A big data-based task allocation management method and system

Abstract

The application discloses a kind of based on big data's task allocation management method and system, it is related to task allocation management technical field, the method includes: obtaining multimodal task demand input, extract cross-domain task element, construct intent graph with weight constraint, and construct resource twin containing implicit ability label by multi-source data fusion, real-time update resource heat map;Based on intent graph weight and resource twin state, using multi-objective collaborative optimization model, solve time, cost and quality Pareto frontier, output cross-domain task chain;Injection extreme disturbance in digital twin environment, combined with reinforcement learning dynamically adjusts resource redundancy strategy, verifies task scheme robustness and outputs emergency plan;Build traceable task flow, real-time monitoring cross-department cooperation state.The application forms "perception-optimization-verification-iteration" closed loop system, significantly improves the overall adaptability and decision reliability of task allocation under complex environment.

Description

Technical Field

[0001] This invention relates to the field of task allocation and management technology, and in particular to a task allocation and management method and system based on big data. Background Technology

[0002] In corporate or factory administration, task allocation in project management aims to coordinate cross-departmental resources and control schedules and costs. Its core is to achieve goal alignment, risk warning, and resource optimization through methods such as task decomposition and responsibility matrices.

[0003] In smart manufacturing and cloud computing environments, systems need to process massive amounts of task requests in real time, including order processing and data analysis, while coordinating human resources from multiple departments or distributed server resources to optimize production efficiency or computing performance. In industries such as retail and finance, task allocation must balance task urgency, resource costs, and the complexity of cross-departmental collaboration; for example, the supply chain department needs to collaborate with production and logistics departments to complete time-sensitive orders. These scenarios require task allocation models to have dynamic responsiveness to cope with fluctuations in task demand, changes in resource status, and unexpected anomalies.

[0004] Existing technologies primarily rely on static rules or simple algorithms for task allocation. For example, traditional systems use basic scheduling algorithms such as First-Come, First-Served (FCFS) and Shortest Job First (SJF), or pre-set fixed task priorities during development, adjusting resource matching strategies through manual intervention. In cross-departmental collaboration, systems allocate tasks proportionally based on departmental workloads, but fail to consider implicit capability differences or dependencies between tasks. Some solutions introduce clustering analysis (such as resource demand grouping under the Hadoop framework) or linear programming to optimize resource distribution, but the calculations are based solely on explicit indicators and do not incorporate real-time behavioral data. Furthermore, centralized optimization requires aggregating global data, which can easily lead to departmental privacy leaks.

[0005] In summary, existing technologies, due to their rigid rules and lack of dynamic analysis capabilities, are unable to support the efficiency and robustness of task allocation in complex scenarios, and there is an urgent need to build a new integrated solution. Summary of the Invention

[0006] Therefore, it is necessary to provide a task allocation management method and system based on big data to address the aforementioned technical problems.

[0007] In a first aspect, the present invention provides a task allocation management method based on big data, the method comprising:

[0008] Acquire multimodal task requirements input, extract cross-domain task elements, construct an intent graph with weight constraints, and construct a resource twin containing implicit capability labels through multi-source data fusion, and update the resource heat map in real time;

[0009] Based on intent graph weights and resource twin states, a multi-objective collaborative optimization model is adopted to solve for the Pareto fronts of time, cost, and quality, and output a cross-domain task chain.

[0010] Injecting extreme perturbations into a digital twin environment, combining reinforcement learning to dynamically adjust resource redundancy strategies, verifying the robustness of the task plan, and outputting contingency plans;

[0011] Construct a traceable task flow, monitor the cross-departmental collaboration status in real time, and trigger cross-domain resource allocation based on zero-knowledge proof when abnormal status exists. Compare the task results with the original intent graph, locate the root cause of the deviation through causal reasoning, and drive incremental training of the intent graph and resource twin.

[0012] Furthermore, obtaining multimodal task requirement input, extracting cross-domain task elements, and constructing an intent graph with weight constraints includes:

[0013] Synchronously capture multimodal task requirements input, generate spatiotemporally aligned raw data streams through timestamp synchronization and spatial coordinate mapping, extract high-order features from the aligned multimodal data, and map them to a unified semantic space to achieve cross-modal semantic alignment;

[0014] The system associates task elements extracted across modalities, dynamically generates weight constraints based on business rules and context, integrates the associated intent elements and weights, generates a constrained intent graph, and verifies its feasibility through logical consistency.

