A warehouse material scheduling method and system for intelligent informatization

Abstract

The application belongs to the technical field of logistics and warehousing informatization, and particularly relates to a warehouse material scheduling method and system for intelligent informatization, which mainly comprises the following steps: step one: constructing a warehousing scenario causality knowledge base, wherein the knowledge base is constructed based on historical scheduling logs and expert experience rules, a Bayesian network model is used to represent the causality strength between elements, the range of core elements is determined, key elements affecting material scheduling are screened, then multi-source historical data are collected, warehouse operation historical data are collected, and the causality between elements and the causality strength are quantified by combining the Bayesian network model, data mining and expert verification, the knowledge base structure is completed, and a graph database is used to store the causality and the core data of the Bayesian network.

Description

Technical Field

[0001] This invention belongs to the field of logistics and warehousing information technology, specifically a method and system for intelligent information-based warehouse material scheduling. Background Technology

[0002] With the rapid development of the logistics industry and the advancement of intelligent transformation in manufacturing, warehouses, as the core hubs for material storage and circulation, directly affect the supply chain response speed and overall cost control of enterprises.

[0003] In recent years, the application of smart information technology in the warehousing field has gradually deepened. The integration of IoT, big data, and artificial intelligence technologies has driven traditional warehouses to upgrade towards automation, intelligence, and digitalization. As a key link in warehouse operation, material scheduling is responsible for coordinating the planning and resource allocation of the entire process of material receiving, storage, picking, outbound, and transfer. Its level of intelligence has become the core factor determining the operational efficiency of smart warehouses.

[0004] However, traditional material scheduling relies heavily on human experience or simple information systems for decision-making. The scheduling logic is rigid and cannot adapt to dynamic changes inside and outside the warehouse in real time. When complex scenarios such as sudden fluctuations in order volume, differences in material attributes, changes in the operating status of warehousing equipment, and dynamic adjustments to inventory occur, manual scheduling is prone to problems such as delayed response and unreasonable scheduling plans, resulting in material backlog, shortages, redundant transportation routes, or idle equipment, which seriously affects warehouse operating efficiency. Therefore, this invention provides a warehouse material scheduling method and system for intelligent information technology. Summary of the Invention

[0005] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0006] The technical solution adopted by this invention to solve its technical problem is: a warehouse material scheduling method and system for intelligent information technology, which mainly includes the following steps: Step 1: Construct a causal relationship knowledge base for warehousing scenarios. The knowledge base is built based on historical scheduling logs and expert experience rules. A Bayesian network model is used to represent the causal strength between each element. The range of core elements is determined by the causal strength, and factors affecting material scheduling are screened. Subsequently, multi-source historical data is collected, including warehouse historical data. A combination of Bayesian network model, data mining, and expert verification is used to mine the causal relationships between elements and quantify the causal strength, thus constructing the knowledge base structure. Step 2: Real-time data acquisition and data preprocessing module simultaneously preprocess the data to provide data support for causal identification. By adopting multi-terminal real-time acquisition, lightweight causal sensing terminals are deployed in the warehouse to clean and standardize the acquired real-time data. Step 3: Real-time causal identification and core driving factor localization. By inputting the preprocessed real-time data into the causal identification module, and based on the constructed Bayesian network model, the causal strength of each candidate causal link is calculated in combination with the real-time data, and the posterior probability of each potential factor is calculated. The factor with the highest posterior probability is determined as the core driving factor. Step 4: Based on the core driving factors and the coping strategies in the causal relationship knowledge base, generate a targeted scheduling plan. By matching the adaptation strategy templates corresponding to the core driving factors from the causal relationship knowledge base, strategy matching and initial plan generation are achieved, and then the feasibility of the initial plan is verified. Step 5: Utilize pilot verification, optimize the solution based on the verification results, and verify the introduction of a causal verification mechanism to avoid scheduling errors caused by misjudgment of causal links and ensure the effectiveness of the solution; Step Six: After the pilot verification is successful, the scheduling scheme will be promoted and deployed across the entire domain, and continuously optimized and dynamically adjusted in combination with real-time data to achieve dynamic iterative updates.

[0007] Preferably, the data preprocessing module adopts a full-process architecture of data acquisition and docking, cleaning and processing, standardization, association and labeling, and quality verification. The data preprocessing module includes multiple functional units, which work together through an internal high-speed data bus and establish data linkage with the front-end acquisition terminal and the back-end cause-effect identification module.

[0008] Preferably, the causal strength calculation and core driving factor screening method is based on the constructed Bayesian network model, combined with real-time data to complete the accurate quantitative calculation of causal strength, and then the core driving factors are located through multi-dimensional screening, avoiding misjudgment of data correlation throughout the process, including the mapping and matching of real-time data and Bayesian network nodes, Bayesian network posterior probability inference calculation, multi-condition link screening, core driving factor location and contribution measurement.

[0009] Preferably, the causal identification module adopts a modular architecture design, including multiple functional units.

[0010] Preferably, the plurality of functional units include a scene matching unit, a data mapping and preprocessing unit, a Bayesian network inference unit, a causal link filtering unit, a core driving factor location and sorting unit, and an inter-unit collaborative scheduling unit.

[0011] Preferably, the feasibility of the initial verification scheme includes: full-process verification of resource adaptability, constraint satisfaction, and cost controllability, with each dimension having clear verification objectives, quantitative indicators, operation procedures, and judgment criteria.

[0012] Preferably, the precise quantification of causal strength is based on a Bayesian network model, employs a joint tree algorithm, and combines real-time data to calculate the posterior probability of each causal link, thereby achieving precise quantification of causal strength.

[0013] Preferably, the cause-effect identification module is uniformly verified by the collaborative scheduling unit. If the condition is met, data flow between units is triggered; if the condition is not met, the process returns to the previous unit or triggers exception handling.

[0014] A warehouse material scheduling system for intelligent information technology, applied to any one of the aforementioned warehouse material scheduling methods for intelligent information technology, wherein the warehouse material scheduling system for intelligent information technology includes: The processor triggers the initialization process upon the first system startup or after a knowledge base iteration and restart. The standardized data acquisition control unit triggers data acquisition when the acquisition timer reaches a preset period in the system standby state. The causal identification and driver location module receives a qualified structured dataset and triggers the causal identification process. The solution generation and verification unit receives a list of core driving factors and triggers the solution generation process. Pilot verification and optimization, receive qualified initial solutions, and trigger the pilot process; The dynamic iteration module is deployed across the entire domain. Once the optimized solution passes verification, the entire domain deployment process is triggered.

[0015] The beneficial effects of this invention are as follows: 1. The present invention provides a warehouse material scheduling method and system for intelligent information technology. By constructing a causal relationship knowledge base for warehousing scenarios, it focuses on the real causal relationships of various elements rather than simple data correlations, thereby eliminating scheduling deviations caused by false relationships from the source. This makes decisions more aligned with the core contradictions of warehouse operations, accurately pinpointing the direct cause and effect of moisture-sensitive materials exceeding humidity standards and failing quality inspection, and avoiding mis-adjusting other irrelevant elements.

[0016] 2. The warehouse material scheduling method and system for intelligent information systems described in this invention systematically sorts out four categories of core elements and causal links, adopts a unified knowledge base structure of nodes, edges, and attributes, and forms a reusable and iterative decision framework with quantitative standards of causal strength grading and conditional probability. This eliminates the need to re-sort out the element relationships for each scheduling, greatly improving the efficiency and consistency of scheduling scheme generation.

[0017] 3. The warehouse material scheduling method and system for intelligent information technology described in this invention, by clearly defining the scope of core elements and collecting multi-source data in the early stage, provides clear element anchors and data specifications for subsequent real-time data collection and causal identification. It can accurately adapt to various warehousing scenarios such as batch orders, emergency picking, and local material backlog, ensuring that dynamic scheduling always revolves around the causal relationship of core elements, and improving the adaptability and implementation effect of the solution. Attached Figure Description

[0018] The invention will now be further described with reference to the accompanying drawings.

