Supply chain intelligent management system based on dynamic collaborative optimization

Abstract

The present invention discloses a supply chain intelligent management system based on dynamic collaborative optimization, which relates to the technical field of hotel supply chain management, including a supply chain intelligent management platform, wherein the supply chain intelligent management platform is communicatively connected to the following modules, wherein: a multimodal data perception module is used to collect multimodal data within the hotel through a LoRaWAN+BLE hybrid sensor network, and access external data sources to obtain real-time information through an API interface; a dynamic demand forecasting module is based on an ASTGNN model and captures time series trends through a bidirectional LSTM. The present invention significantly improves the accuracy of demand forecasting by integrating external dynamic data such as social media public opinion and weather, combined with multimodal data analysis technology, and utilizing the ASTGNN model, combined with bidirectional LSTM and GCN to analyze time series trends and demand correlations between branches, dynamically adjust feature weights, and significantly reduce the demand forecast error rate, providing enterprises with a more reliable basis for demand forecasting, and helping to optimize inventory management and production planning.

Description

Technical Field

[0001] The present invention relates to the technical field of hotel supply chain management, and in particular to a supply chain intelligent management system based on dynamic collaborative optimization. Background Art

[0002] The supply chain of the hotel industry covers multiple links, including procurement, storage, allocation, delivery, sales, etc. Each link is affected by external environment (such as changes in market demand, weather, holidays, etc.) and internal factors (such as inventory, employee efficiency, etc.).

[0003] The existing technology has the following industry pain points:

[0004] 1. Demand forecast bias: Traditional time series models cannot incorporate dynamic factors such as social media sentiment and large-scale events. The forecast error rate during hotel industry promotions can be as high as 25%.

[0005] 2. High inventory costs: The bullwhip effect results in a guest room supplies inventory turnover rate of only 4 times per year, and a stock-out rate exceeding 12%;

[0006] 3. Inefficient collaboration: Information transmission between suppliers, logistics providers, and hotels is delayed by more than 48 hours, and response speed to order changes is insufficient.

[0007] 4. Lack of traceability: Recalls of food, linens, and other supplies require manual verification, which takes an average of 72 hours and costs 8% of sales.

[0008] Therefore, supply chain forecasting models or blockchain traceability are usually used to solve the above problems. However, the existing supply chain forecasting models only rely on historical sales data and do not integrate external dynamic data. Blockchain traceability lacks a real-time collaborative decision-making mechanism and cannot automatically trigger replenishment. To this end, a supply chain intelligent management system based on dynamic collaborative optimization is proposed, which integrates Internet of Things perception, multimodal data analysis, dynamic optimization algorithms and blockchain technology. It can be extended to scenarios such as retail and manufacturing to meet the needs of various industries for efficient, transparent and intelligent supply chain management. Summary of the Invention

[0009] The purpose of the present invention is to provide a supply chain intelligent management system based on dynamic collaborative optimization to solve the problems raised in the above background technology.

[0010] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0011] The supply chain intelligent management system based on dynamic collaborative optimization includes a supply chain intelligent management platform, which is communicatively connected to the following modules:

[0012] The multimodal data perception module is used to collect multimodal data within the hotel through a LoRaWAN+BLE hybrid sensor network, access external data sources such as social media public opinion and weather forecasts, and obtain real-time information through an API interface;

[0013] The dynamic demand forecasting module, based on the ASTGNN model, uses bidirectional LSTM to capture time series trends and combines GCN to analyze demand correlations between branches to forecast demand by time period for the next seven days.

[0014] A multi-objective collaborative optimization module uses the APSO algorithm to balance inventory costs, logistics timeliness, and carbon emissions. It automatically analyzes supplier capacity and logistics route constraints, and dynamically adjusts inertia weights and learning factors to generate a Pareto optimal solution set within 10 minutes.

[0015] The blockchain collaborative execution module is used to convert the optimal solution into a smart contract and deploy it to the blockchain, automatically triggering processes such as order generation and payment. The purchase order hash value is uploaded to the blockchain in real time, and suppliers and logistics providers can synchronize cross-chain data through Hyperledger Fabric to ensure transparent and traceable execution status.

[0016] The execution feedback and model correction module is used to provide real-time feedback on inventory consumption and logistics timeliness through IoT sensors, calculate prediction errors and trigger model corrections. If the error exceeds the preset threshold, the model is automatically retrained and the event tag library is updated to form a closed-loop optimization mechanism.

[0017] A further improvement of the technical solution of the present invention is that the multimodal data perception module specifically includes:

[0018] A LoRaWAN+BLE hybrid sensor network is deployed within the hotel to collect multimodal data, including room temperature, humidity, pressure, and usage frequency. APIs are also used to access external data sources, including social media sentiment and weather forecasts, to obtain real-time information on trending topics and weather changes on social media, ensuring comprehensive and up-to-date data.

[0019] The collected multi-sensor data is fused to improve its accuracy and reliability. A positioning error correction model is used to correct the data to reduce positioning errors. Furthermore, internal and external hotel data are pre-processed and all data is unified to a UTC timestamp with minute-level accuracy to achieve time alignment.

[0020] For the preprocessed data, feature engineering is performed to generate lagged features and event label annotations, and a standardized feature matrix containing 100+ dimensions is output. The standardized feature matrix is ​​then transmitted to the supply chain intelligent management platform through a communication protocol.

[0021] A further improvement of the technical solution of the present invention is that the dynamic demand forecasting module specifically includes:

[0022] The pre-built ASTGNN model is used to extract spatiotemporal features. In the temporal dimension, a bidirectional LSTM is used to capture time series trends and identify seasonal fluctuations and long-term trends. In the spatial dimension, a GCN (graph convolutional network) is used to analyze demand correlations between branches. The spatial relationships between branches are represented by a graph structure, and the dynamic inventory allocation relationships between branches are learned. The ASTGNN model is an attention-enhanced spatiotemporal graph neural network model.

