Material reserve dynamic optimization system based on multi-dimensional data fusion

Abstract

This invention discloses a dynamic optimization system for material reserves based on multi-dimensional data fusion, relating to the field of emergency material management technology. The system includes: a spatiotemporal data integration module outputting a standardized dataset; a material quality modeling module generating quality status data based on the PINN algorithm; a causal relationship networking module constructing a causal knowledge network using a hybrid adjustment algorithm; a counterfactual decision-making computation module generating disposal decision schemes using a physical constraint algorithm; and finally, a closed-loop feedback update module collecting execution data and reversing the model architecture and computational parameters of each module to achieve a fully intelligent closed loop. This invention integrates multi-source, full-domain data to achieve standardized interoperability, utilizes physical information neural networks to accurately predict material quality, constructs a knowledge network using causal algorithms to support scientific decision-making, generates multi-dimensional disposal schemes through counterfactual deduction, and continuously self-optimizes through a closed-loop feedback mechanism, achieving intelligent and refined management and control of emergency materials throughout the entire process, improving dispatch efficiency and guarantee capabilities.

Description

Technical Field

[0001] This invention relates to the field of emergency supplies management technology, specifically a dynamic optimization system for supplies reserves based on multi-dimensional data fusion. Background Technology

[0002] Emergency material reserves are a crucial guarantee for disaster emergency response. As the frequency and scope of disaster events continue to expand, the types of data involved in emergency material reserve management are becoming increasingly diverse. Data sources cover multiple fields such as disaster information, warehouse management, environmental monitoring, meteorology and geology, quality testing, and on-site collection. These data exist in various forms, including structured, semi-structured, and unstructured. There is a lack of unified integration standards and communication channels between different types of data, making it difficult to achieve efficient collection and standardized utilization. Changes in material quality are affected by multiple factors such as their own attributes, storage environment, and storage duration. The quality evolution mechanisms of different types of materials vary significantly. Currently, the industry lacks quality prediction methods that conform to the laws of physical change, and has not established a correlation analysis system covering all business elements. This makes it impossible to provide accurate data support for material reserve management and control, and it is difficult to adapt to the technical requirements of modern dynamic emergency material management.

[0003] Traditional emergency material reserve management models have significant shortcomings. Data management employs a decentralized and independent architecture, with data from different business systems not interconnected. This hinders the integration, processing, and standardization of multi-source data, creating data silos that prevent a comprehensive understanding of the overall business situation. Material quality control relies primarily on periodic manual inspections, failing to track the quality degradation process in real time. There is a lack of precise modeling and dynamic prediction for quality changes in different materials, resulting in lag and errors in quality status assessment. Correlation analysis between data elements remains at a superficial statistical level, failing to uncover deeper transmission relationships and influencing logic. Decision-making relies excessively on human experience, lacking scientific quantitative extrapolation and scenario adaptation capabilities. Material reserve layout, batch rotation, cross-regional allocation, and warehousing control are fragmented, lacking a collaborative management mechanism and feedback optimization from end-to-end operational data. The system cannot self-iterate and upgrade, leading to unreasonable resource allocation, low management efficiency, and an inability to meet actual emergency response and material support needs. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a dynamic optimization system for material reserves based on multi-dimensional data fusion. It standardizes and organizes multi-source heterogeneous data through a spatiotemporal data integration module, completes multi-category material quality evolution modeling and real-time status calculation based on a physical information neural network, builds a causal knowledge network of all business elements using a hybrid adjustment causal quantification algorithm, and generates full-scenario disposal solutions covering reserve layout, batch rotation, allocation scheduling, and warehousing control through a physical constraint counterfactual decision-making algorithm. It also achieves continuous iterative optimization of the system model and parameters through a two-layer closed-loop feedback. This invention effectively breaks down data barriers and solves the shortcomings of traditional management, such as data dispersion, delayed quality prediction, reliance on experience for decision-making, and lack of dynamic optimization. It realizes intelligent and refined management of the entire emergency material reserve process, significantly improving material scheduling efficiency and disaster emergency response capabilities.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a dynamic optimization system for material reserves based on multi-dimensional data fusion, the system comprising: Spatiotemporal data integration module: used to build multi-type data docking channels, acquire raw business data from multiple sources and in multiple formats, complete data regularization and processing according to preset specifications, and output standardized datasets in a unified format; Material quality modeling module: retrieves the standardized dataset output by the spatiotemporal data integration module, uses the PINN physical information neural network quality prediction algorithm to build a quality evolution model adapted to different types of materials, calculates and generates quality status data corresponding to various types of materials, and transmits the quality status data. Causal Relationship Networking Module: Simultaneously receives the standardized dataset output by the spatiotemporal data integration module and the quality status data output by the material quality modeling module, and uses a hybrid adjustment causal quantification algorithm to mine the correlation between data elements, and builds a complete causal knowledge network based on the correlation; Counterfactual decision-making module: It connects the causal knowledge network generated by the causal relationship networking module and the quality status data output by the material quality modeling module, uses the physical constraint counterfactual decision algorithm to carry out multi-dimensional variable calculation and deduction, and generates corresponding material disposal decision schemes based on the deduction results; Closed-loop feedback update module: Collects operational data generated throughout the entire process of implementing the decision-making scheme from the counterfactual decision-making module. Based on the collected operational data, it reverse-corrects the model architecture and operational parameters within the spatiotemporal data integration module, material quality modeling module, causal relationship networking module, and counterfactual decision-making module.