[0019] Figure 1 This is a schematic diagram of the scheduling method in this invention; Figure 2 This is a schematic diagram of the scheduling system process in this invention. Detailed Implementation

[0020] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0021] like Figure 1 As shown in the embodiment of the present invention, a warehouse material scheduling method and system for intelligent information technology mainly includes the following steps: Step 1: Construct a causal relationship knowledge base for warehousing scenarios. This knowledge base is built based on historical scheduling logs and expert experience rules, using a Bayesian network model to represent the causal strength between elements. By determining the scope of core elements, four key elements affecting material scheduling are selected. Subsequently, multi-source historical data is collected, including 3 to 5 years of warehouse operational history data. A combination of Bayesian network modeling, data mining, and expert verification is used to uncover causal relationships between elements and quantify their causal strength, thus completing the knowledge base structure. A graph database is used to store causal relationships and core Bayesian network data. Specific operations include: Define the scope of core elements, including material attribute elements (category, weight, fragility, shelf life, temperature and humidity requirements, etc.), order characteristic elements (order quantity, urgency, delivery time limit, order type, etc.), equipment status elements (AGV load rate, battery life, failure frequency, shelf occupancy rate, sorting machine processing speed, etc.), and environmental and operational elements (warehouse temperature and humidity, energy consumption level, number of operators / skill level, material backlog, etc.). Collect multi-source historical data: Collect warehouse operation history data for the past 3 to 5 years, including material inbound and outbound records, order fulfillment data, equipment operation logs, environmental monitoring data, personnel operation data, and scheduling error cases. The data format is uniformly structured data, such as Excel and JSON. Unstructured data is converted into structured data through OCR and natural language processing (NLP), such as equipment failure pictures and manual scheduling logs. Uncovering causal links: Using a combination of Bayesian network models, data mining, and expert verification, we can uncover causal relationships between elements and quantify the strength of causality. Specifically, this includes Bayesian network model construction, causal strength quantification, potential association identification and verification, and expert verification optimization. Bayesian network model construction: Using the four core elements determined in step one as network nodes, based on multi-source historical data, the causal dependencies between nodes are learned through the K2 algorithm to construct the Bayesian network topology. For example, the humidity of the moisture-prone material storage area is the parent node, the outbound quality inspection failure rate is the child node, and the temperature and humidity control status is the intermediate node. Causal strength quantification: Causal strength is represented by the conditional probability table of the Bayesian network, which is the conditional probability that the child node will be in the corresponding state when the parent node is in a certain state. For example, P(outbound quality inspection failure rate increases by >15%|humidity in the moisture-prone material storage area >60%) = 0.75. The higher the probability value, the stronger the causal strength. Subsequently, the probability value is mapped to three levels: strong, medium, and weak, which can also be understood as three levels: 0.6 to 1.0, 0.3 to 0.6, and 0 to 0.3. Constructing the knowledge base structure: Using a graph database to store causal relationships and core data of Bayesian networks, organized in the form of nodes, edges, and attributes; Nodes: These include core element nodes and Bayesian network nodes. Element nodes store basic element information, while network nodes store the corresponding element's state values. For example, the AGV load rate node can have values ​​of ≤60%, 60%-80%, and >80%. Edges: Mark the causal direction of cause and effect, and also identify the dependency relationship in the Bayesian network; Attributes: In addition to the original causal strength, triggering conditions, impact threshold, and response priority, new Bayesian network parameter attributes are added to store the conditional probability values ​​of the corresponding causal links, such as P(transportation delay|AGV load rate>80%)=0.72, the range of node status values, and the probability update timestamp. At the same time, a linkage update rule between the knowledge base and the Bayesian network is established. It is set to update the conditional probability table every quarter based on the latest operational data through Bayesian network parameter learning, such as the maximum likelihood estimation method, and adjust the causal strength level synchronously to ensure the real-time and accuracy of the causal strength representation. Step Two: Real-time data acquisition and preprocessing module simultaneously preprocess the data to provide data support for causal identification, ensuring data timeliness, integrity, and accuracy. This is achieved through multi-terminal real-time acquisition, deploying lightweight causal sensing terminals within the warehouse, and cleaning and standardizing the acquired real-time data. Multi-terminal real-time data collection: Deploy lightweight causal sensing terminals within the warehouse, including: Material side: Built-in RFID reader / writer, mainly collecting material ID, attributes, location, and entry / exit time; temperature and humidity sensor, mainly collecting temperature and humidity of the material storage environment; Equipment side: The AGV has built-in sensors that mainly collect data on load rate, battery life, running speed, and fault codes; the shelf pressure sensor mainly collects data on occupancy rate; and the sorting machine's operating status sensor mainly collects data on processing speed and failure rate. Order side: Order management system interface, real-time synchronization of order volume, urgency level, and delivery deadline; Personnel side: Workers' clock-in terminal and skill level recognition module, mainly collecting the number of personnel, skill level, and work location; Environmental components: Warehouse-wide temperature and humidity sensors, energy consumption monitors; The data collection frequency is set according to the dynamic nature of the elements. High-frequency elements are collected once every 5 seconds, such as AGV operating status and order volume, while low-frequency elements are collected once every 30 seconds, such as the number of personnel and shelf occupancy rate. Data preprocessing module: Cleans and standardizes the collected real-time data: removes outlier data; fills in missing data using adjacent time point interpolation; standardizes the format, converting heterogeneous data from different terminals into a unified format; and adds data association labels, labeling each data point with associated element IDs. Step 3: Real-time causal identification and core driving factor localization. By inputting the preprocessed real-time data into the causal identification module, and based on the constructed Bayesian network model, the causal strength of each candidate causal link is calculated in combination with the real-time data, the posterior probability of each potential factor is calculated, and the factor with the highest posterior probability is determined as the core driving factor. Scene matching and causal link activation: The pre-processed real-time data is input into the causal identification module. The module matches the warehousing scene type based on the data features, such as batch order inbound scene, emergency order picking scene, and material transfer scene. It also activates the candidate causal links for the corresponding scene from the causal relationship knowledge base. For example, in the emergency order picking scene, the candidate links include high order urgency, insufficient picking time, outbound delay and high AGV load rate, low transfer efficiency, picking waiting, etc. Prioritizing Driving Factors: If multiple core driving factors exist, such as simultaneously high AGV load rate and a high proportion of novice operators, the priority attributes in the causal knowledge base are used for hierarchical prioritization. This is further fine-tuned based on real-time scenario requirements to ensure the prioritization aligns with actual scheduling needs. The core priority rule is as follows: Factors affecting order delivery time are prioritized > factors affecting material safety > factors affecting operating costs. Within each level, factors are further ranked according to their causal strength. Higher causal strength indicates a greater posterior probability and a higher priority. For example, excessive AGV load affects delivery time (causal strength 0.72), which is higher than excessively high proportion of novice operators affecting work efficiency and consequently cost (causal strength 0.61). Within the same level, excessive humidity in the storage area for moisture-sensitive materials affects material safety (causal strength 0.75), which is higher than disorderly stacking of materials affecting material safety (causal strength 0.63). An emergency fine-tuning channel is also provided. If a lower-level factor presents a sudden risk, such as a material safety hazard that may escalate rapidly, it can be temporarily upgraded by one level to ensure the flexibility of scheduling response. Step 4: Based on the core driving factors and the coping strategies in the causal relationship knowledge base, generate a targeted scheduling plan to ensure that the plan can accurately solve the core problems in the causal chain. By matching the corresponding adaptation strategy templates of the core driving factors from the causal relationship knowledge base, strategy matching and initial plan generation are achieved. Then, the feasibility of the initial plan is verified from three dimensions. Strategy matching and initial solution generation: Matching suitable strategy templates corresponding to core driving factors from the causal relationship knowledge base. For example, if the core driving factor is excessive AGV load, the corresponding strategy template is to split the material transfer task and schedule idle AGVs for collaborative operation. Refining template parameters based on real-time data generates an initial solution, such as splitting the material transfer task of order-12345 into two sub-tasks, scheduling AGV-02 and AGV-01 for collaborative operation, with AGV-02 currently at a load of 30%, AGV-01 responsible for transporting light materials (weight < 50 kg), and AGV-02 responsible for transporting heavy materials (50-100 kg). The transfer path is optimized to the shortest path and avoid congested areas, selecting area B. The feasibility of the initial solution is verified in three dimensions: resource adaptability, constraint satisfaction, and cost controllability. If infeasible, returning to the strategy template library to rematch alternative strategies. If there are no idle AGVs, the strategy is adjusted to prioritize the transfer of urgent materials and delay the transfer of non-urgent materials. Step 5: Utilize a pilot test to verify the solution, optimize the solution based on the verification results, and introduce a causal verification mechanism to avoid scheduling errors caused by misjudgment of causal links and ensure the effectiveness of the solution. Pilot scope selection: Based on the coverage of the current scheduling scenario, select 10%-20% of the area / task as the pilot scope. For example, if it is a full warehouse AGV transfer scheduling, select warehouse area A as the pilot, which includes 3 AGVs and 5 order tasks. If it is a single order picking scheduling, directly use that order as the pilot. Pilot Implementation and Effect Monitoring: Deploy the generated scheduling scheme within the pilot area and monitor the implementation effect of the scheme in real time. Key monitoring indicators include problem-solving effectiveness, such as whether the AGV load rate, a core driving factor, has dropped below 80% and whether the outbound delay risk has dropped below 5%; derivative impacts, such as whether it has led to material backlog or equipment conflicts in other areas; causal link verification, such as whether there is a direct causal relationship between the elimination of the core driving factor and the alleviation of problem symptoms after the scheme is implemented, rather than a coincidence; monitoring data is synchronized to the causal identification module in real time. Solution optimization and iteration: If the pilot test results meet the requirements, the problem resolution rate is ≥80%, and there are no serious derivative impacts, then the core logic of the solution will be retained; if the results do not meet the requirements, such as the AGV load rate only drops to 75%, but the order picking waiting time increases, then the causal links will be re-analyzed based on the monitoring data, and the solution parameters will be adjusted, such as optimizing the AGV task splitting ratio and adjusting the transfer path, and a small-scale pilot test will be conducted again until the results meet the requirements. Step Six: After the pilot verification is successful, the scheduling plan will be promoted and deployed across the entire domain, and continuously optimized and dynamically adjusted based on real-time data to achieve dynamic iterative updates; Deployment of the overall solution: The optimized scheduling plan is issued to each execution terminal in the warehouse through scheduling instructions, including the AGV scheduling system, the sorting machine control system, and the personnel operation terminals. The execution tasks, time nodes, and operation standards of each terminal are clearly defined. For example, the AGV needs to complete the transfer after the task is broken down within 10 minutes, and the personnel need to respond to the picking instructions within 5 minutes. During the deployment process, the execution status of each terminal's instructions is monitored in real time to avoid instruction conflicts. Real-time dynamic adjustment: During the implementation of the scheduling scheme, the cause identification module continuously receives real-time data from each terminal. If the core driving factor is detected to reappear or a new driving factor appears, steps three to five are repeated to quickly generate the adjusted scheduling scheme, thus realizing a closed loop of real-time monitoring and dynamic adjustment. Knowledge base iterative updates: The implementation effect data of the scheduling scheme is synchronized to the causal relationship knowledge base. The effect data includes the problem-solving effect, derivative impact, and scheme optimization records. This triggers the Bayesian network submodule to learn parameters and updates the conditional probability table of the corresponding causal link through maximum likelihood estimation. The causal strength level is adjusted, and the network nodes, dependencies, and initial conditional probability values ​​corresponding to newly discovered causal relationships are added. For example, the initial conditional probability P(device conflict|path intersection) = 0.68 for path intersection and equipment conflict during AGV collaborative operation. This continuously optimizes the inference accuracy of the Bayesian network model and improves the accuracy of subsequent scheduling decisions.