[0023] Based on feature extraction, an attention mechanism is introduced to dynamically adjust the importance weight of each data feature. By calculating the attention weight of each data feature (for example, the weight coefficient of activity events is 1.5, and that of regular data is 1.0), the weight is adjusted and a higher weight is given to important features.

[0024] After feature extraction and weight adjustment, the model enters the online learning phase, triggering incremental training every 15 minutes. A loss function is designed to quantify the deviation between the predicted value and the true value, and joint averaging is performed in the time and space dimensions to ensure global optimization. The AdamW optimizer (learning rate = 0.001) is then used to update the parameters to ensure that the model can adapt to the latest data changes in real time. Finally, the ASTGNN model is used to output the demand forecast value for the next 7 days by time period, with the prediction error rate controlled within 8%.

[0025] A further improvement of the technical solution of the present invention is that the calculation process of the loss function is:

[0026] In the ASTGNN model, for each prediction point, the square of the difference between its predicted value and the true value is calculated, the square of the difference of all prediction points is summed up to get the total error, and then divided by the number of prediction points to get the mean square error. The mean square error is a common indicator to measure the difference between the model prediction value and the true value.

[0027] In the ASTGNN model, the regularization parameter is determined, the sum of the squares of all model parameters is calculated, and the sum of the squares of all parameters is multiplied by the regularization parameter to obtain the regularization term;

[0028] The total loss function of the ASTGNN model is obtained by adding the mean square error and the regularization term.

[0029] A further improvement of the technical solution of the present invention is that the multi-objective collaborative optimization module specifically includes:

[0030] The multi-objective collaborative optimization module analyzes the constraints of supplier capacity constraints and logistics providers' vehicle scheduling constraints, including but not limited to the supplier's maximum daily supply, logistics providers' vehicle load limits, and route planning. It then performs initialization operations to generate a certain number of particles, which are candidate solutions. Each particle represents a combination of procurement plans, supplier selections, and transportation routes.

[0031] After initialization, the multi-objective collaborative optimization module executes the APSO algorithm to calculate the fitness value of each particle by simulating the behavior of the particle swarm to find the optimal solution. Each particle updates its own speed and position according to its individual optimal and global optimal positions. During the iteration process, the inertia weight and learning factor are dynamically adjusted to balance the capabilities of global and local searches to maintain the diversity and effectiveness of the search.

[0032] In the multi-objective collaborative optimization process, a carbon emission constraint algorithm is introduced to handle carbon emission constraints through an adaptive penalty mechanism. In logistics path planning, logistics path optimization constraints are preset to ensure that the transportation route meets the constraints.

[0033] After a series of iterations, when there is no significant improvement in fitness for 50 consecutive generations, the algorithm terminates and generates a Pareto optimal solution set, which represents a combination of solutions that achieve the optimal or near-optimal solution under different objectives. It includes multiple optional solutions that balance inventory costs, logistics timeliness and carbon emissions. The top three optimal solutions are retained for manual decision-making. The entire optimization process is finally completed within 10 minutes, and the optimal procurement plan and logistics scheduling plan are output.

[0034] A further improvement of the technical solution of the present invention is that the process of carbon emission constraints and logistics path optimization constraints is as follows:

[0035] In the initial stage of multi-objective collaborative optimization, carbon emission targets are incorporated into the optimization model, an upper limit on carbon emission intensity is set, and constraints are dynamically adjusted through an adaptive penalty mechanism;

[0036] During the logistics route planning phase, logistics route optimization constraints are preset, including hard and soft constraints. A carbon emission constraint algorithm is used to solve the problem, prioritizing the selection of route combinations that meet all constraints. The cost is then optimized based on the carbon emission target.

[0037] The APSO algorithm iteratively optimizes the three objectives of inventory cost, logistics timeliness, and carbon emissions. In each iteration, the degree of constraint violation of the particles (candidate solutions) is dynamically evaluated: if the carbon emission or logistics path constraints are not met, the fitness value is reduced through the penalty function, guiding the APSO algorithm to converge to the feasible solution area.

[0038] A further improvement of the technical solution of the present invention is that the process of outputting the optimal procurement plan and logistics scheduling plan is:

[0039] During the iteration process of the APSO algorithm, the fitness value changes of each generation of particles are continuously monitored. If there is no significant improvement in fitness for 50 consecutive generations (i.e. less than 0.1%), the algorithm is considered to have converged, the iteration is terminated, and the Pareto optimal solution set is selected from all non-dominated solutions.

[0040] From the Pareto optimal solution set, the top three solutions are retained based on target weights or decision maker preferences. A visual comparison report is generated for each optimal solution, showing key indicators such as inventory cost, transportation time, and carbon emissions, and noting whether the constraints are satisfied.

[0041] Within a 10-minute time limit, the system automatically outputs the optimal procurement plan including supplier selection and procurement volume, and the logistics scheduling plan including route planning and vehicle allocation, and presents the output content in the form of structured data and visual charts to ensure executability. If the optimization times out, it will fall back to the previous generation of feasible solutions, giving priority to ensuring carbon emission constraints, and then synchronize the final plan to the blockchain module, triggering the automatic execution of the smart contract.

[0042] A further improvement of the technical solution of the present invention is that the blockchain collaborative execution module specifically includes:

[0043] The optimal procurement and logistics solutions output by the multi-objective collaborative optimization module are converted into smart contracts and deployed on the blockchain platform. The smart contracts clearly define trigger conditions. When inventory levels fall below a preset safety threshold, the order generation process is automatically triggered. The smart contracts also include information on order generation and payment terms, ensuring the automation and accuracy of the order generation and payment processes.

