Intelligent warehouse big data analysis and dynamic scheduling method and system

Abstract

The application discloses a wisdom warehouse big data analysis and dynamic scheduling method and system, and relates to the technical field of warehousing.The wisdom warehouse big data analysis and dynamic scheduling method and system realize dynamic allocation of resources through an improved algorithm, avoid resource overload and idling by combining a load threshold triggering mechanism, significantly improve resource utilization, realize conflict-free optimal path planning by combining a multi-agent collaborative communication and collision detection mechanism, improve operation smoothness, provide reliable data support for scheduling decisions through multi-source data accurate preprocessing and multi-dimensional quantitative analysis, timely correct scheduling abnormalities by combining a closed-loop feedback mechanism, realize self-iterative upgrading of scheduling strategies through a self-learning optimization mechanism, and do not need manual intervention, and can adapt to dynamic change requirements of different warehouse scenes.

Description

Technical Field

[0001] This invention relates to the field of warehousing technology, specifically to a method and system for intelligent warehousing big data analysis and dynamic scheduling. Background Technology

[0002] With the rapid development of the logistics industry, smart warehousing has become a core support for improving warehouse operation efficiency. Its core requirement lies in achieving the rational allocation of warehouse resources and the efficient planning of operational routes. Currently, smart warehouse scheduling methods generally suffer from two major pain points: First, resource allocation lacks dynamic adjustment capabilities, easily leading to uneven distribution with some resources overloaded and others idle, resulting in low resource utilization. Second, route planning often uses fixed algorithms, failing to fully consider factors such as real-time congestion and multi-operator collaboration, easily causing route conflicts and affecting operational progress.

[0003] In existing technologies, warehouse scheduling mostly employs traditional algorithms for resource allocation and path planning. However, these algorithms suffer from slow convergence speeds and poor adaptability, and lack a complete big data analysis and closed-loop feedback mechanism, making it impossible to adaptively optimize scheduling strategies based on real-time warehouse operating status and historical data. Furthermore, existing scheduling methods are not precise enough in processing multi-source data, leading to data redundancy and interference from abnormal data, further impacting the accuracy of scheduling decisions. Therefore, there is an urgent need for a smart warehouse big data analysis and dynamic scheduling method that can address these issues while balancing accuracy and dynamism. Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for intelligent warehousing big data analysis and dynamic scheduling, which solves the problems mentioned in the background section.

[0006] (II) Technical Solution

[0007] To achieve the above objectives, the present invention provides a method for intelligent warehousing big data analysis and dynamic scheduling, comprising the following steps:

[0008] S1: Multi-source data acquisition and preprocessing: Deploy multiple types of sensing devices to collect multi-source data from the entire warehousing scenario, clean, denoise, standardize and fuse the raw data to generate a standardized big data set;

[0009] S2: Big Data Analysis and Status Assessment: Based on standardized big data sets, a multi-dimensional analysis model is built to quantitatively analyze the load of warehouse resources, the priority of operational tasks and the risk of path congestion, and output corresponding assessment reports, priority lists and congestion warnings;

[0010] S3: Dynamic Resource Allocation Optimization: Based on the analysis results, a dynamic resource allocation model is constructed, and an improved genetic algorithm is introduced to realize dynamic matching and allocation of warehouse resources. A load threshold is set to trigger scheduling adjustments.

[0011] S4: Conflict-free path planning: Combining resource allocation results and congestion warnings, a multi-agent collaborative path planning model is constructed, and an improved algorithm is used to plan the optimal path and synchronize path information;

[0012] S5: Real-time scheduling execution and feedback optimization: Distribute scheduling plans to execution terminals, collect dynamic work data in real time and compare it with preset targets, trigger plan correction when anomalies occur, and form a closed-loop scheduling mechanism;

[0013] S6: Scheduling strategy self-learning optimization: A self-learning model is built based on historical and real-time feedback data, and a reinforcement learning algorithm is used to optimize scheduling parameters to achieve self-iterative upgrading of the scheduling strategy.