[0022] like Figure 1 As shown, the data preprocessing module adopts a full-process architecture of data acquisition and docking, cleaning and processing, standardization, association and annotation, and quality verification. The core consists of 6 functional units, each of which works collaboratively through an internal high-speed data bus and establishes data linkage with the front-end acquisition terminal and the back-end cause-effect identification module. Specifically, it includes a multi-source data docking unit, a raw data cleaning unit, a missing data completion unit, a data standardization processing unit, a data association and annotation unit, and a data quality verification and feedback unit. Each unit is independently encapsulated and can be flexibly started and stopped according to data processing needs, adapting to different data processing scenarios. It works in a sequential manner according to the time logic of data access, cleaning and completion, standardization, association and annotation, quality verification, and data output, forming a closed-loop data processing flow. Data access of the multi-source data docking unit: As the front-end entry point of the module, this unit is responsible for establishing stable connections with various types of sensing terminals and system interfaces in the warehouse to achieve unified access to multi-source raw data. First, the unit has a built-in multi-protocol adaptation module that supports mainstream communication protocols of devices / systems such as RFID readers, sensors, AGV control systems, and order management systems, such as MQTT, HTTP, and TCP / IP. It can automatically identify the type of access device and match the corresponding protocol to avoid data transmission interruption caused by protocol heterogeneity. Second, it adopts a time-division multiplexing data reception mechanism, allocating priority data channels for high-frequency data collection, such as AGV running status and order volume, which are collected once every 5 seconds. Low-frequency data uses ordinary channels, such as the number of personnel and shelf occupancy rate, which are collected once every 30 seconds to ensure the real-time transmission of high-frequency data and avoid channel congestion. Finally, it adds metadata tags such as timestamps, device IDs, and collection locations to the received raw data to form a data packet associated with the raw data and metadata. This packet is pushed to the raw data cleaning unit through the internal bus, and a data access log is recorded for subsequent fault tracing. Impurity removal in the raw data cleaning unit: This unit precisely removes abnormal, duplicate, and invalid data from the incoming raw data to ensure data purity. Specific operations include: Abnormal data removal: A dual mechanism of threshold judgment and trend analysis is adopted to preset reasonable value ranges for each type of data, such as the reasonable range of AGV load rate of 0-100% and the reasonable range of warehouse humidity of 30%-80%RH. Data outside the range is directly marked as abnormal. For abnormal trend data, such as temperature and humidity changing by more than 20% within 1 second and no equipment failure signal, combined with the trend judgment of data at adjacent time points, abnormal data caused by sensor failure and signal interference are removed. Duplicate data removal: A unique identifier is constructed based on the data's timestamp, device ID, and core collected value. The duplicate data packets are removed by comparing them using a hash algorithm, such as when the same RFID reader repeatedly collects the same material ID data within the same second. Invalid data removal: Invalid data packets without core collected values ​​or missing metadata (such as no timestamp or device ID) are removed. At the same time, incomplete data fragments caused by communication interruption are filtered to ensure the integrity of data entering the next stage. After cleaning, the cleaned dataset is generated and pushed to the missing data completion unit. Missing data completion unit's vulnerability mitigation: For the small number of missing values ​​remaining in the cleaned data, this unit employs a scenario-based completion strategy to avoid data distortion caused by a single completion method. Specific strategies include: High-frequency data completion: For high-frequency data such as AGV endurance and operating speed, the adjacent time point interpolation method is used to calculate the mean based on the effective data of 3 time points before and after the missing data, fill in the missing values, and ensure the continuity of data time sequence; Low-frequency data completion: For low-frequency data such as the number of personnel and shelf occupancy rate, the historical comparison method is used to retrieve the average historical data of the same time period and the same scenario in the past 7 days as the basis for completion, and at the same time, a completion mark is marked to facilitate subsequent data traceability; Key data completion: If key data affecting scheduling decisions, such as order urgency and material temperature and humidity requirements, are missing, a re-collection instruction is immediately sent to the acquisition terminal through the data feedback interface. If re-collection fails, the data is completed based on the default values ​​in the causal relationship knowledge base, such as defaulting to normal priority for orders and normal temperature storage for materials. The data is also synchronized to the quality verification unit for key marking. After completion, the complete dataset is output to the data standardization processing unit. The data standardization processing unit unifies the format: This unit is responsible for transforming heterogeneous complete datasets into a unified format to meet the input requirements of Bayesian network inference and causal identification modules. Its core includes three types of standardization operations: Data type standardization: All data is uniformly converted into a structured format. Numerical data, such as load rate and temperature and humidity, is retained to two decimal places. Discrete data, such as order type and equipment status, is converted into numerical codes, such as emergency order = 1, ordinary order = 0 and equipment normal = 0, fault = 1. Text data, such as material category, is converted into standardized codes through dictionary mapping, such as fragile items = 001 and ordinary materials = 002. Unit standardization: unify the units of measurement for all types of data, such as temperature and humidity units being standardized to ℃ and %RH, weight units being standardized to kg, and speed units being standardized to m / s, to avoid inference errors caused by differences in units; Format standardization: All data is encapsulated in JSON format and organized according to the field structure of device ID, timestamp, data type, standardized value, and completion identifier to ensure that the data can be directly parsed and called by subsequent modules. After standardization, a standardized dataset is generated and pushed to the data association and annotation unit. Link binding of the data association annotation unit: This unit is responsible for establishing the association between different types of data, providing link support for subsequent causal identification. First, based on the device ID and timestamp in the metadata, related data of the same material, the same order, and the same device are associated. For example, the 85% load rate of AGV-01 is bound to the corresponding transfer order-12345, material-6789, and collection location-area A. Second, the element categories corresponding to the annotation data, such as material attribute elements, equipment status elements, and order feature elements, are associated with the element node IDs in the causal relationship knowledge base to achieve accurate mapping between standardized data and knowledge base element nodes. Finally, a structured real-time dataset containing the association is generated and synchronized to the data quality verification and feedback unit. The data quality verification and feedback unit is the final gatekeeper: As the end unit of the module, it is responsible for performing final quality verification on the structured real-time dataset to ensure that the output data meets the requirements of the causal identification module. The verification dimensions include: Integrity check: Checks whether the dataset contains missing values ​​or whether the relationships are complete; Accuracy verification: Randomly select 10% of the data and compare it with the original collected data to ensure that the cleaning, standardization and correlation operations are not distorted, and the error rate is controlled within 1%. Real-time verification: Check the difference between the data timestamp and the current time. The delay of high-frequency data should not exceed 1 second, and the delay of low-frequency data should not exceed 3 seconds to avoid affecting the accuracy of causal inference due to data lag. If the verification passes, the structured real-time dataset is pushed to the causal identification module in step three through the module output interface. If the verification fails, the unqualified data type and the reason are marked. On the one hand, the data is fed back to the corresponding processing unit for secondary processing, such as abnormal data being fed back to the cleaning unit and delayed data being fed back to the docking unit. On the other hand, data output is suspended until the verification passes after secondary processing, forming a closed-loop control of data processing.

