Warehouse goods supply chain management method and system based on AI big data

Abstract

The application discloses a warehouse cargo supply chain management method and system based on AI big data, relates to the warehouse management technical field, and comprises collecting warehouse multi-source data and preprocessing, predicting cargo warehouse-out frequency based on an LSTM neural network, constructing a multi-objective scheduling model combining pickup path time consumption and warehouse gravity control based on a prediction result, jointly modeling cargo tasks and AGV path planning as an open path multi-start asymmetric TSP problem, using a FWA-GA hybrid optimization algorithm to solve the model, and obtaining an optimal storage location allocation matrix and an optimal AGV scheduling path table. Thus, the application realizes integrated joint optimization of task allocation, path planning and storage location layout, further links a supply chain platform to complete out-of-stock early warning and intelligent supplier selection, and effectively breaks through multiple limitations of traditional methods in prediction accuracy, scheduling coupling, structure control and supply collaboration.

Description

Technical Field

[0001] This invention relates to the field of warehouse management technology, and in particular to a warehouse goods supply chain management method and system based on AI (Artificial Intelligence) big data, a computer device, and a computer-readable storage medium. Background Technology

[0002] With the development of industrial automation and smart logistics, modern warehousing systems are constantly evolving towards intelligence, flexibility, and collaboration. Traditional static manual management models are gradually being replaced by automated equipment, information systems, and artificial intelligence models. Against this backdrop, warehouse management systems, enterprise resource planning systems, IoT sensor networks, and various mobile handling equipment have achieved initial integration, providing fundamental support for realizing cargo inbound and outbound scheduling, route optimization, inventory early warning, and supply chain linkage.

[0003] However, most current systems still treat prediction, scheduling, execution, and supply response modules separately, and have not yet formed a prediction-driven full-chain optimization strategy. At the same time, the global modeling capability and algorithm solution efficiency for complex path scheduling in multi-AGV (Automated Guided Vehicle) scenarios are still insufficient, making it difficult to achieve both real-time performance and system stability. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a warehouse goods supply chain management method and system based on AI big data, which solves the problem that most current systems still process prediction, scheduling, execution, and supply response modules separately, and have not yet formed a prediction-driven full-chain optimization strategy. At the same time, it solves the problem that the global modeling capability and algorithm solution efficiency for complex path scheduling in multi-AGV scenarios are still insufficient, making it difficult to achieve both real-time performance and system stability.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] Firstly, this invention provides a warehouse cargo supply chain management method based on AI big data, comprising: collecting and preprocessing multi-source warehouse data; predicting cargo outbound frequency based on real-time cargo data using an LSTM (Long Short-Term Memory) neural network, wherein the multi-source warehouse data includes basic cargo information and cargo location data; constructing a multi-objective scheduling model combining picking path time and warehouse center of gravity control based on the prediction results, jointly modeling cargo tasks and AGV path planning as an open-path multi-starting-point asymmetric TSP (Traveling Salesman Problem), and solving the model using a hybrid optimization algorithm of FWA-GA (Fireworks Algorithm-Genetic Algorithm) to obtain the optimal cargo location allocation matrix and the optimal AGV scheduling path table; implementing the optimal scheduling scheme; after the scheduling optimization is completed, comparing the predicted cargo outbound frequency with the current inventory, automatically generating a stockout warning and a procurement plan and pushing them to the supply chain control system, and selecting the optimal supplier for cargo response.

[0008] As a preferred embodiment of the AI-based big data-driven warehouse cargo supply chain management method of the present invention, the multi-objective scheduling model constructed based on the prediction results, which combines the picking path time and the warehouse center of gravity control, includes: using the Flying-V layout as the warehouse basic structure, and combining the outbound frequency, cargo quality and structural position, constructing a dual objective function for scheduling optimization, including minimizing path time f1 and minimizing warehouse center of gravity f2.

[0009] As a preferred embodiment of the AI-based big data-driven warehouse goods supply chain management method of the present invention, the method of jointly modeling goods tasks and AGV path planning as an open-path, multi-starting-point asymmetric TSP problem, and using the FWA-GA hybrid optimization algorithm to solve the model, yields the optimal storage location allocation matrix and the optimal AGV scheduling path table, including:

[0010] Extract all cargo tasks that need to be scheduled, forming a task set A. Record the number of currently online AGVs, initialize the current position of each AGV, and calculate the corresponding path cost d for any two points in the task set as the input of the task movement cost.

[0011] A scheduling objective function F is constructed by combining overall efficiency and the response time of a single AGV, and scheduling constraints are added.

[0012] Set constraints that each task must be assigned to only one AGV, and the storage locations for all tasks must meet the physical capacity of the warehouse;

[0013] Construct an initial feasible solution group, setting the total number of individuals and the maximum number of iterations, with each individual representing a scheduling scheme;

[0014] Use the Logistic chaotic map to generate the initial perturbation sequence;

[0015] The first O variables are encoded as integers and mapped to the AGV number to which the task belongs and the execution priority. The last O variables are mapped to the layer number of the goods. The task scheduling result and the cargo location allocation are decoded into a task path table and a cargo location allocation matrix.

[0016] The scheduling objective function is used as the fitness function to calculate the fitness, and individuals are sorted in descending order of fitness. A spark explosion operation is performed on the top U individuals to generate perturbed candidate scheduling solutions.

[0017] Define the maximum disturbance radius of the fireworks explosion and calculate the explosion radius of generation t;

[0018] For each principal individual, calculate the number of sparks S based on its corresponding fitness value and the worst fitness value in the population. Ω ;

[0019] Iterate through each scheduling individual, initialize the spark set, loop through each individual Q times, generate spark individuals in sequence, automatically select the current perturbation strategy according to stage t, compare stage t with the preset threshold, if t is less than the threshold, enter the early global search stage and execute uniform explosion perturbation.