[0014] Preferably, in step S1, the multi-source data specifically includes warehouse resource status, work tasks, environmental status and historical operation-related data, wherein the resource status data covers the occupancy / idle status of storage locations, AGV operating parameters, and the working status of picking personnel and loading / unloading equipment;

[0015] The task data includes order information, task type and priority; environmental status data includes warehouse temperature and humidity, passage congestion and temporary obstacle information; historical operation data includes historical resource allocation, path planning and operation efficiency records; the preprocessing process is as follows: Z-score algorithm is used to remove abnormal data, linear interpolation is used to fill missing data, min-max normalization algorithm is used to standardize the data to the [0,1] interval, and weighted fusion algorithm is used to integrate the same type of data collected by multiple devices.

[0016] Preferably, in step S2, the resource load analysis adopts a dynamic weighted algorithm, and the load coefficient calculation formula is: load coefficient = (real-time workload / maximum resource capacity) × resource importance weight + (historical average utilization rate × historical weight), and the weight coefficient is determined by the analytic hierarchy process.

[0017] The task priorities are determined using the analytic hierarchy process (AHP), categorized into three levels: order urgency, cargo characteristics, and task complexity. The categorization results are divided into three levels: Level 1 (urgent), Level 2 (normal), and Level 3 (ordinary). The path congestion risk is predicted using an LSTM neural network, which combines real-time channel occupancy rate, AGV movement speed, and historical congestion data to output the congestion probability, corresponding to three risk levels: high, medium, and low.

[0018] Preferably, in step S3, the improved genetic algorithm is optimized by adaptively adjusting the crossover probability and mutation probability, specifically as follows:

[0019] When the population fitness variance is large, the crossover probability is 0.7-0.9 and the mutation probability is 0.05-0.1; when the variance is small, the crossover probability is 0.3-0.5 and the mutation probability is 0.01-0.03. The resource load threshold is set to 0.8. When the resource load coefficient exceeds 0.8, overload adjustment is triggered, and when it is below 0.2, idle resource scheduling is triggered.

[0020] Preferably, in step S4, the improved algorithm introduces a congestion weight and conflict weight optimization heuristic function, where the heuristic function = path length weight × actual path length + congestion weight × congestion risk coefficient + conflict weight × conflict probability, and the weights are dynamically adjusted according to real-time congestion warnings.

[0021] The dynamic window method adjusts the path speed and direction in real time by setting speed and angle windows; the multi-agent collaborative communication adopts the MQTT protocol and is equipped with an improved collision detection algorithm. When the distance between two agents is detected to be less than the safety threshold, path adjustment is triggered.

[0022] Preferably, in step S5, the preset optimization targets specifically include resource utilization rate ≥80%, on-time task completion rate ≥95%, and path conflict rate ≤5%;

[0023] The scheduling anomalies include three scenarios: resource load imbalance, path conflict, and task delay. When a scheduling anomaly is detected, the mean square error (MSE) algorithm is used to calculate the deviation between the real-time job data and the preset optimization target. When the deviation exceeds 5%, the scheduling scheme correction process is triggered, and steps S2 to S4 are re-executed.

[0024] Preferably, in step S6, the reinforcement learning algorithm adopts the Q-learning algorithm, wherein the learning rate is set to 0.1, the discount factor is set to 0.9, and the exploration rate is set to 0.2.

[0025] The self-learning model is trained based on historical scheduling data and real-time feedback data, and adaptively optimizes the weight parameters of the big data analysis model, the objective function coefficients of the resource allocation model, and the heuristic function parameters of the path planning algorithm.

