Storage space dynamic management system based on multi-source data fusion and intelligent optimization

Abstract

The invention relates to the technical field of storage space dynamic management, in particular to a storage space dynamic management system based on multi-source data fusion and intelligent optimization. Comprising a real-time data acquisition and fusion module, a dynamic demand prediction and visualization module, a multi-target dynamic optimization decision module, a scheme decision and instruction execution module and a closed-loop feedback and model updating module, the dynamic demand prediction and visualization module predicts demands based on the data set and generates a dynamic demand thermodynamic diagram, the multi-target dynamic optimization decision module constructs a multi-target goods allocation model and adaptively adjusts the weight to solve an optimal scheme set, and the scheme decision and instruction execution module selects a scheme and analyzes the scheme into an operation instruction; and the closed-loop feedback and model updating module forms a closed loop and updates the prediction model. According to the invention, dynamic intelligent optimization of the storage space is realized, the operation efficiency is improved, the management cost is reduced, and modern storage management requirements are met.

Description

Technical Field

[0001] This invention relates to the field of dynamic warehouse space management technology, specifically a dynamic warehouse space management system based on multi-source data fusion and intelligent optimization. Background Technology

[0002] With the rapid development of the logistics industry, warehousing, as a core hub of the supply chain, directly impacts the operational efficiency of the entire logistics chain through the efficiency, accuracy, and flexibility of its space management. Currently available warehouse space management systems generally suffer from numerous unresolved issues, failing to meet the demands of modern dynamic warehouse operations.

[0003] Existing systems rely heavily on simple statistical models or manual experience for demand forecasting, making it difficult to deeply analyze the cyclical trends and product category relationships in historical warehouse operations. They also fail to fully integrate real-time status characteristics with known pre-booked order data, resulting in significant discrepancies between forecasts and actual inbound / outbound demand. Furthermore, the lack of intuitive visualization methods prevents managers from quickly grasping the spatial distribution patterns of future warehouse operations and hinders advance planning of operational resources.

[0004] Traditional warehouse location allocation often focuses on a single objective, such as maximizing space utilization or simply reducing picking costs, neglecting the core requirement of adapting goods to the storage environment. This can easily lead to goods being damaged due to unsuitable conditions. Furthermore, the optimization process fails to adjust objective priorities based on dynamic needs and does not adequately address path optimization in high-frequency operation areas, resulting in an imbalance in warehouse resource allocation and a difficulty in balancing operational efficiency with goods preservation.

[0005] The actual status data after command execution cannot be fed back to the decision-making module in a timely manner, and the prediction model and optimization algorithm cannot be dynamically updated according to actual operational deviations, resulting in poor system adaptability. Faced with dynamic scenarios such as changes in cargo flow patterns and adjustments to operational strategies, it is difficult to quickly adjust the optimization direction, and the accuracy of decision-making is prone to decline after long-term operation.

[0006] Some systems have low levels of automation, requiring manual parsing of storage location adjustment plans into operational instructions, which is prone to human error. Furthermore, the lack of real-time monitoring and safety verification of the instruction execution process makes it impossible to promptly detect abnormalities such as equipment conflicts and exceeding operational limits, potentially leading to operational interruptions, cargo damage, or equipment failure, thus affecting the stability and safety of warehousing operations.

[0007] The existence of these problems leads to low operational efficiency, serious resource waste, and high risks in cargo storage of existing warehouse space management systems, making it difficult to adapt to the high-density, dynamic, and intelligent management needs of modern warehousing. Therefore, there is an urgent need for a dynamic warehouse space management system based on multi-source data fusion and intelligent optimization to solve the above problems. Summary of the Invention

[0008] The purpose of this invention is to provide a dynamic warehouse space management system based on multi-source data fusion and intelligent optimization to solve the problems mentioned in the background art.

[0009] To achieve the above objectives, the present invention provides the following technical solution:

[0010] A dynamic warehouse space management system based on multi-source data fusion and intelligent optimization includes:

[0011] Real-time data acquisition and fusion module: Through a sensor network deployed in the warehouse area, it collects real-time data on the identity of goods, occupancy status, and environmental monitoring data of each storage location; it then performs spatiotemporal alignment and fusion processing with historical order data from the warehouse management system to generate a comprehensive status dataset containing historical and real-time information, with storage locations as the basic unit.

[0012] Dynamic Demand Forecasting and Visualization Module: Based on a comprehensive status dataset, the module uses a forecasting model to predict the inbound and outbound demand of various categories of goods within a preset future time period; based on the forecast results, it generates a dynamic demand heatmap that reflects the probability of future warehousing operation space distribution.

[0013] Multi-objective dynamic optimization decision module: Constructs a storage location allocation model with the optimization objectives of maximizing warehouse space utilization, minimizing expected picking path costs, and minimizing cargo storage environment mismatch;

[0014] Based on the dynamic demand heatmap, the weight coefficients of each objective in the multi-objective optimization model are adaptively adjusted. Among them, the high-frequency operation areas marked in the heatmap are given higher weights for the objective of minimizing the expected picking path cost.

[0015] A multi-objective optimization algorithm was used to solve the model after weight adjustment, resulting in a set of Pareto optimal storage location adjustment schemes;

[0016] Solution Decision and Instruction Execution Module: Based on preset decision rules, select an optimal execution plan from the Pareto optimal storage location adjustment plan set, and parse the optimal execution plan into a specific sequence of operation instructions that can drive the storage equipment to execute.

[0017] Closed-loop feedback and model update module: Executes the sequence of operation instructions and confirms changes in the status of the storage location in real time; feeds back the new storage status data generated after execution to the comprehensive status dataset to start the next optimization cycle; at the same time, updates the parameters of the prediction model based on the difference between the actual inbound and outbound data and the predicted data.

[0018] As a preferred option, the real-time data acquisition and fusion module includes:

[0019] Receive raw data streams uploaded by a sensor network consisting of RFID readers, QR code scanners and weight sensors deployed at each physical storage location. The raw data streams include at least cargo identification codes, storage location occupancy indicators and ambient temperature and humidity readings.

[0020] The raw data stream is cleaned and formatted to remove outliers and null values, and the timestamps and data formats are standardized to generate a standardized real-time data table.

[0021] Based on the electronic map of the storage area and the location coding rules, each record in the standardized real-time data table is associated with its corresponding unique location space coordinates to construct a real-time status snapshot of the location with spatiotemporal labels.

[0022] Extract order execution logs from the order database of the warehouse management system within a preset historical period. The order execution logs must include at least the product category, inbound / outbound time, and coordinates of the source and target storage locations.

[0023] Establish a spatiotemporal index structure with the spatial coordinates of the storage location as the key, associate and align the real-time status snapshots of the storage locations with the order execution logs, and generate the historical operation frequency, associated category sequence and environmental preference characteristics of each storage location;

[0024] Based on the results of association and alignment, the real-time physical status information of each storage location is integrated with its corresponding historical behavioral feature information to generate a comprehensive storage location status record that includes static attributes, dynamic status and historical statistical features.

[0025] Collect all the comprehensive status records of the storage locations to generate a comprehensive status dataset with the storage location as the basic unit. The comprehensive status dataset serves as the input for the dynamic demand forecasting and visualization module.

[0026] As a preferred solution, the dynamic demand forecasting and visualization module specifically includes:

[0027] The data preparation and feature extraction unit is used to extract historical order sequences, cargo turnover frequency and category co-occurrence relationships associated with each storage location from the comprehensive status dataset, and divide the historical order sequences into continuous time segments based on a preset time window to form a time-series feature sample set based on storage location.

[0028] The spatiotemporal demand forecasting unit is used to input the time series feature sample set into the trained ensemble forecasting model. The ensemble forecasting model is used to analyze the periodicity and trend of the historical operation sequence of each storage location and integrate recent real-time status features to predict the probability value of inbound and outbound operations of each storage location and the possible related set of goods categories within a preset future time period.

[0029] The operation intensity and range mapping unit is used to calibrate and overlay the predicted probability values ​​based on the probability values ​​of inbound and outbound operations at each storage location and the set of goods categories that may be associated with them, combined with the known future reservation order data of the warehouse management system, to generate an initial operation intensity distribution map that reflects the expected operation intensity of each storage location within a preset future time period.

[0030] The dynamic heatmap generation unit receives the initial operation intensity distribution map and renders it according to a preset spatial smoothing algorithm and threshold segmentation rules. It maps different levels of operation intensity to different visual colors to generate a dynamic demand heatmap that covers the entire storage area and reflects the probability of future storage operation space distribution.

[0031] As a preferred option, the multi-objective dynamic optimization decision module includes the following for constructing a cargo location allocation model:

[0032] Based on the comprehensive status dataset, the current occupancy status, cargo size attributes, and cargo capacity of each storage location are extracted. Using the storage space occupancy rate and cargo stacking compactness as metrics, a target function for warehouse space utilization is constructed.

[0033] Based on the dynamic demand heatmap, the expected operational intensity of each storage location and the correlation between goods are extracted. The expected length of the picking equipment movement path is used as the measurement index to construct the expected picking path cost objective function.

[0034] Based on the environmental monitoring data in the comprehensive status dataset and the environmental preference characteristics in the comprehensive status records of cargo locations, the difference between the environmental conditions required for each cargo and the actual environmental conditions of the cargo location is calculated, and a cargo storage environment mismatch objective function is constructed.

[0035] Using the storage location allocation scheme as the decision variable and the storage location capacity limit, cargo size matching and storage area division as constraints, the objective functions of storage space utilization, expected picking path cost and cargo storage environment mismatch are integrated into a multi-objective optimization problem to form a storage location allocation model.

