E-commerce inventory dynamic optimization model system based on real-time big data

Abstract

The present invention relates to the field of model optimization technology, and in particular to an e-commerce inventory dynamic optimization model system based on real-time big data, which includes an inventory information sharing unit, an intelligent replenishment algorithm unit and an inventory allocation strategy unit. In the present invention, the inventory information sharing unit realizes real-time collection, processing and secure sharing of inventory data by deploying Internet of Things sensors at key nodes of the supply chain and combining edge computing with blockchain technology. The intelligent replenishment algorithm unit obtains data from a blockchain platform, builds a demand forecasting model using a gated loop module, sets a replenishment threshold value for an automatic replenishment module and adjusts the replenishment quantity in combination with risk factors, selects an optimal supplier using a multi-objective optimization algorithm, and the inventory allocation strategy unit formulates and optimizes inventory allocation rules based on real-time data using a reinforcement learning algorithm, adjusts the inventory allocation ratio and uploads it to a blockchain, thereby improving inventory management efficiency, reducing costs and realizing dynamic inventory optimization.

Description

Technical Field

[0001] The present invention relates to the field of model optimization technology, and in particular to an e-commerce inventory dynamic optimization model system based on real-time big data. Background Art

[0002] E-commerce inventory management is an important technology. With the booming e-commerce industry, inventory management has become a key factor affecting the operational efficiency and economic benefits of e-commerce companies. Traditional e-commerce inventory management relies on manual inventory and regular reporting, which makes it difficult to grasp the location, quantity and status of inventory goods in real time, resulting in a disconnect between inventory data and actual conditions, affecting the timeliness and accuracy of corporate decision-making.

[0003] Traditional methods are mostly based on experience or simple historical data statistics for forecasting and replenishment, which cannot fully consider market dynamics. When consumer preferences change rapidly or social media popularity changes the demand for goods, companies often face the dilemma of out-of-stock or inventory backlogs. Out-of-stock will lead to customer loss, while inventory backlogs increase warehousing costs, which in turn leads to the ineffective use of inventory resources. In order to solve this technical problem, we provide an e-commerce inventory dynamic optimization model system based on real-time big data. Summary of the invention

[0004] The purpose of the present invention is to provide an e-commerce inventory dynamic optimization model system based on real-time big data to solve the problems raised in the above background technology.

[0005] To achieve the above objectives, an e-commerce inventory dynamic optimization model system based on real-time big data is provided, including an inventory information sharing unit, an intelligent replenishment algorithm unit, and an inventory allocation strategy unit;

[0006] The inventory information sharing unit deploys IoT sensors in supplier warehouses, logistics transportation vehicles and e-commerce warehouses in the supply chain, processes the location, quantity and status data of inventory goods collected by the IoT sensors through edge computing devices, and uploads the preliminarily processed data to the alliance chain blockchain platform;

[0007] The intelligent replenishment algorithm unit obtains historical sales data, real-time market trends and supply chain inventory information from the alliance chain blockchain platform, uses a gated loop module including a gated loop unit to build a demand forecasting model, uses pre-processed historical data to train the model, and dynamically adjusts the forecast results according to real-time data changes. Then, the automatic replenishment module sets the inventory safety threshold and replenishment point for each commodity. When the node inventory level is lower than the replenishment point, the smart contract module automatically triggers the replenishment process, introduces risk factors that consider supply chain uncertainty, calculates risk factors based on historical data and real-time monitoring conditions, and dynamically adjusts the replenishment quantity. Finally, a multi-objective optimization algorithm is used to select the optimal supplier in combination with real-time inventory information and the supplier's supply capacity, price and delivery time;

[0008] The inventory allocation strategy unit uses reinforcement learning algorithms to develop and continuously optimize dynamic inventory allocation rules based on real-time sales data, demand forecasts for each region, and inventory costs. It adjusts the inventory allocation ratio in real time according to actual sales data and market change data, and uploads it to the alliance chain blockchain platform.

[0009] As a further improvement of the technical solution, in the inventory information sharing unit, when the edge computing device processes the location, quantity and status data of the inventory goods collected by the IoT sensor, a data fusion algorithm is adopted, and the specific steps are as follows:

[0010] The data collected by different types of IoT sensors are synchronized in time, and the data collected by different sensors at similar time points are processed as a group. The similar time points are the time when the data points were collected. Minutes, data with timestamp differences within the set threshold are considered synchronized data;

[0011] The Kalman filter algorithm is used to process the position data. For the quantity data, the sliding window statistical method is used to calculate the quantity change rate within a certain time window. If the change rate exceeds the set threshold, the data is marked and further verified.