[0026] This invention also discloses a smart warehousing big data analysis and dynamic scheduling system, including:

[0027] Multi-source data module: Deploys multiple types of sensing devices to collect multi-source data from the entire warehousing scenario, cleans, denoises, standardizes and fuses the raw data to generate a standardized big data set;

[0028] Analysis and evaluation module: Based on standardized big data, a multi-dimensional analysis model is built to quantitatively analyze the load of warehouse resources, the priority of operation tasks and the risk of path congestion, and output corresponding evaluation reports, priority lists and congestion warnings;

[0029] Resource optimization module: Based on the analysis results, a dynamic resource allocation model is constructed, an improved genetic algorithm is introduced to realize dynamic matching and allocation of warehouse resources, and a load threshold is set to trigger scheduling adjustments;

[0030] Path planning module: Combining resource allocation results and congestion warnings, a multi-agent collaborative path planning model is constructed, and an improved algorithm is used to plan the optimal path and synchronize path information;

[0031] Scheduling feedback module: Distributes scheduling plans to execution terminals, collects dynamic work data in real time and compares it with preset targets, triggers plan correction when anomalies occur, and forms a closed-loop scheduling mechanism;

[0032] Learning optimization module: Based on historical and real-time feedback data, a self-learning model is built, and a reinforcement learning algorithm is used to optimize scheduling parameters, so as to realize the self-iterative upgrading of scheduling strategy.

[0033] (III) Beneficial Effects

[0034] This invention provides a method and system for intelligent warehousing big data analysis and dynamic scheduling. Compared with existing technologies, it has the following advantages:

[0035] This intelligent warehousing big data analysis and dynamic scheduling method and system achieves dynamic resource allocation through improved algorithms. Combined with a load threshold triggering mechanism, it avoids resource overload and idleness, significantly improving resource utilization. With multi-agent collaborative communication and collision detection mechanisms, it achieves conflict-free optimal path planning, improving operational smoothness. Through accurate preprocessing of multi-source data and multi-dimensional quantitative analysis, it provides reliable data support for scheduling decisions. Combined with a closed-loop feedback mechanism, it promptly corrects scheduling anomalies. Through a self-learning optimization mechanism, it achieves self-iterative upgrading of scheduling strategies without manual intervention and can adapt to the dynamic changing needs of different warehousing scenarios. Attached Figure Description

[0036] Figure 1 This is a block diagram of an intelligent management system for order picking tasks, as shown in an embodiment of this application.

[0037] Figure 2 This is a schematic diagram of the structure of an electronic device shown in an embodiment of this application. Detailed Implementation

[0038] 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.

[0039] Please see Figures 1-2 The present invention provides a technical solution:

[0040] The intelligent warehousing big data analysis and dynamic scheduling method includes the following steps:

[0041] S1: Multi-source data acquisition and preprocessing: Deploy multiple types of sensing devices to collect multi-source data from the entire warehousing scenario. Clean, denoise, standardize, and fuse the raw data to generate a standardized big data set. In step S1, the multi-source data specifically includes warehousing resource status, work tasks, environmental status, and historical operation-related data. Resource status data covers the occupancy / idle status of storage locations, AGV operating parameters, and the working status of picking personnel and loading / unloading equipment. Work task data includes order information, work type, and priority. Environmental status data includes warehousing temperature and humidity, aisle congestion, and information on temporary obstacles. Historical operation data includes historical resource allocation, path planning, and work efficiency records. The preprocessing process specifically involves: using the Z-score algorithm to remove abnormal data, using linear interpolation to fill in missing data, using the min-max normalization algorithm to standardize the data to the [0,1] interval, and integrating the same type of data collected by multiple devices through a weighted fusion algorithm.

[0042] S2: Big Data Analysis and Status Assessment: Based on standardized big data sets, a multi-dimensional analysis model is constructed to quantitatively analyze warehouse resource load, task priority, and path congestion risk, outputting corresponding assessment reports, priority lists, and congestion warnings. In step S2, resource load analysis adopts a dynamic weighted algorithm. The load coefficient calculation formula is: Load coefficient = (real-time workload / maximum resource capacity) × resource importance weight + (historical average utilization rate × historical weight). The weight coefficient is determined by the analytic hierarchy process (AHP). Task priority is classified into three levels using the AHP, based on order urgency, cargo characteristics, and operational complexity. The classification results are divided into Level 1 (urgent), Level 2 (normal), and Level 3 (ordinary). Path congestion risk is predicted using an LSTM neural network. Combining real-time channel occupancy rate, AGV movement speed, and historical congestion data, the congestion probability is output, corresponding to three risk levels: high, medium, and low.