[0044] Once the smart contract is triggered, the relevant purchase order information is converted into a hash value and written to the blockchain in real time. At the same time, cross-chain data synchronization between suppliers and logistics providers is carried out through Hyperledger Fabric's blockchain technology. Key information such as purchase orders, payment status, and logistics progress are synchronized in real time to the blockchain nodes of each participant, ensuring that all participants can obtain the latest execution status. In the cross-chain synchronization process, the SPECTRE protocol delay model is used to optimize the blockchain fork selection rule through GHOST to improve transaction confirmation speed and throughput, thereby evaluating the delay time of transaction confirmation and ensuring the consistency and reliability of data among all participants. GHOST is the ghost protocol.

[0045] The blockchain collaborative execution module continuously monitors the execution status of smart contracts and provides real-time feedback on the execution of orders. At the same time, it records the status information of orders on the blockchain, providing real-time and transparent execution tracking for supply chain management.

[0046] A further improvement of the technical solution of the present invention is that the specific process of cross-chain synchronization combined with the SPECTRE protocol is as follows:

[0047] When the triggering conditions of the smart contract are met, the purchase order information is automatically converted into a hash value using the SHA-256 algorithm and written to the blockchain in real time. Using Hyperledger Fabric's chaincode logic, a cross-chain synchronization event is generated simultaneously, and the hash value is broadcast to the supplier and logistics provider's blockchain nodes through Fabric's channel mechanism.

[0048] Hyperledger Fabric uses its interchain communication technology to synchronize real-time data between supplier and logistics provider nodes. Key information, including purchase orders, payment status, and logistics progress, is transmitted between chains through event logs and independently verified by each node. During this process, the SPECTRE protocol delay model is used to evaluate transaction confirmation delays.

[0049] The blockchain nodes of each participant continuously monitor cross-chain events and verify data consistency through the finality rules of the SPECTRE protocol, verifying whether the hash value of each node matches the latest block of the blockchain main chain. If there is a fork conflict, the GHOST mechanism quickly eliminates the inferior branch based on the block generation rate and security parameters, so that all nodes eventually converge to the same state. Ultimately, the full life cycle data of the purchase order (generation, payment, transportation) is permanently recorded on the blockchain.

[0050] A further improvement of the technical solution of the present invention is that the execution feedback and model correction module specifically includes:

[0051] IoT sensors deployed within the hotel collect real-time inventory consumption and logistics timeliness data, accurately monitoring the usage frequency and inventory status of guest room supplies, as well as the real-time progress of logistics distribution. The collected actual data is compared with demand forecasts, and the forecast error is calculated to ensure that the system can promptly understand the deviation between the forecast and actual situation. If the error is within the preset range of inventory error rate and logistics timeliness deviation, the current plan will continue to be executed. If the error exceeds the preset threshold, the model correction process will be triggered to ensure timely data feedback and problem discovery. The preset threshold for inventory error rate is 5%, and the preset threshold for logistics timeliness deviation is 2 hours.

[0052] When the prediction error exceeds the preset threshold, relevant historical data and newly acquired real-time data are automatically collected to retrain the model. During the training process, the model parameters are adjusted and the constraints of logistics path planning are optimized to adapt to the new transportation environment and needs.

[0053] Based on the model correction results, the event tag library is updated. The tag library contains the characteristics of various events and the corresponding optimization strategies. The new data features and the corrected model strategies are added to the tag library to form a closed-loop optimization mechanism.

[0054] Due to the adoption of the above technical solution, the present invention has the following technical advancements compared to the prior art:

[0055] The present invention provides a supply chain intelligent management system based on dynamic collaborative optimization. By integrating external dynamic data such as social media public opinion and weather, and combining multimodal data analysis technology, it can significantly improve the accuracy of demand forecasting. By using the ASTGNN model, combined with bidirectional LSTM and GCN, time series trends and demand correlations between branches are analyzed, and feature weights are dynamically adjusted, the demand forecast error rate is greatly reduced, providing enterprises with a more reliable demand forecast basis, and helping to optimize inventory management and production planning.

[0056] The present invention provides a supply chain intelligent management system based on dynamic collaborative optimization. Through multi-objective collaborative optimization, it adopts the APSO algorithm to balance inventory costs, logistics timeliness and carbon emissions, automatically analyzes the constraints of supplier production capacity and logistics routes, and generates a Pareto optimal solution set. It effectively reduces inventory backlogs and stock-outs, improves inventory turnover, and reduces out-of-stock rates. While ensuring supply continuity, it can reduce unnecessary inventory holding costs and improve capital utilization efficiency.

[0057] The present invention provides a supply chain intelligent management system based on dynamic collaborative optimization. It utilizes blockchain collaborative execution to convert the optimal solution into a smart contract and deploy it on the blockchain, achieving real-time synchronization and transparent traceability of purchase orders, payment status, and logistics progress. Suppliers and logistics providers achieve cross-chain data synchronization through Hyperledger Fabric, ensuring the timeliness and accuracy of information transmission, greatly shortening the response time for order changes, and improving the overall collaborative efficiency of the supply chain. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments described in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0059] Figure 1 It is a schematic diagram of the workflow of the present invention;

[0060] Figure 2 Schematic diagram of the workflow of the dynamic demand forecasting module of the present invention. DETAILED DESCRIPTION

[0061] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0062] Example 1, as Figure 1 、 Figure 2 As shown, the present invention provides a supply chain intelligent management system based on dynamic collaborative optimization, including a supply chain intelligent management platform, which is communicatively connected to the following modules, wherein:

[0063] The multimodal data perception module is used to collect multimodal data inside the hotel through the LoRaWAN+BLE hybrid sensor network, and access external data sources such as social media public opinion and weather forecasts, obtain real-time information through the API interface, deploy the LoRaWAN+BLE hybrid sensor network inside the hotel, collect multimodal data inside the hotel, including temperature and humidity, pressure and frequency of use of items in the guest rooms, etc. At the same time, use the API interface to access external data sources including social media public opinion (grabbing keywords through platforms such as Weibo and Twitter) and weather forecasts (obtaining real-time weather information from OpenWeatherMap), obtain hot topics on social media, weather changes and other information in real time, ensure the comprehensiveness and real-time nature of the data, and collect The multi-sensor data is fused and processed to improve the accuracy and reliability of the data, and the positioning error correction model is used to correct the data to reduce the positioning error. Then, the hotel's internal and external data are preprocessed, the data is cleaned, and outliers are removed. For missing data, a time series interpolation algorithm is used to fill in the missing data to ensure data integrity. For unstructured text data in external data, it is converted into semantic vectors through the BERT model, and all data is unified to the UTC timestamp with an accuracy of minutes to achieve time alignment. For the preprocessed data, feature engineering is performed to generate lagged features and event label annotations, and a standardized feature matrix containing 100+ dimensions is output. The standardized feature matrix is ​​transmitted to the supply chain intelligent management platform through the communication protocol;

[0064] The multi-sensor data fusion formula is:

[0065] ;

[0066] Where, is the fused sensor data, is the number of sensors, For the The weight of each sensor data is dynamically adjusted through Kalman filtering. The weight reflects the importance of each sensor data in the fusion process. For the The data of the sensors, is the noise term, which represents the random error introduced in the data fusion process;

[0067] The formula of the positioning error correction model is:

[0068] ;

[0069] Where, is the corrected positioning position, is the original positioning position, is the direction correction coefficient, which is used to adjust the correction amount. To calculate the vector on the two-dimensional plane The angle of the result is in the range of arrive , For vector The component on the axis, For vector The component on the axis, is the displacement vector, which represents the vector from the original position to the corrected position;

[0070] The dynamic demand forecasting module is based on the ASTGNN model. It uses bidirectional LSTM to capture time series trends, combines GCN to analyze demand correlations between branches, and conducts demand forecasts for the next 7 days by time period. It uses the pre-built ASTGNN model to extract spatiotemporal features. In the time dimension, it uses bidirectional LSTM to capture time series trends and identify seasonal fluctuations and long-term trends (such as weekend occupancy rate fluctuations). In the spatial dimension, it uses GCN (graph convolutional network) to analyze demand correlations between branches (such as the probability of store B adjusting stock when store A is out of stock). The spatial relationship between branches is represented by a graph structure, and the dynamic stock adjustment relationship between branches is then learned. Among them, the ASTGNN model is an attention-enhanced spatiotemporal graph neural network model. On the basis of feature extraction, it introduces the attention mechanism to dynamically adjust the importance weight of each data feature. By calculating the attention weight of each data feature (for example, a weight coefficient of 1.5 for active events and 1.0 for regular data), adjusting the weights and assigning higher weights to important features to highlight their impact on demand forecasting, the model can better capture the short-term impact of emergencies on demand, thereby improving the accuracy and adaptability of the forecast. After completing feature extraction and weight adjustment, the online learning phase begins, triggering incremental training of the model every 15 minutes. A loss function is designed to quantify the deviation between the predicted value and the true value, and a joint average is performed in the spatiotemporal dimensions to ensure global optimization. The AdamW optimizer (learning rate = 0.001) is then used to update the parameters to ensure that the model can adapt to the latest data changes in real time. Finally, the ASTGNN model is used to output the demand forecast value for the next 7 days by time period, with the forecast error rate controlled within 8%.

[0071] In addition, the calculation process of the loss function is:

[0072] In the ASTGNN model, for each prediction point, the square of the difference between its predicted value and the true value is calculated, the squares of the differences of all prediction points are summed to obtain the total error, and then divided by the number of prediction points to obtain the mean square error, where the mean square error is a common indicator for measuring the difference between the model prediction value and the true value. In the ASTGNN model, the regularization parameter is determined, the sum of the squares of all model parameters is calculated, and the sum of the squares of all parameters is multiplied by the regularization parameter to obtain the regularization term, where the regularization term is used to prevent the model from overfitting. By adding the model parameters to the loss function, the regularization term is used to prevent the model from overfitting. The norm (i.e., the sum of the squares of the parameters) is used to implement the regularization parameter, which controls the importance of the regularization term in the total loss. The mean square error and the regularization term are added together to obtain the total loss function of the ASTGNN model. The total loss function comprehensively considers the accuracy of the model's predictions (measured by the mean square error) and the complexity of the model (controlled by the regularization term), thereby seeking a balance between prediction accuracy and model generalization ability during the training process. By minimizing the loss function, the ASTGNN model can learn a spatiotemporal feature representation that can accurately predict demand without being overly complex.