[0023] like Figure 1 As shown, the causal strength calculation and core driving factor screening method is mainly based on the constructed Bayesian network model. It combines real-time data to complete the accurate quantitative calculation of causal strength, and then locates the core driving factors through multi-dimensional screening, avoiding misjudgment of data correlation throughout the process. The specific operations include mapping and matching of real-time data and Bayesian network nodes, Bayesian network posterior probability inference calculation, multi-condition link screening, core driving factor positioning and contribution measurement. Mapping and matching of real-time data with Bayesian network nodes: First, the preprocessed real-time data is discretized to transform continuous data into discrete state values ​​preset by Bayesian network nodes, such as AGV load rate and temperature and humidity. For example, the continuous value of AGV load rate is mapped to three states: low load (≤60%), medium load (60%-80%), and high load (>80%). The warehouse humidity is mapped to three states: suitable (≤60%RH), slightly high (60%-70%RH), and excessively high (>70%RH). For discrete data, such as order urgency and equipment failure status, the corresponding state label of the network node is directly matched. For example, urgent orders are matched with high priority status, and normal equipment is matched with no failure status. During the mapping process, the timestamp and associated element ID of the data are recorded synchronously to ensure that each piece of real-time data can be accurately associated with the target node of the Bayesian network and avoid inference errors caused by misalignment between data and nodes. Posterior probability inference calculation for Bayesian networks: The joint tree algorithm is used as the core algorithm for precise inference of Bayesian networks. The specific inference process is as follows: First, the topology of the Bayesian network is transformed into a joint tree. The joint tree integrates the conditional probability tables of each node into the potential function of the clique node. Then, the mapped real-time data is input into the joint tree as evidence variables. The potential function of the clique node is updated through a message passing mechanism to complete the posterior probability calculation for all nodes in the network. Finally, the posterior probability of the result node in each candidate causal link is extracted, serving as the quantification value of the causal strength of that link. A higher posterior probability indicates a stronger driving effect of the current cause node state on the result node state, and a more significant causal association. For example, if the real-time data shows AGV-01 with a load rate of 85%, indicating a high load state, after inputting into the Bayesian network, the posterior probability of the result node with a transportation delay probability increase > 20% is 0.72, meaning the causal strength quantification value of this causal link is 0.72, corresponding to a strong level. If the real-time data shows a 10% decrease in sorting machine processing speed, the posterior probability of an order delay risk increase > 15% is 0.45, corresponding to a medium level. Multi-condition link filtering: A three-dimensional filtering mechanism is adopted, which includes causal strength threshold, trigger condition matching, and link validity verification, to eliminate invalid or weakly correlated links; The first screening step is the causal strength threshold: a posterior probability ≥ 0.3 is set as the screening threshold, corresponding to the medium level. Weak causal links with a posterior probability < 0.3 are directly eliminated. For example, the posterior probability of slightly higher warehouse energy consumption and order delay is 0.18, so they are eliminated. The second layer of screening is trigger condition matching: verify whether the status of the cause node in the filtered link meets the preset trigger conditions in the causal knowledge base. For example, the trigger condition for humidity > 60% in the storage area of ​​moisture-sensitive materials and the increase in the quality inspection failure rate is humidity > 60% and the material is a moisture-sensitive category. If the material in the real-time data is not a moisture-sensitive category, even if the humidity exceeds the standard and the posterior probability meets the standard, it is determined that the trigger condition does not match and is removed. The third screening step is link validity verification: By using the causal independence test of Bayesian networks, it is determined whether there is a direct causal relationship between the cause node and the result node. False links caused by third-party nodes are eliminated, such as the link from staff lunch break to reduced staff to decreased picking efficiency to order delay. The link from staff lunch break to order delay is an indirect link. Only direct links such as reduced staff to decreased picking efficiency to order delay and decreased picking efficiency to order delay are retained. Core Driver Identification and Contribution Measurement: Employing reverse tracing and probability contribution calculation, core drivers are accurately identified. Starting with currently monitored problem symptoms, i.e., the status of result nodes in the post-processing chain is filtered (e.g., outbound delay risk > 20%), all parent and grandparent nodes pointing to this result node in the Bayesian network are traced back to construct a causal tracing chain of problem symptoms and multi-level causes. Subsequently, the marginal probability contribution algorithm is used to calculate the contribution ratio of each cause node to the posterior probability of the result node. The formula is: The contribution of a cause node = (posterior probability of the result node when the cause node exists - posterior probability of the result node when the cause node does not exist) / total posterior probability of the result node × 100%; Finally, the top 3 contributing factors with a contribution rate ≥ 30% are selected as core driving factors. If there are nodes with significantly different contribution rates, such as a node with a contribution rate ≥ 50% and other nodes < 30%, then that node is judged as a core driving factor. For example, if the total posterior probability of outbound delay risk > 20% is 0.78, where the contribution of AGV load rate > 80% is 72%, the contribution of insufficient picking personnel is 28%, and the contribution of high warehouse humidity is 15%, then only the excessive AGV load rate is judged as a core driving factor. If the total posterior probability of quality inspection failure rate increase > 15% is 0.65, the contribution of humidity > 60% in the moisture-prone material storage area is 45%, the contribution of messy material stacking is 35%, and the contribution of decreased sensitivity of quality inspection equipment is 20%, then the first two are both judged as core driving factors.