[0006] Furthermore, the data connection channels built by the spatiotemporal data integration module are respectively connected to a database that stores information on the location, scale, and scope of past disasters, a warehouse management system that records the quantity and location of stored materials, a sensor network that collects environmental data by being placed on different shelves and storage areas inside the warehouse, a real-time monitoring platform that continuously outputs regional meteorological elements and geological change data, a professional quality inspection system that records various quality inspection values ​​and inspection times of materials, and a mobile information collection terminal for disaster site workers to enter on-site real-time information.

[0007] Furthermore, the spatiotemporal data integration module performs multiple processing steps sequentially on the data normalization process, including: For structured raw data with well-organized field layout, the following steps are performed sequentially: removing duplicate data entries, correcting abnormal values, and filling in missing values ​​in empty positions. For semi-structured raw data with nested hierarchical arrangement, perform overall structure decomposition and layer-by-layer parsing of internal recorded information; For unstructured raw data in the form of text, images, and voice, perform effective information filtering and extraction. After all data processing is completed, the Gregorian calendar time standard is uniformly adopted to define the time statistical benchmark, the plane coordinate system is uniformly adopted to define the spatial positioning coordinates, and all data entries are classified and grouped according to the pre-set classification labels in the system.

[0008] Furthermore, the material quality modeling module extracts material category attributes, production batch parameters, outer packaging structure parameters, and warehousing and transportation environment parameters from a standardized dataset. It distinguishes the material change mechanisms corresponding to food materials, pharmaceutical materials, medical device materials, and emergency equipment materials, and matches the change rules corresponding to microbial proliferation changes, effective component degradation changes, metal structure corrosion deformation changes, and polymer material aging changes. The material change rules are embedded into the calculation framework of the quality evolution model. The PINN physical information neural network quality prediction algorithm is used to complete the iterative fitting of the model's internal parameters. The module calculates the real-time quality level, quality decay rate, and remaining usable time of a single batch of materials at fixed time intervals to obtain the quality status data.

[0009] Furthermore, the data element relationships mined by the causal relationship networking module include the relationships formed between disaster-specific characteristics and material reserve distribution, the relationships formed between material storage environment and material quality decay, the relationships formed between material quality status and on-site available material quantity, the relationships formed between available material quantity and disaster relief and disposal, as well as the relationships formed by the direct effects of each element and the relationships formed by the multi-layer transmission of each element.

[0010] Furthermore, the complete causal knowledge network built by the causal relationship networking module uses various independent data elements as the basic network nodes, and uses the intensity of causal effects calculated between elements as the connection weight between nodes. The arrangement of all nodes is arranged according to the disaster evolution process, material management process, and rescue and disposal process. The corresponding network nodes are connected according to the transmission order of the effects of the elements, forming an overall network structure that can present the influence of the elements on each other.

[0011] Furthermore, the material disposal decision scheme generated by the counterfactual decision calculation module includes regional material reserve layout configuration, inventory material batch rotation management, cross-regional emergency material allocation and dispatch, warehousing environment adaptation and control, and disaster scenario material graded dispatch. The configuration of the regional material reserve layout includes material category allocation parameters, material inventory ratio parameters, and material layout allocation parameters for each reserve warehouse. The inventory batch rotation control includes rotation cycle parameters for different categories of materials, batch rotation sorting rule parameters, and outbound and inbound ratio parameters for rotation materials. The cross-regional emergency material allocation and dispatch includes material transfer route planning parameters, multi-batch material allocation timing parameters, transportation capacity allocation parameters, and material distribution parameters in disaster-stricken areas. The warehouse environment adaptation and control content includes warehouse temperature setting parameters, humidity setting parameters, and environmental ventilation control parameters adapted to the storage requirements of different material qualities. The disaster scenario material tiered allocation content includes material allocation level parameters matching the material quality status, material allocation quantity parameters adapting to different disaster scales, and multi-category material combination allocation ratio parameters.

[0012] Furthermore, the counterfactual decision-making module performs multi-dimensional variable calculations and deductions, including calculations of material reserves, material rotation, warehousing environment, material scheduling, and disaster scenarios. Material reserves calculations cover numerical calculations of the total amount of material reserves in a single region, the ratio of reserves of multiple categories of materials, and the proportion of material allocation at reserve locations. Material rotation calculations cover numerical calculations of the interval between material batch rotations, the execution order of batch rotations, and the ratio of new and old material rotations. Warehouse environment calculations cover numerical calculations of warehouse temperature, warehouse humidity, and warehouse ventilation frequency. Material scheduling calculations cover numerical calculations of cross-regional material transportation routes, material allocation timing nodes, and the proportion of transportation resource allocation. Disaster scenario calculations cover numerical calculations of disaster intensity, disaster impact range, and disaster duration. The calculations and deductions support independent assignment of single variables and cross-coupled assignment of multiple variables. Based on the causal structure parameters of the causal knowledge network and the quality status data output by the material quality modeling module, the numerical deduction calculations of the combination of variables across the entire domain are completed.