[0015] Furthermore, by associating task elements extracted across modalities, dynamically generating weight constraints based on business rules and context, and integrating the associated intent elements and weights, a constrained intent graph is generated, including:

[0016] Task elements, including text event elements and visual event elements, are extracted from multimodal data respectively. A cross-modal link matrix is ​​constructed, and text event element features, visual event element features, and cross-modal link features are fused to generate a unified event element embedding vector.

[0017] Based on a pre-defined business rule base and real-time context, dynamic weights are assigned to different task elements. The aligned task elements are used as nodes, and the dynamic weights are used as edge attributes. The graph nodes are filled based on the intent quintuple to construct a constrained intent graph.

[0018] Furthermore, by fusing multi-source data, a resource twin containing implicit capability tags is constructed, and the resource heatmap is updated in real time, including:

[0019] Real-time acquisition of full-dimensional status data of physical resources, fusion of three types of data streams—physical, biological, and environmental—after spatiotemporal alignment, extraction of high-order features, and generation of enhanced labels through feature cross-referencing;

[0020] Construct a dynamically evolving latent capability map, integrate collected real-time data with latent capability tags, build an interactive resource twin, and generate a dynamic heat map with predictive capabilities.

[0021] By comparing resource twins with real data in real time, hidden capability degradation or resource conflicts can be identified, triggering self-correction of dynamic heatmap parameters.

[0022] Furthermore, a dynamically evolving latent capability map is constructed, integrating collected real-time data with latent capability tags to create an interactive resource twin, and generating a dynamic heatmap with predictive capabilities, including:

[0023] Based on enhanced labels, a latent capability graph is constructed with task-performing individuals as nodes, quantifying unstructured behaviors, capturing behavioral data in real time, and dynamically adjusting the weights of latent capabilities through reinforcement learning.

[0024] By combining real-time data of physical resources with the embedding vectors of implicit capability graph nodes, an interactive resource twin is constructed, and a real-time message channel is established to trigger the update of resource twin attributes.

[0025] Based on the current state and historical behavior data of the resource twin, a spatiotemporal graph convolutional network is used to predict the future resource load distribution and generate a dynamic heatmap with a time dimension.

[0026] Furthermore, based on intent graph weights and resource twin states, a multi-objective collaborative optimization model is employed to solve for the Pareto fronts of time, cost, and quality, outputting a cross-domain task chain including:

[0027] The global multi-objective optimization is decomposed into distributed solvable subproblems, generating a set of departmental-level subproblems, and a dual-population co-evolutionary framework is adopted to achieve privacy protection.

[0028] Based on the real-time state of the resource twin, an adaptive weight adjustment strategy is adopted to correct the weights of the objective function, solve for the diverse Pareto solution set, and maintain the uniform distribution of the solution set in the objective-resource dual space.

[0029] The non-dominated solutions in the Pareto solution set are analyzed to generate cross-domain task chains with time-series dependencies and resource reservation instructions. The task node order is optimized through reinforcement learning to minimize inter-department waiting time and reserve dynamic buffer resources to cope with task uncertainty.

[0030] Furthermore, the global multi-objective optimization is decomposed into distributively solvable subproblems, generating a departmental-level subproblem set. A dual-population co-evolutionary framework is employed to achieve privacy protection, including:

[0031] Based on the node weight constraints of the intent graph, local objective functions for each department are constructed, the priority of objective functions is dynamically divided, and a set of department-level sub-problems is generated.

[0032] Using homomorphic encryption technology, each department calculates the ciphertext form of the constraints locally, and uses the central node to aggregate the ciphertext constraints to generate the global feasible domain boundary without decrypting the original data;

[0033] Set up a main population and an auxiliary population. The main population initializes a set of solutions locally for each department and solves the objective function. The auxiliary population uses the central node to generate a relaxed constraint problem and searches for the potential feasible region.

[0034] The federated averaging framework is adopted, with each department performing multi-objective optimization locally, uploading the gradient mean of the solution set, and the central node aggregating the noise gradient, updating the global reference solution and broadcasting it to the main population or auxiliary population to guide the search direction.