[0036] As a preferred approach, the weight coefficients of each objective in the multi-objective optimization model are adaptively adjusted based on the dynamic demand heatmap, including:

[0037] Receive dynamic demand heatmaps, analyze the spatial coordinates of the storage location corresponding to each pixel in the heatmap and the corresponding operation intensity probability value; based on the preset operation intensity threshold, identify the set of storage locations with operation intensity higher than the preset threshold as high-frequency operation areas;

[0038] Traverse each storage location in the high-frequency operation area and read its operation intensity probability value; based on the operation intensity probability value and the actual turnover rate of the storage location in historical orders, calculate the local path optimization importance factor corresponding to the storage location; normalize the local path optimization importance factors of all high-frequency operation storage locations to generate the overall path optimization weight enhancement benchmark value of the high-frequency area.

[0039] Obtain current warehousing operation strategy instructions, which include at least the priority level of space utilization, operation efficiency, or cargo preservation; based on the priority level, strategically adjust the baseline value of the overall route optimization weight enhancement to obtain the adjusted route cost target weight enhancement value.

[0040] Using the revised path cost target weight enhancement value as the core adjustment parameter, and based on the preset weight allocation rules, the weights of the warehouse space utilization objective function and the cargo storage environment mismatch objective function are adjusted in reverse coordination to ensure that the sum of all weight coefficients remains constant, and finally generate a set of adaptive multi-objective weight coefficients that match the current dynamic needs and warehousing strategy.

[0041] As a preferred approach, a multi-objective optimization algorithm is used to solve the model after weight adjustment, resulting in a set of Pareto optimal storage location adjustment schemes; specifically including:

[0042] Based on the adaptive multi-objective weight coefficient set and the storage location allocation model, the optimization objective function and constraints are determined, and an initial population containing multiple storage location adjustment schemes is constructed according to the actual layout of the current storage locations and the distribution status of goods.

[0043] Based on the objective functions of warehouse space utilization, expected picking route cost, and cargo storage environment mismatch, and using an adaptive multi-objective weight coefficient set for weighted aggregation, the fitness evaluation value corresponding to each individual storage location adjustment plan is calculated.

[0044] Based on fitness evaluation values, individuals in the initial population with different storage locations were ranked by non-dominance and crowding, and individuals with high Pareto non-dominance levels and low distribution density were selected to form the first generation elite set.

[0045] Genetic operations are performed on the individual storage location adjustment schemes in the first-generation elite set. The genetic operations include crossover operators based on storage operation logic and mutation operators based on storage location state constraints to generate new storage location adjustment scheme individuals and form a progeny population.

[0046] The first generation of elites is merged with the offspring population, and the merged population is subjected to non-dominated sorting and crowding calculation again to select a new generation of elites, thus completing one iteration of optimization.

[0047] Repeat the iterative process of population merging, sorting, selection and genetic operations described above until the preset iteration termination condition is met, and obtain the final generation of elites.

[0048] The final generation of elite collections contains all individual storage location adjustment schemes, which are output as a set of Pareto optimal storage location adjustment schemes.

[0049] As a preferred option, the scheme decision and instruction execution module selects an optimal execution scheme from the Pareto optimal storage location adjustment scheme set according to preset decision rules, and parses the optimal execution scheme into a specific sequence of operation instructions that can drive the storage equipment to execute, specifically including:

[0050] The scheme evaluation and ranking unit is used to receive the Pareto optimal storage location adjustment scheme set and to give a comprehensive score to each storage location adjustment scheme based on multiple preset evaluation dimensions. The evaluation dimensions include at least the scheme implementation cost, expected target benefits and the degree of interference with the current on-the-go operations, and generate a sorted list of schemes with comprehensive scores.

[0051] The strategy matching and filtering unit is used to receive the solution ranking list and obtain the real-time operation status and strategy preference configuration of the current warehousing system; it matches the attributes of each solution in the solution ranking list with the real-time operation status and strategy preference configuration, filters out solutions that do not meet the current operation constraints and strategy preferences, and generates a filtered subset of candidate solutions.

[0052] The final solution selection unit is used to conduct a secondary evaluation of the selected subset of candidate solutions. The secondary evaluation is based on a weighted calculation of the urgency of the solution execution and the resource readiness status, and the storage location adjustment solution with the highest comprehensive score is selected from the subset of candidate solutions as the optimal execution solution.

[0053] The instruction decomposition and serialization unit is used to parse the optimal execution plan and extract each cargo movement task to be adjusted. The cargo movement task includes at least cargo identifier, source location coordinates, target location coordinates and cargo handling priority. Based on the preset equipment scheduling rules and path planning algorithm, each cargo movement task is decomposed and mapped into one or more atomic operation steps that can be executed by a specific warehousing equipment, generating a set of atomic operation steps.

[0054] The instruction sequence generation and verification unit receives a set of atomic operation steps and, based on the logical dependencies between tasks and the results of device resource conflict detection, sorts and schedules the steps in the set of atomic operation steps to form a conflict-free, time-sequential sequence of specific operation instructions; it then performs logical correctness verification and security boundary checks on the specific operation instruction sequence to generate the final executable instruction sequence.

[0055] The instruction distribution and status monitoring unit is used to distribute the executable instruction sequence to the corresponding warehouse equipment controller and monitor the execution status of each warehouse equipment in real time; it collects and summarizes the status feedback of execution start, execution completion or execution exception, and generates an instruction execution status report.

[0056] As a preferred solution, the closed-loop feedback and model update module includes:

[0057] The instruction execution status monitoring unit is used to receive the executable instruction sequence output by the scheme decision and instruction execution module, drive the warehousing equipment to execute the executable instruction sequence, collect the execution feedback signal of the warehousing equipment in real time, and generate an instruction execution status report containing the execution status of each instruction and abnormal events.

[0058] The location status change confirmation unit is used to receive the instruction execution status report, parse the goods movement results corresponding to the completed instructions, verify the actual status of goods location changes based on the real-time physical data of the location collected by the sensor network in the storage area, and generate a location status change confirmation record that reflects the changes in the distribution and occupancy status of goods in each location.

[0059] The warehouse status data feedback unit is used to receive the confirmation records of changes in the status of the storage location, associate the goods location, storage location occupancy status and environmental data in the confirmation records with the timestamp, format them into standard data entries, and add them to the comprehensive status dataset in real time to generate an updated comprehensive status dataset to trigger the start of the next optimization cycle.

[0060] The prediction error calculation unit is used to extract the actual inbound and outbound operation logs within a preset time period from the updated comprehensive status dataset, and to perform spatiotemporal alignment and comparison with the prediction results of the corresponding time period generated by the dynamic demand prediction and visualization module. It calculates the difference between the predicted and actual values ​​of the operation intensity and cargo category of each storage location, and generates a set of prediction error indicators.

[0061] The model parameter adjustment unit receives a set of prediction error indicators and, based on a preset machine learning update algorithm, dynamically adjusts the weight parameters and structural hyperparameters within the prediction model according to the values ​​and distribution characteristics of each indicator in the prediction error indicator set, generating parameter update instructions.

[0062] The incremental update unit for the prediction model is used to receive parameter update instructions, apply the parameter update instructions to the integrated prediction model in the dynamic demand prediction and visualization module, complete the hot update of parameters without interrupting the online service of the model, and generate the updated integrated prediction model for demand prediction in subsequent optimization cycles.

[0063] As can be seen from the technical solution provided by the present invention above, the present invention provides a dynamic warehouse space management system based on multi-source data fusion and intelligent optimization, which has the following beneficial effects:

[0064] Through the real-time data acquisition and fusion module, real-time data from sensor networks and historical order data from the warehouse management system are integrated. After cleaning, formatting, spatiotemporal alignment, and deep fusion, a comprehensive status dataset containing static attributes, dynamic states, and historical statistical features is generated. This effectively solves problems such as heterogeneous data distortion from multiple sources, ensuring the integrity, consistency, and timeliness of the data. It provides a high-quality data foundation for subsequent demand forecasting, optimization, and decision-making, thereby improving the scientific nature and accuracy of decision-making.

[0065] The multi-objective dynamic optimization decision module constructs a multi-dimensional optimization objective function, adaptively adjusts the objective weights by combining a dynamic demand heatmap, and obtains the Pareto optimal solution set through a multi-objective optimization algorithm. The solution maximizes the utilization of warehouse space while minimizing the expected picking path cost and the mismatch between the goods storage environment. In particular, it strengthens path optimization for high-frequency operation areas, significantly shortens picking time, reduces resource consumption, and ensures the safety of goods storage, thereby achieving a dual improvement in warehouse operation efficiency and effectiveness.

[0066] The closed-loop feedback and model update module form a complete closed loop for decision execution feedback optimization. It feeds back the warehouse status data after execution to the comprehensive status dataset in real time, triggering the next optimization cycle. At the same time, based on the difference between the predicted and actual data, it dynamically updates the prediction model parameters, so that the model can continuously adapt to changes in warehouse demand. The system can achieve self-optimization and iterative upgrades without manual intervention, which significantly enhances the system's adaptability to complex warehouse environments and dynamic demands, and maintains long-term stable and efficient operation.