[0013] Furthermore, the operational data collected by the closed-loop feedback update module fully covers the entire process of material procurement verification and registration, storage in designated warehouse locations, classification and distribution in designated disaster-stricken areas, and on-site rescue operations. At the same time, it fully covers the entire process of disaster occurrence time and location recording, emergency response level confirmation and activation, step-by-step advancement of various on-site rescue operations, and disaster situation stabilization and handling.

[0014] Furthermore, the closed-loop feedback update module is divided into two independent feedback levels. The first feedback level adjusts the internal model structure and various calculation parameters of the system in real time based on the numerical fluctuations of the operational data obtained from real-time monitoring. The second feedback level performs a unified and comprehensive revision and correction operation on the internal model structure and various calculation parameters of the system after all the handling operations corresponding to a single disaster have been completed.

[0015] Compared with existing technologies, this dynamic optimization system for material reserves based on multi-dimensional data fusion has the following beneficial effects: I. This invention integrates multi-source, multi-format business data across the entire domain, unifying and processing structured, semi-structured, and unstructured data. It establishes a data interface system connecting multiple scenarios, including disaster information, warehouse management, environmental monitoring, and quality inspection, breaking down barriers between different data sources and achieving efficient data resource interoperability and standardized application. Based on a physical information neural network, it constructs a quality evolution model adaptable to multiple categories of materials, accurately reproducing the quality change patterns of various materials and outputting real-time information related to material quality status. This solves the problems of lagging and inaccurate predictions in traditional material quality control. Furthermore, it utilizes a hybrid adjustment causal quantification algorithm to mine multi-layered transmission relationships between data elements, building a complete causal knowledge network that clearly presents the impact of disasters, warehousing, materials, and rescue efforts. This allows material reserve management to break free from experience dependence, using data logic to support accurate judgment, comprehensively improving the reliability of material quality supervision and element correlation analysis, and laying a solid data foundation for emergency material management.

[0016] Second, this invention conducts multi-dimensional counterfactual decision-making simulations based on causal knowledge networks and material quality status, covering all business processes such as reserve layout and configuration, batch rotation management, cross-regional allocation, warehousing environment control, and disaster scenario deployment. It generates material disposal plans tailored to actual needs, comprehensively considering management effectiveness, resource consumption, and execution risks to achieve scientific adaptation of decision-making plans. It constructs a two-layer, independently operating closed-loop feedback mechanism, comprehensively collecting operational data from the entire process of material flow and disaster response. Through a combination of real-time adjustments and comprehensive corrections, it continuously optimizes the system model architecture and computational logic, forming a self-iterable dynamic optimization system. This breaks free from the limitations of traditional static management models, achieving intelligent and refined management of the entire material reserve process. It significantly improves the efficiency of emergency material dispatch and reserve management, strengthens the response speed and guarantee capabilities for disaster emergency response, and makes material reserve management more aligned with the actual needs of emergency response.

[0017] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0019] Figure 1 A flowchart for a dynamic optimization system for material reserves based on multi-dimensional data fusion; Figure 2 This is a schematic diagram illustrating the data transmission between different steps in a dynamic optimization system for material reserves based on multi-dimensional data fusion. Figure 3 This is a schematic diagram of data transmission within the causal relationship networking module. Detailed Implementation

[0020] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0021] Reference Figure 1 One embodiment of this invention proposes a dynamic optimization system for material reserves based on multi-dimensional data fusion. It employs a collaborative operational architecture that integrates standardized multi-source data, models the quality evolution of materials by category, networks causal relationships among data elements, performs multi-variable counterfactual decision-making, and implements a two-layer closed-loop feedback iterative update. This system can perform full-dimensional fusion processing of multi-source, multi-format data, including disaster information, warehouse management, environmental sensing, meteorology and geology, quality testing, and on-site collection. It continuously calculates the quality status of different categories of emergency supplies, such as food, medicine, medical devices, and emergency equipment. It uncovers the relationships and transmission logic between various data elements throughout the entire material reserve process, generating material reserve layout, batch rotation, cross-regional allocation, environmental control, and tiered dispatch and disposal plans adapted to different disaster scenarios. Simultaneously, through a two-layer feedback mechanism of real-time adjustment and post-disaster comprehensive correction, it continuously optimizes the model architecture and computational parameters of each module within the system, ensuring stable adaptation to the dynamic optimization needs of emergency material reserve management and disaster emergency response throughout the entire lifecycle.