[0035] Furthermore, extreme perturbations are injected into the digital twin environment, and a resource redundancy strategy is dynamically adjusted using reinforcement learning to verify the robustness of the task plan and output contingency plans, including:

[0036] Based on historical fault data, generate multi-dimensional disturbance scenarios that conform to the actual fault distribution;

[0037] Perturbations are injected layer by layer into the resource twin to construct the perturbation propagation impact chain, quantify the risk of cascading failures, and dynamically adjust the amount of backup resources and real-time redistribution task paths.

[0038] We set robustness quantitative indicators, construct a three-dimensional evaluation space by combining cost change rate, screen the Pareto optimal solution set that balances disturbance resistance and economy, and eliminate fragile solutions.

[0039] The behavior trajectory of the robust optimal solution is analyzed, key response actions are extracted as emergency nodes, and a contingency plan graph with weighted constraints is constructed to support multi-level response triggering.

[0040] Furthermore, a traceable task flow is constructed to monitor the cross-departmental collaboration status in real time. When an abnormal state exists, cross-domain resource reallocation based on zero-knowledge proofs is triggered, including:

[0041] Based on cross-domain task chains, a task flow graph with spatiotemporal stamps is constructed, encapsulating task nodes, resource dependencies, and execution constraints into verifiable units, supporting end-to-end traceability for cross-departmental collaboration;

[0042] By using resource twins to collect cross-departmental collaboration indicators in real time, and through multi-dimensional monitoring, we can predict potential anomalies in the task execution process and trigger zero-knowledge proof verification.

[0043] When an anomaly is detected, the responsible department is required to generate a zero-knowledge proof to verify the authenticity of the resource status without disclosing the original data. Based on the verification results, the resource allocation strategy is dynamically adjusted to generate a minimum disturbance relocation plan, which is then injected into the task flow for execution.

[0044] Secondly, the present invention also provides a task allocation management system based on big data, the system comprising:

[0045] The cross-modal perception module is used to acquire multimodal task requirement input, extract cross-domain task elements, construct an intent graph with weight constraints, and construct a resource twin containing implicit capability labels through multi-source data fusion, and update the resource heat map in real time.

[0046] The collaborative optimization module is used to solve for the Pareto fronts of time, cost, and quality based on intent graph weights and resource twin states, using a multi-objective collaborative optimization model, and outputting a cross-domain task chain.

[0047] The scheduling and simulation module is used to inject extreme disturbances into the digital twin environment, combine reinforcement learning to dynamically adjust resource redundancy strategies, verify the robustness of the task plan, and output emergency plans.

[0048] The execution traceability module is used to build a traceable task flow, monitor the cross-departmental collaboration status in real time, and trigger cross-domain resource allocation based on zero-knowledge proof when an abnormal status exists. It also compares the task results with the original intent graph, locates the root cause of the deviation through causal reasoning, and drives incremental training of the intent graph and resource twin.

[0049] The beneficial effects of this invention are as follows: Intelligent decision-making and dynamic adaptation for cross-domain tasks are achieved through multimodal fusion and multi-objective collaborative mechanisms; in the task planning stage, multimodal task requirements are spatiotemporally aligned and semantically mapped to generate a weighted intent graph, solving the strategy rigidity problem caused by static weights in traditional systems; in the resource modeling stage, implicit capability graphs are constructed by integrating physical, biological, and environmental data, generating resource twins with predictive heatmaps, breaking through the limitations of traditional monitoring that only focuses on explicit indicators; in the execution stage, resilient task chains are generated by combining Pareto front solving and perturbation pre-simulation, and privacy-preserving cross-domain resource allocation is achieved through zero-knowledge proofs; finally, incremental training of the intent graph and resource twins is driven by causal reasoning feedback, forming a closed-loop system of "perception-optimization-verification-iteration," significantly improving the overall adaptability and decision reliability of task allocation in complex environments. Attached Figure Description

[0050] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0051] Figure 1 This is a flowchart of a task allocation and management method based on big data according to an embodiment of the present invention;

[0052] Figure 2 This is a system principle block diagram of a task allocation and management system based on big data according to an embodiment of the present invention.

[0053] The reference numerals are: 1. Cross-modal perception module; 2. Collaborative optimization module; 3. Scheduling pre-simulation module; 4. Execution tracing module. Detailed Implementation

[0054] 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 and not intended to limit the invention.