[0020] Otherwise, it will enter the later local convergence stage, using a Cauchy-Gaussian hybrid perturbation.

[0021] A Cauchy-Gaussian hybrid perturbation strategy is used to globally perturb the structure of the initial main population, providing new scheduling solution candidates for elite selection;

[0022] Each time a spark individual is generated, a legality check and boundary violation repair are performed immediately.

[0023] All valid spark individuals are stored in the spark set of this entity to form a spark individual set;

[0024] For each spark individual, the fitness function is recalculated, and the spark individual is bound to the new fitness value to form an expanded population;

[0025] Sort all individuals in the extended population in ascending order of fitness, select the top n individuals to form the main population of generation t+1, retain the current global best individual, randomly select M pairs of individuals, and perform partial mapping crossover on their task path encoding.

[0026] Update the path structure of the two individuals after the intersection to form a new set of intermediate individuals X.pmx ;

[0027] For X pmx For each individual, a perturbation probability is set, and a portion of tasks are randomly selected to perform mutation operations.

[0028] During the genetic evolution stage, a Cauchy-Gaussian (a hybrid model combining Cauchy and Gaussian distributions) perturbation strategy is applied to crossover individuals to further optimize the task hierarchy ranking and fine layout of cargo locations.

[0029] If the perturbation causes variables to go out of bounds, boundary repair should be performed immediately to obtain the perturbation-optimized population X. mut ;

[0030] For X mut For each individual, extract the task path of each AGV. Based on the Flying-V storage structure, perform path planning checks to obtain an optimized scheduling solution group X1. Sort X1 in ascending order of fitness value, extract the individual w with the smallest fitness, and compare w with the previously recorded best individual w'. If the fitness value of w is less than that of the best individual w', then update w as the global best individual.

[0031] As a preferred embodiment of the AI-based big data-driven warehouse cargo supply chain management method of the present invention, the implementation of the optimal scheduling scheme refers to extracting the optimal task location allocation matrix, AGV path scheduling table, total task path time and maximum AGV task time from the optimal individual code, converting the location matrix into a visual structure in a three-dimensional coordinate system for inbound / outbound area instruction matching, and converting the AGV path scheduling table into a scheduling queue to realize task push.

[0032] As a preferred embodiment of the AI-based big data-driven warehouse goods supply chain management method of the present invention, the following steps are taken: after the scheduling optimization is completed, the predicted goods outbound frequency is compared with the current inventory, and a stockout warning and procurement plan are automatically generated and pushed to the supply chain control system. The optimal supplier is selected for goods response. The prediction results and real-time data are compared to analyze the risk of goods stockout. For each stockout item, the pre-registered list of available suppliers in the supply chain platform is automatically called. The historical average delivery cycle, fulfillment success rate and unit supply price of the supplier are weighted and summed. The suppliers are sorted in descending order based on the summation score. The supplier with the highest score is selected, and the corresponding replenishment task order is immediately generated.

[0033] As a preferred embodiment of the AI-based big data-driven warehouse goods supply chain management method of the present invention, the step of collecting and preprocessing multi-source warehouse data includes:

[0034] Before any goods are put into storage, a unique RFID (Radio Frequency Identification) tag is attached to each item, and the unique item number, storage / outbound timestamp, and current location coordinate triplet (x, y, z) are obtained in real time. i ,y i ,z i (This corresponds to the shelf row number, shelf number, and column number;)

[0035] By collecting the current weight of goods and real-time inventory, and combining the inbound / outbound time, an inbound / outbound frequency sequence and the current storage coordinates of the goods can be generated.

[0036] By connecting with the data interface of ERP (Enterprise Resource Planning) and WMS (Warehouse Management System), each item can obtain corresponding supplier attributes, including delivery cycle, fulfillment rate and unit procurement cost;

[0037] The collected data is processed for missing values ​​and normalized.

[0038] As a preferred embodiment of the warehouse cargo supply chain management method based on AI big data described in this invention, the prediction of cargo outbound frequency based on LSTM neural network refers to inputting real-time cargo outbound data into the LSTM model to obtain the future cargo outbound frequency.

[0039] Secondly, this invention provides a warehouse goods supply chain management system based on AI big data, including:

[0040] The data acquisition module is used to collect real-time cargo data and perform preprocessing.

[0041] The forecasting module is used to predict the frequency of goods leaving the warehouse;

[0042] The scheduling optimization module is used for multi-objective modeling.

[0043] The scheduling solution module is used to solve for the optimal scheduling result using FWA-GA.

[0044] The implementation module is used for real-time task scheduling based on the optimal scheduling results.

[0045] The response module is used for out-of-stock alerts and supplier responses.

[0046] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the warehouse goods supply chain management method based on AI big data as described in the first aspect of the present invention.

[0047] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the warehouse goods supply chain management method based on AI big data as described in the first aspect of the present invention.

[0048] Secondly, this invention provides a warehouse cargo supply chain management system based on AI big data, comprising: a data acquisition module for collecting and preprocessing real-time cargo data; a prediction module for predicting cargo outbound frequency based on real-time cargo data using an LSTM neural network; a scheduling optimization module for constructing a multi-objective scheduling model combining pick-up path time and warehouse center of gravity control based on the cargo outbound frequency, jointly modeling cargo tasks and AGV path planning as an open-path multi-starting-point asymmetric TSP problem; a scheduling solution module for solving the multi-objective scheduling model using a FWA-GA hybrid optimization algorithm to obtain an optimal cargo location allocation matrix and an optimal AGV scheduling path table; an implementation module for implementing the optimal scheduling scheme based on the optimal task cargo location allocation matrix and the optimal AGV path scheduling table; and a response module for comparing the predicted cargo outbound frequency with the current inventory, automatically generating stockout warnings and purchasing plans, pushing them to the supply chain control system, and selecting the optimal supplier for cargo response.