[0067] The solution decision-making and instruction execution module automatically parses the optimal solution into a standardized sequence of operation instructions, realizing an automated flow from decision-making to execution and reducing manual operation and human error. Real-time monitoring logic verification and security checks during instruction execution effectively prevent safety accidents such as equipment damage and cargo breakage. At the same time, the intelligent operation of the system reduces reliance on professional management personnel, reduces the time cost of troubleshooting and maintenance, and comprehensively reduces the labor costs and operational risks of warehouse management. Attached Figure Description

[0068] Figure 1This is a schematic diagram of the structure of a dynamic warehouse space management system based on multi-source data fusion and intelligent optimization according to the present invention. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0070] To better understand the above technical solutions, the following will provide a detailed description of the technical solutions in conjunction with the accompanying drawings and specific embodiments.

[0071] like Figure 1 As shown, this embodiment of the invention provides a dynamic warehouse space management system based on multi-source data fusion and intelligent optimization, comprising:

[0072] Real-time data acquisition and fusion module: Through a sensor network deployed in the warehouse area, it collects real-time data on the identity of goods, occupancy status, and environmental monitoring data of each storage location; it then performs spatiotemporal alignment and fusion processing with historical order data from the warehouse management system to generate a comprehensive status dataset containing historical and real-time information, with storage locations as the basic unit.

[0073] Dynamic Demand Forecasting and Visualization Module: Based on a comprehensive status dataset, the module uses a forecasting model to predict the inbound and outbound demand of various categories of goods within a preset future time period; based on the forecast results, it generates a dynamic demand heatmap that reflects the probability of future warehousing operation space distribution.

[0074] Multi-objective dynamic optimization decision module: Constructs a storage location allocation model with the optimization objectives of maximizing warehouse space utilization, minimizing expected picking path costs, and minimizing cargo storage environment mismatch;

[0075] Based on the dynamic demand heatmap, the weight coefficients of each objective in the multi-objective optimization model are adaptively adjusted. Among them, the high-frequency operation areas marked in the heatmap are given higher weights for the objective of minimizing the expected picking path cost.

[0076] A multi-objective optimization algorithm was used to solve the model after weight adjustment, resulting in a set of Pareto optimal storage location adjustment schemes;

[0077] Solution Decision and Instruction Execution Module: Based on preset decision rules, select an optimal execution plan from the Pareto optimal storage location adjustment plan set, and parse the optimal execution plan into a specific sequence of operation instructions that can drive the storage equipment to execute.

[0078] Closed-loop feedback and model update module: Executes the sequence of operation instructions and confirms changes in the status of the storage location in real time; feeds back the new storage status data generated after execution to the comprehensive status dataset to start the next optimization cycle; at the same time, updates the parameters of the prediction model based on the difference between the actual inbound and outbound data and the predicted data.

[0079] In this embodiment, the real-time data acquisition and fusion module is the core data foundation of the dynamic warehouse space management system based on multi-source data fusion and intelligent optimization. It captures real-time data through a sensor network deployed in the warehouse area, and performs multi-dimensional processing and deep fusion with historical order data of the warehouse management system to generate a comprehensive status dataset with the storage location as the basic unit. This provides comprehensive and accurate data support with spatiotemporal characteristics for subsequent modules such as dynamic demand forecasting and intelligent optimization decision-making, ensuring the accurate operation and intelligent decision-making of the entire warehouse management system.

[0080] The real-time data acquisition and fusion module is primarily responsible for collecting real-time cargo identification data, occupancy status data, and environmental monitoring data from each physical storage location. It cleans and formats the collected raw data, assigns spatiotemporal tags to the data, and extracts order execution logs from the warehouse management system within a preset historical period. By constructing a spatiotemporal index structure, it achieves correlation and alignment between real-time and historical data, fusing the real-time physical status and historical behavioral characteristics of the storage locations. Ultimately, it generates a comprehensive status dataset containing static attributes, dynamic states, and historical statistical features, providing high-quality data input for subsequent demand forecasting and optimization decisions, ensuring data integrity, consistency, and timeliness. The real-time data acquisition and fusion module includes:

[0081] Raw data receiving and preprocessing unit:

[0082] Data reception: Real-time reception of raw data streams uploaded by a sensor network consisting of RFID readers, QR code scanning devices and weight sensors deployed in each physical storage location. The raw data streams contain key information such as cargo identification codes, storage location occupancy marks and environmental temperature and humidity readings, enabling comprehensive capture of real-time data related to storage locations.

[0083] Data cleaning: The received raw data stream is processed to remove outliers and null values. By using preset data validity judgment rules, invalid data caused by factors such as equipment failure and signal interference is identified and filtered to ensure data accuracy.

[0084] Data formatting: unify the timestamp format and data record format of the raw data stream, convert heterogeneous data uploaded from different devices into a standardized data format that the system can recognize, and generate standardized real-time data tables, laying the foundation for subsequent data processing and integration;

[0085] Cargo location space-time tag association unit:

[0086] Based on the electronic map and location coding rules of the storage area, each record in the standardized real-time data table is associated with its corresponding unique location spatial coordinates. Through the mapping relationship between location codes and spatial coordinates, each real-time data is given a clear spatial identifier, and a real-time status snapshot of the location with spatiotemporal labels is constructed, so that the data can be accurately mapped to the specific physical location and realize the spatial positioning of the data.

[0087] Historical order data extraction unit:

[0088] The order execution logs within a preset historical period are extracted from the order database of the warehouse management system. The order execution logs contain key information such as goods category, inbound and outbound time, source location and target location coordinates, etc., comprehensively collecting historical data related to location operations, and providing data support for analyzing the historical behavioral characteristics of the locations.

[0089] Spatiotemporal index construction and data alignment unit:

[0090] Spatiotemporal index construction: Establish a spatiotemporal index structure with the spatial coordinates of the cargo location as the key. This index structure can quickly associate the spatial location of the cargo location with the corresponding real-time data and historical data, thereby improving the efficiency of data retrieval and association.

[0091] Data association and alignment: Based on the constructed spatiotemporal index structure, the real-time status snapshot of the storage location is associated with the order execution log, and precise alignment is performed according to the time and space dimensions. The spatiotemporal alignment algorithm is used to match real-time data with historical data in time and space, generating the historical operation frequency, associated category sequence and environmental preference characteristics of each storage location, and mining the historical behavior patterns of the storage location;

[0092] Multi-dimensional information fusion unit:

[0093] Based on the spatiotemporal index association and alignment results, the real-time physical status information of each storage location is integrated with its corresponding historical behavioral feature information. The real-time physical status information includes cargo identification code, storage location occupancy mark, environmental temperature and humidity readings, etc., while the historical behavioral feature information includes historical operation frequency, associated category sequence, environmental preference features, etc. The multi-dimensional data is integrated through information fusion algorithm to generate a comprehensive storage location status record that includes static attributes, dynamic status and historical statistical features, thereby achieving deep data fusion.

[0094] Comprehensive state dataset generation unit:

[0095] Collect all the comprehensive status records of the storage locations, organize them according to a unified data format and storage specifications, and generate a comprehensive status dataset with storage locations as the basic unit. This dataset contains complete information about all storage locations and serves as input data for the dynamic demand forecasting and visualization module, providing comprehensive data support for the operation of subsequent modules.

[0096] Furthermore, the multi-source data standardization processing technology is based on data cleaning and format unification theory. For heterogeneous raw data uploaded from different sensor devices, it identifies and removes invalid data through outlier detection algorithms, unifies the time signature of data through timestamp synchronization algorithms, and converts data with different structures into a standardized format using data format conversion rules. This technology can eliminate noise and interference in the data, solve the problem of heterogeneous multi-source data, ensure the accuracy and consistency of input data, and provide reliable basic data for subsequent data association and fusion.

[0097] The spatiotemporal indexing and association technology is based on spatial indexing and time alignment theory. It constructs an index key with the spatial coordinates of the cargo location as the core, and associates the spatial location information of the cargo location with real-time data and historical data. Through spatial indexing, it can quickly locate all data corresponding to a specific cargo location, and use time alignment algorithms to match real-time data with historical order data in the time dimension, so as to achieve accurate spatiotemporal data association. This technology can significantly improve the efficiency of data retrieval and association, ensure the effective connection between real-time data and historical data, and provide technical support for mining the spatiotemporal characteristics of cargo locations.

[0098] Multi-dimensional information fusion technology, based on information fusion theory, uses a weighted fusion algorithm to integrate real-time physical status information and historical behavioral characteristic information of cargo locations. By assigning reasonable weight coefficients to information in different dimensions, which are determined according to the importance and reliability of the data, multi-dimensional data is mapped to a unified feature space, generating feature records that comprehensively reflect the status of cargo locations. This technology can fully explore the potential correlations between data in various dimensions, integrate the advantages of data, improve the value density of data, and provide high-quality comprehensive data for subsequent demand forecasting and optimization decisions.

[0099] In this embodiment, the dynamic demand forecasting and visualization module is the core of demand perception and visualization in the dynamic warehouse space management system based on multi-source data fusion and intelligent optimization. It receives the comprehensive status dataset output by the real-time data acquisition and fusion module, and generates a dynamic demand heat map reflecting the probability of future warehouse operation space distribution through a series of operations such as feature extraction, time series analysis, demand forecast intensity calibration, and visualization rendering. This provides the multi-objective dynamic optimization decision module with accurate demand guidance and intuitive spatial distribution basis, and builds a key bridge between data support and intelligent decision-making.