[0073] The formula of the loss function is:

[0074] ;

[0075] Where, is the value of the loss function, which represents the overall error of the model prediction. is the total number of samples, that is, the number of data points used in model training, For the model The predicted value of the sample, For the The true value of the sample, is a regularization parameter used to control the weight of the regularization term to prevent overfitting. is the set of model parameters, including all weights and biases, For model parameters The square of the norm, that is, the sum of the squares of all parameters;

[0076] The multi-objective collaborative optimization module is used to balance inventory costs, logistics timeliness and carbon emissions using the APSO algorithm, automatically analyze the constraints of supplier production capacity and logistics routes, and dynamically adjust the inertia weight and learning factor to generate a Pareto optimal solution set within 10 minutes. The multi-objective collaborative optimization module analyzes the constraints of supplier production capacity constraints (such as daily maximum supply) and logistics vehicle scheduling constraints (such as load restrictions and route planning), including but not limited to the supplier's daily maximum supply, logistics vehicle load restrictions and route planning, and performs initialization operations to generate a certain number of particles, which are candidate solutions. Each particle represents a combination of procurement plans, supplier selections and transportation routes. After initialization is completed, the multi-objective collaborative optimization module executes the APSO algorithm and calculates the fitness value of each particle by simulating the behavior of the particle swarm to find the optimal solution. Each particle is based on Its individual optimal and global optimal positions update their own speed and position, and during the iteration process, dynamically adjust the inertia weight and learning factor to balance the capabilities of global search and local search to maintain the diversity and effectiveness of the search. In the multi-objective collaborative optimization process, a carbon emission constraint algorithm is introduced to handle carbon emission constraints through an adaptive penalty mechanism. In logistics path planning, logistics path optimization constraints are preset to ensure that the transportation route meets the constraints. After a series of iterations, when there is no significant improvement in fitness for 50 consecutive generations, the algorithm terminates and generates a Pareto optimal solution set, which represents a combination of solutions that achieve the optimal or near-optimal solution under different objectives. It contains multiple optional solutions that strike a balance between inventory costs, logistics timeliness and carbon emissions. The top three optimal solutions are retained for manual decision-making. The entire optimization process is finally completed within 10 minutes, and the optimal procurement plan and logistics scheduling plan are output.

[0077] The formula of the APSO algorithm is:

[0078] ;

[0079] Where, is the particle In the The speed of the generation is the inertia weight, which controls the influence of the particle's previous speed. and is the learning factor, also known as the acceleration constant, which is used to adjust the speed at which particles move toward the individual optimal and global optimal positions. and For random numbers in the range [0,1], increase randomness, For particles In the The optimal position of individuals in the generation, For the The global optimal position of the generation, For particles In the The position of the generation;

[0080] In addition, the process of carbon emission constraints and logistics path optimization constraints is as follows:

[0081] In the initial stage of multi-objective collaborative optimization, carbon emission targets are incorporated into the optimization model, an upper limit on carbon emission intensity is set, and constraints are dynamically adjusted through an adaptive penalty mechanism. In the logistics path planning stage, logistics path optimization constraints are preset, including hard and soft constraints, and the carbon emission constraint algorithm is used to solve them. Path combinations that meet all constraints are prioritized, and then the cost is optimized in combination with the carbon emission target. The three objectives of inventory cost, logistics timeliness, and carbon emission are iteratively optimized based on the APSO algorithm. In each iteration, the degree of constraint violation of the particle (candidate solution) is dynamically evaluated: if the carbon emission or logistics path constraints are not met, the fitness value is reduced through the penalty function to guide the APSO algorithm to converge to the feasible solution area. At the same time, the constraint parameters are dynamically adjusted according to real-time data to ensure that the solution remains effective in the face of environmental and operational changes.

[0082] The formula for the carbon emission constraint algorithm is:

[0083] ;

[0084] Where, is the adjusted fitness value, used in the selection mechanism of the particle swarm algorithm, is the original multi-objective optimization function, is the penalty coefficient, which is dynamically adjusted through the adaptive mechanism. is the actual carbon emission intensity (unit: kg / km·t), is the upper limit of carbon emission intensity, when When the penalty item is triggered Otherwise, the penalty term is 0, which does not affect the original objective function. The optimization goal is to minimize , that is, optimizing inventory costs and logistics time at the same time, and ensuring that carbon emissions do not exceed the limit. As the number of iterations increases, Exponential growth, allowing a certain degree of constraint violation in the early stage and enforcing constraint satisfaction in the later stage;

[0085] The formula for logistics path optimization constraints is:

[0086] ;

[0087] ;

[0088] Where, is the set of all nodes, Gather at the departure point, is the set of destinations, decision variables (usually binary variables), Indicates whether to select a slave node To Node Path;

[0089] The process of outputting the optimal procurement plan and logistics scheduling plan is as follows:

[0090] During the iteration of the APSO algorithm, the fitness value changes of each generation of particles are continuously monitored. If there is no significant improvement in fitness for 50 consecutive generations (i.e. less than 0.1%), the algorithm is determined to have converged, the iteration is terminated, and the Pareto optimal solution set is screened from all non-dominated solutions. At the same time, the fast non-dominated sorting technology is used to ensure that the solution set covers the most diverse solutions. From the Pareto optimal solution set, the top three optimal solutions are retained based on the target weight or the decision maker's preference indicator. Among them, if the company prioritizes low-carbon goals, the solution with the lowest carbon emissions and moderate timeliness and cost is selected. If it is an urgent replenishment, then The system prioritizes the most efficient solution and generates a visual comparison report for each optimal solution, displaying key indicators such as inventory cost, transportation time, and carbon emissions, and noting whether constraints are satisfied. Within a 10-minute time limit, the system automatically outputs the optimal procurement plan including supplier selection and procurement volume, as well as a logistics scheduling plan including route planning and vehicle allocation. The output is presented in structured data and visual charts to ensure executability. If the optimization times out, it falls back to the previous generation of feasible solutions, prioritizing carbon emission constraints. The final solution is then synchronized to the blockchain module, triggering the automatic execution of the smart contract.

[0091] The blockchain collaborative execution module is used to convert the optimal solution into a smart contract and deploy it to the blockchain, automatically triggering processes such as order generation and payment. The purchase order hash value is uploaded to the blockchain in real time, and suppliers and logistics providers can synchronize cross-chain data through Hyperledger Fabric to ensure transparent and traceable execution status.