[0049] Thirdly, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the warehouse goods supply chain management method based on AI big data as described in the first aspect of the present invention.

[0050] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the warehouse goods supply chain management method based on AI big data as described in the first aspect of the present invention.

[0051] The beneficial effects of this invention are as follows: Based on AI-driven outbound frequency prediction, this invention constructs a dual-objective scheduling model that takes into account both path time and warehouse center of gravity control. It introduces the FWA-GA hybrid optimization algorithm to solve the open path multi-starting point asymmetric TSP problem, and integrates the Cauchy-Gaussian perturbation mechanism to enhance local convergence capability. Finally, it realizes the integrated joint optimization of task allocation, path planning and storage location layout, and further links the supply chain platform to complete stockout early warning and intelligent supplier selection. This effectively breaks through the multiple limitations of traditional methods in terms of prediction accuracy, scheduling coupling, structural control and supply coordination. Attached Figure Description

[0052] Figure 1 This is a flowchart of a warehouse goods supply chain management method based on AI big data according to an embodiment of the present invention.

[0053] Figure 2 This is a schematic diagram of the Flying-V warehouse layout and AGV path according to an embodiment of the present invention.

[0054] Figure 3 This is a structural diagram of a warehouse goods supply chain management system based on AI big data according to an embodiment of the present invention. Detailed Implementation

[0055] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0056] Figure 1 This is a flowchart of a warehouse goods supply chain management method based on AI big data according to an embodiment of the present invention, such as... Figure 1 As shown, the method includes the following steps:

[0057] S1 collects and preprocesses multi-source warehouse data, and predicts the outbound frequency of goods based on real-time cargo data using an LSTM neural network. The multi-source warehouse data includes basic cargo information and cargo location data.

[0058] In one specific embodiment of the present invention, collecting and preprocessing multi-source warehouse data includes: binding a unique RFID tag to each item before all goods enter the warehouse. This can be done through batch initialization of UHF (Ultra High Frequency, referring to radio waves with frequencies between 300MHz and 3000MHz) electronic tags, and deploying RFID reader antennas at warehouse entry and exit points, main passage intersections, and key picking points to obtain in real time the unique item number, entry / exit timestamp, and current location coordinate triplet (x, y, x). i ,y i ,z i This corresponds to the shelf row number, shelf level, and column number, as well as the operation type (inbound / outbound / transfer) and other operation events; each shelf unit is equipped with a weight sensor and a location coding device to collect the following parameters: current goods weight m i Real-time inventory s i (Indicates the number of items currently at this location), and by combining the inbound / outbound times, an inbound / outbound frequency sequence r can be generated. i (t), (where r) i (t) represents the number of times goods i leave the warehouse at time t), and the current storage coordinates of the goods; through data interface with ERP and WMS systems, each goods can obtain corresponding supplier attributes, including the supply cycle τ. q , fulfillment rate ρ q (Reflecting supplier order fulfillment success rate) and unit procurement cost ε q Where q refers to the q-th supplier; missing value processing is performed on the collected data, logarithmic normalization is performed on the outbound frequency, Min-Max (deviation standardization) normalization is performed on the quality of goods, inventory, and delivery cycle, and Z-score (standard score) standardization is performed on the procurement cost.

[0059] Specifically, by deploying RFID readers at key warehouse nodes (such as entry / exit points, junction channels, and picking stations), and combining this with a unique electronic tag attached to each item, the system can collect key data in real time, including item number, location, transfer time, and operation type, constructing a high-resolution trajectory chain for the flow of goods. Compared to traditional barcode scanning methods, this solution offers significant advantages such as non-contact, automation, and high reliability, enabling continuous data collection without human intervention. Simultaneously, the deployment of weight sensors combined with location coding devices allows for real-time acquisition of information such as changes in goods weight, real-time inventory levels at each location, and space utilization, thereby achieving dynamic warehouse status monitoring and refined inventory management. By statistically analyzing the inbound and outbound behavior of goods at various time points, a time-series outbound frequency sequence can be constructed, serving as an important input variable for training predictive models. Through the above preprocessing, each item's data is ultimately organized into a unified format multi-dimensional vector, providing a consistent, operable, and learnable input interface for subsequent LSTM predictive models and optimization scheduling algorithms.

[0060] Furthermore, predicting the frequency of goods outbound based on LSTM neural network refers to using an LSTM model to predict the frequency of goods outbound. Historical goods outbound data is used as the training set and input into the LSTM model for model training. A loss function and Adam (an optimization algorithm based on gradient descent) optimizer are defined to iteratively optimize the model parameters. When the loss of the LSTM model no longer decreases significantly during continuous iteration, the iteration stops, the model parameters are output, and the LSTM model is updated. Real-time goods outbound data is input into the LSTM model to obtain the future frequency of goods outbound.

[0061] Embedding the LSTM prediction output into the scheduling and replenishment decision chain forms a closed loop of "prediction-optimization-execution-response", which is significantly better than the low-precision prediction strategies such as static rules or moving averages in existing technologies. Through multi-stage fine modeling and optimization, the accuracy of outbound trend prediction and the adaptability of the scheduling model are significantly improved.