[0022] The specific operation flow of the system in this embodiment is as follows: Figure 2 As shown: The spatiotemporal data integration module establishes multi-type data docking channels, acquires raw business data from multiple sources and in multiple formats, completes hierarchical classification and processing according to preset specifications, and outputs a standardized dataset in a unified format. The material quality modeling module retrieves a standardized dataset and uses the PINN physical information neural network quality prediction algorithm to build a quality evolution model that adapts to different types of materials, and calculates and generates quality status data corresponding to various types of materials. The causal relationship networking module simultaneously receives standardized datasets and quality status data, uses a hybrid adjustment causal quantification algorithm to mine the relationships between data elements, and builds a complete causal knowledge network based on these relationships. The counterfactual decision-making module connects to the causal knowledge network and quality status data, uses a physical constraint counterfactual decision-making algorithm to perform multi-dimensional variable calculation and deduction, and generates corresponding material disposal decision-making schemes based on the deduction results. The closed-loop feedback update module collects operational data generated throughout the entire process of implementing the decision-making plan. Based on the collected operational data, it reverse-corrects the model architecture and operation parameters within each module, including spatiotemporal data integration, material quality modeling, causal relationship networking, and counterfactual decision calculation, forming a complete dynamic optimization operation closed loop.

[0023] Spatiotemporal data integration module: Specifically, the spatiotemporal data integration module first establishes multi-type data connection channels. Through standardized data interfaces, it establishes stable data transmission links with six major categories of external data systems, monitoring equipment, and information terminals, achieving full-domain, real-time, and stable collection of raw business data. The external data sources include: a disaster information database storing information on the location, scale, affected area, and duration of past disasters; a warehouse management system that records the quantity, location, and distribution of materials in each reserve warehouse in real time; and a sensor network evenly distributed within different shelves, storage areas, and temperature and humidity control zones within the warehouse, collecting environmental values ​​such as temperature, humidity, and ventilation status in real time. A real-time monitoring platform that outputs regional meteorological and geological data such as temperature, precipitation, wind, and geological changes 24 hours a day; a professional quality inspection system that regularly conducts material quality inspections and records various quality inspection values, inspection times, and inspection results; and a mobile information collection terminal that allows disaster site workers and material delivery personnel to input on-site disaster conditions, material needs, road access status, and material consumption in real time.

[0024] Optionally, the spatiotemporal data integration module performs multiple targeted processing steps in sequence for data normalization, adopting differentiated processing procedures for raw data of different structural types to ensure that all data is converted into a unified and callable standard format: For structured raw data with well-organized fields and uniform format, the following processes are performed in sequence: automatic removal of duplicate data entries, automatic correction of abnormal values ​​that exceed reasonable ranges, and intelligent supplementation of missing values ​​in data field empty positions. This process eliminates data redundancy, numerical errors, and information gaps, ensuring the accuracy and integrity of the basic data. For semi-structured raw data with hierarchical nesting and inconsistent formats, the process of splitting the overall structure layer by layer, parsing the internal recorded information in a targeted manner, and unifying the field format is performed to transform the complex data with nested structure into a flat, clear, and directly callable standard data format. For unstructured raw data in the form of images, text, and voice, we perform effective information feature extraction, semantic recognition and parsing, and filtering of invalid and redundant content to accurately screen and retain effective information related to material reserve management, disaster emergency response, and material quality monitoring, while eliminating irrelevant and redundant content.

[0025] Specifically, after all the raw data has undergone differential normalization processing, the spatiotemporal data integration module uniformly adopts the Gregorian calendar time standard to define a unified time statistical benchmark for all data entries, ensuring that the time dimension of all data remains consistent. It also uniformly adopts a planar coordinate system to define a unified spatial positioning coordinate for all data entries involving spatial location, ensuring that the spatial dimension of all data corresponds accurately. Furthermore, according to the pre-set classification and labeling entries such as material category, disaster type, storage area, testing batch, and data source, all data entries are classified, labeled, and grouped. Finally, a standardized dataset with unified format, standardized time sequence, accurate spatial positioning, and complete information dimensions is output, providing stable and reliable data support for subsequent module model calculations, data analysis, and decision generation.

[0026] For example, when processing the raw data of the regional emergency material reserve system, for the structured material inventory data output by the warehouse management system, the system automatically scans, identifies, and removes duplicate inventory entries, corrects abnormal values ​​that do not conform to the actual storage logic, and supplements missing warehousing and storage location information through adjacent data correlation fitting. For the semi-structured hierarchical data output by the disaster monitoring platform, the system decomposes the nested disaster information layer by layer, transforms it into independent and clear standard fields, and completes the format unification. For unstructured data such as pictures and voice uploaded by on-site personnel, the system uses feature extraction and semantic recognition to filter out effective information such as disaster status, material demand, and road conditions, while filtering out irrelevant background content and redundant information. After all data processing is completed, the system unifies and standardizes time statistics standards and spatial positioning coordinates, completes the labeling and grouping of all data according to preset classification rules, and finally generates a standardized dataset with unified format and complete information, which is synchronously transmitted to the material quality modeling module and the causal relationship networking module to provide a data foundation for subsequent calculations.

[0027] Material Quality Modeling Module: Specifically, the material quality modeling module retrieves the standardized dataset output by the spatiotemporal data integration module, and extracts four types of feature information from the dataset: material category attributes, production batch parameters, outer packaging structure parameters, and warehousing and transportation environment parameters. Based on the different material categories, it distinguishes the corresponding material change mechanisms and matches appropriate physicochemical change rules: food materials match rules related to microbial proliferation and change, pharmaceutical materials match rules related to active ingredient degradation and change, medical device materials match rules related to metal structure corrosion and deformation and change, and emergency equipment materials match rules related to polymer material aging and change.

