That is, the linear regression model will predict which products will enter the recycling cycle and estimate the recycling quantity within a period of time based on historical data and real-time data.

[0107] In this embodiment, before inputting the real-time sales data into the linear regression model, the third intelligent agent can clean (handle missing values and outliers) and standardize the data to make the data suitable for model input.

[0108] In this embodiment, the third AI model is a pre-trained model. For example, the real-time sales data can be input into the trained linear regression model, and the model generates the predicted recycling volume and recycled product categories, and outputs the recycled product data (recycling target quantity, product model, region, recycling time).

[0109] Step S166: The third intelligent agent generates the recycling plan data based on the recycled product data.

[0110] In this embodiment, after obtaining the recycled product data, the third intelligent agent can further generate corresponding recycling plan data based on the recycled product data. Specifically, the third intelligent agent can first analyze the recycled product data using a data analysis model to identify recycling targets and requirements (e.g., which products are approaching the recycling time point, where the recycling volume is the largest, etc.). Then, according to the recycling targets, the third intelligent agent can use optimization algorithms such as integer programming for resource scheduling to determine the recycling task allocation for each region and each time period. For example, determine which time period has a large recycling volume and prioritize the arrangement of transportation and processing resources. Next, the third intelligent agent can use algorithms such as TSP or VRP to plan the optimal route for the recycling activity to ensure the minimization of transportation costs and time during the recycling process. For example, arrange the shortest path or the recycling routes of multiple fleets to maximize efficiency. Finally, the third intelligent agent can integrate information such as the optimized recycling tasks, resource allocation, and time arrangement to generate specific recycling plan data.

[0111] This embodiment, where the third intelligent agent generates recycled product data based on real-time sales data and further generates recycling plan data, helps to achieve dynamic optimization of the recycling link in the supply chain. Specifically, real-time sales data reflects the latest market demands and product life cycles, enabling the third intelligent agent to accurately identify which products enter the recycling stage, thereby timely planning recycling activities. By using the third AI model, it is possible to accurately predict the recycling timing, recycling volume, and recycling priorities of different products, providing data support for subsequent recycling resource scheduling and disassembly work. Ultimately, the generated recycling plan data ensures that the recycling process is more efficient, flexible, and forward-looking, reducing recycling costs, minimizing resource waste, and enhancing the overall sustainability and response speed of the supply chain.

[0112] In some embodiments, the step in which the preset fourth intelligent agent receives the fourth input data and outputs the fourth output data includes:

[0113] Step S182: Through a preset message queue, the fourth intelligent agent receives the recycling plan data from the third intelligent agent.

[0114] Step S184: The fourth intelligent agent generates the disassembly plan data through a preset fourth AI model and the recycling plan data.

[0115] In this embodiment, the fourth AI model can be a random forest model to predict the disassembly quantity.

[0116] In some embodiments, the fourth intelligent agent generates disassembly plan data by receiving recycling plan data from the third intelligent agent and processing this data using a preset fourth AI model. The beneficial effect of this process is that it realizes the intelligent and systematic processing of recycled products in the supply chain. Specifically, by analyzing the recycling plan data, the fourth intelligent agent can accurately determine which recycled products need to be disassembled, how to disassemble them, and the priority and sequence of disassembly, thereby optimizing the disassembly process, improving the resource reuse rate and processing efficiency. Through the generation of this intelligent disassembly plan, manual intervention can be reduced, costs can be lowered, the accuracy and efficiency of disassembly operations can be improved, and at the same time, sustainable development goals can be supported to achieve more efficient resource recycling.

[0117] In some embodiments, the steps of the preset fifth intelligent agent receiving fifth input data and outputting fifth output data include:

[0118] Step S1102: Through a preset message queue, the fifth intelligent agent receives the disassembly plan data from the fourth intelligent agent.

[0119] Step S1104: The fifth intelligent agent generates reusable data through a preset fifth AI model and the disassembly plan data.

[0120] In this embodiment, the fifth AI model can be a random forest model, a support vector machine regression model, or a deep learning model, etc.

[0121] In some embodiments, the fifth intelligent agent generates reusable data by receiving disassembly plan data from the fourth intelligent agent and using a preset fifth AI model. The beneficial effect of this process is reflected in the efficient recycling and utilization of resources. Specifically, based on the disassembly plan data, the fifth intelligent agent can intelligently determine which disassembled parts or materials can be reused and generate corresponding reusable data. This not only optimizes the reuse process but also improves the recycling rate of waste materials. Through this intelligent processing, it can ensure that the production link can obtain reusable parts or materials in a timely and accurate manner, thereby reducing production costs, reducing dependence on new raw materials, and at the same time supporting environmental protection goals and promoting the recycling of resources.