[0100] The dynamic demand forecasting and visualization module is primarily responsible for extracting time-series and correlation features related to storage locations from the comprehensive status dataset, constructing a time-series feature sample set, analyzing historical operational patterns and real-time status features using a trained ensemble prediction model, predicting the probability of inbound and outbound operations for each storage location and the associated product categories within a preset future time period, calibrating and overlaying the prediction results with known pre-booked order data, generating an initial operation intensity distribution map, and then mapping the operation intensity to visual colors through spatial smoothing and threshold segmentation, ultimately generating a dynamic demand heatmap covering the entire warehousing area. This visually presents the spatial distribution probability of future warehousing operations, providing accurate demand references for optimizing storage location allocation. The dynamic demand forecasting and visualization module specifically includes:

[0101] Data preparation and feature extraction unit:

[0102] Key information associated with each storage location is comprehensively extracted from the integrated status dataset, including the turnover frequency of goods in historical order sequences and the co-occurrence relationship of product categories. The historical order sequence records the past inbound and outbound operation trajectory of the storage location. The turnover frequency reflects the activity level of the storage location, and the co-occurrence relationship of product categories reflects the association pattern of different product categories in the same or related storage locations. Based on a preset time window, the historical order sequence is divided into continuous time-series segments. Each time-series segment contains the complete operation record of the storage location within that time window. This division method preserves the temporal characteristics and local patterns of the order sequence, ultimately forming a temporal feature sample set based on storage location, providing structured input data for subsequent demand forecasting.

[0103] Spatiotemporal demand forecasting unit:

[0104] The time-series feature sample set is input into the trained ensemble prediction model. The ensemble prediction model integrates the advantages of multiple prediction algorithms, enabling in-depth analysis of the periodicity and trend of historical operation sequences of each storage location. Periodicity reflects the recurring operation patterns of storage locations within a specific time period, while trend reflects the overall direction of the intensity of storage location operations over time. Simultaneously, the model incorporates recent real-time status features, including immediate information such as the current occupancy status and environmental conditions of the storage location. By comprehensively analyzing historical patterns and real-time status, the model accurately predicts the probability of inbound and outbound operations occurring at each storage location within a preset future time period, as well as the set of possible associated goods categories. The probability value quantifies the likelihood of an operation occurring at a storage location, and the set of goods categories clarifies the types of goods that the storage location may handle in the future, providing core data for subsequent operation intensity assessment.

[0105] Operational strength and range mapping unit:

[0106] Based on the probability values ​​of inbound and outbound operations at each storage location and the set of potentially associated product categories, combined with known future pre-order data from the warehouse management system, the predicted probability values ​​are calibrated and overlaid. Known future pre-order data represents definite future operational demands, correcting potential biases arising from predictions based solely on historical data. During calibration, the predicted probability values ​​for the corresponding storage locations are adjusted according to the storage locations and product categories involved in the pre-orders. The overlay operation incorporates the operational demands brought by the pre-orders into the prediction results, ultimately generating an initial operational intensity distribution map reflecting the expected operational intensity of each storage location within a preset future time period. This initial operational intensity distribution map presents the expected operational busyness of each storage location in a spatial distribution format, providing foundational data for subsequent visualization processing.

[0107] Dynamic heatmap generation unit:

[0108] The system receives an initial operation intensity distribution map and renders it based on a preset spatial smoothing algorithm and threshold segmentation rules. The spatial smoothing algorithm eliminates potential local data fluctuations in the initial distribution map, making the spatial distribution of operation intensity more continuous and reasonable. The threshold segmentation rules divide the operation intensity into different levels, each corresponding to a specific numerical range. Subsequently, different levels of operation intensity are mapped to different visual colors, with higher operation intensity corresponding to more prominent colors. The color differences visually distinguish the operational busyness of each area, ultimately generating a dynamic demand heatmap covering the entire warehouse area. This clearly presents the probability of future warehouse operation spatial distribution, providing warehouse managers and subsequent optimization modules with an intuitive demand reference.

[0109] Furthermore, the time-series feature extraction technology, based on time series analysis theory, segments historical order sequences through a preset time window, transforming long time-series data into multiple short time-series segments. Each segment retains local time-series correlation information while facilitating the extraction of periodic and trend features. Simultaneously, by combining non-time-series features such as cargo turnover frequency and category co-occurrence relationships, a multi-dimensional time-series feature sample set is constructed. This sample set comprehensively reflects the historical operational characteristics and correlation patterns of cargo locations, providing rich and effective input information for demand forecasting models and improving the accuracy of forecast results.

[0110] Integrated predictive model technology is based on the idea of ​​multi-model fusion, which integrates the advantages of multiple single prediction algorithms. Through fusion strategies such as weighted voting or stacking, it reduces the prediction bias and variance of a single model. The model first conducts in-depth analysis of historical order sequences to identify periodic patterns, such as daily and weekly operation peaks, while capturing long-term trend changes in operation intensity. On this basis, it integrates recent real-time status features and dynamically adjusts the prediction results through feature interaction and weight allocation. This makes the prediction not only based on historical patterns but also adaptable to real-time environmental changes, thereby accurately outputting the probability values ​​of inbound and outbound operations for each storage location and the associated set of goods categories.

[0111] Operational intensity calibration and mapping technology is based on data fusion theory, which organically integrates predicted probability values ​​with known pre-booked order data. Known pre-booked order data is deterministic, and by superimposing it with predicted data, it can correct possible deviations in the prediction process and improve the accuracy of operational intensity assessment. The mapping process establishes a correspondence between operational intensity values ​​and spatial distribution, transforming abstract numerical data into intuitive spatial distribution graphics, so that the expected operational intensity of each storage location can be clearly presented through spatial location, laying the foundation for subsequent visualization processing.

[0112] Dynamic heatmap rendering technology, based on computer graphics and data visualization theory, first processes the initial operation intensity distribution map using a spatial smoothing algorithm. This algorithm eliminates discontinuities caused by local data fluctuations through a weighted average of neighborhood data, making the spatial distribution of operation intensity smoother and more reasonable. Subsequently, operation intensity is divided into multiple levels according to threshold segmentation rules, with each level corresponding to a visual color. The difference in brightness or hue of the color intuitively reflects the level of operation intensity. Finally, the processed operation intensity data is combined with an electronic map of the warehouse area to render a dynamic demand heatmap, realizing the visualization of the spatial distribution of operation intensity.

[0113] In this embodiment, the multi-objective dynamic optimization decision module is the core decision-making hub of the warehouse space dynamic management system based on multi-source data fusion and intelligent optimization. It focuses on three core demands: improving the efficiency of warehouse space utilization and ensuring the safety of goods storage. It constructs a multi-objective storage location allocation model, adaptively adjusts the target weights by combining a dynamic demand heatmap, and obtains the Pareto optimal storage location adjustment scheme set through a multi-objective optimization algorithm. This provides a scientific and reasonable optimization direction for subsequent scheme decisions and instruction execution, thereby realizing the dynamic and intelligent optimization configuration of warehouse space.

[0114] The multi-objective dynamic optimization decision module is primarily responsible for constructing a location allocation model that maximizes warehouse space utilization, minimizes expected picking path costs, and minimizes cargo storage environment mismatch. Based on the location operation intensity distribution analyzed from the dynamic demand heatmap, it adaptively adjusts the weight coefficients of each objective in the model, emphasizing the weight of the picking path cost optimization objective in high-frequency operation areas. A multi-objective optimization algorithm is used to iteratively solve the model after weight adjustment, selecting a set of non-dominated and evenly distributed Pareto optimal location adjustment schemes. This provides comprehensive optimization support for the selection of the subsequent optimal execution scheme, ensuring that location allocation meets the multi-dimensional requirements of efficient space utilization, cost savings, and suitable cargo storage. The location allocation model constructed in the multi-objective dynamic optimization decision module includes:

[0115] Storage location allocation model building unit:

[0116] Construction of the objective function for warehouse space utilization: Based on the comprehensive status dataset, extract the current occupancy status of goods size attributes and warehouse capacity for each storage location; use storage location space occupancy rate and goods stacking compactness as metrics, where storage location space occupancy rate reflects the ratio of actual used space to total capacity, and goods stacking compactness reflects the regularity and density of goods stacked within the storage location; construct the objective function for warehouse space utilization: ,in, The objective function value for warehouse space utilization is... These are the weighting coefficients. This represents the total number of storage locations. For the first The occupied space of each storage location For the first Total space of each storage location For the first The compactness of goods stacking in each storage location, This function aims to maximize the theoretical stacking compactness and maximize the effective utilization efficiency of storage space.

[0117] Construction of the target function for expected picking path cost: Based on the dynamic demand heatmap, the expected operational intensity of each storage location and the correlation between goods are extracted; the expected operational intensity reflects the future busyness of the storage location, and the correlation between goods reflects the collaborative relationship between different goods in the picking operation; using the expected length of the picking equipment movement path as a metric, the target function for expected picking path cost is constructed. ,in, The objective function value for the expected picking path cost. These are the weighting coefficients. The total number of picking tasks is expected. For the first The expected operational intensity of the storage location corresponding to each picking task. For the first The expected length of movement of the picking equipment for each picking task; the path consumption during the picking operation is minimized through this function;

[0118] Construction of the objective function for cargo storage environment mismatch: Based on environmental monitoring data in the comprehensive status dataset and environmental preference characteristics in the comprehensive status records of cargo locations, the difference between the required environmental conditions for each cargo and the actual environmental conditions of the cargo location is calculated; the difference covers the degree of deviation of key environmental parameters such as temperature and humidity, and the objective function for cargo storage environment mismatch is constructed as follows: ,in, The objective function value for the mismatch between the cargo storage environment and the cargo storage environment. These are the weighting coefficients. Total number of environmental monitoring indicators For the first The demand value of goods for each environmental indicator. For the first Actual values ​​of each environmental indicator for the cargo location. For the first The maximum permissible value for each environmental indicator For the first The minimum allowable value for each environmental indicator; the function minimizes the adaptation deviation between goods and storage environment;

[0119] Model Integration: The storage location allocation scheme is used as the decision variable, which defines the specific storage location of each good within the storage area. Storage location capacity limits, good size matching, and storage area division are used as constraints. Storage location capacity limits ensure that the storage of goods does not exceed the storage location's capacity limit, good size matching ensures that the goods can fit the space size of the storage location, and storage area division constraints clearly state that the goods must be stored in the designated area. The objective functions of storage space utilization, expected picking path cost, and goods storage environment mismatch are integrated into a multi-objective optimization problem to form the storage location allocation model.