[0028] The module embeds the material change rules corresponding to the different categories of materials into the calculation framework of the quality evolution model. It relies on the PINN physical information neural network quality prediction algorithm to complete the iterative fitting of the model's internal parameters. It continuously performs calculations at fixed time intervals to generate quality status data corresponding to the real-time quality level, quality decay rate, and remaining usable time of a single batch of materials. The generated quality status data is synchronously transmitted to the causal relationship networking module and the counterfactual decision calculation module to provide data support for subsequent causal analysis and decision calculation.

[0029] The mathematical expression for the PINN (Physical Information Neural Network) quality prediction algorithm used in this module is:

[0030] in: This represents the overall loss value obtained from the neural network model calculation. This represents the numerical value corresponding to the material quality obtained from the model calculation. This indicates the numerical value corresponding to the material quality obtained from the actual testing operation. This represents the loss term generated by the model fitting the actual data. This represents the loss item resulting from conforming to the physical changes of materials. This represents the gradient change in mass value over spatial orientation and time. This indicates the parameters corresponding to the spatial location where the materials are stored. This indicates the time parameter corresponding to the storage duration of materials. This represents the loss term corresponding to the model boundary constraints. , , Each of the three types of loss items is matched with a set weight ratio value; Weighting coefficient The normalized weighted calculation method is used to determine the result, and the mathematical expression for the calculation is as follows: ; ; ;in To fit the contribution values ​​of the loss term to the actual data in the model, To align the contribution value of the loss item with the physical changes of materials, This represents the contribution value of the loss term due to the model's boundary constraints. , , This is obtained by statistically fitting historical standardized datasets and quality inspection data, and is used to quantify the impact of the three types of loss terms on the overall loss of the model.

[0031] Specifically, during the calculation process, the model combines the material quality data obtained from actual detection with the physical and chemical change patterns of the materials themselves, and continuously adjusts the weight parameters inside the algorithm to continuously reduce the deviation between the calculated values ​​of the model and the actual detection values. When the overall loss value of the model is reduced to below the reasonable threshold preset by the system, the fitting optimization of the model parameters is completed to ensure the accuracy and stability of the quality status data calculation.

[0032] For example, when performing quality modeling on medical drugs in a storage warehouse, the module extracts relevant information such as the drug's category attributes, production batch, outer packaging structure, and storage environment from a standardized dataset, matches the rules of degradation of the drug's active ingredients, and embeds them into a quality evolution model; it iteratively fits the model parameters using the Physical Information Neural Network (PINN) quality prediction algorithm, allowing the model to fully learn the patterns of drug quality changes with storage time and storage environment; it continuously performs calculations at fixed time intervals set by the system, generating quality status data such as the quality grade, quality decay rate, and remaining usable time of the batch of drugs in real time, and synchronously transmits the data to subsequent modules for causal relationship mining and emergency decision generation.

[0033] Causal relationship networking module: Specifically, the causal relationship networking module simultaneously receives standardized datasets from the spatiotemporal data integration module and quality status data from the material quality modeling module. It employs a hybrid adjusted causal quantification algorithm to mine the relationships between data elements. These relationships include: the relationship between disaster-specific characteristics and material reserve distribution; the relationship between material storage environment and material quality degradation; the relationship between material quality status and the quantity of available materials on-site; and the relationship between the quantity of available materials and disaster relief and response. It also covers the direct effects of each element and the multi-layered transmission of these elements, comprehensively reconstructing the influence logic and transmission paths of elements in material reserves, quality evolution, and disaster response. Figure 3 As shown.

[0034] The module uses various independent data elements as basic network nodes and the calculated causal effect strength between elements as the connection weight between nodes. It arranges the positions of all nodes according to the sequence of disaster evolution, material control, and rescue and disposal processes, and connects the corresponding network nodes according to the sequence of the transmission of the effects of the elements. This forms a complete causal knowledge network that can clearly show the influence and transmission path of the elements. The completed causal knowledge network is then transmitted to the counterfactual decision-making module to provide logical support for decision-making.

[0035] The mathematical expression for the hybrid adjusted causal quantization algorithm used in this module is:

[0036] in This represents the numerical value of the causal effect strength calculated between two sets of statistical variables. This indicates the preceding variable that triggers the related change. This indicates a subsequent variable that changes in accordance with the preceding variable. This indicates the value of the variable after manual adjustment. This represents the original baseline value before the variable was adjusted. This represents the set of interference variables identified after screening and selection. Represents the set of interference variables The distribution of the operation operator to obtain the expectation, inner layer It represents the conditional expectation given the values ​​of the variable.

[0037] Specifically, during the computation process, the algorithm first filters out and eliminates the influence of interfering variables, accurately calculates the strength of the true causal effect of the preceding variables on the subsequent variables, avoids misjudgment of the association caused by confounding factors, and improves the accuracy and rationality of the causal knowledge network construction.