[0024] like Figure 1As shown, the causal identification module adopts a modular architecture design, including multiple functional units. The causal identification module is the core decision-making unit of the entire scheduling system. Its core function is to accurately identify the causal relationships of warehouse scheduling problems and pinpoint the core driving factors based on qualified real-time data and a causal knowledge base, providing a scientific basis for the generation of subsequent scheduling schemes and replacing the traditional decision-making mode that relies on misjudgments of correlation. Specific functions include: Scene matching and link activation: After receiving the standardized dataset, the scene matching degree is calculated through a quantitative formula to accurately activate the candidate causal links of the corresponding scene in the knowledge base, ensuring that the analysis scope fits the actual scheduling scenario. Data node mapping verification: Accurately map real-time data to Bayesian network nodes, and eliminate unreasonable mappings through accuracy verification to provide effective data input for subsequent inference calculations; Bayesian network inference computation: calling network parameters from the knowledge base, quantifying the causal strength of each link through the joint tree algorithm and Bayesian formula, and generating a traceable causal strength comparison table; Effective link screening: Through a triple screening process of strength threshold, triggering conditions, and validity criteria, indirect links and weakly related links are eliminated, and core effective causal links are retained; Core factor identification and ranking: Calculate the contribution of each factor, screen the core driving factors with high contribution, sort them according to priority rules and output a list to directly guide the targeted generation of scheduling plans; In addition, this module plays a synergistic role, receiving qualified data from the real-time data acquisition module, linking with the causal knowledge base module to call models and parameters, and directly pushing the output list of core driving factors to the solution generation module. At the same time, it receives pilot verification data to back-verify the accuracy of the causal link and help the knowledge base iterate.

[0025] like Figure 1 As shown, the multiple functional units include a scene matching unit, a data mapping and preprocessing unit, a Bayesian network inference unit, a causal link filtering unit, a core driving factor localization and sorting unit, and an inter-unit collaborative scheduling unit. Each functional unit works collaboratively according to the temporal logic of data input, scene matching, inference calculation, filtering and localization, and result output. The specific principle is as follows: Initialization and data reception of the collaborative scheduling unit: After the module starts, the collaborative scheduling unit first completes the initialization, loads the scene configuration rules and Bayesian network basic parameters in the causal relationship knowledge base, and receives the structured real-time dataset after preprocessing in step two. It verifies the data integrity through the data verification interface, such as whether the key element ID and timestamp are missing. If the data is missing, it is fed back to the data preprocessing module in step two for completion. After the data is qualified, it is distributed to the scene matching unit and the data mapping and preprocessing unit through the internal data bus. Scene recognition and link activation of the scene matching unit: After receiving real-time data distributed by the collaborative scheduling unit, the scene matching unit performs scene matching based on the preset scene feature rule base, which is stored in the causal relationship knowledge base. The feature rule base contains the core data features of different warehousing scenarios. For example, the features of the batch order inbound scenario are ≥50 orders per order and ≥10 times / hour for material inbound frequency. The features of the emergency order picking scenario are order priority code=1 and delivery time limit≤30 minutes. The matching degree between real-time data and each scenario feature is calculated by the feature matching algorithm. The scenario with a matching degree of ≥85% is selected as the current target scenario. Then, according to the target scenario, the corresponding candidate causal link list is called from the causal relationship knowledge base. The list is pushed to the causal link filtering unit through the link activation interface and synchronized to the collaborative scheduling unit to record the time sequence node. Data mapping and node association with the preprocessing unit: This unit works in conjunction with the data preprocessing module in step two. After receiving the preprocessed real-time data, it further completes the accurate mapping between the real-time data and the Bayesian network nodes. For continuous data, such as AGV load rate and temperature and humidity, the state mapping is completed through preset discretization rules, such as mapping an AGV load rate of 85% to a high load state. The reasonableness of the mapping results is verified to avoid data that exceeds the range of node state values. For discrete data, such as the urgency of orders, the state labels of the Bayesian network nodes are directly matched. Finally, a data, node, and state association mapping table is generated and pushed to the Bayesian network inference unit through the internal data bus and synchronized to the collaborative scheduling unit for record. Causal strength calculation of the Bayesian network inference unit: After receiving the data mapping table, the inference unit is triggered by the collaborative scheduling unit to perform the inference calculation process. First, the Bayesian network topology and conditional probability table matching the current scenario are loaded from the causal relationship knowledge base. The mapped real-time data is input into the network as evidence variables. The joint tree algorithm is used to perform posterior probability inference. The specific process is as follows: First, the Bayesian network is transformed into an undirected joint tree structure. The CPT of each node is integrated into the potential function of the clique node in the joint tree. The posterior probability calculation of all network nodes is completed through the message passing mechanism, that is, the potential function update between clique nodes. Finally, the posterior probability (i.e., causal strength quantification value) of the result node (problem symptom node) in each candidate causal link is extracted. A causal link and posterior probability comparison table is generated and pushed to the causal link filtering unit. The calculation process log is fed back to the collaborative scheduling unit. Multi-dimensional filtering of causal link filtering unit: After receiving the candidate causal link list pushed by the scene matching unit and the causal link and posterior probability comparison table pushed by the Bayesian network inference unit, this unit starts the three-dimensional filtering process. After the filtering is completed, a list of valid causal links is generated and pushed to the core driving factor localization and ranking unit. Precise positioning of the core driving factor location and ranking unit: After receiving the list of valid causal links, this unit first extracts all result nodes from the list, that is, the currently monitored problem symptoms, such as outbound delay risk >20%. Through the reverse tracing interface, it traces all parent and grandparent nodes pointing to the result node from the Bayesian network, constructing the tracing link of problem symptoms and multi-level causes. Then, it calculates the contribution ratio of each cause node to the result node through the marginal probability contribution algorithm. Based on the contribution, cause nodes with a contribution of ≥30% are selected as candidate driving factors. Then, it completes the ranking by combining the response priority attribute in the causal relationship knowledge base, generating a core driving factor list, including priority, contribution, and corresponding causal link. Finally, the list is pushed to the collaborative scheduling unit through the result output interface. Results summary and output of the collaborative scheduling unit: After receiving the list of core driving factors, the collaborative scheduling unit integrates the work logs of each unit, namely the scene matching results, reasoning calculation process, and filtering records, to form a causal identification process report. The core driving factor list is pushed to the driving factor priority sorting stage in step three through the module output interface. At the same time, the process report is archived to the log storage module of the causal relationship knowledge base for subsequent knowledge base iteration and optimization. After completing the identification of the current round, the collaborative scheduling unit switches to standby mode to wait for the next round of real-time data input, realizing cyclical work. This module automates the entire process from data processing to inference and calculation to filtering and positioning through unit-based decomposition and collaborative scheduling. Furthermore, through deep integration with the causal relationship knowledge base, it ensures the accuracy of causal inference and avoids the problem of misjudging data correlation as causation in traditional scheduling.