[0120] Adaptive adjustment unit for weight coefficients:

[0121] Heatmap Analysis and High-Frequency Area Identification: Receives dynamic demand heatmaps, analyzes the spatial coordinates of each pixel in the heatmap and the corresponding operation intensity probability value; based on a preset operation intensity threshold, identifies the set of storage locations with operation intensity exceeding the threshold as high-frequency operation areas, which are the core active areas for future warehousing operations;

[0122] Calculation of the baseline value for enhancing the weight of path optimization: Traverse each storage location in the high-frequency operation area and read its operation intensity probability value; Based on the operation intensity probability value and the actual turnover rate of the storage location in historical orders, calculate the local path optimization importance factor corresponding to the storage location; The local path optimization importance factor quantifies the priority of a single high-frequency storage location in path optimization, and normalize the local path optimization importance factors of all high-frequency operation storage locations to generate the overall path optimization weight enhancement baseline value for the high-frequency area;

[0123] Route cost target weight enhancement value correction: Obtain current warehousing operation strategy indication information, which clarifies the priority level of space utilization, operation efficiency, or cargo preservation; based on the priority level, strategically adjust the overall route optimization weight enhancement benchmark value. If the operation strategy prioritizes operation efficiency, the benchmark value is further increased; if the priority is space utilization or cargo preservation, the benchmark value is appropriately decreased, resulting in the corrected route cost target weight enhancement value.

[0124] Multi-objective weight coefficient coordinated adjustment: Using the modified path cost target weight enhancement value as the core adjustment parameter, the weights of the warehouse space utilization objective function and the cargo storage environment mismatch objective function are adjusted in reverse coordination according to the preset weight allocation rules; when the path cost target weight is increased, the weights of the other two objectives are reduced accordingly to ensure that the sum of all weight coefficients remains constant, and finally an adaptive multi-objective weight coefficient set that matches the current dynamic demand and warehousing strategy is generated.

[0125] Multi-objective optimization algorithm solution unit:

[0126] Initial population construction: Based on the adaptive multi-objective weight coefficient set and storage location allocation model, the optimization objective function and constraints are clearly defined; according to the actual layout of the current storage locations and the distribution of goods, an initial population containing multiple storage location adjustment schemes is constructed, with each scheme representing a potential storage location allocation adjustment method, ensuring that the initial population has a certain degree of diversity and feasibility.

[0127] Fitness evaluation value calculation: Based on the objective function of warehouse space utilization, the objective function of expected picking path cost, and the objective function of mismatch between goods storage environment, an adaptive multi-objective weight coefficient set is used for weighted aggregation to calculate the fitness evaluation value corresponding to each individual storage location adjustment scheme. The fitness evaluation value comprehensively reflects the overall performance of the individual scheme on multiple optimization objectives, providing a quantitative basis for scheme selection.

[0128] Elite set selection: Based on fitness evaluation values, non-dominated ranking and crowding calculation are performed on individuals in the initial population for the storage location adjustment schemes; non-dominated ranking is used to distinguish the superiority and inferiority of individual schemes, with individuals having higher non-dominated rankings performing better on multiple objectives; crowding calculation is used to measure the distribution density of individuals in the population, with individuals having low distribution density ensuring the diversity of the solution set; individuals with high Pareto non-dominated rankings and low distribution density are selected to form the first generation elite set;

[0129] Offspring population generation: Genetic operations are performed on the individuals of the storage location adjustment scheme in the first generation elite set. The genetic operations include crossover operators based on storage operation logic and mutation operators based on storage location status constraints. The crossover operator generates a new scheme individual that combines the advantages of both by exchanging and combining the storage location allocation information of two parent scheme individuals. The mutation operator randomly adjusts some of the storage location allocation information of the scheme individuals to introduce new genetic characteristics, thereby generating new storage location adjustment scheme individuals and forming the offspring population.

[0130] Iterative optimization: Merge the first generation elite set with the offspring population, perform non-dominated sorting and crowding calculation on the merged population again, select the new generation elite set, and complete one iteration optimization; repeat the above iterative process of population merging, sorting, selection and genetic operations until the preset iteration termination condition is reached. The iteration termination condition can be set to the number of iterations reaching a threshold or the fitness evaluation value of the elite set tending to stabilize.

[0131] Pareto optimal solution set output: All individual storage location adjustment solutions contained in the final generation elite set are output as a set of Pareto optimal storage location adjustment solutions. This solution set covers the optimal solutions under different objective trade-offs, providing comprehensive support for the selection of the subsequent optimal execution solution.

[0132] Furthermore, the construction principle of the multi-objective storage location allocation model is based on multi-objective optimization theory, comprehensively considering the core contradictions and needs in warehouse management. By extracting key attribute data of storage locations and goods, the three objectives of space utilization efficiency, operation path cost, and goods environment adaptation are quantified into mathematical functions. At the same time, physical constraints and rule constraints in actual warehouse operations are used as boundary conditions to form a complete multi-objective optimization model. The core of this principle lies in transforming the complex warehouse storage location allocation problem into a computable and optimizable mathematical problem through mathematical modeling, realizing the synergistic optimization of multiple conflicting objectives, and providing a clear mathematical framework for subsequent algorithm solutions.

[0133] The principle of adaptive weight coefficient adjustment technology is based on the concept of dynamic demand perception and strategy matching. By analyzing the dynamic demand heat map, it obtains the spatial distribution characteristics of the intensity of warehouse operations, identifies high-frequency operation areas, and quantifies the importance of path optimization. Combined with the priority direction of the current warehousing operation strategy, the weight of each target is dynamically adjusted so that the weight allocation can accurately match real-time demand and operational strategy. This technology breaks the limitations of fixed weights and realizes the adaptive adjustment of weight coefficients as the warehousing changes dynamically, ensuring that the optimization model can focus on the current core demand and improve the pertinence and effectiveness of optimization decisions.

[0134] The multi-objective optimization algorithm is based on evolutionary algorithms and Pareto optimality theory. It constructs an initial population to simulate the initial group of biological evolution, uses fitness evaluation values ​​to quantify the quality of individuals, selects dominant individuals through non-dominated sorting, and uses genetic operations such as crossover and mutation to simulate the reproduction and mutation process of organisms to achieve iterative evolution of the population. During the iteration process, it takes into account both the quality and diversity of solutions, and finally converges to the Pareto optimal solution set. The core of this principle is to find the optimal trade-off solution among multiple conflicting objectives by simulating the survival of the fittest mechanism of natural evolution, ensuring that the output Pareto optimal solution set not only contains the optimal solution under each objective, but also has good distribution characteristics to meet the decision-making needs of different scenarios.

[0135] In this embodiment, the scheme decision and instruction execution module is the execution center of the warehouse space dynamic management system based on multi-source data fusion and intelligent optimization. It receives the Pareto optimal storage location adjustment scheme set output by the multi-objective dynamic optimization decision module, determines the optimal execution scheme through multi-dimensional evaluation and screening, parses it into an executable operation instruction sequence for the storage equipment, and monitors the instruction execution status in real time to ensure that the storage location adjustment operation is implemented accurately and efficiently, thus building a key bridge between intelligent decision-making and actual warehousing operations.

[0136] The solution decision-making and instruction execution module is primarily responsible for selecting the optimal execution plan from the Pareto optimal storage location adjustment plan set, based on preset decision rules, and adapting it to the current warehousing operation status and strategy preferences. It extracts the goods movement task by parsing the optimal execution plan, decomposes it into atomic operation steps, and generates a conflict-free, time-sequential operation instruction sequence by combining task logic dependencies and equipment resource conflict detection. The instruction sequence is then distributed to the corresponding warehousing equipment, and the execution status is monitored in real time. Execution feedback information is collected to generate an instruction execution status report, ensuring that storage location adjustment operations are executed in an orderly, accurate, and safe manner, thus realizing the transformation of optimized decisions into actual operations. The solution decision-making and instruction execution module specifically includes:

[0137] Solution evaluation and ranking unit:

[0138] The system receives the Pareto optimal location adjustment scheme set output by the multi-objective dynamic optimization decision-making module and comprehensively scores each scheme based on multiple preset evaluation dimensions. These dimensions include implementation cost, expected target benefits, and the degree of interference with current operations. Implementation cost covers resource consumption such as energy consumption for cargo handling and equipment occupancy time. Expected target benefits reflect the scheme's anticipated effects on space utilization improvement, picking path cost reduction, and environmental adaptability optimization. The degree of interference with current operations reflects whether the scheme's execution will affect ongoing warehousing operations. A comprehensive scoring formula is used: ,in, For the comprehensive evaluation of the proposal, Assign weights to the implementation costs of the plan. Weighted by the expected target return. To assign weights to the degree of interference with current on-the-go operations, To score the implementation cost of the plan, Score the expected target return. To score the degree of interference with current operations, a comprehensive score is calculated for each solution using this formula, generating a ranked list of solutions with comprehensive scores, providing a quantitative basis for subsequent selection.