[0092] The execution feedback and model correction module is used to provide real-time feedback on inventory consumption and logistics timeliness through IoT sensors, calculate prediction errors and trigger model corrections. If the error exceeds the preset threshold, the model is automatically retrained and the event tag library is updated to form a closed-loop optimization mechanism.

[0093] Example 2, as Figure 1 、 Figure 2 As shown, based on Example 1, the present invention provides a technical solution: preferably, the blockchain collaborative execution module specifically includes:

[0094] The optimal procurement and logistics plan output by the multi-objective collaborative optimization module is converted into a smart contract and deployed to the blockchain platform. The trigger conditions are clearly set in the smart contract. When the inventory level falls below the preset safety threshold, the order generation process is automatically triggered. At the same time, the smart contract contains information on order generation and payment terms to ensure the automation and accuracy of the order generation and payment processes. Once the smart contract is triggered, the relevant purchase order information is converted into a hash value and written into the blockchain in real time. At the same time, through Hyperledger Fabric's blockchain technology synchronizes cross-chain data between suppliers and logistics providers, synchronizing key information such as purchase orders, payment status, and logistics progress to the blockchain nodes of each participant in real time, ensuring that all participants can obtain the latest execution status. Furthermore, during the cross-chain synchronization process, the SPECTRE protocol delay model is used to optimize the blockchain fork selection rule through GHOST, improving transaction confirmation speed and throughput, assessing the delay time of transaction confirmation, and ensuring data consistency and reliability among all participants. GHOST is the Ghost Protocol. The blockchain collaborative execution module continuously monitors the execution status of smart contracts and provides real-time feedback on order execution. At the same time, order status information is recorded on the blockchain, providing real-time and transparent execution tracking for supply chain management.

[0095] In addition, the specific process of cross-chain synchronization combined with the SPECTRE protocol is as follows:

[0096] When the triggering conditions of the smart contract are met, the purchase order information is automatically converted into a hash value through the SHA-256 algorithm and written to the blockchain in real time. The chain code logic of Hyperledger Fabric is used to generate cross-chain synchronization events at the same time, and the hash value is broadcast to the blockchain nodes of suppliers and logistics providers through Fabric's channel mechanism. At the same time, each node quickly selects the main chain branch based on the "ghost protocol" rules of the SPECTRE protocol to ensure the initial consistency of the hash value in the distributed network. Through the parallel block processing capability of GHOST, the delay caused by forks is reduced and information synchronization is accelerated. Hyperledger Fabric uses its cross-chain technology (Interchain Communication) to synchronize real-time data between supplier and logistics provider nodes, and transmits key information including purchase orders, payment status and logistics progress through event logs (Event Logs). The SPECTRE protocol delay model is used to evaluate transaction confirmation delays during this process. GHOST prioritizes branches containing the most subblocks, optimizes fork selection rules, improves throughput, and shortens confirmation times, ensuring cross-chain data consensus within seconds. Each participant's blockchain node continuously monitors cross-chain events and verifies data consistency using the SPECTRE protocol's finality rules, verifying that each node's hash value matches the latest block on the main blockchain. If a fork conflict occurs, the GHOST mechanism quickly eliminates the inferior branch based on block generation rate and security parameters, ultimately converging all nodes to the same state. Ultimately, the entire lifecycle data of the purchase order (generation, payment, and transportation) is permanently recorded on the blockchain, allowing each participant to query the real-time status through the node, achieving transparent and traceable supply chain execution tracking.

[0097] The formula for the SPECTRE protocol delay model is:

[0098] ;

[0099] Where, The transaction confirmation delay is the time required for the transaction to be confirmed by the network consensus. is the error probability, that is, the upper limit of the probability that the transaction is not correctly confirmed, usually set to 0.01, security parameters (such as the required block confirmation depth or number of nodes), is the block generation rate, that is, the number of new blocks generated by the entire network per unit time, is a probability parameter (e.g., the probability that a transaction is not included in enough blocks);

[0100] The execution feedback and model correction module specifically includes:

[0101] By deploying IoT sensors inside the hotel, real-time data on inventory consumption and logistics timeliness are collected, the frequency of use and inventory status of guest room supplies, as well as the real-time progress of logistics distribution, are accurately monitored. The collected actual data is compared with the demand forecast, and the forecast error is calculated to ensure that the system can timely understand the deviation between the forecast and the actual situation. If the error is within the preset range of inventory error rate and logistics timeliness deviation, the current plan will continue to be executed. If the error exceeds the preset threshold, the model correction process will be triggered to ensure timely data feedback and problem discovery. Among them, the preset threshold of inventory error rate is 5%, and the preset threshold of logistics timeliness deviation is 2 hours. When the forecast error exceeds the preset threshold, the relevant historical data and newly collected The model is retrained based on real-time data, and during the training process, the model parameters are adjusted, including the inertia weight of the particle swarm algorithm, the learning factor, and the penalty coefficient in the carbon emission constraint algorithm. At the same time, the constraints of the logistics path planning are optimized to adapt to the new transportation environment and demand. According to the model correction results, the event label library is updated. The label library contains the characteristics of various events and the corresponding optimization strategies, including labels and response plans for events such as supplier delivery delays and logistics route congestion. The new data features and the revised model strategy are added to the label library to form a closed-loop optimization mechanism to continuously improve the performance of the model, so that it can better balance inventory costs, logistics timeliness and carbon emissions, and improve the overall efficiency and reliability of supply chain management.