[0012] The status data is classified and encoded, and different statuses are converted into numerical codes, including normal, damaged, and out of stock. Finally, the processed location, quantity, and status data are integrated to form a comprehensive inventory data record and uploaded to the alliance chain blockchain platform.

[0013] As a further improvement of the technical solution, the gated loop module adopts the following training and adjustment methods when constructing the demand forecasting model:

[0014] The historical sales data obtained from the alliance chain blockchain platform is cleaned, and the missing values ​​are filled by linear interpolation. Time features, seasonal features, and promotional activity features are extracted from the historical sales data, and these features are used together with the sales data as the input of the model;

[0015] The gated recurrent module containing the gated recurrent unit is trained using preprocessed historical data, and the model parameters are updated using the stochastic gradient descent algorithm. The model is then adjusted using online learning based on real-time data changes. When new real-time sales data arrives, it is added to the training data, the gradient is recalculated, and the model parameters are updated.

[0016] As a further improvement of the technical solution, the automatic replenishment module adopts the following method when setting the inventory safety threshold and replenishment point:

[0017] Calculate the average daily demand of each commodity based on historical sales data, and calculate the standard deviation of demand based on the average daily demand, where the standard deviation of demand is used to measure the degree of fluctuation of demand;

[0018] Based on the supplier's historical delivery data, the average replenishment lead time is calculated, and the safety stock is calculated based on the average replenishment lead time. Finally, the average daily demand, average replenishment lead time and safety stock are substituted into the formula to calculate the replenishment point. When the node inventory level is lower than the replenishment point, the replenishment process is triggered.

[0019] As a further improvement of the technical solution, the smart contract module calculates the risk factor and adjusts the replenishment quantity using the following method:

[0020] Count the number of supply interruptions and total supply times of suppliers in history, and calculate the supply chain interruption risk factor based on the number of supply interruptions and total supply times;

[0021] Then, the ratio of the standard deviation of the demand to the average daily demand is taken as the coefficient of variation of the demand, and the coefficient of variation of the demand is taken as the demand fluctuation risk factor;

[0022] Statistics are collected on the number of delayed deliveries and the total number of transportations by logistics transport vehicles, and the transport delay risk factor is calculated based on the number of delayed deliveries and the total number of transportations. The above three risk factors are comprehensively considered to calculate the comprehensive risk factor, and the initial replenishment quantity is adjusted based on the comprehensive risk factor to obtain the adjusted replenishment quantity.

[0023] As a further improvement of the technical solution, the intelligent replenishment algorithm unit adopts the following steps when selecting the optimal supplier using the multi-objective optimization algorithm:

[0024] Supplier The supply capacity is , the price is , delivery date is , the inventory holding cost is , the out-of-stock cost is , then the objective function for:

[0025] ;

[0026] in is the weight coefficient, and , To suppliers The order quantity, is the predicted demand, order quantity Need to meet ,and ,in is the number of suppliers;

[0027] A genetic algorithm is introduced to solve the multi-objective optimization problem. A population containing multiple chromosomes is initialized. Each chromosome represents a supplier selection scheme and the corresponding order quantity distribution. The fitness value of each chromosome is calculated. The fitness value is the objective function value. A new population is generated through selection, crossover and mutation operations. The iteration is repeated until the termination condition is met.

[0028] The supplier and order quantity allocation corresponding to the chromosome with the largest fitness value are selected as the optimal solution, and replenishment orders are automatically sent to the selected supplier.

[0029] As a further improvement of the technical solution, the specific method steps of introducing a genetic algorithm to solve the multi-objective optimization problem in the intelligent replenishment algorithm unit are as follows:

[0030] Create a population containing multiple chromosomes, each chromosome represents a supplier selection scheme and the corresponding order quantity allocation;

[0031] For each chromosome in the population, its corresponding order quantity distribution is substituted into the objective function. The calculated objective function value is the fitness value of the chromosome. The supplier selection scheme and order quantity distribution are proportional to the fitness value.