[0062] S2. Based on the frequency of goods leaving the warehouse, a multi-objective scheduling model is constructed that combines the picking path time and the warehouse center of gravity control. The goods task and AGV path planning are jointly modeled as an open path multi-starting point asymmetric TSP problem. The FWA-GA hybrid optimization algorithm is used to solve the multi-objective scheduling model to obtain the optimal cargo location allocation matrix and the optimal AGV scheduling path table.

[0063] Among them, the multi-objective scheduling model based on the prediction results, which combines the picking path time and the warehouse center of gravity control, includes: taking the Flying-V layout as the warehouse basic structure, and combining the outbound frequency, cargo quality and structural position, constructing a dual objective function for scheduling optimization, including minimizing path time f1 and minimizing warehouse center of gravity f2.

[0064] Specifically, such as Figure 2 As shown, the Flying-V warehouse structure is V-shaped. In the Flying-V warehouse structure, the AGV needs to travel in the main aisle and then enter each sub-aisle before performing lifting operations to complete the cargo handling. The movement time of various types of goods on the planar path and the vertical lifting time are weighted and superimposed to construct the following path time objective function f1:

[0065]

[0066] In the formula, R i It is the predicted frequency of goods leaving the warehouse, α i The scheduling priority weights, L, are derived by fitting historical order urgency with predicted trends. x It is the basic length of the main channel, y i These are the row numbers of the goods within the aisle; 'l' represents the row spacing on the shelf; and 'z' represents the row number. i The number of shelves is the layer number of the goods, multiplied by h to represent the height, v1 and v2 are the AGV's lateral movement and vertical lifting speeds, N is the total number of goods types or tasks in the warehouse, and h is the shelf height.

[0067] Unlike traditional models, this invention uses AI-driven priority coefficient α i By embedding path distance weights, the system not only considers the location of goods but also the urgency of their actual scheduling, thus better aligning with "AI prediction-driven scheduling scenarios".

[0068] To avoid increased handling load or rack instability caused by frequently outbound goods being concentrated on higher floors, a center of gravity control mechanism was introduced. Considering that some high-frequency goods may have uneven weight distribution, a frequency ranking influence factor β was introduced into the center of gravity model to enhance the comprehensiveness of the scheduling model. j Reconstruct the objective function f2 as follows:

[0069]

[0070] In the formula, m i It is the weight of the goods, β i It is the quantile value of the goods in the predicted outbound frequency ranking.

[0071] Considering the limitations of the actual warehouse structure and AGV operational capabilities, hard boundaries need to be set for the spatial locations where goods can be allocated to ensure that all scheduling results can be physically implemented. The constraint expressions are as follows:

[0072] x i ≤x max ,y i ≤y max ,z i ≤z max

[0073] In the formula, x i y i , z i Let x represent the column, row, and layer positions of the i-th item, respectively. max y max , z max These represent the maximum capacity of the warehouse's three-dimensional structure in each direction;

[0074] In summary, the structure of the multi-objective scheduling model has been completed, and the following optimization objective structure has been formed:

[0075] Minimize{f1,f2}Subject to:x i ≤x max ,y i ≤y max ,z i ≤z max ;

[0076] Minimize is the parameter that minimizes the objective function.

[0077] The model is guided by AI prediction results, constrained by warehouse structure characteristics, and evaluated by scheduling time and structural stability, forming an evaluation objective function for the scheduling algorithm.

[0078] By combining predicted outbound frequency with path cost, this approach considers not only physical distance but also introduces a dynamic scheduling dimension of "time urgency." Compared to traditional scheduling models that only optimize distance or operation time, this method achieves a fusion of "frequency-driven + location-oriented" modeling, more closely reflecting the management logic of prioritizing frequent tasks in real-world business operations, significantly improving the intelligence and responsiveness of scheduling. Traditional scheduling models rarely consider the load distribution along the vertical direction of the racks, especially in high-bay warehouses where frequent high-level operations lead to increased energy consumption and decreased stability of AGV lifting. This invention establishes a mass-heat coupling model by incorporating cargo weight and frequency quantile values, constraining heavy or high-frequency goods to concentrate in the middle and lower levels, achieving optimal warehouse center of gravity, and effectively preventing structural safety issues such as rack tilting and AGV lifting conflicts. By explicitly setting the maximum column / row / layer position that goods can be allocated, hard constraints are applied to the spatial range of all variables during the optimization process to prevent scheduling tasks from failing due to out-of-bounds results of the optimization function. The coordinated optimization of path time minimization and center-of-gravity control objectives achieves a balance between execution efficiency and system safety in the scheduling results. Especially in complex scheduling scenarios with multiple AGVs and multiple tasks, this invention provides a scheduling control architecture with unified global objective modeling and dynamic optimization of local conflicts, exhibiting higher operational stability and scalability.

[0079] In one embodiment of the present invention, the step of jointly modeling the cargo task and AGV path planning as an open-path multi-starting-point asymmetric TSP problem, and using the FWA-GA hybrid optimization algorithm to solve the model to obtain the optimal cargo location allocation matrix and the optimal AGV scheduling path table includes the following steps S21-S220:

[0080] S21, extract all cargo tasks that need to be scheduled, form a task set A, record the number of currently online AGVs, initialize the current position of each AGV, and for any two points in the task set, calculate the corresponding path cost d as the input of the task movement cost.

[0081]

[0082] Calculate the path cost from the origin to the task for each AGV, and sum the costs of all AGVs completing their respective task paths to obtain the total path cost. Within each AGV path, the tasks are sequentially encoded into a set P = {p1, p2, ..., p...}. k}, where p k It is the kth task;

[0083] S22, combine overall efficiency and single AGV response time to construct scheduling objective function F and add scheduling constraints.