[0139] Strategy matching and filtering unit:

[0140] The system receives a sorted list of solutions and obtains the real-time operational status and strategy preference configuration of the current warehousing system. The real-time operational status includes ongoing tasks, warehousing equipment operating status, and dynamic storage space occupancy. The strategy preference configuration clarifies the priority of the current warehousing operation for objectives such as space utilization, operational efficiency, and cargo preservation. The system matches the attributes of each solution in the sorted list with the real-time operational status and strategy preference configuration, filtering out solutions that do not meet the current operational constraints (such as equipment occupancy conflicts or severe interference from in-transit operations) and strategy preferences (such as prioritizing efficiency but choosing a solution that focuses on space optimization). This generates a subset of candidate solutions, ensuring that the candidate solutions are feasible for actual execution.

[0141] Final solution selected unit:

[0142] A secondary evaluation is conducted on the selected subset of candidate solutions. This secondary evaluation is based on a weighted calculation of the urgency of the solution's execution and the resource readiness status. The urgency of the solution's execution reflects the pressing need for warehouse location adjustments, such as prioritizing warehouse location optimization in high-frequency operation areas. The resource readiness status reflects whether the equipment, manpower, and other resources required for the implementation of the solution are in place. The secondary evaluation formula is as follows: ,in, For the second assessment score, Assigning weights based on urgency, Weights for resource ready states. Score the urgency of the plan's implementation. The resource readiness status is scored; based on the secondary evaluation score, the storage location adjustment plan with the highest comprehensive score is selected from the subset of candidate plans as the optimal execution plan, ensuring that the plan is both suitable for the current situation and can be implemented quickly;

[0143] Instruction decomposition and serialization unit:

[0144] The optimal execution plan is analyzed, and each cargo movement task to be adjusted is extracted. Each cargo movement task contains key information such as cargo identifier, source location coordinates, target location coordinates, and cargo handling priority. Based on the preset equipment scheduling rules and path planning algorithm, each cargo movement task is decomposed and mapped into one or more atomic operation steps that can be executed by specific warehousing equipment. The equipment scheduling rules specify the appropriate warehousing equipment for different types of cargo movement tasks, and the path planning algorithm plans the optimal handling path for each movement task. An atomic operation step is the smallest operation unit that the warehousing equipment can directly execute, such as moving the equipment to the source location, grabbing cargo, moving it to the target location, or placing cargo, thus generating a set of atomic operation steps.

[0145] Instruction sequence generation and verification unit:

[0146] The system receives a set of atomic operation steps and sorts and schedules these steps based on the logical dependencies between tasks and the results of equipment resource conflict detection. Logical dependencies between tasks refer to operations that can only be executed after other operations are completed, such as unloading a cargo location after clearing it. Equipment resource conflict detection identifies the simultaneous occupancy requirements of different operation steps on the same equipment. Through sorting and scheduling, a conflict-free, time-sequential sequence of specific operation instructions is formed. The logical correctness of the specific operation instruction sequence is verified to ensure that the instruction execution flow conforms to the warehouse operation logic. Safety boundary checks are performed to ensure that the instruction operations do not exceed safety boundaries such as equipment operating range and cargo location capacity limits, generating the final executable instruction sequence.

[0147] Command distribution and status monitoring unit:

[0148] The executable instruction sequence is distributed to the corresponding warehouse equipment controller to ensure that each equipment receives and executes its own corresponding operation instruction; the execution status of each warehouse equipment in response to the instruction is monitored in real time, including the status of execution start, execution in progress, execution completion, or execution exception; the status feedback information of each equipment is collected, the feedback information is summarized and organized, and an instruction execution status report containing the execution status of each instruction, execution time, and description of the exception event is generated, providing execution data support for the closed-loop feedback and model update module;

[0149] Furthermore, the multi-dimensional solution evaluation technology is based on the weighted comprehensive evaluation theory. By constructing an evaluation system covering key dimensions such as solution implementation cost, expected target benefits, and the degree of interference with in-transit operations, a reasonable weight is assigned to each dimension. The weight is determined based on the actual needs and target priorities of warehousing operations. A linear weighted summation formula is used to comprehensively score each solution, transforming qualitative and quantitative indicators into a unified score value. This enables the quantitative ranking of multiple Pareto optimal solutions, providing an objective and scientific basis for selecting the optimal solution and avoiding biases caused by subjective decision-making.

[0150] The strategy matching and dynamic screening technology is based on the concepts of state adaptation and preference matching. It collects real-time operational status data of the warehousing system to construct an operational status feature model and extracts strategy preference parameters for warehousing operations. It performs multi-dimensional adaptation calculations on the solution attributes, state feature model, and strategy preference parameters, and filters out solutions that meet the current operational constraints and strategy preferences by setting adaptation thresholds. This technology can dynamically respond to changes in warehousing operational status, ensuring that the selected candidate solutions have the actual execution conditions, and improving the feasibility and effectiveness of solution execution.

[0151] The task decomposition and atomic operation mapping technology is based on the modular decomposition concept. It decomposes complex cargo movement tasks into a series of independent atomic operation steps that can be directly executed by the equipment, according to the warehousing operation process and equipment operation logic. By establishing a mapping relationship between cargo movement tasks and atomic operation steps, and combining equipment scheduling rules and path planning algorithms, it ensures that each atomic operation step can accurately correspond to specific warehousing equipment and operation path, realizing the standardized and normalized decomposition of complex tasks, and laying the foundation for the generation of subsequent instruction sequences.

[0152] The instruction sequence optimization and verification technology is based on logical dependency analysis and conflict detection theory. It identifies the execution order constraints of operation steps by constructing a logical dependency graph between tasks; it uses a resource conflict detection algorithm to identify conflicting demands on resources such as equipment and storage space for different operation steps; it sorts and schedules the operation steps based on the analysis and detection results to generate a conflict-free time-sequential instruction sequence; at the same time, it verifies the rationality of the instruction execution process through a logical correctness verification algorithm, and ensures that the instruction operation does not exceed the safety threshold through a safety boundary verification model, thus guaranteeing the executability and security of the instruction sequence.

[0153] The instruction distribution and status monitoring technology is based on a distributed communication and real-time feedback mechanism. It uses point-to-point communication to accurately distribute the sequence of executable instructions to the corresponding warehouse equipment controllers, ensuring the timeliness and accuracy of instruction transmission. By deploying status monitoring sensors and data acquisition interfaces, it collects the execution status data of the equipment in real time, and uses a status feedback protocol to transmit the execution status information to the module. The status information is then parsed and summarized to generate an instruction execution status report. This technology realizes closed-loop management of instruction distribution and execution status monitoring, ensuring full control over the entire operation execution process.

[0154] In this embodiment, the closed-loop feedback and model update module is the core of the closed-loop optimization of the dynamic warehouse space management system based on multi-source data fusion and intelligent optimization. It takes over the operation instructions and execution feedback from the scheme decision and instruction execution module. By monitoring the instruction execution status in real time, confirming changes in the storage location status, and feeding back the warehouse status data, it starts the next optimization cycle. At the same time, it dynamically updates the prediction model based on the difference between the prediction and the actual data, ensuring that the system forms a complete closed loop of "decision-execution-feedback-optimization", and continuously improving the system's adaptability and decision accuracy.

[0155] The closed-loop feedback and model update module is primarily responsible for driving the warehousing equipment to execute executable instruction sequences, monitoring the instruction execution status in real time and generating execution reports; parsing the execution results, verifying changes in the storage location status using sensor network data, formatting the new warehousing status data and feeding it back to the comprehensive status dataset, triggering the next optimization cycle; extracting actual inbound and outbound data within a preset time period, comparing it with the prediction results for the corresponding time period to calculate the prediction error, dynamically adjusting the parameters of the prediction model based on the error, and completing incremental model updates without interrupting system services. This ensures that the prediction model always aligns with actual warehousing needs, maintaining the long-term stable and efficient operation of the system. The closed-loop feedback and model update module includes:

[0156] Instruction execution status monitoring unit:

[0157] The system receives the executable instruction sequence output by the scheme decision and instruction execution module, and sends the instruction sequence to the corresponding storage equipment to drive the storage equipment to perform operations according to the instructions; it collects the execution feedback signals of each storage equipment in real time, including information such as equipment operating status, instruction execution progress, and whether any abnormalities have occurred; it analyzes and summarizes the feedback signals in real time to generate an instruction execution status report containing the execution status of each instruction (execution start, execution in progress, execution complete, execution abnormality), execution time, description of abnormal events, and handling suggestions, providing a basis for confirming changes in storage location status and subsequent data processing;

[0158] Cargo location status change confirmation unit:

[0159] The system receives instruction execution status reports, filters out instructions that have been completed, analyzes the cargo movement results corresponding to each instruction, and clarifies key information such as cargo identification, source location coordinates, and target location coordinates. Through the sensor network of the storage area, it collects real-time physical data of each location, including location occupancy status, cargo identification, and ambient temperature and humidity. The analyzed cargo movement results are compared and verified with the physical data of the locations collected by the sensors to confirm whether the actual status of cargo location changes is consistent with the instruction requirements. Finally, it generates a location status change confirmation record reflecting changes in cargo distribution, occupancy status, and environmental conditions at each location, ensuring the authenticity and accuracy of the status data.