[0122] In a possible and specific implementation scheme, the method may include:

[0123] (1) Data collection: Data collection is a crucial step in optimizing the supply chain of home appliance products. First of all, it is necessary to identify data sources at all links in the supply chain, including manufacturing data, sales data, recycling data, disassembly data, reuse data, and market trends, etc. Manufacturing data mainly comes from production equipment and covers information such as product models, materials, production volume, and production efficiency. Sales data includes order quantity, sales volume, and customer feedback. Recycling data involves product recovery rates and the effectiveness of recycling channels. Disassembly data records the efficiency and costs during the disassembly process, while reuse data provides information on the performance and market demand of reusable components. In addition, market trend data is sourced from industry reports and competitor analysis to help enterprises grasp market dynamics.

[0124] (2) Process modeling: Process modeling is an important part of optimizing the supply chain of home appliance products. By constructing a dynamic model based on the collected data, the relationships and dependencies between various links such as manufacturing, sales, recycling, disassembly, and reuse can be clearly described, thus achieving more efficient supply chain management.

[0125] First of all, the first step in constructing a dynamic model is to identify each link and its key activities. The manufacturing link involves activities such as raw material procurement, production planning, equipment operation, and quality control; the sales link includes order processing, customer delivery, and market feedback; the recycling link involves product recovery, evaluation, and classification; the disassembly link includes disassembly operations, component inspection, and waste treatment; while the reuse link focuses on the overhaul, market launch, and performance evaluation of reusable components.

[0126] Then, using these activities and their data, establish a relationship model between each link. In the supply chain of home appliance products, there are complex interrelationships and dependencies between each link, as Figure 4 shown. The efficiency of the manufacturing link directly affects the delivery time of the sales link, while sales data can in turn affect the adjustment of production plans to ensure that production matches market demand. Sales data not only affects production plans but also the quantity of recycling. For example, the recovery rate of best-selling products may be affected by sales strategies. The efficiency and quantity of the recycling link will affect the workload and costs of the disassembly link, and recycling data can help adjust the disassembly process to improve efficiency and reduce costs. The performance of the disassembly link (such as speed and quality) directly affects the feasibility and economy of the reuse link, and at the same time, the quality and quantity of components after disassembly will also affect reuse decisions. The market performance of the reuse link in turn affects sales strategies and promotes the recycling of more products, forming a virtuous cycle. The entire model requires a real-time data feedback mechanism to quickly adjust the strategies of each link according to market changes, ensuring that management can make timely decisions, thereby optimizing the overall operating efficiency of the supply chain.

[0127] (3) AI Agent Design: The agent can automatically identify the best-matching supply chain processes based on real-time and historical data and propose optimization suggestions, such as predicting sales volume and recovery volume, optimizing production plans and resource allocation, etc. The data mainly involved in this design is as follows.

[0128] Manufacturing Data: Product model, materials, quantity produced, etc.

[0129] Sales Data: Product model, historical sales records, order data, customer feedback, etc.

[0130] Recovery Data: Production efficiency, equipment operation time, production cost, etc.

[0131] Disassembly Data: Material data during the disassembly process, including product model, types, quantities, weights of different materials, etc. Information on various materials during the disassembly process.

[0132] Reuse Data: Historical reuse data, including types, quantities, and application scenarios of reused materials.

[0133] Market Data: Industry trends, competitor analysis, consumer behavior.

[0134] Recovery Data: Recovery rate, recovery channels, and product status.

[0135] Data preprocessing is a crucial step to ensure the quality of analysis and modeling. This process mainly performs data cleaning, transformation, feature engineering, and data splitting, etc., to improve the convergence speed and performance of the model. It mainly deals with missing values and outliers. Data transformation converts categorical variables into numerical forms, etc. After data processing, the data is split into a training set and a test set to ensure the representativeness of model training and evaluation.

[0136] After data processing is completed, the AI agent will conduct sales forecasting and optimize supply chain management. Sales forecasting uses the time series analysis method Long Short-Term Memory Network (LSTM) to effectively capture seasonal and trend changes in sales data, and then accurately predict future sales volume. By analyzing historical sales data, identifying periodic fluctuations in sales, and considering external factors such as promotional activities, market environment changes, and consumer behavior, the accuracy of sales forecasting is ensured, enabling enterprises to better adjust production plans and inventory levels and avoid inventory overstock or shortage caused by demand fluctuations.