[0084] F = ω1·SOC + ω2·MS + ω3·GC

[0085] In the formula, ω1, ω2 and ω3 are weighting coefficients, SOC is the total time of all AGV task paths, MS is the maximum task execution time, and GC is the total impact of all tasks on the center of gravity.

[0086]

[0087]

[0088] In the formula, n j f1(p) is the total number of tasks performed by the j-th AGV. k f2(g) is the path time corresponding to the kth task, e is the total number of AGVs, g∈E is the gth cargo task in the current scheduling task set E, f2(g) is the single impact of the gth cargo on the warehouse center of gravity, j∈[1,e] is the jth AGV, and there are a total of e AGVs in the system.

[0089] S23 sets constraints that each task must be assigned to only one AGV, and the storage locations for all tasks must meet the physical capacity of the warehouse.

[0090] S24, construct the initial feasible solution group, set the total number of individuals and the maximum number of iterations, and each individual is represented as a scheduling scheme;

[0091] S25, use Logistic chaotic mapping to generate initial perturbation sequence.

[0092] z o+1 =μ·z o ·(1-z o )

[0093] In the formula, z o+1 This is the next chaotic value, μ is the control parameter of the Logistic mapping, and z o It is the value of the chaotic variable generated in the o-th iteration.

[0094] S26. Encode the first O variables into integers and map them to the AGV number and execution priority of the task. For the last O variables, map them to the layer number of the goods. Decode the task scheduling result and the cargo location allocation into a task path table and a cargo location allocation matrix.

[0095] S27. The scheduling objective function is used as the fitness function to calculate the fitness, and individuals are sorted in descending order according to their fitness. A spark explosion operation is performed on the top U individuals to generate candidate scheduling solutions after perturbation.

[0096] To expand the search space and improve the global optimal solution finding capability of the scheduling model, it is necessary to perform a spark explosion operation on individuals with better fitness to generate candidate scheduling solutions after some disturbance.

[0097] S28, set the maximum disturbance radius of the fireworks explosion, and calculate the explosion radius of the t-th generation.

[0098] Define the maximum disturbance radius of the fireworks explosion and calculate the explosion radius of generation t:

[0099]

[0100] In the formula, r t r is the explosion radius of the current t-th iteration, used to control the perturbation amplitude. initial It is the initial maximum disturbance radius, manually set, r end It is the final convergence perturbation radius, which limits the minimum granularity of the perturbation. It is manually set and is often an empirical value. t is the current iteration round number, T is the maximum iteration round number, and ξ is the power factor, which controls the convergence speed and is a constant.

[0101] S29, For each principal individual, calculate the number of sparks S based on the corresponding fitness value and the worst fitness in the population. Ω .

[0102]

[0103] In the formula, S Ωf(X) is the number of sparks that the Ω-th scheduling individual will generate, M is the total spark budget, and f(X) is the number of sparks that will be generated. Ω f is the fitness value of the Ωth scheduled individual. max It is the worst fitness value in the current population, ε is a small constant to prevent the denominator from being zero, and Y is the total number of individuals scheduled. It is the first The fitness value of each individual being scheduled.

[0104] Round the number of sparks down to the nearest integer and set a minimum and maximum number to control the complexity of a single explosion. Pair the number of each spark with the explosion radius.

[0105] S210: Traverse each scheduling individual, initialize the spark set, loop Q times for each individual, generate spark individuals in sequence, automatically select the current perturbation strategy according to stage t, compare stage t with the preset threshold, if t is less than the threshold, enter the early global search stage and execute uniform explosion perturbation, otherwise enter the later local convergence stage and use Cauchy-Gaussian hybrid perturbation.

[0106]

[0107] In the formula, The new value of the y-th variable obtained under the current perturbation, x Ωy It is the y-th variable value of the Ω-th scheduling individual, rand(0,r) t ) is from 0 to r t Random numbers generated between [a certain number of points].

[0108]

[0109] In the formula, It is the new value of the Ω-th individual in dimension a after the mixed perturbation, used to generate a new solution individual, x k is the k-th variable to be disturbed, which may be a discrete value such as task number, storage layer number, AGV number, etc. Gaussian(1,1) is a standard normal distribution sample (mean 1, variance 1), used for small-amplitude, symmetrical disturbances. Cauchy(1,1) is a Cauchy distribution sample (position parameter 1, scale 1), which has a heavy-tailed characteristic and is used for large-amplitude jump disturbances. λ is a stage control factor that determines the degree of dominance of the disturbance type. Round() is a rounding function to ensure that the result is an integer, which meets the requirement that the warehouse scheduling variable is a discrete attribute of the number class.

[0110]

[0111] In the formula, t is the current iteration number, and T is the total maximum number of iterations.

[0112] S211 employs a Cauchy-Gaussian hybrid perturbation strategy to globally perturb the initial main population individuals, providing new scheduling solution candidates for elite selection.

[0113] S212: For each spark individual generated, a legality check and boundary correction are performed immediately.

[0114] S213 stores all legal spark individuals into the spark set of the subject, forming a spark individual set.

[0115] S214. For each spark individual, the fitness function is recalculated, and the spark individual is bound to the new fitness value to form an expanded population.

[0116] S215: Sort all individuals in the extended population in ascending order of fitness, select the top n individuals to form the main population of generation t+1, retain the current global best individual, randomly select M pairs of individuals, and perform partial mapping crossover on their task path encoding to exchange the partial task segments and their order allocated by the AGV, ensuring that each task is still uniquely assigned after crossover, all tasks are fully covered, and the crossover range length is randomly set, with a maximum not exceeding 30% of the number of tasks.