[0043] S3: Dynamic Resource Allocation Optimization: Based on the analysis results, a dynamic resource allocation model is constructed, and an improved genetic algorithm is introduced to realize dynamic matching and allocation of warehousing resources. A load threshold is set to trigger scheduling adjustments. In step S3, the improved genetic algorithm achieves optimization by adaptively adjusting the crossover probability and mutation probability. Specifically, when the population fitness variance is large, the crossover probability is 0.7-0.9 and the mutation probability is 0.05-0.1; when the variance is small, the crossover probability is 0.3-0.5 and the mutation probability is 0.01-0.03. The resource load threshold is set to 0.8. When the resource load coefficient exceeds 0.8, overload adjustment is triggered, and when it is below 0.2, idle resource scheduling is triggered.

[0044] S4: Conflict-Free Path Planning: Combining resource allocation results and congestion warnings, a multi-agent collaborative path planning model is constructed. An improved algorithm is used to plan the optimal path and synchronize path information. In step S4, the improved algorithm introduces congestion weight and conflict weight to optimize the heuristic function. The heuristic function = path length weight × actual path length + congestion weight × congestion risk coefficient + conflict weight × conflict probability. The weights are dynamically adjusted based on real-time congestion warnings. The dynamic window method adjusts the path speed and direction in real time by setting speed and angle windows. Multi-agent collaborative communication uses the MQTT protocol, combined with an improved collision detection algorithm. When the distance between two agents is detected to be less than a safety threshold, path adjustment is triggered.

[0045] S5: Real-time scheduling execution and feedback optimization: The scheduling plan is sent to the execution terminal, and the dynamic data of the operation is collected in real time and compared with the preset target. When an anomaly occurs, the plan is corrected to form a closed-loop scheduling mechanism. In step S5, the preset optimization target specifically includes resource utilization rate ≥80%, on-time task completion rate ≥95%, and path conflict rate ≤5%. The scheduling anomaly includes three scenarios: resource load imbalance, path conflict, and task delay. When a scheduling anomaly is detected, the mean square error (MSE) algorithm is used to calculate the deviation between the real-time operation data and the preset optimization target. When the deviation exceeds 5%, the scheduling plan correction process is triggered, and steps S2 to S4 are re-executed.

[0046] S6: Scheduling Strategy Self-Learning Optimization: A self-learning model is constructed based on historical and real-time feedback data. A reinforcement learning algorithm is used to optimize scheduling parameters, enabling the scheduling strategy to iteratively upgrade itself. In step S6, the reinforcement learning algorithm uses the Q-learning algorithm, with a learning rate of 0.1, a discount factor of 0.9, and an exploration rate of 0.2. The self-learning model is trained based on historical scheduling data and real-time feedback data, adaptively optimizing the weight parameters of the big data analysis model, the objective function coefficients of the resource allocation model, and the heuristic function parameters of the path planning algorithm.

[0047] The multi-dimensional analysis model is built on the TensorFlow framework and trained using the Adam optimization algorithm. The training iterations are set to 500, and the learning rate is 0.001. The LSTM neural network has the following structure: an input layer with 128 dimensions, 3 hidden layers using ReLU activation, and an output layer using Sigmoid activation. The objective function of the resource dynamic allocation model is constructed using a weighted summation method, with resource utilization, task completion efficiency, and load balancing having weights of 0.4, 0.3, and 0.3, respectively. The improved genetic algorithm encodes the resource allocation scheme into chromosomes using binary encoding, with each gene corresponding to a resource-task matching relationship.