[0102] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A supply chain intelligent management system based on dynamic collaborative optimization, including a supply chain intelligent management platform, characterized by: The supply chain intelligent management platform is connected to the following modules: The multimodal data perception module is used to collect multimodal data within the hotel through a LoRaWAN+BLE hybrid sensor network, connect to external data sources, and obtain real-time information through an API interface; The dynamic demand forecasting module, based on the ASTGNN model, uses bidirectional LSTM to capture time series trends and combines GCN to analyze demand correlations between branches to forecast demand by time period for the next seven days. A multi-objective collaborative optimization module uses the APSO algorithm to balance inventory costs, logistics timeliness, and carbon emissions. It automatically analyzes supplier capacity and logistics route constraints, and dynamically adjusts inertia weights and learning factors to generate a Pareto optimal solution set within 10 minutes. The blockchain collaborative execution module is used to convert the optimal solution into a smart contract and deploy it to the blockchain, and enable suppliers and logistics providers to synchronize cross-chain data through Hyperledger Fabric; The execution feedback and model correction module is used to provide real-time feedback on inventory consumption and logistics timeliness through IoT sensors, calculate prediction errors, and trigger model corrections. If the error exceeds a preset threshold, the model is automatically retrained and the event tag library is updated. The multi-objective collaborative optimization module specifically includes: The multi-objective collaborative optimization module analyzes the constraints of supplier capacity limitations and logistics providers' vehicle scheduling constraints, performs initialization operations, and generates a certain number of particles, namely candidate solutions. Each particle represents a combination of procurement plans, supplier selections, and transportation routes. After initialization, the multi-objective collaborative optimization module executes the APSO algorithm to calculate the fitness value of each particle by simulating the behavior of the particle swarm to find the optimal solution. Each particle updates its own speed and position based on its individual optimal and global optimal positions. During the iteration process, the inertia weight and learning factor are dynamically adjusted to balance the capabilities of global and local search. In the multi-objective collaborative optimization process, a carbon emission constraint algorithm is introduced to handle carbon emission constraints through an adaptive penalty mechanism. In logistics path planning, logistics path optimization constraints are preset to ensure that the transportation route meets the constraints. After a series of iterations, if there is no significant improvement in fitness for 50 consecutive generations, the algorithm terminates and generates a Pareto optimal solution set. This set represents a combination of solutions that achieve optimal or near-optimal results under different objectives. It includes multiple options that balance inventory costs, logistics timeliness, and carbon emissions. The top three optimal solutions are retained for manual decision-making. The entire optimization process is completed within 10 minutes, and the optimal procurement plan and logistics scheduling plan are output. The formula for the carbon emission constraint algorithm is: ； Where, is the adjusted fitness value, used in the selection mechanism of the particle swarm algorithm, is the original multi-objective optimization function, is the penalty coefficient, which is dynamically adjusted through the adaptive mechanism. is the actual carbon emission intensity (unit: kg / km·t), is the upper limit of carbon emission intensity, when When the penalty item is triggered Otherwise, the penalty term is 0, which does not affect the original objective function. The optimization goal is to minimize , that is, optimizing inventory costs and logistics time at the same time, and ensuring that carbon emissions do not exceed the limit. As the number of iterations increases, Exponential growth allows a certain degree of constraint violation in the early stage and enforces constraint satisfaction in the later stage.

2. The supply chain intelligent management system based on dynamic collaborative optimization according to claim 1 is characterized by: The multimodal data perception module specifically includes: A LoRaWAN+BLE hybrid sensor network is deployed within the hotel to collect multimodal data, including room temperature, humidity, pressure, and item usage frequency. APIs are also used to access external data sources, including social media sentiment and weather forecasts, to obtain real-time information on trending topics and weather changes on social media. The collected multi-sensor data is fused and corrected using a positioning error correction model. Furthermore, internal and external hotel data are pre-processed and all data is unified to a UTC timestamp with minute-level accuracy to achieve time alignment. For the preprocessed data, feature engineering is performed to generate lagged features and event label annotations, and a standardized feature matrix containing 100+ dimensions is output. The standardized feature matrix is ​​then transmitted to the supply chain intelligent management platform through a communication protocol.

3. The supply chain intelligent management system based on dynamic collaborative optimization according to claim 1 is characterized by: The dynamic demand forecasting module specifically includes: The pre-built ASTGNN model is used to extract spatiotemporal features. In the temporal dimension, a bidirectional LSTM is used to capture time series trends and identify seasonal fluctuations and long-term trends. In the spatial dimension, a GCN is used to analyze demand correlations between branches. The spatial relationships between branches are represented by a graph structure, and the dynamic inventory transfer relationships between branches are learned. The ASTGNN model is an attention-enhanced spatiotemporal graph neural network model. Based on feature extraction, the attention mechanism is introduced to dynamically adjust the importance weight of each data feature. By calculating the attention weight of each data feature, the weight is adjusted and a higher weight is given to important features. After feature extraction and weight adjustment, the system enters the online learning phase, triggering incremental training of the model every 15 minutes. A loss function is designed to quantify the deviation between the predicted value and the true value, and joint averaging is performed in the spatiotemporal dimensions. The AdamW optimizer is then used to update the parameters. Finally, the ASTGNN model is used to output the demand forecast value for the next seven days by time period, with the forecast error rate controlled within 8%.

4. The supply chain intelligent management system based on dynamic collaborative optimization according to claim 3 is characterized by: The calculation process of the loss function is: In the ASTGNN model, for each prediction point, the square of the difference between its predicted value and the true value is calculated, the square of the difference of all prediction points is summed up to get the total error, and then divided by the number of prediction points to get the mean square error; In the ASTGNN model, the regularization parameter is determined, the sum of squares of all model parameters is calculated, and the sum of squares of all parameters is multiplied by the regularization parameter to obtain the regularization term; The total loss function of the ASTGNN model is obtained by adding the mean square error and the regularization term.