[0117] S216, update the path structure of the two individuals after the crossover, forming a new set of intermediate individuals X. pmx .

[0118] S217, for X pmx For each individual, a perturbation probability is set, and a portion of the tasks are randomly selected to perform mutation operations.

[0119]

[0120] In the formula, These are the new variable values ​​after the perturbation, used to generate new individual solutions.

[0121] S218 employs a Cauchy-Gaussian perturbation strategy on crossover individuals during the genetic evolution phase to further optimize task hierarchy ranking and fine-grained cargo location layout. This enhances local convergence stability and mutation depth control.

[0122] S219, if the perturbation causes variables to go out of bounds, immediately perform boundary repair to obtain the perturbation-optimized population X. mut This results in a better distribution of storage locations and a more efficient local task sequencing.

[0123] S220, for X mutFor each individual, extract the task path for each AGV. Based on the Flying-V storage structure, perform path planning checks: use the improved A* algorithm to simulate AGV movement and detect issues such as path intersection, reverse conflict, and simultaneous lifting deadlock. If a conflict is found, prioritize adjusting the task order (swapping local tasks). If there is still no solution, insert waiting time or backtracking path to obtain an optimized scheduling solution group X1. Sort X1 in ascending order of fitness value, extract the individual with the smallest fitness w, and compare w with the previously recorded best individual w'. If the fitness value of w is less than that of the best individual w', then update w as the globally best individual.

[0124] Specifically, the "Open Path Multi-Start Asymmetric TSP Problem" (OP-MATSP) is a scheduling modeling approach for multi-start, multi-task, and asymmetric path costs. Unlike the traditional TSP problem's assumptions of "closed-loop paths" and "symmetric path distances," OP-MATSP allows each AGV to start from different locations without needing to return to the starting point along the path, and the path cost can vary depending on the spatial structure or task order, making it particularly suitable for multi-channel warehouse structures in Flying-V layouts. This modeling approach can realistically reflect the complex coupling characteristics of task allocation and path sequencing in multi-AGV systems, significantly improving the expressiveness and practicality of the scheduling model. The FWA-GA hybrid optimization algorithm combines the advantages of the "fireworks algorithm" and the "genetic algorithm," generating search solutions with widespread perturbations through the spark explosion mechanism, while simultaneously improving the evolutionary efficiency of the population through genetic crossover and mutation. Among them, the "Logistic chaotic mapping" is used to generate perturbation variables with chaotic distribution characteristics during the population initialization stage, effectively preventing the initial population from getting trapped in local optima; the "spark explosion radius" adopts a dynamic convergence function to control the perturbation amplitude, combined with a fitness-guided spark quantity calculation mechanism, to ensure a dynamic balance between early global exploration and later local convergence. In addition, the "Cauchy-Gaussian hybrid perturbation strategy" is introduced, which utilizes the long tail and high jump characteristics of the Cauchy distribution combined with the local smoothing characteristics of Gaussian perturbation to achieve deep fine-tuning of the local space and jump exploration of the global space, greatly enhancing the diversity of the optimization algorithm and the solution space coverage capability.

[0125] The scheduling objective function not only considers the traditional total path time (SOC) and maximum single-vehicle task time (MS), but also introduces the warehouse center of gravity control index (GC). By modeling the stability of the center of gravity of goods in the warehouse structure through frequency and weight weighting, it avoids the problem of high-frequency heavy items concentrating on upper floors, which could lead to rack instability or increased handling costs. This three-objective scheduling function balances efficiency, safety, and scheduling equilibrium, achieving a reasonable balance between executability, system stability, and operational timeliness in the final optimization result. In terms of scheduling execution, the system incorporates both "task allocation" and "location mapping" substructures into individual codes, and uses evolutionary operations to control the crossover and perturbation behavior of the codes, achieving joint optimization of task path sequencing and location layer numbers. An improved A* path checking mechanism is introduced during the optimization process to perform conflict detection and feasibility verification on the generated AGV paths, automatically handling path intersections, reverse conflicts, and deadlock issues related to lifting resources, further improving the stability and controllability of the scheduling scheme in actual deployment.

[0126] S3. Implement the optimal scheduling scheme. After the scheduling optimization is completed, compare the predicted outbound frequency with the current inventory level, automatically generate a stockout warning and push the procurement plan to the supply chain control system, and select the best supplier to respond to the goods.

[0127] In one specific embodiment of the present invention, implementing the optimal scheduling scheme refers to extracting the optimal task location allocation matrix, AGV path scheduling table, total task path time and maximum AGV task time from the optimal individual code, converting the location matrix into a visual structure in a three-dimensional coordinate system for inbound / outbound area instruction matching, and converting the AGV path scheduling table into a scheduling queue to realize task push.

[0128] Specifically, by directly parsing the scheduling structure from individual codes, additional data conversion and secondary calculations are avoided, improving scheduling efficiency. Secondly, the combination of 3D coordinate mapping and path instruction generation ensures that the scheduling results have high execution feasibility. Thirdly, this structure supports standardized integration with digital twin systems and AGV scheduling control systems, providing a good interface foundation for the integration of multi-platform heterogeneous systems. Fourthly, by extracting the total path time and maximum task time, a refined evaluation of the scheduling strategy under different target dimensions can be achieved, providing support for dynamic strategy adjustment and system load prediction. Finally, the AGV task scheduling queue directly generated based on the scheduling solution can achieve millisecond-level task distribution and path updates, significantly improving the task response speed and path conflict avoidance capability of the warehousing system.