[0032] According to the fitness value of the chromosome, a certain number of chromosomes are selected from the current population as parent chromosomes using the roulette wheel selection method for subsequent crossover and mutation operations. Two chromosomes are randomly selected from the selected parent chromosomes and a single-point crossover method is used to generate new daughter chromosomes. In the crossover process, the generated daughter chromosomes must meet the constraints.

[0033] The offspring chromosomes generated by crossover and mutation operations are merged with the parent chromosomes to form a new population. The above steps of selection, crossover, mutation and generation of a new population are repeated until the preset fitness threshold is reached. The supplier and order quantity allocation corresponding to the chromosome with the largest fitness value are selected as the optimal solution, and replenishment orders are automatically sent to the selected suppliers.

[0034] As a further improvement of the technical solution, the inventory allocation strategy unit adopts the following steps when formulating and optimizing inventory allocation rules using a reinforcement learning algorithm:

[0035] defining a state space, the state space comprising a state vector, the state vector comprising real-time sales data, demand forecasts, inventory levels, and inventory costs for each region;

[0036] defining an action space, the action space including an action vector, the action vector representing the inventory allocation ratio of different regions, defining a reward function, the reward function including a reward, the reward being related to the sales profit and inventory cost of each region;

[0037] A deep Q network algorithm is used to construct a Q network, wherein the Q network includes an experience replay pool and an intelligent agent, the experience replay pool is initialized, samples obtained by the interaction between the intelligent agent and the inventory-related data environment are stored in the experience replay pool, a batch of samples are randomly sampled from the experience replay pool, a loss function is calculated, and the parameters of the Q network are updated using a gradient descent algorithm;

[0038] Finally, the inventory allocation ratio is adjusted in real time according to actual sales data and market change data, and uploaded to the alliance chain blockchain platform.

[0039] Compared with the prior art, the present invention has the following beneficial effects:

[0040] In the dynamic optimization model system of e-commerce inventory based on real-time big data, the demand forecasting model constructed by the gated loop module of the intelligent replenishment algorithm unit integrates multi-source data and can be adjusted in real time with high accuracy. The automatic replenishment module sets reasonable thresholds and replenishment points, and adjusts the replenishment quantity in combination with risk factors to avoid out-of-stock and backlog. The multi-objective optimization algorithm selects the best supplier, reduces costs, and ensures supply timeliness. The inventory allocation strategy unit uses a reinforcement learning algorithm to dynamically optimize the allocation rules based on real-time sales data, demand forecasts, and inventory costs, thereby improving inventory utilization, reducing cost waste, and improving enterprise operational efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 It is an overall block diagram of the present invention.

[0042] The meaning of each number in the figure is:

[0043] 1. Inventory information sharing unit; 2. Intelligent replenishment algorithm unit; 21. Gated loop module; 22. Automatic replenishment module; 23. Smart contract module; 3. Inventory allocation strategy unit. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in 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. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0045] The present invention provides an e-commerce inventory dynamic optimization model system based on real-time big data, please refer to Figure 1 As shown, it includes an inventory information sharing unit 1, an intelligent replenishment algorithm unit 2 and an inventory allocation strategy unit 3;

[0046] The inventory information sharing unit 1 deploys IoT sensors in the supplier warehouses, logistics transportation vehicles and e-commerce warehouses of the supply chain, processes the location, quantity and status data of inventory goods collected by the IoT sensors through edge computing devices, and uploads the preliminarily processed data to the alliance chain blockchain platform.

[0047] In the inventory information sharing unit 1, the edge computing device uses a data fusion algorithm to process the location, quantity, and status data of inventory items collected by the IoT sensor. The specific steps are as follows:

[0048] The data collected by different types of IoT sensors are synchronized to improve the accuracy of subsequent data fusion and analysis. The threshold of timestamp difference is set. For the data collected by different sensors, the timestamp information is extracted to determine whether the timestamp difference is within the range of If the data is within the threshold, the data will be regarded as synchronous data and processed as a group.

[0049] The Kalman filter algorithm is used to process the location data. The Kalman filter algorithm is an optimal estimation algorithm that can make the best estimate of the system state based on the system's dynamic model and observation data. For quantity data, the sliding window statistical method is used to calculate the quantity change rate within a certain time window. If the change rate exceeds the set threshold, the data is marked and further verified, which provides more accurate information for the positioning and tracking of inventory goods, and helps to optimize inventory layout and logistics distribution path planning.