[0048] This invention also discloses a smart warehousing big data analysis and dynamic scheduling system, including:

[0049] Multi-source data module: Deploys multiple types of sensing devices to collect multi-source data from the entire warehousing scenario, cleans, denoises, standardizes and fuses the raw data to generate a standardized big data set;

[0050] Analysis and evaluation module: Based on standardized big data, a multi-dimensional analysis model is built to quantitatively analyze the load of warehouse resources, the priority of operation tasks and the risk of path congestion, and output corresponding evaluation reports, priority lists and congestion warnings;

[0051] Resource optimization module: Based on the analysis results, a dynamic resource allocation model is constructed, an improved genetic algorithm is introduced to realize dynamic matching and allocation of warehouse resources, and a load threshold is set to trigger scheduling adjustments;

[0052] Path planning module: Combining resource allocation results and congestion warnings, a multi-agent collaborative path planning model is constructed, and an improved algorithm is used to plan the optimal path and synchronize path information;

[0053] Scheduling feedback module: Distributes scheduling plans to execution terminals, collects dynamic work data in real time and compares it with preset targets, triggers plan correction when anomalies occur, and forms a closed-loop scheduling mechanism;

[0054] Learning optimization module: Based on historical and real-time feedback data, a self-learning model is built, and a reinforcement learning algorithm is used to optimize scheduling parameters, so as to realize the self-iterative upgrading of scheduling strategy.

[0055] After adopting the method of this embodiment, the utilization rate of warehousing resources has been increased from 65% in the prior art to more than 82%, the on-time completion rate of tasks has reached more than 96%, and the path conflict rate has been controlled below 4%. This effectively solves the problems of uneven distribution of warehousing resources and path planning conflicts, significantly improves the efficiency and intelligence level of warehousing operations, and reduces the cost of manual intervention.

[0056] Example 1: E-commerce warehousing peak season scheduling scenario

[0057] Application scenarios

[0058] A large warehousing center of an e-commerce company covers an area of ​​20,000 square meters and is equipped with 50 AGVs, 10 loading and unloading ports, and 2,000 storage locations. During peak season, the daily order volume exceeds 100,000 orders, including ordinary orders, emergency orders, and cold chain orders. The traditional scheduling mode has problems such as severely uneven resource allocation, frequent AGV path conflicts, and operation delay rate of over 15%.

[0059] Implementation steps

[0060] S1. Multi-source data acquisition and preprocessing

[0061] Deploy RFID readers (to collect location data at a frequency of 5Hz), infrared sensors (to collect information on aisle congestion), AGV status sensors (to collect information on AGV position, battery level, and load), high-definition cameras (to collect information on temporary obstacles), order collection terminals (to collect order information), and temperature and humidity sensors (to collect information on the warehouse environment) to comprehensively collect four types of data:

[0062] Resource status data: storage location occupancy / idle status, AGV real-time power (minimum 10%) and load, loading and unloading port occupancy status, and picking personnel's work progress;

[0063] Task data: 100,000 order records (including order number, type of goods, quantity, urgency level), and task type (picking, replenishment, loading and unloading);

[0064] Environmental status data: passageway congestion rate, location of temporary obstacles, warehouse temperature and humidity;

[0065] Historical operational data: resource allocation records, path planning records, and task completion efficiency data for the past 3 months.

[0066] Preprocessing: The Z-score algorithm is used to remove abnormal AGV positioning data caused by signal interference (such as data with a deviation of more than 2 meters); the missing AGV power data is filled by linear interpolation; all data are normalized to the [0,1] interval by the min-max normalization algorithm; the channel congestion data collected by multiple cameras are integrated by a weighted fusion algorithm to eliminate data redundancy and generate a standardized large dataset.

[0067] S2. Big Data Analysis and Status Assessment

[0068] A multi-dimensional analysis model was built based on the TensorFlow framework and trained using the Adam optimization algorithm (500 iterations, learning rate 0.001) to complete three types of quantitative analysis:

[0069] Resource load analysis: The load coefficient is calculated using a dynamic weighted algorithm. For example, AGV1 has a real-time workload of 80 boxes, a maximum capacity of 100 boxes, an importance weight of 0.6, a historical average utilization rate of 70%, and a historical weight of 0.4. Therefore, the load coefficient is (80 / 100) × 0.6 + (0.7 × 0.4) = 0.76, which does not exceed the threshold of 0.8. The load coefficient of loading and unloading port 3 is 0.85, triggering an overload warning.

[0070] Task priority classification: Using the analytic hierarchy process, order A is an urgent order (urgency level 0.4 + goods characteristics 0.3 + operation complexity 0.3 = 1), and is classified as a level 1 task; order B is a regular order for ordinary goods, and is classified as a level 3 task.