[0129] S4. After the scheduling optimization is completed, the predicted frequency of goods leaving the warehouse is compared with the current inventory level, and a stockout warning and procurement plan are automatically generated and pushed to the supply chain control system to select the best supplier for goods response.

[0130] Specifically, after scheduling optimization, the predicted outbound frequency is compared with the current inventory level, automatically generating a stockout warning and pushing the procurement plan to the supply chain control system. The optimal supplier is selected for the goods response. The predicted outbound frequency is multiplied by the prediction period to obtain v, and the current inventory quantity is multiplied by the safety stock redundancy (set as a multiple of the standard deviation of the outbound frequency) to obtain l. If v is greater than l, it indicates a stockout risk, achieving early quantitative identification of potential stockout risks, significantly improving the warehousing system's early warning capabilities, and reducing logistics delays and production line stoppages caused by inventory shortages. For each stockout item, the pre-registered list of available suppliers in the supply chain platform is automatically called. The supplier's historical average delivery cycle, fulfillment success rate, and unit supply price are weighted and summed. This mechanism comprehensively considers supply stability and cost control, and also supports dynamic weight adjustment based on material urgency to meet the intelligent adaptation of multi-level inventory response strategies. Suppliers are sorted in descending order based on the summed score, selecting the supplier with the highest score, and immediately generating a corresponding replenishment task order, including the replenishment item ID, quantity, latest arrival time, and executing supplier.

[0131] By generating task orders containing "goods ID, required quantity, latest delivery time and designated executor" for the highest-scoring supplier, the system can achieve closed-loop control from warehousing to suppliers, including "prediction-judgment-response-execution", breaking down data barriers between warehousing operations and supply chain control systems, and realizing intelligent and real-time supply response.

[0132] This embodiment also provides a warehouse goods supply chain management system based on AI big data, such as... Figure 3 As shown, the AI-based big data warehouse cargo supply chain management system includes: a data acquisition module for collecting and preprocessing multi-source warehouse data, including basic cargo information and cargo location data; a prediction module for predicting cargo outbound frequency based on real-time cargo data using an LSTM neural network; a scheduling optimization module for constructing a multi-objective scheduling model combining pick-up path time and warehouse center of gravity control based on the prediction results; a scheduling solution module for jointly modeling cargo tasks and AGV path planning as an open-path multi-starting-point asymmetric TSP problem, using the FWA-GA hybrid optimization algorithm to solve the model, obtaining the optimal cargo location allocation matrix and the optimal AGV scheduling path table; and an implementation module for comparing the predicted cargo outbound frequency with the current inventory after scheduling optimization, automatically generating stockout warnings and purchasing plans, pushing them to the supply chain control system, and selecting the optimal supplier for cargo response.

[0133] This embodiment also provides a computer device applicable to the warehouse goods supply chain management method based on AI big data, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the warehouse goods supply chain management method based on AI big data proposed in the above embodiment.

[0134] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0135] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the warehouse goods supply chain management method based on AI big data as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0136] In summary, this invention, based on AI-driven outbound frequency prediction, constructs a dual-objective scheduling model that considers both path time and warehouse center of gravity control. It introduces the FWA-GA hybrid optimization algorithm to solve the open path multi-starting-point asymmetric TSP problem, while integrating the Cauchy-Gaussian perturbation mechanism to enhance local convergence capability. Ultimately, it achieves integrated joint optimization of task allocation, path planning, and warehouse location layout, further linking with the supply chain platform to complete stockout early warning and intelligent supplier selection. This effectively overcomes the multiple limitations of traditional methods in terms of prediction accuracy, scheduling coupling, structural control, and supply coordination.

[0137] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0138] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0139] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of the invention pertain.

[0140] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0141] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system.

[0142] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0143] Furthermore, the functional units in the various embodiments of the present invention can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0144] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present invention.