[0050] Define the status coding rules, encode the "normal" status as 1, the "damaged" status as 2, and the "out of stock" status as 3. For the collected status data, convert it into the corresponding numerical value according to the coding rule, digitize the status data, and facilitate computer processing and analysis, thereby improving the efficiency and accuracy of data processing. Integrate the processed location, quantity, and status data in a certain format to form a comprehensive data record containing product ID, location information, quantity information, and status code. Upload the data record to the blockchain platform through the interface between the edge computing device and the alliance chain blockchain platform. All parties in the supply chain can obtain accurate and reliable inventory information in real time, which is conducive to collaborative inventory management.

[0051] The intelligent replenishment algorithm unit 2 obtains historical sales data, real-time market trends and supply chain inventory information from the alliance chain blockchain platform, and uses a gated loop module 21 containing a gated loop unit to build a demand forecasting model. The model is trained using pre-processed historical data, and the forecast results are dynamically adjusted according to real-time data changes. The automatic replenishment module 22 sets inventory safety thresholds and replenishment points for each commodity. When the node inventory level is lower than the replenishment point, the smart contract module 23 automatically triggers the replenishment process, introduces risk factors that consider supply chain uncertainty, calculates risk factors based on historical data and real-time monitoring conditions, and dynamically adjusts the replenishment quantity. Finally, a multi-objective optimization algorithm is used to select the optimal supplier based on real-time inventory information and the supplier's supply capacity, price and delivery time.

[0052] When the gated loop module 21 constructs the demand forecasting model, the following training and adjustment methods are used:

[0053] The historical sales data obtained from the alliance chain blockchain platform has problems such as noise, outliers and missing values. The historical sales data obtained from the alliance chain blockchain platform is cleaned, and the missing values ​​are filled by linear interpolation, which improves the quality and reliability of the data.

[0054] Time features are extracted from historical sales data. Months are encoded as integers from 1 to 12, days of the week are encoded as integers from 1 to 7, and seasons are divided into spring, summer, autumn, and winter according to the month, encoded as integers from 1 to 4 respectively. Check whether there are records of promotional activities in the sales data. If so, mark it as 1, otherwise mark it as 0. This enriches the input information of the model, enables the model to take into account more factors that affect commodity demand, and improves the model's expressive and predictive capabilities.

[0055] The gated recurrent module 21 including the gated recurrent unit is trained using the preprocessed historical data. Suppose the input sequence is , then the reset gate of the gated recurrent unit , the mathematical expression of the update gate is , candidate hidden states , the mathematical expression of the hidden state is ,in is the sigmoid function, is the hyperbolic tangent function, and is a learnable weight matrix, It represents element-by-element multiplication, uses mean square error as the loss function, and adopts stochastic gradient descent algorithm to update model parameters so as to quickly converge to the local optimal solution, so that the model can learn the rules of the data in a shorter time. The trained model can better fit the historical sales data and make more accurate predictions on commodity demand.

[0056] When new real-time sales data is received, it is merged with historical sales data to form a new training data set. The gradient of the loss function is recalculated using the new training data set, and the parameters of the model are updated according to the stochastic gradient descent algorithm, providing more reliable support for the company's inventory management and production decisions.

[0057] The automatic replenishment module 22 sets the inventory safety threshold and replenishment point in the following manner:

[0058] Suppose historical sales data records The sales volume of goods per day is , then the average daily demand The calculation formula is , demand standard deviation It is used to measure the fluctuation of demand. The calculation formula is: The average daily demand can directly reflect the regular sales level of the commodity, and the standard deviation of demand supplements the information of demand fluctuation, so that inventory management can be more in line with the actual sales situation.

[0059] Replenishment lead time refers to the time required from the issuance of a replenishment order to the actual arrival of the goods at the warehouse. The lead time data for the deliveries are , average replenishment lead time The calculation formula is The average replenishment lead time provides a reasonable time expectation for inventory management, helping companies better plan inventory levels and safety stocks. It is the additional inventory reserved to cope with demand fluctuations and the uncertainty of replenishment lead time. The setting of safety stock provides the enterprise with a certain risk buffer and can effectively cope with the uncertainty of demand and replenishment lead time.