[0026] like Figure 1 As shown, the feasibility verification of the initial plan includes full-process verification across the dimensions of resource adaptability, constraint satisfaction, and cost controllability. Each dimension has clearly defined verification objectives, quantitative indicators, operation procedures, and judgment criteria, forming a closed-loop verification mechanism to ensure that the plan is implementable and free from derivative risks. If all three dimensions pass the verification, the initial plan is deemed feasible and proceeds to step five, the pilot verification stage. If any dimension fails the verification, the system immediately returns to the strategy template library to rematch alternative strategies. For example, if there are no idle AGVs, the system prioritizes the transfer of urgent materials and delays the transfer of non-urgent materials; if energy consumption exceeds the standard, the system optimizes the path to shorten the running time; if there is a shortage of skilled personnel, the system allocates cross-regional skilled personnel for support. After the alternative strategy is generated, the above verification process must be repeated until all three dimensions meet the standards to ensure the feasibility of the plan, as detailed below: Resource compatibility verification aims to ensure that the resources required by the solution are available in real time and that the matching degree meets the standards. The verification scope covers three core resources: equipment resources, personnel resources, and space resources. All of them use the structured real-time data preprocessed in step two and the data from the warehouse resource management system for verification. Equipment resource verification: For equipment such as AGVs, sorting machines, and shelves involved in the solution, verify the matching degree between real-time status, performance parameters and solution requirements. For example, the idle AGVs scheduled must meet the following requirements: battery life ≥ 1.2 times the solution execution time and load capacity covering the weight of the assigned tasks; the sorting machine processing speed ≥ the order material sorting requirements; and the shelves must meet the material storage attributes and the current occupancy rate < 90%. Storage attributes include temperature, humidity, and load-bearing capacity. The operation process is to associate real-time data with equipment IDs, compare equipment parameters with solution requirements one by one, and generate an equipment compatibility list. The judgment standard is that all involved equipment meets the compatibility requirements, with no faults and no performance failures. Otherwise, the resource compatibility is deemed unqualified. Personnel resource verification: For the picking, transfer, and quality inspection personnel involved in the solution, verify the real-time on-duty quantity, skill level and match the solution requirements. For example, fragile item picking tasks require personnel with a skill level of ≥3, and batch material transfer requires the number of on-duty personnel to be ≥ the number required by the solution, and the distance between the personnel's working position and the task area is ≤50 meters to ensure response timeliness. The operation process is to combine personnel terminal data and skill level knowledge base to compare personnel configuration with solution requirements, mark personnel with mismatched skills or too far away. The judgment criteria are sufficient number of personnel, all skill levels meet the requirements, and response distance meets the requirements; otherwise, it is judged as unqualified. Space resource verification: For the space involved in material storage and transfer routes, verify the matching degree between the area occupancy status, access capacity and the plan requirements. For example: the transfer route must be free of congestion, the number of equipment operating simultaneously in the area must be less than or equal to the route capacity, there must be no temporary barriers, the storage area must reserve space greater than or equal to 1.1 times the volume of materials required by the plan, and the temperature and humidity of the area must meet the material property requirements. The operation process is to retrieve real-time data from the warehouse space management system, simulate the space occupancy of the plan execution route, and identify conflict points. The judgment criteria are no space occupancy conflicts and the access capacity and storage space meet the requirements; otherwise, it is judged as unqualified. Constraint satisfaction verification aims to ensure that the solution complies with material properties, order requirements, and safety specifications, thereby mitigating potential risks. Material attribute constraint verification: Strictly compare the material attribute rules in the causal relationship knowledge base to verify whether the solution meets the special requirements of material storage, transfer and picking. For example, the transfer speed of fragile items must be ≤2m / s and the stacking height must be ≤3 layers. The temperature and humidity of cold chain materials during storage and transfer must be maintained at 0-4℃. Perishable materials must be transferred separately and avoid being mixed with other materials. The operation process is to associate the material ID with the attribute rules in the knowledge base, check the material handling parameters in the solution one by one, and verify the temperature and humidity conditions in combination with real-time environmental sensor data. The judgment standard is that there is no violation of the material attribute rules. Otherwise, it is judged as unqualified. Order time limit constraint verification: Taking the order delivery time limit as the core, combined with the execution time of each stage of the solution, the total time is verified to see if it is within the allowable range, while reserving emergency buffer time. For example, for urgent orders, the total execution time of the solution must be ≤ 80% of the delivery time limit, and a 20% buffer time must be reserved to deal with emergencies. For batch orders, the completion time of each sub-task must be ensured not to affect the overall delivery rhythm. The operation process is to build a time prediction model based on historical execution data, calculate the time of each stage of the solution and the total time, and compare it with the order delivery time limit. The judgment standard is that the total time limit is ≤ the delivery time limit, including the buffer time, and there is no risk of timeout in a single stage. Otherwise, it is judged as unqualified. Safety compliance verification: Verify whether the solution complies with warehouse safety operation rules and avoids risks such as equipment collisions, material damage, and personnel safety. For example, the path spacing of AGVs during collaborative operation must be ≥1.5 meters, the weight of material transfer must not exceed 90% of the equipment's rated load, there should be no cross-operation of equipment in the picking area, and the operation process should simulate the execution scenario of the solution to identify safety hazards such as equipment path conflicts, overload, and cross-operation in different areas. The judgment standard is that there are no safety hazards and the solution complies with warehouse safety operation standards; otherwise, it is deemed unqualified. Cost controllability verification aims to ensure that the implementation cost of the solution is within a reasonable range and to avoid additional losses. Three quantitative indicators were set: energy consumption, equipment wear and tear, and labor costs. All of them were verified based on the cost of similar tasks in the same period in history. Energy consumption cost verification: Calculate the total energy consumption required for the implementation of the plan, including AGV operation, temperature and humidity control, etc., and compare it with the average energy consumption of similar tasks in history. The allowable fluctuation range is ≤±5%. The operation process is to calculate the theoretical energy consumption based on the equipment power parameters and execution time, and correct it with real-time energy consumption monitoring data. It is then compared with the benchmark value. The judgment standard is that the energy consumption fluctuation is within the allowable range and there is no abnormal increase in energy consumption. Otherwise, it is judged as unqualified. Equipment wear verification: Assess the degree of wear on the equipment caused by the solution to avoid overuse leading to increased failure risk. For example, the continuous running time of the AGV should not exceed 80% of the rated continuous working time of the equipment, and the load rate of the sorting machine should not exceed 90% of the rated value. The operation process is to combine the equipment operation log and wear model to estimate the equipment wear rate after the implementation of the solution and compare it with the historical benchmark value. The judgment standard is that the wear rate ≤ the historical benchmark value, there is no risk of excessive wear; otherwise, it is judged as unqualified. Labor cost verification: Calculate the required man-hours for the proposed solution and compare them with the average man-hours for similar tasks in the past. The allowable increase is ≤10%, except for reasonable increases due to optimization needs. The operation process is to calculate the man-hours based on the personnel skill level and task complexity, and compare them with the benchmark value. The judgment standard is that the increase in man-hours is within the allowable range and the labor cost is controllable; otherwise, it is judged as unqualified.

[0027] like Figure 1As shown, the precise quantification of causal strength is based on a Bayesian network model, employs a joint tree algorithm, and combines real-time data to calculate the posterior probability of each causal link, thereby achieving precise quantification of causal strength. The specific process is as follows: First, the Bayesian network topology is transformed into an undirected joint tree structure, and the conditional probability tables (CPTs) of each node are integrated into the potential function of the clique node in the joint tree. The preprocessed, discretized real-time data is used as evidence variables and input into the joint tree. Through the message passing mechanism, the potential function between cluster nodes is updated, the posterior probability calculation of all network nodes is completed, and finally the posterior probability of the result node in each causal link is extracted as the causal strength quantification value. The core formula is: the basic formula for causal strength (single-link posterior probability) (Bayesian formula): ; illustrate: The cause node status is indicated by factors such as AGV load rate > 80%. For example, the probability of transportation delay increases by >20%; Let be the likelihood probability, the conditional probability of the cause node occurring when the result node occurs. Let be the prior probability of the result node. The prior probabilities of the cause nodes are all taken from the conditional probability table (CPT) of the causal knowledge base. Multi-node joint posterior probability calculation (core of the joint tree algorithm): the potential function of clique nodes in the joint tree. ; After updating the potential function via message passing, the posterior probability of the node is... ; illustrate: For a clique node in a union tree, The updated potential function is used to eliminate irrelevant variables through marginalization operations to obtain the posterior probability of the target node.