Claims

1. A warehouse goods supply chain management method based on AI big data, characterized in that: include, Collect and preprocess multi-source warehouse data, and predict the frequency of goods leaving the warehouse based on LSTM neural network; The warehouse multi-source data includes basic cargo information and cargo location data; Based on the prediction results, a multi-objective scheduling model combining picking path time and warehouse center of gravity control is constructed. The cargo task and AGV path planning are jointly modeled as an open path multi-starting point asymmetric TSP problem. The FWA-GA hybrid optimization algorithm is used to solve the model to obtain the optimal cargo location allocation matrix and the optimal AGV scheduling path table. Implement the optimal scheduling scheme. After the scheduling optimization is completed, compare the predicted frequency of goods leaving the warehouse with the current inventory level, automatically generate stockout warnings and purchase plans and push them to the supply chain control system to select the best supplier for goods response. The multi-objective scheduling model based on prediction results, which combines pick-up path time and warehouse center of gravity control, includes: using a Flying-V layout as the warehouse infrastructure, and combining outbound frequency, cargo quality, and structural location to construct a dual-objective function for scheduling optimization, including minimizing path time. Minimize the center of gravity of the warehouse ; The process of jointly modeling cargo tasks and AGV path planning as an open-path, multi-starting-point asymmetric TSP problem, and using the FWA-GA hybrid optimization algorithm to solve the model, yields the optimal cargo location allocation matrix and the optimal AGV scheduling path table, including: Extract all cargo tasks that need to be scheduled, forming a task set A. Record the number of currently online AGVs, initialize the current position of each AGV, and calculate the corresponding path cost d for any two points in the task set as the input of the task movement cost. Construct a scheduling objective function by combining overall efficiency and the response time of a single AGV. And add scheduling constraints; Set constraints that each task must be assigned to only one AGV, and the storage locations for all tasks must meet the physical capacity of the warehouse; Construct an initial feasible solution group, setting the total number of individuals and the maximum number of iterations, with each individual representing a scheduling scheme; Use the Logistic chaotic map to generate the initial perturbation sequence; The first O variables are encoded as integers and mapped to the AGV number to which the task belongs and the execution priority. The last O variables are mapped to the layer number of the goods. The task scheduling result and the cargo location allocation are decoded into a task path table and a cargo location allocation matrix. The scheduling objective function is used as the fitness function to calculate the fitness, and individuals are sorted in descending order of fitness. Spark explosion operation is performed on the top U individuals to generate perturbed candidate scheduling solutions. Define the maximum disturbance radius of the fireworks explosion and calculate the explosion radius of generation t; For each principal individual, calculate the number of sparks based on its corresponding fitness value and the worst fitness value in the population. ; Iterate through each scheduling individual, initialize the spark set, loop through each individual Q times, generate spark individuals in sequence, automatically select the current perturbation strategy according to stage t, compare stage t with the preset threshold, if t is less than the threshold, enter the early global search stage and execute uniform explosion perturbation. Otherwise, it will enter the later local convergence stage, using a Cauchy-Gaussian hybrid perturbation. A Cauchy-Gaussian hybrid perturbation strategy is used to globally perturb the individuals in the initial main population, providing new scheduling solution candidates for elite selection; Each time a spark individual is generated, a legality check and boundary violation repair are performed immediately. All valid spark individuals are stored in the spark set of this entity to form a spark individual set; For each spark individual, the fitness function is recalculated, and the spark individual is bound to the new fitness value to form an expanded population; Sort all individuals in the extended population in ascending order of fitness, select the top n individuals to form the main population of generation t+1, retain the current global best individual, randomly select M pairs of individuals, and perform partial mapping crossover on their task path encoding. Update the path structure of the two individuals after the intersection to form a new set of intermediate individuals. ; right For each individual, a perturbation probability is set, and a portion of tasks are randomly selected to perform mutation operations. During the genetic evolution stage, a Cauchy-Gaussian perturbation strategy is applied to crossover individuals to further optimize the task hierarchy and fine layout of cargo locations. If the perturbation causes variables to go out of bounds, boundary repair should be performed immediately to obtain the perturbation-optimized population. ; right For each individual AGV, extract its task path, perform path planning checks based on the Flying-V warehouse structure, and obtain the optimized scheduling solution for the AGV group. ,right Sort the individuals by fitness value in ascending order, extract the individual with the lowest fitness value w, and compare w with the previously recorded best individual w'. If the fitness value of w is less than that of the best individual w', then update w as the global best individual.

2. The warehouse cargo supply chain management method based on AI big data as described in claim 1, characterized in that: The implementation of the optimal scheduling scheme refers to extracting the optimal task location allocation matrix, AGV path scheduling table, total task path time and maximum AGV task time from the optimal individual code, converting the location matrix into a visual structure in a three-dimensional coordinate system for inbound / outbound area instruction matching, and converting the AGV path scheduling table into a scheduling queue to realize task push.

3. The warehouse cargo supply chain management method based on AI big data as described in claim 2, characterized in that: After the scheduling optimization is completed, the predicted outbound frequency of goods is compared with the current inventory, and a stockout warning and procurement plan are automatically generated and pushed to the supply chain control system. The optimal supplier is selected for goods response. The system compares the prediction results with real-time data to analyze the risk of stockouts. For each stockout item, the system automatically calls the pre-registered list of available suppliers in the supply chain platform, performs a weighted sum of the supplier's historical average delivery cycle, fulfillment success rate and unit supply price, sorts the suppliers in descending order based on the sum score, selects the supplier with the highest score, and immediately generates the corresponding replenishment task order.

4. The warehouse cargo supply chain management method based on AI big data as described in claim 3, characterized in that: The process of collecting and preprocessing multi-source data from the warehouse includes: Before all goods are put into storage, a unique RFID tag is attached to each item to obtain a triplet of the item's unique number, storage / outbound timestamp, and current location coordinates in real time. This corresponds to the shelf row number, shelf number, and column number; By collecting the current weight of goods and real-time inventory, and combining the inbound / outbound time, an inbound / outbound frequency sequence and the current storage coordinates of the goods can be generated. By connecting with the data interface of ERP and WMS systems, each item can obtain corresponding supplier attributes, including delivery cycle, fulfillment rate and unit procurement cost; The collected data is processed for missing values ​​and normalized.

5. The warehouse cargo supply chain management method based on AI big data as described in claim 4, characterized in that: The prediction of cargo outbound frequency based on LSTM neural network refers to inputting real-time cargo outbound data into the LSTM model to obtain the future cargo outbound frequency.

6. A warehouse goods supply chain management system based on AI big data, based on the warehouse goods supply chain management method based on AI big data as described in any one of claims 1 to 5, characterized in that: include, The data acquisition module is used to collect real-time cargo data and perform preprocessing. The forecasting module is used to predict the frequency of goods leaving the warehouse; The scheduling optimization module is used for multi-objective modeling. The scheduling solution module is used to solve for the optimal scheduling result using FWA-GA. The implementation module is used for real-time task scheduling based on the optimal scheduling results. The response module is used for out-of-stock alerts and supplier responses.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the processor executes the computer program, it implements the steps of the warehouse goods supply chain management method based on AI big data as described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it implements the steps of the warehouse goods supply chain management method based on AI big data as described in any one of claims 1 to 5.

Images