[0060] The replenishment point refers to the point when the inventory level drops to which the company needs to issue a replenishment order to replenish the inventory. By combining the average daily demand, average replenishment lead time and safety stock to calculate the replenishment point, it can ensure that the goods are replenished in time when they are consumed to a certain extent, ensuring the continuity and stability of the inventory. The calculation formula is In actual inventory management, the inventory level of the node is monitored in real time. When the inventory level is lower than When the replenishment process is triggered automatically, a replenishment order is issued to the supplier. The calculation of the replenishment point takes into account factors such as demand, replenishment lead time and safety stock, making the replenishment decision more scientific and reasonable and realizing dynamic management of inventory.

[0061] When the smart contract module 23 calculates the risk factor and adjusts the replenishment quantity, the following method is used:

[0062] Assume that the number of times the supplier interrupted supply in history is , the total number of deliveries is , then the supply chain disruption risk factor The calculation formula is Quantifying the risk of supply chain disruptions enables companies to intuitively understand the stability of supplier supply, helping companies reduce the risk of out-of-stocks caused by supplier disruptions.

[0063] Market demand is often unstable, and fluctuations in demand will have a significant impact on inventory management. The average daily demand is known to be The standard deviation of demand is , the coefficient of variation of demand The calculation formula is , the coefficient of variation of demand As a demand volatility risk factor ,but ,The demand fluctuation risk factor can accurately reflect the uncertainty of demand and reduce the inventory management problems caused by demand fluctuations.

[0064] Logistics and transportation are an important link in the supply chain. Transportation delays may cause goods to fail to arrive at the warehouse on time, affecting the company's inventory level and sales plan. Suppose the number of delayed deliveries by logistics vehicles is The total number of transports is , then the transportation delay risk factor The calculation formula is , quantifying the risk of transportation delays, enabling companies to clearly understand the reliability of logistics transportation and reducing the risk of out-of-stock situations caused by transportation delays.

[0065] Supply chain disruption, demand fluctuations, and transportation delays are not isolated risks. They will affect each other and jointly affect the inventory management of enterprises. Suppose the weight of the supply chain disruption risk factor is , the weight of the demand fluctuation risk factor is , the weight of the transportation delay risk factor is , then the comprehensive risk factor The calculation formula is , assuming the initial replenishment quantity is , adjusted replenishment quantity The calculation formula is The comprehensive risk factor takes into account the combined impact of multiple risk factors, enabling enterprises to more comprehensively evaluate the risks in inventory management and make the replenishment quantity more reasonable, which can not only meet market demand but also cope with various risk factors, thereby reducing the inventory costs and operational risks of enterprises.

[0066] When the intelligent replenishment algorithm unit 2 uses the multi-objective optimization algorithm to select the best supplier, the following steps are adopted:

[0067] When selecting the best supplier, companies usually need to consider multiple factors, such as the supplier's supply capacity, price, delivery time, inventory holding costs and out-of-stock costs. The supply capacity is , the price is , delivery date is , the inventory holding cost is , the out-of-stock cost is , then the objective function for:

[0068] ;

[0069] in is the weight coefficient, and , To suppliers The order quantity, is the predicted demand, order quantity Need to meet ,and ,in is the number of suppliers. The objective function takes into account multiple key factors and can more comprehensively evaluate the pros and cons of supplier selection options.

[0070] This multi-objective optimization problem involves multiple variables and complex objective functions. Traditional optimization methods may find it difficult to find the global optimal solution. Genetic algorithm is an optimization algorithm that simulates the biological evolution process and has strong global search capabilities. It randomly generates a Chromosome population , each chromosome Represents a supplier selection scheme and the corresponding order quantity allocation. For each chromosome , according to the objective function Calculate its fitness value , using the roulette wheel selection method, calculate each chromosome The probability of selection , and then generate a random number in the interval [0, 1] , and accumulate the selection probability of each chromosome in turn ,when When selecting chromosomes as parent chromosomes, and repeat this process to select a sufficient number of parent chromosomes for subsequent crossover and mutation operations.

[0071] Randomly select two chromosomes from the selected parent chromosome and , randomly select an intersection point ,Will The previous Gene and After Gene combination, generate a daughter chromosome ,Will The previous Gene and After Gene combination to create another daughter chromosome At the same time, we must ensure that the offspring chromosomes meet the constraints and , if it is not satisfied, it will be adjusted. For the generated offspring chromosomes, a gene in the chromosome is mutated with a certain mutation probability. Under the premise of meeting the constraints, it is replaced with a new order quantity. The offspring chromosomes generated by the crossover and mutation operations replace part of the chromosomes in the population to form a new population. Repeat the above selection, crossover and mutation operations until the termination conditions are met. The genetic algorithm has a strong global search capability and can find a better solution in a complex search space, avoiding falling into the local optimal solution.