[0028] like Figure 1 As shown, the causal identification module is uniformly verified by the collaborative scheduling unit. If the conditions are met, data flow between units is triggered; otherwise, it returns to the previous unit or triggers exception handling. The core functions are data validity, matching degree threshold, and probability threshold, as detailed below: Coordinated scheduling unit, scene matching unit / data mapping and preprocessing unit: Judgment criteria: The structured real-time dataset meets the integrity standard, the missing rate of core elements is ≤3%, and the data latency meets the real-time requirements, with high-frequency data latency ≤1s and low-frequency data latency ≤3s. Quantification formula: ; Data delay =Current system time - Data timestamp , For the corresponding data type, the delay threshold is used; high frequency... =1s, low frequency =3s; Triggering logic: After receiving the structured real-time dataset from step 2, the cooperative scheduling unit calculates... and At the same time satisfy ≤3% and ≤ If the data is deemed valid, it is distributed to the scene matching unit and the data mapping and preprocessing unit respectively; if If >3%, return to step 2 preprocessing module to complete the data; if > This triggers a data refresh command, causing the corresponding data to be collected again. Scene matching unit, causal link filtering unit, and scene matching validity determination: Judgment criteria: Target scene matching degree ≥ 85%, and the number of activated candidate causal links ≥ 3, ensuring screening redundancy. Quantification formula: Scene matching = (Number of matching features between real-time data features and target scene features / Total number of features in the target scene) × Feature weight The weight of the i-th feature is used, with core features such as order volume and equipment status. =0.8, secondary characteristics such as ambient temperature and humidity =0.2, =1; Triggering logic: The scene matching unit calculates the cosine similarity algorithm. ,when When ≥85% and the number of activated candidate causal links is ≥3, the link list is pushed to the causal link filtering unit; if If the accuracy is less than 85%, the scene matching range is expanded, switching from exact matching to fuzzy matching, and the results are recalculated. If the number of candidate links is less than 3, supplementary links for similar scenarios are called from the causal knowledge base to ensure link redundancy. Data mapping and preprocessing unit, Bayesian network inference unit, and data node mapping validity determination: Judgment criteria: Data-node mapping accuracy ≥ 98%, no mapping data exceeding the range of node state values, quantification formula: Mapping accuracy = (Number of valid mapped data entries / Total number of mapped data entries) × 100% ≥ 98%; Triggering logic: After the data mapping unit completes the mapping between real-time data and Bayesian network nodes, the collaborative scheduling unit verifies... And the reasonableness of mapping, such as AGV load rate data should not be mapped to temperature and humidity nodes, when When ≥98% accuracy and no unreasonable mappings are found, push the data, node, and state association mapping table to the inference unit; if... If the accuracy rate is less than 98%, the mapping process will be re-executed to remove invalid mapping data; if unreasonable mappings exist, the abnormal data will be marked and the process will be returned to the mapping unit for correction. Bayesian network inference unit, causal link filtering unit, causal strength validity determination: Judgment criteria: The posterior probability (causal strength) of the causal link derived from the inference is calculated with a completion rate of 100%, and there are no failed inference links. Quantification formula: Reasoning completion rate = (Number of links with successfully calculated posterior probabilities / Total number of candidate causal links) × 100% = 100%; Single-link causal strength (posterior probability) ; Bayes' theorem For the state of the cause node, The state of the result node; Triggering logic: The inference unit calculates each link using the joint tree algorithm. Afterwards, the coordinated scheduling unit verifies... ,when When the probability is 100%, push the causal link and posterior probability lookup table to the filtering unit; if... If the inference success rate is less than 100%, for links that fail inference, reload the Bayesian network topology and conditional probability table (CPT) and perform inference again; if multiple inference failures occur, mark the link as abnormal and remove it. Causal link filtering unit, core driving factor location and ranking unit, effective link filtering and judgment: Judgment criteria: After triple screening, the number of valid causal links is ≥1, and each valid link meets the three requirements of posterior probability ≥0.3, trigger condition matching, and direct link association. Quantification formula and supplementary conditions: Intensity threshold screening: ≥0.3 corresponds to a level in causal strength; values ​​below this level are discarded. Trigger condition matching: Trigger condition satisfaction T=1, T=1 means that the preset trigger condition is fully met, T=0 means that it is not met. For example, the moisture-sensitive material humidity exceeding the standard link must simultaneously meet the conditions of humidity >60% and material being a moisture-sensitive category. If both conditions are met, then T=1. Link validity: Determined by the d-separation criterion; if a third-party node Z exists, such that... ( and In a given If the conditions are independent, then it is an indirect link and should be eliminated; Triggering logic: After the screening unit completes the triple screening, the collaborative scheduling unit verifies the number of valid links and the compliance of the screening. When the number of valid links is ≥1, the list is pushed to the positioning and sorting unit; if the number of valid links is 0, it returns to the scenario matching unit, reactivates the candidate links of the same scenario, and repeats the reasoning and screening process. Core driving factor positioning and ranking unit, collaborative scheduling unit, core factor output judgment; Judgment criteria: The core driving factors contributing ≥30% to the positioning, and the number of factors ≤3, ensuring focus on the core contradictions. Quantification formula: For the i-th cause node, The cause node does not exhibit a state. This represents the total posterior probability of the result node. Triggering logic: The location and sorting unit calculates the contribution of each cause node. After sorting, the collaborative scheduling unit verifies and selects... ≥30% and the top 3 nodes are considered the core driving factors; if If ≥30% of the nodes are 0, return to the causal link filtering unit, lower the strength threshold to 0.2 (only a temporary adjustment), and re-filter; if the number of nodes is >3, only the top 3 and... The highest level ensures focused decision-making.

[0029] A warehouse material scheduling system for intelligent information technology, such as Figure 2 As shown, the warehouse material scheduling method for intelligent information technology described in any of the above-mentioned embodiments, wherein the warehouse material scheduling system for intelligent information technology includes: The processor, upon system startup or restart after knowledge base iteration, triggers an initialization process. Its core functions are building a causal reasoning knowledge base and configuring terminal communication to ensure smooth data and instruction flow in subsequent processes. By verifying the integrity of the knowledge base and the connectivity of the terminal link, it ensures the initialization achieves its goals and avoids potential operational risks. Core control logic: Based on historical data and expert experience, the K2 algorithm is used to learn the causal dependencies of elements, construct a Bayesian network topology, generate a conditional probability table, preset causal strength level mapping rules, and simultaneously input scene feature rules and adaptation strategy templates. Configure the terminal communication protocol, enter the information of the registered terminal and complete the communication link test, and preset the data collection frequency and data cleaning and standardization rules; Judgment and Flow: Verify the integrity of the knowledge base and the connectivity of the terminal link = 100%. If the standard is met, the system enters standby mode and waits for real-time data input. If the standard is not met, return to reconstruct the knowledge base or check the terminal link. The standardized data acquisition control unit triggers data acquisition when the acquisition timer reaches a preset cycle in standby mode. Data quality is improved through cleaning, completion, and standardization operations, providing reliable input for causal identification. Qualified data is filtered according to a triple quality verification rule; invalid data triggers an alarm after correction, ensuring accuracy in downstream processes. Core control logic: Collect raw data from multiple sources according to terminal type, automatically add timestamps, device IDs, and collection location metadata, and then perform cleaning and completion after forming data packets; Perform standardization processing, unify data types, units, and formats, and then pass triple quality checks: Core element missing rate = (Number of missing items / Total number of items) × 100% ≤ 3%; Data delay = Current time - Data timestamp, high frequency ≤ 1s, low frequency ≤ 3s; Randomly compared 10% of the data, the error rate is ≤1%; Judgment and Flow: If all three checks are passed, the data and knowledge base element node IDs are associated and bound, a structured dataset is generated and pushed to the cause-effect identification stage. If any one check is not passed, data correction is triggered. If the data still fails after two retries, an alarm is triggered and the downstream process is suspended. The causal identification and driver localization module receives a qualified structured dataset, triggers the causal identification process, quantifies causal strength through scene matching and Bayesian inference, accurately locates core driving factors, and filters effective links and factors according to multiple judgment criteria to ensure that the output results match actual scheduling requirements. If no effective link is found, the process is backtracked and re-executed. Core control logic: Scene matching: Calculate scene matching degree (Number of feature-fitting terms / Total number of feature terms) × Weight ,core =0.8, minor =0.2, If the success rate is ≥85% and there are ≥3 active candidate links, proceed to the next step; otherwise, expand the matching range or supplement the links. Data mapping: Calculating mapping accuracy If (number of valid mappings / total number of mappings) × 100% ≥ 98% and there are no unreasonable mappings, map the data to the Bayesian network node state; otherwise, re-execute the mapping. Inference computation: The joint tree algorithm is used to input mapping data and calculate the inference completion rate. (Number of successful inference links / Total number of links) × 100% = 100%, calculated using Bayes' theorem. Generate a causal strength comparison table. Then reload the network parameters and try again; Link screening and factor localization: Screening For valid links (≥1) that meet the trigger condition satisfaction T=1 and are verified by the d-separation criterion, calculate the factor contribution. Select ≥30% and the top 3 core factors, ranked in order of delivery time > material safety > operating cost, and within the same level, according to Detailing; Judgment and Flow: Generate a list of core driving factors and push it to the solution generation stage. If there is no effective link, backtrack to the scenario matching stage, adjust the parameters and re-execute. The solution generation and verification unit receives a list of core driving factors, triggers the solution generation process, generates a targeted scheduling solution based on the core driving factors, and ensures the solution is implementable and risk-free through three-dimensional verification. If the verification is successful, it is pushed to the pilot stage; if it fails, it is iteratively optimized. After multiple ineffective attempts, manual intervention is triggered to avoid potential problems in solution implementation. Core control logic: Match the corresponding strategy template, refine the parameters to generate an initial scheduling plan, such as splitting tasks and scheduling idle devices; Triple feasibility verification: Resource adaptation, equipment performance, personnel skills, and space capacity matching solution requirements; Constraints are met, conforming to material attributes, order deadline ≤ 80% of delivery deadline, and safety regulations; Costs are controllable, and energy consumption fluctuates. Equipment wear and tear and labor costs meet the standards; Judgment and Flow: If all dimensions are verified and the standard is met, the solution is pushed to the pilot verification stage. If any standard is not met, an alternative strategy is rematched and the verification is repeated. If it still fails after 3 retries, manual intervention is triggered. Pilot verification and optimization: Upon receiving a qualified initial plan, the pilot process is triggered. The effectiveness of the plan is verified through a small-scale pilot, and the pilot scope is scientifically defined to ensure the representativeness of the results. Key indicators are monitored and the plan is iteratively optimized until the standards are met before output. This mitigates the potential risks of full-domain deployment. Core control logic: Define the scope of the pilot program. Simultaneously, the pilot program must meet the following conditions: the impact of the pilot program on non-pilot areas is ≤5% and the cost accounts for ≤20% of the total cost; otherwise, the scope of the pilot program will be adjusted. Deploy pilot programs and monitor key indicators in real time: problem resolution effectiveness ≥80%, no serious derivative impact, and valid causal link verification; Judgment and Flow: If the indicators meet the standards, the optimized solution is output; if they do not meet the standards, the solution parameters are adjusted, such as optimizing the path and adjusting the task ratio, and the pilot test is repeated until the standards are met. A dynamic iteration module is deployed across the entire domain. Once the optimized solution is validated, the entire domain deployment process is triggered, ensuring the optimized solution is implemented nationwide. Simultaneously, real-time monitoring and regular iterations maintain system adaptability. This forms a closed-loop operation mechanism, dynamically adjusting the solution and updating the knowledge base to ensure continuous optimization of scheduling effectiveness. Core control logic: The solution is transformed into terminal-specific instructions for issuance, and the execution status of the instructions is monitored in real time to avoid terminal conflicts. Terminal execution data is continuously collected. If the core driving factors reappear or new factors appear, the solution is automatically traced back to the causal identification stage, and the identification, generation, verification, and pilot processes are repeated. The solution is dynamically adjusted, and the implementation effect data is synchronized to the knowledge base every quarter. The CPT and causal strength level are updated through the maximum likelihood estimation method, and new causal relationships are added. Judgment and Flow: A closed loop of deployment, monitoring, adjustment, and iteration is formed, the system runs continuously and dynamically, and alarms are triggered in real time for abnormal situations, triggering corresponding handling logic.