[0137] Recovery prediction focuses on analyzing historical sales and product lifecycle data. The AI agent utilizes the random forest algorithm to identify the key factors influencing recovery. These factors may include the usage frequency of the product, consumers' willingness to repurchase, product performance, and market feedback, etc. Through in-depth analysis of relevant variables, a prediction model is established to accurately predict the recovery volume of different products and notify the recycling department, assisting enterprises in formulating effective recycling strategies, optimizing the reuse of resources, and enhancing environmental sustainability.

[0138] After the recovery prediction is completed, the AI agent will infer and disassemble relevant information based on the results of the recovery prediction to further optimize the management and reuse of resources. The AI agent will analyze the predicted recovery volume data, identify which products are most likely to be recycled, and evaluate the economy and feasibility of their disassembly.

[0139] The AI agent will formulate corresponding disassembly strategies according to the recovery potential of different products, including selecting the most suitable disassembly methods and processes to ensure the maximum utilization of resources. For products that are easy to disassemble and have high recycling value, the AI agent may recommend more efficient mechanical disassembly methods; while for complex products, more delicate manual disassembly may be required. The AI agent will also analyze the key factors affecting disassembly efficiency, such as material composition, product design complexity, and changes in market demand, identify potential bottlenecks in the disassembly process, and put forward improvement suggestions. By combining the disassembly strategy with resource allocation, the scheduling of disassembly operations is optimized to improve disassembly efficiency.

[0140] After obtaining the disassembly information, the AI agent will further deduce the reuse information to optimize the reuse of resources and reduce environmental impact. The AI agent first analyzes the data obtained during the disassembly process, including the type, quantity, and quality of the materials after disassembly, to identify which components can be reused and evaluate their market demand and potential value. For example, the recyclability of metals, plastics, and electronic components will be focused on. Through the analysis of disassembly data, the AI agent will identify reusable components and materials. And through the classification and evaluation of disassembled components, determine which components meet the reuse standards. The AI agent will evaluate the market demand for reusable components by combining market trends and consumer preferences, enabling enterprises to better formulate reuse strategies.

[0141] During the implementation process, the AI agent continuously monitors the operating status of the home appliance product supply chain and collects feedback information. According to the feedback information, the process matching algorithm is adjusted and optimized to improve the accuracy and efficiency of matching, ensuring the coordinated operation of each link.

[0142] The supply chain management of home appliance products consists of five AI agents: manufacturing agent, sales agent, recycling agent, disassembly agent, and reuse agent, which are respectively responsible for different supply chain links to achieve efficient and intelligent supply chain management of home appliance products.

[0143] (1) The manufacturing agent is responsible for managing the production process of household appliances. It formulates and adjusts the production plan based on the market demand forecast provided by the sales agent to ensure the effective utilization of production resources.

[0144] (2) The sales agent focuses on market analysis and the processing of sales data. By analyzing historical sales data and market trends, it forecasts future product demands, thus providing accurate market information to the manufacturing agent.

[0145] The sales agent is also responsible for formulating sales strategies, adjusting pricing and promotional activities according to market feedback and demand changes, and enhancing sales efficiency.

[0146] (3) The recycling agent is responsible for managing the recycling process of household appliances to ensure the effective recycling of resources and environmental protection. Based on the sales data and market feedback provided by the sales agent, it formulates a recycling plan, establishes relationships with recycling partners, and ensures the smoothness of the recycling channels.

[0147] (4) The disassembly agent is responsible for disassembling and classifying the recycled household appliances. It formulates disassembly standards and processes to ensure the efficient and safe handling of recycled products.

[0148] (5) The reuse agent focuses on evaluating the reusability of disassembled components and formulating reuse strategies.

[0149] Asynchronous communication is achieved among the various AI agents using a message queue, and data exchange among the agents is carried out through the publish / subscribe mode. The workflow among the agents is as Figure 5 shown.