[0038] For example, when exploring the causal relationship between the storage environment and the quality degradation of food products, the module sets storage environment-related factors as antecedent variables and food quality degradation-related factors as consequent variables, filtering and eliminating the influence of other irrelevant interfering variables; it accurately calculates the strength of the causal effect between the two through a hybrid adjustment causal quantification algorithm, and uses the calculation results as node connection weights; it treats the storage environment and food quality degradation as independent network nodes, connects the nodes according to the transmission logic of environmental changes affecting material quality changes, and simultaneously completes the causal association construction of multiple sets of factors such as disaster characteristics, reserve arrangement, and scheduling timeliness, ultimately forming a complete causal knowledge network covering the entire process of material reserves.

[0039] Counterfactual decision-making module: Specifically, the counterfactual decision-making module connects to the causal knowledge network generated by the causal relationship networking module and the quality status data output by the material quality modeling module. It uses a physical constraint counterfactual decision-making algorithm to perform multi-dimensional variable calculation and deduction. The calculation dimensions cover material reserve variables, material rotation variables, storage environment variables, material scheduling variables, and disaster scenario variables. The calculation and deduction supports independent assignment of single variables and cross-coupled assignment of multiple variables. It completes the numerical deduction calculation of the combination of variables across the entire domain based on the causal structure parameters of the causal knowledge network and the material quality status data.

[0040] Among them, the calculation of material reserve variables covers the numerical calculation of the total amount of material reserves in a single region, the allocation ratio of reserves of multiple categories of materials, and the proportion of material allocation at reserve locations; the calculation of material rotation variables covers the numerical calculation of the interval between material batch rotations, the execution order of batch rotations, and the ratio of new and old material rotations; the calculation of warehousing environment variables covers the numerical calculation of warehousing temperature, warehousing humidity, and warehouse ventilation frequency; the calculation of material dispatch variables covers the numerical calculation of cross-regional material transportation routes, material allocation time nodes, and the proportion of transportation resource allocation; and the calculation of disaster scenario variables covers the numerical calculation of disaster intensity, disaster impact range, and disaster duration.

[0041] Based on the algorithm's deduction results, the module generates a complete material disposal decision plan. The plan specifically includes five categories: regional material reserve layout and configuration, batch rotation management of inventory materials, cross-regional emergency material allocation and scheduling, warehouse environment adaptation and control, and graded material deployment in disaster scenarios. Each category includes corresponding specific control parameters and execution standards.

[0042] The mathematical expression for the physical constraint counterfactual decision-making algorithm used in this module is:

[0043] in: This represents the statistical value obtained through comprehensive calculation of a single decision-making option. This represents a variable vector formed by the combination of multiple decision-making and control parameters. This represents the numerical value corresponding to the material quality calculated by the model. This represents the numerical value used to measure the strength of the causal effect between variables. The function representing the calculation of the effectiveness of the disposal work. This represents the calculation function corresponding to the total cost of material storage and allocation. This represents the calculation function corresponding to the potential fluctuations during the implementation of the plan. , , Each of the three types of calculation functions corresponds to a set weight ratio value. Weighting coefficient , , The calculation method, using analytic hierarchy process (AHP) normalization, is used to determine the mathematical expression: ; ; ,in The importance score is calculated using the function to determine the effectiveness of the work. The importance score for the function used to calculate the cost of material storage and allocation. The importance score for the calculation function of the creeping wave in the scheme. , , Combining actual emergency response needs with reserve management cost constraints, and through expert scoring and historical data statistics, the following functions were used to quantify the impact weights of three types of calculation functions on the comprehensive statistical value of the decision-making scheme; the calculation function for the effectiveness of the response work. The mathematical expression is: ,in , The weighting coefficients for the sub-items of performance and , To improve the coverage of disaster relief supplies, For emergency response timeliness compliance rate; calculation function for the entire process of material storage and allocation costs. The mathematical expression is: ,in , , The weighting coefficients of each item are as follows: , For material storage management costs, To reduce cross-regional transportation and distribution costs, For labor costs in material scheduling; calculation function for potential fluctuations during plan execution. The mathematical expression is: ,in , , The component fluctuation weighting coefficient and , This represents the fluctuation value of the transportation route. This represents the fluctuation value of the demand for supplies in the disaster area. This represents the fluctuation value of the material quality status.

[0044] Specifically, during the computation process, the algorithm comprehensively considers three dimensions: the effectiveness of the disposal work, the cost of material storage and allocation, and the potential fluctuations in the execution of the plan. It comprehensively scores and ranks different decision-making plans, and selects the plan with the best comprehensive statistical value as the final output material disposal decision plan, ensuring the feasibility, efficiency and stability of the plan.

[0045] For example, in decision-making calculations for responding to regional rainstorm and flood disasters, the module integrates disaster scenario-related variables, material quality status data, and causal knowledge networks to conduct multi-variable cross-coupled calculations and deductions. It comprehensively considers multiple factors such as material reserve allocation, batch rotation rules, cross-regional allocation paths, storage environment parameters, and on-site call standards, and uses algorithms to perform comprehensive calculations and comparisons of multiple schemes. The scheme with the best overall performance is selected to generate a complete material disposal decision-making scheme that includes reserve layout configuration, batch rotation control, cross-regional allocation scheduling, storage environment regulation, and disaster scenario-based hierarchical call, which is then distributed to various reserve warehouses and on-site rescue command terminals for execution.