[0055] Please see Figure 1 This paper provides a task allocation and management method based on big data, which includes:

[0056] S1. Obtain multimodal task requirement input, extract cross-domain task elements, construct an intent graph with weight constraints, and construct a resource twin containing implicit capability labels through multi-source data fusion, and update the resource heat map in real time.

[0057] In the description of this invention, the process of obtaining multimodal task requirement input, extracting cross-domain task elements, constructing an intent graph with weight constraints, and constructing a resource twin containing implicit capability tags through multi-source data fusion, and updating the resource heatmap in real time includes:

[0058] S11. Synchronously capture the input of multimodal task requirements, generate a spatiotemporally aligned raw data stream by synchronizing timestamps and mapping spatial coordinates, extract high-order features from the aligned multimodal data, and map them to a unified semantic space to achieve cross-modal semantic alignment.

[0059] Specifically, it simultaneously captures multimodal inputs such as natural language, images, and speech to solve the problems of data heterogeneity and temporal misalignment, generating spatiotemporally aligned raw data streams. For example, in a meeting, a user's voice command "Complete the launch of a new product in a certain region within two weeks" (voice modality) needs to be aligned on the timeline with the market plan chart (image modality) and task list document (text modality) in the PPT.

[0060] S12. Associate the task elements extracted across modalities, dynamically generate weight constraints based on business rules and context, integrate the associated intent elements and weights, generate a constrained intent graph, and verify its feasibility through logical consistency.

[0061] In the description of this invention, associating task elements extracted across modalities, dynamically generating weight constraints based on business rules and context, and integrating the associated intent elements and weights to generate a constrained intent graph includes:

[0062] S121. Extract the task elements from the multimodal data, including text event elements (including trigger words, physical device entities, etc.) and visual event elements (including target location box coordinates, event types, etc.), construct a cross-modal link matrix, and integrate the text event element features, visual event element features and cross-modal link features. Eliminate modal differences through common semantic space projection to generate a unified event element embedding vector, providing an alignment basis for dynamic weight allocation.

[0063] Specifically, in the process of constructing the cross-modal link matrix, the image is first divided into blocks, and the input image is segmented into blocks of size. p ×size p The image is divided into pixel blocks, and the ViT (Vision Transformer) feature vector of each block is extracted. Then the text sequence is encoded, and word example feature vectors are generated through BERT word embedding. Finally, a binary link matrix is ​​constructed with image blocks as rows and text examples as columns. The matrix value is 1 when there is a semantic association between the pixel block and the example, and 0 otherwise, realizing cross-modal association between image and text.

[0064] S122. Based on the preset business rule base and real-time context, assign dynamic weights to different task elements, use the aligned task elements as nodes and the dynamic weights as edge attributes, fill the graph nodes based on the intent quintuple, and construct a constrained intent graph.

[0065] Specifically, the intent graph is a structured semantic network model with hyperedges as its core design element. It integrates entities, actions, and associated constraints from multimodal task requirements to form a computable intent logic framework. Nodes represent entities or actions related to task execution, edges describe the static relationships between entities, and hyperedges aggregate multiple nodes to form a complete intent unit (e.g., the intent "emergency fault diagnosis" might include nodes such as device number, fault code, and processing time limit). The design aims to solve the problems of semantic fragmentation and implicit constraints in traditional task requirement descriptions. Through structured associations, it transforms loosely coupled multimodal inputs into machine-parseable and logically complete intent expressions, supporting dynamic weight adjustment and conflict resolution.

[0066] In this invention, the intent graph serves a dual role as a carrier of task logic and a hub for dynamic decision-making, converging the complexity of multimodal inputs into a quantifiable and evolvable intent network. This preserves the semantic integrity of the original requirements while providing constrained optimization objectives for task allocation.

[0067] S13. Real-time collection of full-dimensional status data of physical resources (equipment, manpower, environment), fusion of physical, biological and environmental data streams after spatiotemporal alignment, extraction of high-order features (including equipment health, employee innovation potential, etc.), and generation of enhanced labels through feature cross-referencing.