[0072] After iterative optimization of the genetic algorithm, the supplier selection plan and order quantity allocation corresponding to the chromosome with the largest fitness value in the population is the optimal plan currently found. Selecting this plan and automatically sending replenishment orders to the selected suppliers can realize the automation of supplier selection and replenishment processes and improve the company's operational efficiency. After meeting the termination conditions, the chromosome with the largest fitness value is selected from the final population, and the corresponding supplier selection information and order quantity analysis from the largest chromosome are extracted. For suppliers with order quantities greater than 0, replenishment orders are automatically generated, including product information and order quantity content, and replenishment orders are sent to selected suppliers through the company's supply chain management system, realizing the automation of supplier selection and replenishment processes, reducing manual intervention, and improving work efficiency and accuracy.

[0073] Inventory allocation strategy unit 3 uses reinforcement learning algorithms to develop and continuously optimize dynamic inventory allocation rules based on real-time sales data, demand forecasts for each region, and inventory costs. It adjusts the inventory allocation ratio in real time according to actual sales data and market change data, and uploads it to the alliance chain blockchain platform.

[0074] When the inventory allocation strategy unit 3 uses the reinforcement learning algorithm to formulate and optimize the inventory allocation rules, the following steps are adopted:

[0075] When using reinforcement learning algorithms to formulate inventory allocation rules, it is necessary to clarify the state of the environment in which the algorithm is located so that the agent can make appropriate decisions based on different states. Regions, state vector ,in arrive Real-time sales data for each region, arrive Demand forecast for each region, arrive is the inventory level in each region, arrive It not only provides inventory costs for each region, but also describes the inventory allocation environment in detail, enabling the intelligent agent to obtain rich information and make more accurate and reasonable decisions.

[0076] The action space defines the actions that the agent can take. In the inventory allocation problem, the inventory allocation ratio of different regions is the key decision variable, so the action vector is defined as the inventory allocation ratio of different regions. ,in Indicates that the The inventory ratio of each region, The product price in this region is , sales volume is , the inventory cost is , the inventory level is , then reward ,This reward function reflects the sales profit of each region minus the inventory cost, reflecting the overall economic benefits.

[0077] The deep Q network algorithm combines the advantages of deep learning and reinforcement learning, and can handle high-dimensional state space and complex decision-making problems. ,in is the state vector, is the action vector, are the parameters of the network, and the input of the Q network is the state-action pair , the output is the Q value corresponding to the state-action pair, indicating that in state Take action Expected cumulative reward.

[0078] Create an experience replay pool , used to store samples obtained by the interaction between the agent and the environment, the sample is in the form of ,in is the current state, is the action taken, It is the reward obtained. is the next state, the agent in the environment according to the current state Select an action , after executing this action, you will get a reward And enter the next state , the sample Store to experience replay pool , from the experience replay pool A batch of samples are randomly sampled, the loss function is calculated, and the parameters of the Q network are updated using the gradient descent algorithm. The intelligent agent can gradually optimize the inventory allocation strategy, increase sales profits and reduce inventory costs during the continuous learning process.

[0079] Market conditions and sales data are changing dynamically. In order to make inventory allocation always adapt to the actual situation, it is necessary to adjust the inventory allocation ratio in real time according to the actual sales data and market change data. In the continuous interaction with the environment, the intelligent agent continuously updates the parameters of the Q network according to the latest state information, thereby adjusting the action vector in real time. , that is, the inventory allocation ratio in different regions. The adjusted inventory allocation ratio is uploaded to the alliance chain blockchain platform through the interface. Real-time adjustment can enable the inventory allocation strategy to adapt to market changes in a timely manner, improve the company's resilience, and improve the efficiency and competitiveness of the entire supply chain.