[0071] Path congestion prediction: Using an LSTM neural network (128-dimensional input layer, 3 hidden layers, ReLU activation, Sigmoid output), combined with channel occupancy rate and AGV movement speed data, the probability of channel congestion in the next 10 minutes is predicted. For example, if the congestion probability of channel 5 is 75%, it is designated as a high-risk channel.

[0072] The final output includes a resource load assessment report, a task priority list, and a channel congestion warning, which are then sent to the scheduling core module.

[0073] S3. Dynamic Resource Allocation Optimization

[0074] A dynamic resource allocation model is constructed, with objective function weights of resource utilization (0.4), task completion efficiency (0.3), and load balancing (0.3). An improved genetic algorithm is introduced, using binary encoding to encode the resource allocation scheme into chromosomes (each gene corresponds to the matching relationship between AGVs and order tasks), and the population size is set to 80.

[0075] Adjust parameters based on population fitness variance:

[0076] When the variance is large (insufficient population diversity in the early stage), the crossover probability is set to 0.8 and the mutation probability is set to 0.08 to improve population diversity;

[0077] When the variance is small (it tends to stabilize in the later stages of iteration), the crossover probability is set to 0.4 and the mutation probability is set to 0.02 to avoid premature convergence.

[0078] Set a resource load threshold of 0.8. For loading and unloading port 3, which has an excessive load factor, allocate 30% of its loading and unloading tasks to the idle loading and unloading port 5 with a load factor of 0.25. For AGV30 with a load factor of 0.18, allocate it to replenishment tasks in high-load areas to achieve resource load balancing and solve the problem of uneven allocation.

[0079] S4. Conflict-free path planning

[0080] By combining resource allocation results with congestion warnings, a multi-agent cooperative path planning model is constructed:

[0081] Improved A* algorithm optimization: Heuristic function = path length weight × actual length + congestion weight × risk coefficient + conflict weight × conflict probability. The high-risk channel 5 has a congestion weight of 0.45 and a conflict weight of 0.45, while the regular channel has a weight of 0.25. Priority is given to planning paths with no congestion and low conflict.

[0082] Dynamic window method adaptation: Set the speed window to 0.5-1.2m / s and the angle window to -10°~10°. When the AGV12 detects temporarily stacked goods in the channel, it adjusts the path and speed in real time to avoid obstacles.

[0083] Collaborative communication and collision detection: The MQTT protocol (communication frequency 10Hz) is used to synchronize the path and position information of 50 AGVs; with an improved collision detection algorithm, the safety threshold is 0.5m. When the distance between AGV12 and AGV23 is less than 0.5m, the path of the lower priority AGV23 is adjusted first to avoid collision.

[0084] S5. Real-time scheduling execution and feedback optimization

[0085] Resource allocation and route planning schemes are distributed to execution terminals such as AGVs, loading / unloading points, and picking terminals to control operation execution. Real-time operation data is collected, including AGV operating status, loading / unloading point completion progress, and aisle congestion.

[0086] Preset optimization goals: resource utilization rate ≥ 80%, on-time task completion rate ≥ 95%, path conflict rate ≤ 5%.

[0087] The mean squared error (MSE) algorithm is used to calculate the deviation. For example, if the task delay rate of AGV15 is 8% (exceeding the 5% threshold), the scheme is corrected, and S2-S4 are re-executed to adjust the task allocation and path planning of AGV15, forming a closed-loop scheduling to ensure the normal progress of the operation.

[0088] S6. Scheduling strategy self-learning optimization

[0089] Based on historical scheduling data (1 million scheduling records from the 10 days before the peak season) and real-time feedback data, a Q-learning self-learning model (learning rate 0.1, discount factor 0.9, exploration rate 0.2) is constructed.

[0090] Model continuously optimizes parameters:

[0091] Adjust the "channel congestion weight" in the big data analysis model, and reduce the weight ratio for high-frequency congested channels;

[0092] Optimize the objective function weights of the resource allocation model to increase the load balancing weight to 0.35;

[0093] Adjust the "conflict weight" in the heuristic function of the A* algorithm to enhance the priority of conflict avoidance.