[0068] Specifically, physical resources are collected through the deployment of multimodal sensor networks, such as vibration sensors and infrared thermal imagers at key nodes of equipment, and the transmission of real-time data via the MQTT protocol; the energy ratio of employees' beta waves to theta waves is monitored by wearable EEG caps to define fatigue coefficients, such as fatigue index = theta wave energy / beta wave energy, thus obtaining employees' biosignals; and key constraints are extracted by accessing government policy APIs and using NLP entity recognition to convert them into resource availability correction factors, thereby obtaining environmental risks.

[0069] By integrating physical, biological, and environmental data streams, high-order features are extracted, and enhanced labels are generated through feature cross-referencing, providing structured input for the construction of resource twins.

[0070] For example, for physical resources, wavelet packet decomposition of equipment vibration signals is used to extract kurtosis coefficient and envelope entropy, and the equipment health score is synthesized. The formula is: Health score = 1 - (0.6 × kurtosis + 0.4 × envelope entropy).

[0071] For human resources, we analyze the frequency of GitHub code commits × the frequency of innovative words in conference speeches, and combine this with the energy of alpha waves from brainwaves to generate an innovation potential index. The formula is: Innovation Index = 0.4 × Code Innovation + 0.3 × Speech Innovation + 0.3 × Alpha Wave Energy.

[0072] Ultimately, environmental policy factors are linked with physical characteristics to generate enhanced labels with environmental modifications.

[0073] S14. Construct a dynamically evolving latent capability map, integrate the collected real-time data with latent capability tags, build an interactive resource twin, and generate a dynamic heat map with predictive capabilities.

[0074] In the description of this invention, constructing a dynamically evolving latent capability map, integrating collected real-time data with latent capability tags, constructing an interactive resource twin, and generating a dynamic heatmap with predictive capabilities includes:

[0075] S141. Based on enhanced labels, construct a latent capability graph with task-performing individuals as nodes, quantify unstructured behaviors, capture behavioral data in real time, and dynamically adjust the weights of latent capabilities through reinforcement learning.

[0076] Specifically, the implicit capability map is a dynamic graph structure that uses individual task performers as nodes to build a capability relationship network by quantifying unstructured behaviors (such as teamwork, innovation potential, and fault response agility). The design aims to transform implicit capability elements (such as dynamic changes in employee skill proficiency and equipment operation fluency) that are difficult for traditional monitoring systems to capture into structured and quantifiable indicators, thus overcoming the limitations of explicit indicators.

[0077] In this invention, the implicit capability graph serves as the cognitive basis for resource scheduling. By dynamically adjusting node weights through reinforcement learning, it provides an accurate capability profile for task allocation, ensuring that high-difficulty tasks are automatically assigned to individuals with high innovation capabilities, and reflecting capability decline in real time.

[0078] The nodes and relationships in the latent capability graph are defined as follows:

[0079] 1. Node: The task execution individual (personnel / equipment) serves as the core node, with attributes including static attributes (skill certificates, job responsibilities) and dynamic attributes (real-time generated enhanced tags).

[0080] 2. Relationship edges: Collaborative relationships, with weights determined by co-occurrence frequency and task success rate; Capability inheritance relationships, with similarity quantified through graph embedding technology.

[0081] By employing deep reinforcement learning and designing a reward function, such as increasing skill weights for improved task completion speed, the node weight parameters are dynamically adjusted so that the implicit ability graph reflects the trend of ability changes in real time.

[0082] S142. By combining real-time data of physical resources with the embedding vectors of implicit capability graph nodes, an interactive resource twin is constructed, and a real-time message channel is established to trigger the update of resource twin attributes.

[0083] Specifically, the resource twin is a virtual model constructed by fusing real-time physical resource data with node embedding vectors from an implicit capability graph. The design aims to break down the barriers between physical resource status and capability data, forming a dynamically controllable virtual-physical fusion that supports real-time interaction and strategy verification.

[0084] In this invention, the resource twin serves as a virtual-physical collaborative hub for task execution: a real-time message channel is established through the Internet of Things protocol, and the twin's attributes are automatically updated when the physical resource status changes or the implicit capabilities fluctuate; at the same time, users are allowed to manually adjust parameters through a visual interface to achieve two-way closed-loop control of "physical status-virtual strategy".

[0085] S143. Based on the current state and historical behavior data of the resource twin, a spatiotemporal graph convolutional network is used to predict the future resource load distribution and generate a dynamic heat map with a time dimension.