[0150] The sales agent analyzes historical sales data, forecasts future demands, and publishes the results to the message queue. After receiving the demand forecast message, the manufacturing agent adjusts the production plan. After the production plan is updated, the manufacturing agent notifies the sales agent through the API. Based on the sales data, the recycling agent forecasts the quantity of products about to expire, formulates a recycling plan, and sends it to the disassembly agent. After receiving the recycled product information, the disassembly agent disassembles and classifies them, evaluating the quality and reusability of each component. The disassembly results are sent to the reuse agent through the message queue. The reuse agent analyzes the disassembly results, formulates reusable strategies, and feeds back the results to the manufacturing agent and the recycling agent. Each agent makes synchronous requests and responses through the RESTful API. The sales agent requests the manufacturing agent through the API to obtain the production plan. The recycling agent can obtain the latest market feedback from the sales agent through the API.

[0151] The model adopted internally by the sales agent is a bidirectional LSTM model, which has the ability to learn from both past and future contexts, enabling the model to more comprehensively understand time series data. The bidirectional LSTM model sets the number of hidden layer units to 12 and the number of layers to 3 to balance the complexity of the model and the training efficiency. The mean squared error (MSE) is used as the loss function to effectively measure the difference between the predicted value and the actual value. At the same time, the Adam optimizer is selected to update the model parameters to improve the training speed and performance. The model is trained using historical sales data, performing forward propagation, calculating the loss, and performing backpropagation. By continuously adjusting the model parameters, the prediction accuracy of the model is gradually improved. After training is completed, the model is used to predict the training set and the test set, evaluate the performance of the model in actual applications, and apply it to the sales agent. When the performance of the sales agent declines, the model is retrained or adjusted using new data, and the new data is used as part of the training set to enhance the generalization ability of the model.

[0152] A linear regression model is adopted in the recycling AI agent to predict the recycling volume. First, historical sales data, recycling records, and market characteristic data are analyzed to comprehensively understand the factors affecting the recycling volume. Features related to recycling volume prediction are extracted from the original data, including sales quantity, product type, region, and seasonal factors, etc. Sales quantity is a direct factor affecting the recycling volume, and consumer behaviors may also vary for different product types and regions. The data is preprocessed, and numerical features are standardized or normalized to improve the training effect of the model. In the model construction stage, a linear regression model is selected as the basic model. The training set data is used to train the model to fit the relationship between features such as sales quantity and the recycling volume. The test set is used to validate the model, and metrics such as the coefficient of determination R2 and the mean squared error (MSE) are calculated to determine the performance on new data. Once the model is evaluated and confirmed to be effective, it is applied to the home appliance recycling AI agent. The AI agent regularly feeds new recycling data back to the model for retraining to ensure that the model can adapt to market changes and new data trends and maintain the accuracy of the prediction.

[0153] The disassembly AI agent and the reuse agent mainly adopt a random forest model to predict the quantity of disassembled parts and reusable materials. The disassembly material prediction is mainly based on the recycling product type, degree of newness and quantity to predict the quantity of each material to be disassembled. The recycling quantity of each product and the proportion of each material in the total weight of the product (such as metal, plastic, electronic components, etc.) are mainly predicted using a random forest model. Based on the predicted disassembly quantity, material procurement and inventory management strategies are formulated.

[0154] The reusable material agent predicts the quantity of reusable materials based on the disassembled material type, the quantity of disassembled materials, and the quality of the materials, providing decision-making support for manufacturing supply. The random forest is used to predict the quantity of reusable materials, which is predicted based on the disassembled material type, the quantity of disassembled materials, and the quality of the materials, and the prediction results are returned. The AI agent retrains the model regularly to ensure that the model can adapt to market changes and new data trends and maintain the accuracy of the prediction.

[0155] As Figure 6 shown, according to an embodiment of the present invention, a supply chain data processing device is provided, and the device at least includes:

[0156] A first agent module, which is used to receive first input data and output first output data; wherein, the first agent is an agent representing the manufacturing link in the supply chain link, the first input data at least includes market demand prediction data, and the first output data at least includes production plan prediction data;

[0157] A second agent module, which is used to receive second input data and output second output data; wherein, the first agent is an agent representing the sales link in the supply chain link, the second input data at least includes historical sales data, and the second output data at least includes the market demand prediction data;

[0158] A third agent module, which is used to receive third input data and output third output data; wherein, the third agent is an agent representing the recycling link in the supply chain link, the third input data at least includes real-time sales data, and the third output data at least includes recycling plan data.

[0159] According to an embodiment of the present invention, an electronic device is provided. Please refer to Figure 7 . The electronic device in this embodiment may include one or more of the following components: a processor, a network interface, a memory, a non-volatile memory, and one or more application programs. One or more application programs may be stored in the non-volatile memory and configured to be executed by one or more processors. One or more programs are configured to execute the methods described in the foregoing method embodiments.