[0080] In the present invention, the inventory information sharing unit 1 deploys IoT sensors at key nodes of the supply chain and combines edge computing with blockchain technology to achieve real-time collection, processing and secure sharing of inventory data. The intelligent replenishment algorithm unit 2 obtains data from the blockchain platform and uses the gated loop module 21 to build a demand forecasting model. The automatic replenishment module 22 sets the replenishment threshold and adjusts the replenishment quantity based on risk factors. The multi-objective optimization algorithm selects the optimal supplier. The inventory allocation strategy unit 3 uses a reinforcement learning algorithm based on real-time data to formulate and optimize inventory allocation rules, adjusts the inventory allocation ratio and uploads it to the blockchain, thereby improving inventory management efficiency, reducing costs, and achieving dynamic inventory optimization.

[0081] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. The above embodiments and descriptions are only preferred examples of the present invention and are not intended to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. The e-commerce inventory dynamic optimization model system based on real-time big data is characterized by: It includes an inventory information sharing unit (1), an intelligent replenishment algorithm unit (2), and an inventory allocation strategy unit (3); The inventory information sharing unit (1) deploys IoT sensors in supplier warehouses, logistics transportation vehicles and e-commerce warehouses in the supply chain, processes the location, quantity and status data of inventory goods collected by the IoT sensors through edge computing devices, and uploads the preliminarily processed data to the alliance chain blockchain platform; The intelligent replenishment algorithm unit (2) obtains historical sales data, real-time market trends and supply chain inventory information from the alliance chain blockchain platform, uses a gated loop module (21) including a gated loop unit to build a demand forecasting model, uses pre-processed historical data to train the model, and dynamically adjusts the forecast results according to real-time data changes. Then, the automatic replenishment module (22) sets an inventory safety threshold and replenishment point for each commodity. When the node inventory level is lower than the replenishment point, the intelligent contract module (23) automatically triggers the replenishment process, introduces a risk factor that considers supply chain uncertainty, calculates the risk factor based on historical data and real-time monitoring conditions, and dynamically adjusts the replenishment quantity. Finally, a multi-objective optimization algorithm is used to select the optimal supplier in combination with real-time inventory information and the supplier's supply capacity, price and delivery time; The inventory allocation strategy unit (3) uses a reinforcement learning algorithm to formulate and continuously optimize dynamic inventory allocation rules based on real-time sales data, demand forecasts for each region, and inventory costs, adjusts the inventory allocation ratio in real time according to actual sales data and market change data, and uploads it to the alliance chain blockchain platform, using the following steps: defining a state space, the state space comprising a state vector, the state vector comprising real-time sales data, demand forecasts, inventory levels, and inventory costs for each region; defining an action space, the action space including an action vector, the action vector representing the inventory allocation ratio of different regions, defining a reward function, the reward function including a reward, the reward being related to the sales profit and inventory cost of each region; A deep Q network algorithm is used to construct a Q network, wherein the Q network includes an experience replay pool and an intelligent agent, the experience replay pool is initialized, samples obtained by the interaction between the intelligent agent and the inventory-related data environment are stored in the experience replay pool, a batch of samples are randomly sampled from the experience replay pool, a loss function is calculated, and the parameters of the Q network are updated using a gradient descent algorithm; Finally, the inventory allocation ratio is adjusted in real time according to actual sales data and market change data, and uploaded to the alliance chain blockchain platform.

2. The e-commerce inventory dynamic optimization model system based on real-time big data according to claim 1 is characterized by: In the inventory information sharing unit (1), the edge computing device uses a data fusion algorithm when processing the location, quantity and status data of inventory items collected by the IoT sensor. The specific steps are as follows: The data collected by different types of IoT sensors are synchronized in time, and the data collected by different sensors at similar time points are processed as a group. The similar time points are the time when the data points were collected. Minutes, data with timestamp differences within the set threshold are considered synchronized data; The Kalman filter algorithm is used to process the position data. For the quantity data, the sliding window statistical method is used to calculate the quantity change rate within a certain time window. If the change rate exceeds the set threshold, the data is marked and further verified. The status data is classified and encoded, and different statuses are converted into numerical codes, including normal, damaged, and out of stock. Finally, the processed location, quantity, and status data are integrated to form a comprehensive inventory data record and uploaded to the alliance chain blockchain platform.