[0160] Warehouse status data feedback unit:

[0161] The system receives confirmation records of changes in warehouse location status, extracts core information such as cargo location, warehouse occupancy status, and environmental data from these records, and associates this information with corresponding timestamps to ensure the data has time-series characteristics. Following the data format specifications of the comprehensive status dataset, the system formats the information after the associated timestamps into standard data entries. These standard data entries must maintain consistency with the structure of the comprehensive status dataset to facilitate data fusion. The system adds these standard data entries to the comprehensive status dataset in real time, generating an updated comprehensive status dataset containing the latest warehouse status information. This triggers the start of the next optimization cycle, providing real-time and effective data support for subsequent modules.

[0162] Prediction error calculation unit:

[0163] From the updated comprehensive status dataset, extract the actual inbound and outbound operation logs for the preset time period. The logs contain key data such as the intensity of actual inbound and outbound operations at storage locations, associated goods categories, and operation times. Obtain the corresponding time period prediction results generated by the dynamic demand forecasting and visualization module, including the predicted operation intensity and predicted associated goods category set for each storage location. Align the actual inbound and outbound operation logs with the corresponding time period prediction results in time and space to ensure a one-to-one correspondence between the actual and predicted data for the same storage location and time period. Calculate the difference between the predicted and actual values ​​of the operation intensity and goods category for each storage location using the mean absolute error formula: ,in, The mean absolute error, This refers to the total number of storage locations. For the first The actual operational intensity of each storage location. For the first The forecasting intensity of each cargo location is calculated; at the same time, indicators such as the forecasting accuracy of cargo categories are calculated, and a set of forecasting error indicators containing multiple error indicators is generated to comprehensively reflect the deviation of the forecasting model.

[0164] Model parameter tuning unit:

[0165] The system receives a set of prediction error indicators, analyzes the magnitude and distribution characteristics of each indicator, and identifies which storage locations and product categories have the largest prediction deviations. Based on a pre-defined machine learning update algorithm, including gradient descent and adaptive learning rate algorithms, the system dynamically adjusts the weight parameters and structural hyperparameters within the prediction model according to the characteristics of the error indicators. The weight parameter adjustment is used to correct the degree of influence of different features on the prediction results, and the structural hyperparameter adjustment is used to optimize the network structure or model complexity of the model. The system generates parameter update instructions containing the adjustment values ​​of each parameter to ensure that the model parameters can specifically correct the prediction deviations.

[0166] Incremental update unit for prediction model:

[0167] Receive parameter update instructions and apply the parameter adjustment values ​​in the instructions to the integrated prediction model in the dynamic demand forecasting and visualization module; use incremental update technology to hot update the model parameters without interrupting the online service of the model, so as to avoid the system from stopping due to model updates; after the update is completed, an updated integrated prediction model is generated, which has a prediction capability that is more in line with the actual warehousing needs, and is used for demand forecasting in subsequent optimization cycles to continuously improve prediction accuracy.

[0168] Furthermore, the instruction execution status monitoring technology is based on real-time communication and signal analysis theory. By establishing a stable communication link between the storage equipment and the module, it receives the execution feedback signals from the equipment in real time. The signal analysis algorithm is used to decode and analyze the feedback signals, extracting key information such as equipment operating status and instruction execution progress. Through status mapping rules, the signals are converted into intuitive execution status descriptions, generating instruction execution status reports. This technology ensures that the instruction execution status can be grasped in a timely manner, execution anomalies can be quickly detected, and timely support can be provided for subsequent processing.

[0169] The principle of the storage location status verification technology is based on the multi-source data cross-validation theory. It combines the execution feedback results of the storage equipment with the physical data of the storage location collected by the sensor network to construct a dual verification mechanism. By comparing the target status of the goods movement command with the actual status collected by the sensors, the accuracy of the goods location change is verified. When the two are consistent, the status change is confirmed to be valid. When a deviation occurs, an anomaly is marked and a secondary verification is triggered to ensure the authenticity and reliability of the storage location status data and provide an accurate basis for subsequent data feedback.

[0170] The warehouse status data feedback technology is based on the theory of data standardization and real-time integration. It establishes a unified data format specification to convert heterogeneous data in the location status change confirmation record into standard data entries. Through real-time data writing technology, the standard data entries are quickly added to the comprehensive status dataset, realizing real-time updates of warehouse status data. This technology ensures that the updated comprehensive status dataset can reflect the latest warehouse status in a timely manner, providing accurate data input for the next optimization cycle and ensuring the smooth connection of the closed-loop process.

[0171] The prediction error calculation technology is based on the theory of spatiotemporal alignment and difference quantification. It uses a spatiotemporal alignment algorithm to accurately match actual data and predicted data in time and space dimensions to ensure the effectiveness of the comparison. It adopts a variety of error quantification indicators, such as mean absolute error, mean square error, and category prediction accuracy, to quantify prediction deviation from different dimensions. This technology can comprehensively and objectively reflect the performance of the prediction model and provide a scientific quantitative basis for adjusting model parameters.

[0172] The principle of dynamic model parameter adjustment technology is based on machine learning optimization theory. It adopts optimization algorithms such as gradient descent to minimize prediction error and calculate the optimal adjustment direction and magnitude of model parameters. According to the distribution characteristics of error indicators, the model parameters corresponding to the storage locations or product categories with large deviations are adjusted in a focused manner. At the same time, the step size of parameter updates is controlled by adaptive learning rate to avoid over-adjustment or under-adjustment. This technology ensures that the model parameters can specifically correct prediction deviations and improve the model's adaptability to actual warehousing needs.

[0173] The incremental model update technique is based on the theory of hot update and incremental learning. Without stopping the online model service, the adjustment values ​​in the parameter update instructions are gradually applied to the existing model. By retaining the effective features already learned by the model, only the parameters related to the prediction error are updated, avoiding the model "forgetting" historical effective information. This technique enables the parallel execution of model updates and system operation, ensuring the continuity and stability of the system, while rapidly improving the model's prediction accuracy.

[0174] 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.