[0094] Through self-iteration, the scheduling strategy gradually adapts to high-concurrency scenarios during peak seasons, achieving self-upgrading of the strategy without manual intervention.

[0095] Implementation effect

[0096] After adopting the scheduling method of this embodiment, the scheduling effect of e-commerce warehousing during peak season is significantly improved:

[0097] Resource utilization rate increased from 62% to 85%, reducing resource idle costs by 300,000 yuan per month;

[0098] The on-time completion rate of tasks increased from 85% to 97.2%, and the order delay rate during peak season was kept below 2.8%.

[0099] The path conflict rate decreased from 18% to 3.5%, and the AGV equipment loss rate decreased by 40%.

[0100] The number of manual scheduling interventions was reduced from 50 times per day to less than 5 times, and scheduling efficiency was improved by 80%, completely solving the problems of uneven resource allocation and path conflict.

[0101] Furthermore, any content not described in detail in this specification is existing technology known to those skilled in the art.

[0102] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0103] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

[0104] Those skilled in the art will also understand that the various exemplary logic blocks, modules, circuits, and algorithm steps described in connection with the present application can be implemented as electronic hardware, computer software, or a combination of both.

[0105] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems and methods according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0106] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A smart warehousing big data analysis and dynamic scheduling method, characterized in that, Includes the following steps: S1: Multi-source data acquisition and preprocessing: Deploy multiple types of sensing devices to collect multi-source data from the entire warehousing scenario, clean, denoise, standardize and fuse the raw data to generate a standardized big data set; S2: Big Data Analysis and Status Assessment: Based on standardized big data sets, a multi-dimensional analysis model is built to quantitatively analyze the load of warehouse resources, the priority of operational tasks and the risk of path congestion, and output corresponding assessment reports, priority lists and congestion warnings; S3: Dynamic Resource Allocation Optimization: Based on the analysis results, a dynamic resource allocation model is constructed, and an improved genetic algorithm is introduced to realize dynamic matching and allocation of warehouse resources. A load threshold is set to trigger scheduling adjustments. S4: Conflict-free path planning: Combining resource allocation results and congestion warnings, a multi-agent collaborative path planning model is constructed, and an improved algorithm is used to plan the optimal path and synchronize path information; S5: Real-time scheduling execution and feedback optimization: Distribute scheduling plans to execution terminals, collect dynamic work data in real time and compare it with preset targets, trigger plan correction when anomalies occur, and form a closed-loop scheduling mechanism; S6: Scheduling strategy self-learning optimization: A self-learning model is built based on historical and real-time feedback data, and a reinforcement learning algorithm is used to optimize scheduling parameters to achieve self-iterative upgrading of the scheduling strategy.

2. The intelligent warehousing big data analysis and dynamic scheduling method and system according to claim 1, characterized in that: In step S1, the multi-source data specifically includes warehouse resource status, work tasks, environmental status and historical operation-related data, wherein the resource status data covers the occupancy / idle status of storage locations, AGV operating parameters, and the working status of picking personnel and loading and unloading equipment. The task data includes order information, task type and priority; environmental status data includes warehouse temperature and humidity, passage congestion and temporary obstacle information; historical operation data includes historical resource allocation, path planning and operation efficiency records; the preprocessing process is as follows: Z-score algorithm is used to remove abnormal data, linear interpolation is used to fill missing data, min-max normalization algorithm is used to standardize the data to the [0,1] interval, and weighted fusion algorithm is used to integrate the same type of data collected by multiple devices.

3. The intelligent warehousing big data analysis and dynamic scheduling method and system according to claim 1, characterized in that: In step S2, the resource load analysis adopts a dynamic weighted algorithm. The load coefficient calculation formula is: load coefficient = (real-time workload / maximum resource capacity) × resource importance weight + (historical average utilization rate × historical weight). The weight coefficient is determined by the analytic hierarchy process. The task priorities are determined using the analytic hierarchy process (AHP), categorized into three levels: order urgency, cargo characteristics, and task complexity. The categorization results are divided into three levels: Level 1 (urgent), Level 2 (normal), and Level 3 (ordinary). The path congestion risk is predicted using an LSTM neural network, which combines real-time channel occupancy rate, AGV movement speed, and historical congestion data to output the congestion probability, corresponding to three risk levels: high, medium, and low.