[0046] Closed-loop feedback update module: Specifically, the closed-loop feedback update module collects operational data generated throughout the entire process of implementing the decision-making scheme, which is output by the counterfactual decision-making calculation module. The collected data fully covers two complete processes: the first is the data of the entire process of material circulation, which includes data on material procurement verification and registration, data on storage in designated warehouse locations, data on the classification and distribution in designated disaster-stricken areas, and data on the consumption of materials used in on-site rescue operations; the second is the data of the entire process of disaster response, which includes data on the time and location of the disaster, data on the confirmation and activation of the emergency response level, data on the step-by-step progress of various on-site rescue operations, and data on the conclusion and handling of the disaster situation.

[0047] The system is divided into two independent feedback layers to correct the model architecture and calculation parameters of each module: The first feedback layer adjusts the model architecture and calculation parameters of the spatiotemporal data integration module, material quality modeling module, causal relationship networking module, and counterfactual decision-making module in real time based on the numerical fluctuations of the operational data acquired through real-time monitoring, ensuring the real-time adaptability of the system; The second feedback layer performs a unified and comprehensive revision and correction of the model architecture and calculation parameters of each module after all the handling procedures for a single disaster have been completed, solidifying optimization experience and improving the accuracy of subsequent system operations.

[0048] For example, during regional earthquake disaster emergency response, the first-level feedback layer monitors in real time changes in road traffic conditions in the disaster area, and the original material allocation routes become impassable. It immediately collects relevant operational data such as on-site road conditions, transport vehicle operation, and material delivery progress, and adjusts the relevant parameters of transport routes, allocation sequence, and capacity allocation in the counterfactual decision-making module in real time to ensure that materials can be delivered smoothly to the disaster area. After all the response operations corresponding to this earthquake disaster are completed, the second-level feedback layer comprehensively summarizes the entire process of material flow, quality degradation, dispatch efficiency, on-site use, and environmental adaptation in this disaster response. It conducts a comprehensive and unified revision and correction of the quality evolution parameters of the material quality modeling module, the correlation weights of the causal relationship networking module, the variable ratios of the counterfactual decision-making module, and the data processing rules of the spatiotemporal data integration module, continuously optimizing the accuracy and adaptability of the overall system operation.

[0049] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A dynamic optimization system for material reserves based on multi-dimensional data fusion, characterized in that, The system includes: Spatiotemporal data integration module: used to build multi-type data docking channels, acquire raw business data from multiple sources and in multiple formats, complete data regularization and processing according to preset specifications, and output standardized datasets in a unified format; Material quality modeling module: retrieves the standardized dataset output by the spatiotemporal data integration module, uses the PINN physical information neural network quality prediction algorithm to build a quality evolution model adapted to different categories of materials, calculates and generates quality status data corresponding to various materials, and transmits the quality status data. Causal Relationship Networking Module: Simultaneously receives the standardized dataset output by the spatiotemporal data integration module and the quality status data output by the material quality modeling module, and uses a hybrid adjustment causal quantification algorithm to mine the correlation between data elements, and builds a complete causal knowledge network based on the correlation; Counterfactual decision-making module: It connects the causal knowledge network generated by the causal relationship networking module and the quality status data output by the material quality modeling module, and uses the physical constraint counterfactual decision algorithm to carry out multi-dimensional variable calculation and deduction, and generates corresponding material disposal decision schemes based on the deduction results; Closed-loop feedback update module: Collects operational data generated throughout the entire process of implementing the decision-making scheme from the counterfactual decision-making module. Based on the collected operational data, it reverse-corrects the model architecture and operational parameters within the spatiotemporal data integration module, material quality modeling module, causal relationship networking module, and counterfactual decision-making module.

2. The dynamic optimization system for material reserves based on multi-dimensional data fusion according to claim 1, characterized in that, The spatiotemporal data integration module establishes data connection channels that connect to databases storing information on the location, scale, and affected area of ​​past disasters, as well as a warehouse management system that records the quantity and location of stored materials; a sensor network deployed in different shelves and storage areas within the warehouse that collects environmental data; a real-time monitoring platform that continuously outputs regional meteorological elements and geological change data; a professional quality inspection system that records various quality inspection values ​​and inspection times of materials; and mobile information collection terminals for disaster site workers to input real-time information.

3. The dynamic optimization system for material reserves based on multi-dimensional data fusion according to claim 1, characterized in that, The spatiotemporal data integration module performs data normalization and processing through multiple sequential processing steps, including: For structured raw data with well-organized field layout, the following steps are performed sequentially: removing duplicate data entries, correcting abnormal values, and filling in missing values ​​in empty positions. For semi-structured raw data with nested hierarchical arrangement, perform overall structure decomposition and layer-by-layer parsing of internal recorded information; For unstructured raw data in the form of text, images, and voice, perform effective information filtering and extraction. After all data processing is completed, the Gregorian calendar time standard is uniformly adopted to define the time statistical benchmark, the plane coordinate system is uniformly adopted to define the spatial positioning coordinates, and all data entries are classified and grouped according to the pre-set classification labels in the system.