[0160] According to an embodiment of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by the computer, the computer executes the methods described in any of the foregoing embodiments.

[0161] According to an embodiment of the present invention, a computer program product including instructions is further provided. When the instructions are executed by the computer, the computer executes a method in any of the foregoing embodiments.

[0162] It should be noted that the terms "first", "second", etc. in the description, claims and above-mentioned drawings of this application are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances so that the embodiments of the present application described here can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device comprising a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0163] Optionally, the specific examples in this embodiment may refer to the examples described in the above embodiments, and will not be elaborated here.

[0164] The serial numbers of the above embodiments of the present application are only for description and do not represent the advantages or disadvantages of the embodiments.

[0165] In the above embodiments of the present application, the descriptions of the respective embodiments have their own emphases. For the parts not detailed in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0166] The above are only the preferred embodiments of the present application. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present application, several improvements and refinements can still be made, and these improvements and refinements should also be regarded as the protection scope of the present application.

Claims

1. A supply chain data processing method, characterized in that: The method at least comprises: The preset first agent receives first input data and outputs first output data; wherein the first agent is an agent representing a manufacturing link in a supply chain link, the first input data at least includes market demand forecast data, and the first output data at least includes production plan forecast data; The preset second agent receives the second input data and outputs the second output data; wherein the first agent is an agent representing the sales link in the supply chain link, the second input data at least includes the historical sales data, and the second output data at least includes the market demand forecast data; The preset third agent receives third input data and outputs third output data; wherein, the third agent is an agent representing the recycling link in the supply chain link, the third input data at least includes real-time sales data, and the third output data at least includes recycling plan data.

2. The method according to claim 1, characterized in that The method further comprises: The preset fourth agent receives fourth input data and outputs fourth output data; wherein the fourth input data at least includes the recycling plan data, and the fourth output data at least includes the disassembly plan data; The preset fifth agent receives fifth input data and outputs fifth output data; wherein the fifth input data at least includes the disassembly plan data, and the fifth output data at least includes reusable data.

3. The method according to claim 2, characterized in that The first input data also includes: reusable data; the step of the preset first agent receiving the first input data and outputting the first output data includes: The first agent receives the market demand forecast data from the second agent and receives reusable data from the fifth agent through a preset message queue; The first intelligent agent generates the production plan forecast data through a preset first AI model and the market demand forecast data and reusable data.

4. The method according to claim 2, characterized in that: The step of the preset second agent receiving the second input data and outputting the second output data comprises: The preset second intelligent agent obtains the historical sales data from the storage device and collects the real-time sales data; The second intelligent agent generates the market demand forecast data through a preset second AI model and the historical sales data and real-time sales data.

5. The method according to claim 2, characterized in that: The step of the preset third agent receiving the third input data and outputting the third output data comprises: The third intelligent agent collects the real-time sales data; The third agent generates recycled product data through a preset third AI model and the real-time sales data; The third agent generates the recycling plan data based on the recycled product data.

6. The method according to claim 2, characterized in that The step of the preset fourth agent receiving fourth input data and outputting fourth output data comprises: The fourth agent receives the recycling plan data from the third agent through a preset message queue; The fourth intelligent agent generates the disassembly plan data through a preset fourth AI model and the recycling plan data.

7. The method according to claim 2, characterized in that: The step of the preset fifth agent receiving fifth input data and outputting fifth output data comprises: The fifth agent receives the disassembly plan data from the fourth agent through a preset message queue; The fifth agent generates reusable data through the preset fifth AI model and the disassembly plan data.

8. A supply chain data processing device, characterized in that: The device at least comprises: A first agent module, which is used to receive first input data and output first output data; wherein the first agent is an agent representing a manufacturing link in a supply chain link, the first input data at least includes market demand forecast data, and the first output data at least includes production plan forecast data; A second agent module, which is used to receive second input data and output second output data; wherein the first agent is an agent representing a sales link in a supply chain link, the second input data at least includes historical sales data, and the second output data at least includes the market demand forecast data; A third agent module is used to receive third input data and output third output data; wherein, the third agent is an agent representing the recycling link in the supply chain link, the third input data at least includes real-time sales data, and the third output data at least includes recycling plan data.

9. An electronic device, characterized in that: include: a memory, and one or more processors communicatively coupled to the memory; The memory stores instructions that can be executed by the one or more processors. The instructions are executed by the one or more processors to enable the one or more processors to implement the method according to any one of claims 1 to 7.

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

Images