Claims

1. A warehouse space dynamic management system based on multi-source data fusion and intelligent optimization, characterized in that: Comprise: Real-time data acquisition and fusion module: through the deployment of sensor network in the warehouse area, real-time acquisition of each goods identity data, occupancy state data and environmental monitoring data; the data and historical order data of warehouse management system are spatio-temporally aligned and fused, and the comprehensive state data set containing historical and real-time information is generated with goods location as the basic unit; Dynamic demand prediction and visualization module: based on the comprehensive state data set, the prediction model is used to predict the warehouse demand of each category of goods in the future preset period; according to the prediction result, the dynamic demand heat map reflecting the future warehouse operation space distribution probability is generated; Multi-objective dynamic optimization decision module: a goods location allocation model is constructed, which takes maximizing warehouse space utilization, minimizing expected picking path cost and minimizing goods storage environment mismatch degree as optimization objectives; According to the dynamic demand heat map, the weight coefficients of each objective in the multi-objective optimization model are adaptively adjusted, wherein the high-frequency operation area identified in the heat map is given a higher weight for the minimum expected picking path cost objective; The model with adjusted weight is solved by using multi-objective optimization algorithm, and a set of Pareto optimal goods location adjustment scheme set is obtained; Scheme decision and instruction execution module: according to the preset decision rule, an optimal execution scheme is selected from the Pareto optimal goods location adjustment scheme set, and the optimal execution scheme is parsed into specific operation instruction sequence which can be driven by warehouse equipment for execution; Closed-loop feedback and model updating module: execute operation instruction sequence and confirm goods location state change in real time; the new warehouse state data generated after execution is fed back to the comprehensive state data set to start the next optimization cycle; at the same time, the prediction model is updated based on the difference between the actual warehouse-in and warehouse-out data and the prediction data. 2.The warehouse space dynamic management system based on multi-source data fusion and intelligent optimization according to claim 1, characterized in that: The real-time data acquisition and fusion module comprises: Receiving the original data stream uploaded by the sensor network composed of RFID reader, two-dimensional code scanning device and weight sensor deployed in each physical goods location, the original data stream at least including goods identity code, goods location occupancy flag and environmental temperature and humidity readings; The original data stream is cleaned and formatted, and the abnormal values and null values are removed, and the time stamp and data format are unified to generate a standardized real-time data table; According to the electronic map of the warehouse area and the goods location coding rules, the unique goods location space coordinates of each record in the standardized real-time data table are associated, and a goods location real-time state snapshot with space-time label is constructed; Extract the order execution log in the preset historical period from the order database of the warehouse management system, the order execution log at least contains goods category, warehouse-in and warehouse-out time, source goods location and target goods location coordinates; Establish a space-time index structure with goods location space coordinates as the key, associate and align the goods location real-time state snapshot with the order execution log, and generate the historical operation frequency, associated category sequence and environmental preference characteristics of each goods location; Based on the results of association and alignment, the real-time physical state information and the corresponding historical behavior characteristic information of each goods location are fused to generate a goods location comprehensive state record containing static attributes, dynamic state and historical statistical characteristics; The comprehensive status records of all the storage spaces are collected to generate a comprehensive status dataset taking the storage spaces as basic units, and the comprehensive status dataset is taken as an input of the dynamic demand prediction and visualization module. 3.The warehouse space dynamic management system based on multi-source data fusion and intelligent optimization according to claim 1, characterized in that: The dynamic demand prediction and visualization module specifically comprises: A data preparation and feature extraction unit is configured to extract, from the comprehensive status dataset, a historical order sequence, a turnover frequency and a category co-occurrence relationship associated with each storage space, and divide the historical order sequence into continuous time sequence segments based on a preset time window to form a time sequence feature sample set taking the storage spaces as units. A space-time demand prediction unit is configured to input the time sequence feature sample set into a trained integrated prediction model, and the integrated prediction model is configured to analyze periodicity and trend of the historical operation sequence of each storage space, and fuse a recent real-time status feature to predict a probability value of each storage space to have an in-out operation and a possible associated set of goods categories in a future preset time period. An operation intensity and range mapping unit is configured to calibrate and superimpose the predicted probability value based on the probability value of each storage space to have an in-out operation and the possible associated set of goods categories, in combination with future known reservation order data of the warehouse management system, to generate an initial operation intensity distribution map reflecting an expected operation intensity of each storage space in the future preset time period. A dynamic heat map generation unit is configured to receive the initial operation intensity distribution map, and render the initial operation intensity distribution map based on a preset spatial smoothing algorithm and threshold segmentation rule to map operation intensities of different levels to different visual colors, and generate a dynamic demand heat map covering all the warehouse areas and reflecting a spatial distribution probability of future warehouse operations. 4.The warehouse space dynamic management system based on multi-source data fusion and intelligent optimization of claim 1, characterized in that: The storage space allocation model in the multi-objective dynamic optimization decision module comprises: Based on the comprehensive status dataset, the current occupancy state, the goods size attribute and the storage space carrying capacity of each storage space are extracted, and a warehouse space utilization rate objective function is constructed taking the storage space occupancy rate and the goods stacking compactness as the measurement indicators. Based on the dynamic demand heat map, the expected operation intensity of each storage space and the association between goods are extracted, and an expected picking path cost objective function is constructed taking the expected length of the picking equipment moving path as the measurement indicator. Based on the environmental monitoring data in the comprehensive status dataset and the environmental preference features in the storage space comprehensive status record, the difference between the required environmental conditions of each goods and the actual storage space environmental conditions is calculated, and a goods storage environment mismatch degree objective function is constructed. Taking the storage space allocation scheme as the decision variable, taking the storage space capacity limit, the goods size matching and the warehouse area division as the constraint conditions, the warehouse space utilization rate objective function, the expected picking path cost objective function and the goods storage environment mismatch degree objective function are integrated into a multi-objective optimization problem to form the storage space allocation model. 5.The warehouse space dynamic management system based on multi-source data fusion and intelligent optimization according to claim 4, characterized in that: According to the dynamic demand heat map, the weight coefficients of each objective in the multi-objective optimization model are adaptively adjusted, including: The dynamic demand heat map is received, and the storage space coordinates corresponding to each pixel point in the heat map and the operation intensity probability value mapped therefrom are analyzed; and according to a preset operation intensity threshold, a set of storage spaces with an operation intensity higher than the preset threshold are identified as high-frequency operation areas. Traverse each storage location in the high-frequency operation area, read its operation intensity probability value; based on the operation intensity probability value and the actual turnover rate of the storage location in the historical order, the local path optimization importance factor corresponding to the storage location is calculated; the local path optimization importance factors of all high-frequency operation storage locations are normalized to generate the overall path optimization weight enhancement benchmark value of the high-frequency area; Obtain the current warehouse operation strategy indication information, and the warehouse operation strategy indication information at least includes the priority level of space utilization, operation efficiency or goods preservation; according to the priority level, the overall path optimization weight enhancement benchmark value is corrected to obtain the corrected path cost target weight enhancement value; Taking the corrected path cost target weight enhancement value as the core adjustment parameter, and according to the preset weight distribution rule, the weights of the warehouse space utilization target function and the goods storage environment mismatch degree target function are adjusted in reverse cooperation to ensure that the sum of all weight coefficients remains constant, and finally a set of adaptive multi-objective weight coefficient set matched with the current dynamic demand and warehouse strategy is generated. 6.The warehouse space dynamic management system based on multi-source data fusion and intelligent optimization according to claim 5, characterized in that: A multi-objective optimization algorithm is used to solve the model after adjusting the weight, and a set of Pareto optimal storage location adjustment scheme set is obtained; specifically including: Based on the adaptive multi-objective weight coefficient set and the storage location allocation model, the optimization target function and the constraint condition are determined, and according to the actual layout and goods distribution state of the current warehouse storage location, an initial population containing multiple storage location adjustment scheme individuals is constructed; Based on the warehouse space utilization target function, the expected picking path cost target function and the goods storage environment mismatch degree target function, and using the adaptive multi-objective weight coefficient set for weighted aggregation, the fitness evaluation value corresponding to each storage location adjustment scheme individual is calculated; According to the fitness evaluation value, the storage location adjustment scheme individuals in the initial population are non-dominated sorted and crowdedness calculated, and the individuals with high Pareto non-dominated level and low distribution density are selected to form the first generation elite set; Perform genetic operation on the storage location adjustment scheme individuals in the first generation elite set, including crossover operator based on warehouse operation logic and mutation operator based on storage location state constraint, to generate new storage location adjustment scheme individuals and form a sub-population; Merge the first generation elite set and the sub-population, and perform non-dominated sorting and crowdedness calculation on the merged population again, and select a new generation of elite set from it to complete one iteration optimization; Repeat the iteration process of population merging, sorting, selection and genetic operation until the preset iteration termination condition is reached, and obtain the final generation elite set; All storage location adjustment scheme individuals contained in the final generation elite set are output as a set of Pareto optimal storage location adjustment scheme set. 7.The warehouse space dynamic management system based on multi-source data fusion and intelligent optimization according to claim 1, characterized in that: The scheme decision and instruction execution module selects an optimal execution scheme from the Pareto optimal storage location adjustment scheme set according to the preset decision rule, and parses the optimal execution scheme into a specific operation instruction sequence that can be driven by the warehouse equipment, specifically including: The scheme evaluation and ranking unit is configured to receive the set of Pareto optimal slot adjustment schemes, and perform comprehensive scoring on each slot adjustment scheme based on a plurality of preset evaluation dimensions, the evaluation dimensions at least including scheme implementation cost, expected target revenue, and interference degree to current in-transit operations, to generate a scheme ranking list with comprehensive scores; The strategy matching and screening unit is configured to receive the scheme ranking list, and acquire real-time operation status and strategy preference configuration of the current warehouse system; match attributes of each scheme in the scheme ranking list with the real-time operation status and the strategy preference configuration, and screen out schemes that do not meet current operation constraints and strategy inclination, to generate a screened candidate scheme subset; The final scheme selection unit is configured to perform secondary evaluation on the screened candidate scheme subset, the secondary evaluation being based on weighted calculation of emergency degree of scheme execution and resource readiness status, to select one slot adjustment scheme with the highest comprehensive score from the candidate scheme subset as an optimal execution scheme; The instruction decomposition and serialization unit is configured to analyze the optimal execution scheme, extract each to-be-adjusted goods movement task contained in the optimal execution scheme, the goods movement task at least including goods identifier, source slot coordinates, target slot coordinates, and goods handling priority; decompose and map each goods movement task into one or more atomic operation steps executable by a specific warehouse device according to a preset device scheduling rule and path planning algorithm, to generate an atomic operation step set; The instruction sequence generation and verification unit is configured to receive the atomic operation step set, and perform sorting and scheduling on steps in the atomic operation step set according to logical dependency relationship between tasks and device resource conflict detection result, to form a conflict-free and time-sequenced specific operation instruction sequence; perform logical correctness verification and safety boundary check on the specific operation instruction sequence, to generate a final executable instruction sequence; The instruction distribution and state monitoring unit is configured to distribute the executable instruction sequence to a corresponding warehouse device controller, and monitor execution state of the instruction by the warehouse device in real time; collect and aggregate state feedback of execution start, execution completion, or execution exception, to generate an instruction execution state report. 8.The warehouse space dynamic management system based on multi-source data fusion and intelligent optimization of claim 1, characterized in that: The closed-loop feedback and model updating module includes: The instruction execution state monitoring unit is configured to receive the executable instruction sequence output by the scheme decision and instruction execution module, drive the warehouse device to execute the executable instruction sequence, and collect execution feedback signals of the warehouse device in real time, to generate an instruction execution state report containing execution state and abnormal events of each instruction; The slot state change confirmation unit is configured to receive the instruction execution state report, analyze goods movement results corresponding to completed instructions, verify actual state of goods position change according to slot physical data collected by a sensor network of the warehouse area in real time, to generate a slot state change confirmation record reflecting changes in goods distribution and occupancy state of each slot; The closed-loop feedback and model updating module includes: The warehouse state data feedback unit is configured to receive the storage location state change confirmation record, associate the storage location, the storage location occupancy state and the environment data in the storage location state change confirmation record with a time stamp, format the data into a standard data entry, and add the data to the comprehensive state data set in real time to generate an updated comprehensive state data set, so as to trigger the start of the next optimization cycle. The prediction error calculation unit is configured to extract the actual warehouse-in and warehouse-out operation log in a preset time period from the updated comprehensive state data set, perform spatio-temporal alignment and comparison between the actual warehouse-in and warehouse-out operation log and the corresponding time period prediction result generated by the dynamic demand prediction and visualization module, calculate the difference between the predicted value and the actual value of each storage location operation intensity and each type of goods, and generate a prediction error index set. The model parameter adjustment unit is configured to receive the prediction error index set, dynamically adjust the weight parameters and structural hyperparameters in the prediction model based on a preset machine learning update algorithm and according to the numerical value and distribution characteristics of each index in the prediction error index set, and generate a parameter update instruction. The prediction model incremental update unit is configured to receive the parameter update instruction, apply the parameter update instruction to the integrated prediction model in the dynamic demand prediction and visualization module, complete the parameter hot update without interrupting the online service of the model, generate an updated integrated prediction model, and use the updated integrated prediction model for demand prediction in the subsequent optimization cycle.

Images