5. The supply chain intelligent management system based on dynamic collaborative optimization according to claim 1 is characterized by: The process of carbon emission constraints and logistics path optimization constraints is as follows: In the initial stage of multi-objective collaborative optimization, carbon emission targets are incorporated into the optimization model, an upper limit on carbon emission intensity is set, and constraints are dynamically adjusted through an adaptive penalty mechanism; During the logistics route planning phase, logistics route optimization constraints are preset, including hard and soft constraints. A carbon emission constraint algorithm is used to solve the problem, prioritizing the selection of route combinations that meet all constraints. The cost is then optimized based on the carbon emission target. The APSO algorithm is used to iteratively optimize the three objectives of inventory cost, logistics timeliness, and carbon emissions. In each iteration, the degree of constraint violation of the particle is dynamically evaluated: if the carbon emission or logistics path constraints are not met, the fitness value is reduced through the penalty function, guiding the APSO algorithm to converge to the feasible solution area.

6. The supply chain intelligent management system based on dynamic collaborative optimization according to claim 5 is characterized by: The process of outputting the optimal procurement plan and logistics scheduling plan is as follows: During the iteration process of the APSO algorithm, the fitness value changes of each generation of particles are continuously monitored. If there is no significant improvement in fitness for 50 consecutive generations, the algorithm is considered to have converged, the iteration is terminated, and the Pareto optimal solution set is selected from all non-dominated solutions. From the Pareto optimal solution set, the top three solutions are retained based on target weights or decision maker preferences. A visual comparison report is generated for each optimal solution, showing key indicators such as inventory cost, transportation time, and carbon emissions, and noting whether the constraints are satisfied. Within a 10-minute time limit, the system automatically outputs the optimal procurement plan including supplier selection and procurement volume, and the logistics scheduling plan including route planning and vehicle allocation, and presents the output content in the form of structured data and visual charts. If the optimization times out, it will fall back to the previous generation of feasible solutions, giving priority to ensuring carbon emission constraints, and then synchronize the final plan to the blockchain module, triggering the automatic execution of the smart contract.

7. The supply chain intelligent management system based on dynamic collaborative optimization according to claim 6 is characterized by: The blockchain collaborative execution module specifically includes: The optimal procurement and logistics solutions output by the multi-objective collaborative optimization module are converted into smart contracts and deployed on the blockchain platform. The smart contracts clearly define the triggering conditions and also include information on order generation and payment terms. Once the smart contract is triggered, the relevant purchase order information is converted into a hash value and written to the blockchain in real time. At the same time, cross-chain data synchronization between suppliers and logistics providers is carried out through Hyperledger Fabric's blockchain technology. Key information such as purchase orders, payment status, and logistics progress are synchronized in real time to the blockchain nodes of each participant. During the cross-chain synchronization process, the SPECTRE protocol delay model is used to optimize the blockchain fork selection rule through GHOST to evaluate the delay time of transaction confirmation. GHOST is the Ghost Protocol. The blockchain collaborative execution module continuously monitors the execution status of smart contracts, provides real-time feedback on the execution of orders, and records the status information of orders on the blockchain.

8. The supply chain intelligent management system based on dynamic collaborative optimization according to claim 7 is characterized by: The specific process of cross-chain synchronization combined with the SPECTRE protocol is as follows: When the triggering conditions of the smart contract are met, the purchase order information is automatically converted into a hash value using the SHA-256 algorithm and written to the blockchain in real time. Using Hyperledger Fabric's chain code logic, a cross-chain synchronization event is generated simultaneously, and the hash value is broadcast to the supplier and logistics provider's blockchain nodes through Fabric's channel mechanism. Hyperledger Fabric uses its cross-chain technology to synchronize real-time data between supplier and logistics provider nodes. Key information, including purchase orders, payment status, and logistics progress, is transmitted between chains through event logs and independently verified by each node. During this process, the SPECTRE protocol delay model is used to evaluate transaction confirmation delays. Each participant's blockchain node continuously monitors cross-chain events and verifies data consistency using the finality rules of the SPECTRE protocol, verifying whether the hash value of each node matches the latest block of the blockchain main chain. If there is a fork conflict, the GHOST mechanism quickly eliminates the inferior branch based on the block generation rate and security parameters, so that all nodes eventually converge to the same state. Ultimately, the entire life cycle data of the purchase order is permanently recorded on the blockchain. The formula for the SPECTRE protocol delay model is: ； Where, The transaction confirmation delay is the time required for the transaction to be confirmed by the network consensus. is the error probability, that is, the upper limit of the probability that the transaction is not correctly confirmed, usually set to 0.01, is a safety parameter, is the block generation rate, that is, the number of new blocks generated by the entire network per unit time, is the probability parameter.

9. The supply chain intelligent management system based on dynamic collaborative optimization according to claim 1, characterized in that: The execution feedback and model correction module specifically includes: IoT sensors deployed within the hotel collect real-time inventory consumption and logistics timeliness data, monitor the usage frequency and inventory status of guest room supplies, and the real-time progress of logistics distribution. The collected actual data is compared with the demand forecast and the forecast error is calculated. If the error is within the preset range of inventory error rate and logistics timeliness deviation, the current plan will continue to be implemented. If the error exceeds the preset threshold, the model correction process will be triggered. The preset threshold for inventory error rate is 5%, and the preset threshold for logistics timeliness deviation is 2 hours. When the prediction error exceeds the preset threshold, relevant historical data and newly acquired real-time data are automatically collected to retrain the model. During the training process, the model parameters are adjusted and the constraints of logistics path planning are optimized. Based on the model correction results, the event tag library is updated. The tag library contains the characteristics of various events and the corresponding optimization strategies. The new data features and the corrected model strategies are added to the tag library to form a closed-loop optimization mechanism.

Images