3. The e-commerce inventory dynamic optimization model system based on real-time big data according to claim 2 is characterized by: When the gated loop module (21) constructs the demand prediction model, the following training and adjustment methods are used: The historical sales data obtained from the alliance chain blockchain platform is cleaned, and the missing values ​​are filled by linear interpolation. Time features, seasonal features, and promotional activity features are extracted from the historical sales data, and these features are used together with the sales data as the input of the model; The gated recurrent module (21) including the gated recurrent unit is trained using preprocessed historical data, and the model parameters are updated using a stochastic gradient descent algorithm. The model is then adjusted using online learning based on changes in real-time data. When new real-time sales data arrives, it is added to the training data, the gradient is recalculated, and the model parameters are updated.

4. The e-commerce inventory dynamic optimization model system based on real-time big data according to claim 3 is characterized by: The automatic replenishment module (22) uses the following method to set the inventory safety threshold and replenishment point: Calculate the average daily demand of each commodity based on historical sales data, and calculate the standard deviation of demand based on the average daily demand, where the standard deviation of demand is used to measure the degree of fluctuation of demand; Based on the supplier's historical delivery data, the average replenishment lead time is calculated, and the safety stock is calculated based on the average replenishment lead time. Finally, the average daily demand, average replenishment lead time and safety stock are substituted into the formula to calculate the replenishment point. When the node inventory level is lower than the replenishment point, the replenishment process is triggered.

5. The e-commerce inventory dynamic optimization model system based on real-time big data according to claim 4 is characterized by: The smart contract module (23) calculates the risk factor and adjusts the replenishment quantity using the following method: Count the number of supply interruptions and total supply times of suppliers in history, and calculate the supply chain interruption risk factor based on the number of supply interruptions and total supply times; Then, the ratio of the standard deviation of the demand to the average daily demand is taken as the coefficient of variation of the demand, and the coefficient of variation of the demand is taken as the demand fluctuation risk factor; Statistics are collected on the number of delayed deliveries and the total number of transportations by logistics transport vehicles, and the transport delay risk factor is calculated based on the number of delayed deliveries and the total number of transportations. The above three risk factors are comprehensively considered to calculate the comprehensive risk factor, and the initial replenishment quantity is adjusted based on the comprehensive risk factor to obtain the adjusted replenishment quantity.

6. The e-commerce inventory dynamic optimization model system based on real-time big data according to claim 5 is characterized by: When the intelligent replenishment algorithm unit (2) uses the multi-objective optimization algorithm to select the optimal supplier, the following steps are adopted: set up For price, For supply capacity, is the delivery date, then the supplier The supply capacity is , the price is , delivery date is , the inventory holding cost is , the out-of-stock cost is , then the objective function for: ； in is the weight coefficient, and , To suppliers The order quantity, is the predicted demand, order quantity Need to meet ,and ,in is the number of suppliers; A genetic algorithm is introduced to solve the multi-objective optimization problem. A population containing multiple chromosomes is initialized. Each chromosome represents a supplier selection scheme and the corresponding order quantity distribution. The fitness value of each chromosome is calculated. The fitness value is the objective function value. A new population is generated through selection, crossover and mutation operations. The iteration is repeated until the termination condition is met. The supplier and order quantity allocation corresponding to the chromosome with the largest fitness value are selected as the optimal solution, and replenishment orders are automatically sent to the selected supplier.

7. The e-commerce inventory dynamic optimization model system based on real-time big data according to claim 6 is characterized by: The specific method steps of introducing the genetic algorithm to solve the multi-objective optimization problem in the intelligent replenishment algorithm unit (2) are as follows: Create a population containing multiple chromosomes, each chromosome represents a supplier selection scheme and the corresponding order quantity allocation; For each chromosome in the population, its corresponding order quantity distribution is substituted into the objective function. The calculated objective function value is the fitness value of the chromosome. The supplier selection scheme and order quantity distribution are proportional to the fitness value. According to the fitness value of the chromosome, a certain number of chromosomes are selected from the current population as parent chromosomes using the roulette wheel selection method for subsequent crossover and mutation operations. Two chromosomes are randomly selected from the selected parent chromosomes and a single-point crossover method is used to generate new daughter chromosomes. In the crossover process, the generated daughter chromosomes must meet the constraints. The offspring chromosomes generated by crossover and mutation operations are merged with the parent chromosomes to form a new population. The above steps of selection, crossover, mutation and generation of a new population are repeated until the preset fitness threshold is reached. The supplier and order quantity allocation corresponding to the chromosome with the largest fitness value are selected as the optimal solution, and replenishment orders are automatically sent to the selected suppliers.

Images