4. The intelligent warehousing big data analysis and dynamic scheduling method and system according to claim 1, characterized in that: In step S3, the improved genetic algorithm is optimized by adaptively adjusting the crossover probability and mutation probability, specifically as follows: When the population fitness variance is large, the crossover probability is 0.7-0.9 and the mutation probability is 0.05-0.1; when the variance is small, the crossover probability is 0.3-0.5 and the mutation probability is 0.01-0.

03. The resource load threshold is set to 0.

8. When the resource load coefficient exceeds 0.8, overload adjustment is triggered, and when it is below 0.2, idle resource scheduling is triggered.

5. The intelligent warehousing big data analysis and dynamic scheduling method and system according to claim 1, characterized in that: In step S4, the improved algorithm introduces a congestion weight and a conflict weight optimization heuristic function. The heuristic function is calculated as: path length weight × actual path length + congestion weight × congestion risk coefficient + conflict weight × conflict probability. The weights are dynamically adjusted based on real-time congestion warnings. The dynamic window method adjusts the path speed and direction in real time by setting speed and angle windows; the multi-agent collaborative communication adopts the MQTT protocol and is equipped with an improved collision detection algorithm. When the distance between two agents is detected to be less than the safety threshold, path adjustment is triggered.

6. The intelligent warehousing big data analysis and dynamic scheduling method and system according to claim 1, characterized in that: In step S5, the preset optimization targets specifically include resource utilization rate ≥80%, on-time task completion rate ≥95%, and path conflict rate ≤5%. The scheduling anomalies include three scenarios: resource load imbalance, path conflict, and task delay. When a scheduling anomaly is detected, the mean square error (MSE) algorithm is used to calculate the deviation between the real-time job data and the preset optimization target. When the deviation exceeds 5%, the scheduling scheme correction process is triggered, and steps S2 to S4 are re-executed.

7. The intelligent warehousing big data analysis and dynamic scheduling method according to claim 1, characterized in that: In step S6, the reinforcement learning algorithm adopts the Q-learning algorithm, wherein the learning rate is set to 0.1, the discount factor is set to 0.9, and the exploration rate is set to 0.

2. The self-learning model is trained based on historical scheduling data and real-time feedback data, and adaptively optimizes the weight parameters of the big data analysis model, the objective function coefficients of the resource allocation model, and the heuristic function parameters of the path planning algorithm.

8. A smart warehousing big data analysis and dynamic scheduling system, characterized in that: include: Multi-source data module: Deploys multiple types of sensing devices to collect multi-source data from the entire warehousing scenario, cleans, denoises, standardizes and fuses the raw data to generate a standardized big data set; Analysis and evaluation module: Based on standardized big data, a multi-dimensional analysis model is built to quantitatively analyze the load of warehouse resources, the priority of operation tasks and the risk of path congestion, and output corresponding evaluation reports, priority lists and congestion warnings; Resource optimization module: Based on the analysis results, a dynamic resource allocation model is constructed, an improved genetic algorithm is introduced to realize dynamic matching and allocation of warehouse resources, and a load threshold is set to trigger scheduling adjustments; Path planning module: Combining resource allocation results and congestion warnings, a multi-agent collaborative path planning model is constructed, and an improved algorithm is used to plan the optimal path and synchronize path information; Scheduling feedback module: Distributes scheduling plans to execution terminals, collects dynamic work data in real time and compares it with preset targets, triggers plan correction when anomalies occur, and forms a closed-loop scheduling mechanism; Learning optimization module: Based on historical and real-time feedback data, a self-learning model is built, and a reinforcement learning algorithm is used to optimize scheduling parameters, so as to realize the self-iterative upgrading of scheduling strategy.

Images