[0030] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A warehouse material scheduling method for intelligent information systems, characterized in that, The scheduling method mainly includes the following steps: Step 1: Construct a causal relationship knowledge base for warehousing scenarios. The knowledge base is built based on historical scheduling logs and expert experience rules. A Bayesian network model is used to represent the causal strength between each element. The range of core elements is determined by the causal strength, and factors affecting material scheduling are screened. Subsequently, multi-source historical data is collected, including warehouse historical data. A combination of Bayesian network model, data mining, and expert verification is used to mine the causal relationships between elements and quantify the causal strength, thus constructing the knowledge base structure. Step 2: Real-time data acquisition and data preprocessing module simultaneously preprocess the data to provide data support for causal identification. By adopting multi-terminal real-time acquisition, lightweight causal sensing terminals are deployed in the warehouse to clean and standardize the acquired real-time data. Step 3: Real-time causal identification and core driving factor localization. By inputting the preprocessed real-time data into the causal identification module, and based on the constructed Bayesian network model, the causal strength of each candidate causal link is calculated in combination with the real-time data, and the posterior probability of each potential factor is calculated. The factor with the highest posterior probability is determined as the core driving factor. Step 4: Based on the core driving factors and the coping strategies in the causal relationship knowledge base, generate a targeted scheduling plan. By matching the adaptation strategy templates corresponding to the core driving factors from the causal relationship knowledge base, strategy matching and initial plan generation are achieved, and then the feasibility of the initial plan is verified. Step 5: Utilize pilot verification, optimize the solution based on the verification results, and verify the introduction of a causal verification mechanism to avoid scheduling errors caused by misjudgment of causal links and ensure the effectiveness of the solution; Step Six: After the pilot verification is successful, the scheduling scheme will be promoted and deployed across the entire domain, and continuously optimized and dynamically adjusted in combination with real-time data to achieve dynamic iterative updates.

2. The warehouse material scheduling method for intelligent information systems according to claim 1, characterized in that: The data preprocessing module adopts a full-process architecture of data acquisition and docking, cleaning and processing, standardization, association and labeling, and quality verification. The data preprocessing module includes multiple functional units, which work together through an internal high-speed data bus and establish data linkage with the front-end acquisition terminal and the back-end cause-effect identification module.

3. The warehouse material scheduling method for intelligent information systems according to claim 1, characterized in that: The method for calculating causal strength and selecting core driving factors is based on the constructed Bayesian network model. It combines real-time data to complete the accurate quantitative calculation of causal strength, and then uses multi-dimensional screening to locate core driving factors, avoiding misjudgment of data correlation throughout the process. This includes mapping and matching real-time data with Bayesian network nodes, Bayesian network posterior probability inference calculation, multi-condition link screening, core driving factor location and contribution quantification.

4. A warehouse material scheduling method for intelligent information systems according to claim 1, characterized in that: The cause-effect identification module adopts a modular architecture design, which includes multiple functional units.

5. A warehouse material scheduling method for intelligent information systems according to claim 4, characterized in that: The multiple functional units include a scene matching unit, a data mapping and preprocessing unit, a Bayesian network inference unit, a causal link filtering unit, a core driving factor location and sorting unit, and an inter-unit collaborative scheduling unit.

6. A warehouse material scheduling method for intelligent information systems according to claim 1, characterized in that: The feasibility of the initial verification scheme includes full-process verification of resource adaptability, constraint satisfaction, and cost controllability. Each dimension has clear verification objectives, quantitative indicators, operation procedures, and judgment criteria.

7. A warehouse material scheduling method for intelligent information systems according to claim 3, characterized in that: The precise quantification of causal strength is based on a Bayesian network model, employs a joint tree algorithm, and combines real-time data to calculate the posterior probability of each causal link, thereby achieving precise quantification of causal strength.

8. A warehouse material scheduling method for intelligent information systems according to claim 4, characterized in that: The cause-effect identification module is uniformly verified by the collaborative scheduling unit. If the condition is met, data flow between units is triggered; otherwise, it returns to the previous unit or triggers exception handling.

9. A warehouse material scheduling system for intelligent information technology, applied to the warehouse material scheduling method for intelligent information technology as described in any one of claims 1-8, characterized in that: The warehouse material scheduling system for intelligent information technology includes: The processor triggers the initialization process upon the first system startup or after a knowledge base iteration and restart. The standardized data acquisition control unit triggers data acquisition when the acquisition timer reaches a preset period in the system standby state. The causal identification and driver location module receives a qualified structured dataset and triggers the causal identification process. The solution generation and verification unit receives a list of core driving factors and triggers the solution generation process. Pilot verification and optimization, receive qualified initial solutions, and trigger the pilot process; The dynamic iteration module is deployed across the entire domain. Once the optimized solution passes verification, the entire domain deployment process is triggered.

Images