4. The dynamic optimization system for material reserves based on multi-dimensional data fusion according to claim 1, characterized in that, The material quality modeling module extracts material category attributes, production batch parameters, outer packaging structure parameters, and warehousing and transportation environment parameters from a standardized dataset. It distinguishes the material change mechanisms corresponding to food materials, pharmaceutical materials, medical device materials, and emergency equipment materials, and matches the change rules corresponding to microbial proliferation changes, effective component degradation changes, metal structure corrosion and deformation changes, and polymer material aging changes. The material change rules are embedded into the calculation framework of the quality evolution model. The PINN physical information neural network quality prediction algorithm is used to complete the iterative fitting of the model's internal parameters. The module calculates the real-time quality level, quality decay rate, and remaining usable time of a single batch of materials at fixed time intervals to obtain the quality status data.

5. The dynamic optimization system for material reserves based on multi-dimensional data fusion according to claim 1, characterized in that, The data element relationships mined by the causal relationship networking module include the relationships formed between disaster characteristics and material reserve layout, the relationships formed between material storage environment and material quality decay, the relationships formed between material quality status and on-site available material quantity, the relationships formed between available material quantity and disaster relief and disposal, as well as the relationships formed by the direct effects of each element and the relationships formed by the multi-layer transmission of each element.

6. The dynamic optimization system for material reserves based on multi-dimensional data fusion according to claim 1, characterized in that, The complete causal knowledge network built by the causal relationship networking module uses various independent data elements as the basic nodes of the network, and uses the intensity of causal effects calculated between elements as the connection weight between nodes. The arrangement of all nodes is arranged according to the disaster evolution process, material control process, and rescue and disposal process. The corresponding network nodes are connected according to the transmission order of the effects of elements, forming an overall network structure that can present the influence of elements on each other.

7. The dynamic optimization system for material reserves based on multi-dimensional data fusion according to claim 1, characterized in that, The material disposal decision-making scheme generated by the counterfactual decision-making module includes regional material reserve layout configuration, inventory material batch rotation management, cross-regional emergency material allocation and dispatch, warehousing environment adaptation and control, and disaster scenario material graded dispatch. The configuration of the regional material reserve layout includes material category allocation parameters, material inventory ratio parameters, and material layout allocation parameters for each reserve warehouse. The inventory batch rotation control includes rotation cycle parameters for different categories of materials, batch rotation sorting rule parameters, and outbound and inbound ratio parameters for rotation materials. The cross-regional emergency material allocation and dispatch includes material transfer route planning parameters, multi-batch material allocation timing parameters, transportation capacity allocation parameters, and material distribution parameters in disaster-stricken areas. The warehouse environment adaptation and control content includes warehouse temperature setting parameters, humidity setting parameters, and environmental ventilation control parameters adapted to the storage requirements of different material qualities. The disaster scenario material tiered allocation content includes material allocation level parameters matching the material quality status, material allocation quantity parameters adapting to different disaster scales, and multi-category material combination allocation ratio parameters.

8. The dynamic optimization system for material reserves based on multi-dimensional data fusion according to claim 1, characterized in that, The counterfactual decision-making module performs multi-dimensional variable calculations and inferences, including calculations of material reserves, material rotation, warehousing environment, material scheduling, and disaster scenarios. Material reserve calculations cover numerical calculations of the total amount of material reserves in a single region, the proportion of reserves of multiple categories of materials, and the allocation ratio of materials at reserve locations. Material rotation calculations cover numerical calculations of the interval between material batch rotations, the execution order of batch rotations, and the ratio of new and old material rotations. Warehouse environment calculations cover numerical calculations of warehouse temperature, warehouse humidity, and warehouse ventilation frequency. Material scheduling calculations cover numerical calculations of cross-regional material transportation routes, material allocation timing nodes, and the allocation ratio of transportation resources. Disaster scenario calculations cover numerical calculations of disaster intensity, disaster impact range, and disaster duration. The calculations and inferences support independent assignment of single variables and cross-coupled assignment of multiple variables. Based on the causal structure parameters of the causal knowledge network and the quality status data output by the material quality modeling module, it completes the numerical inference calculations of the entire domain variable combination.

9. The dynamic optimization system for material reserves based on multi-dimensional data fusion according to claim 1, characterized in that, The closed-loop feedback update module collects operational data that fully covers the entire process of material procurement verification and registration, storage in designated warehouse locations, classification and distribution in designated disaster-stricken areas, and on-site rescue operations. It also fully covers the entire process of disaster occurrence time and location recording, emergency response level confirmation and activation, step-by-step advancement of various on-site rescue operations, and disaster situation stabilization and handling.

10. The dynamic optimization system for material reserves based on multi-dimensional data fusion according to claim 1, characterized in that, The closed-loop feedback update module is divided into two independent feedback levels. The first feedback level adjusts the internal model structure and various calculation parameters of the system in real time based on the numerical fluctuations of the operational data obtained by real-time monitoring. The second feedback level performs a unified and comprehensive revision and correction operation on the internal model structure and various calculation parameters of the system after all the handling operations corresponding to a single disaster have been completed.

Images