[0086] Specifically, the dynamic heatmap is a resource load distribution prediction map rendered based on spatiotemporal dimensions. It uses color gradients (green → red) to map resource utilization and conflict probability in future time slices. Its design aims to integrate spatial location and temporal decay factors to intuitively display resource bottleneck risks.

[0087] In this invention, dynamic heatmaps serve as predictive decision-making tools: based on the real-time status and historical behavior data of resource twins, spatiotemporal graph convolutional networks are used to model node adjacency relationships (spatial) and time-series loads (temporal), outputting future load distributions; when the predicted value exceeds a safety threshold, high-risk areas are automatically marked and optimization suggestions are pushed, driving dynamic buffer resource allocation, thereby balancing resource utilization.

[0088] S15. By comparing the resource twin with real data in real time, identify hidden capability degradation or resource conflicts, and trigger the self-correction of dynamic heatmap parameters.

[0089] S2. Based on intent graph weights and resource twin states, a multi-objective collaborative optimization model is adopted to solve for the Pareto fronts of time, cost, and quality, and output a cross-domain task chain.

[0090] Specifically, in step S2, the weight constraints of task elements are extracted from the intent map first, and then converted into the objective function priority of departmental sub-problems. For example, high-weight tasks are assigned higher cost or time tolerance in the optimization process to ensure that critical tasks are executed first.

[0091] The system acquires the physical state and implicit capability status of the resource twin in real time, and dynamically adjusts the weights of the objective function. For example, when the twin detects a decline in the health of a device, it automatically reduces the "quality" weight of the tasks assigned to it to avoid overloading.

[0092] In the solution process, time, cost, and quality are the core conflicting objectives of cross-domain tasks. Among them, the conflict between time and cost means that shortening the task cycle requires increased resource investment, which drives up costs; conversely, it will delay delivery. The conflict between quality and time / cost means that high-precision tasks require more verification time and high-cost equipment, while compressing time or cost may reduce the pass rate.

[0093] Therefore, by solving the Pareto front (i.e., the boundary formed by all non-dominated solutions), a set of optimal trade-offs can be provided, avoiding the local optimum trap of traditional single-objective optimization and ensuring that the solution set covers all possible trade-off choices.

[0094] In the description of this invention, based on intent graph weights and resource twin states, a multi-objective collaborative optimization model is employed to solve for the Pareto fronts of time, cost, and quality, outputting a cross-domain task chain including:

[0095] S21. Decompose the global multi-objective optimization into distributed solvable subproblems, generate a department-level subproblem set, and adopt a dual-population co-evolutionary framework to achieve privacy protection.

[0096] In the description of this invention, global multi-objective optimization is decomposed into distributively solvable subproblems, generating a departmental-level subproblem set, and a dual-population co-evolutionary framework is adopted to achieve privacy protection, including:

[0097] S211. Based on the node weight constraints of the intent graph, construct the local objective function for each department, dynamically divide the priority of the objective function, and generate a set of department-level sub-problems.

[0098] S212. Using homomorphic encryption technology, each department calculates the ciphertext form of the constraints locally, and uses the central node to aggregate the ciphertext constraints to generate the global feasible domain boundary without decrypting the original data.

[0099] S213. Define a primary population and an auxiliary population. The primary population initializes a set of solutions locally for each department and solves the objective function. The auxiliary population uses the central node to generate a relaxed constraint problem and searches for potential feasible regions.

[0100] S214. Using a federated averaging framework, each department performs multi-objective optimization locally, uploads the gradient mean of the solution set, aggregates the noise gradient at the central node, updates the global reference solution, and broadcasts it to the main or auxiliary population to guide the search direction.

[0101] S22. Based on the real-time state of the resource twin, an adaptive weight adjustment strategy is adopted to correct the weights of the objective function, solve for the diversity Pareto solution set, and maintain the uniform distribution of the solution set in the objective-resource dual space.

[0102] S23. Analyze the non-dominated solutions in the Pareto solution set, generate a cross-domain task chain with time-series dependencies and resource reservation instructions, optimize the task node order through reinforcement learning, minimize the waiting time between departments, and reserve dynamic buffer resources to cope with task uncertainty.

[0103] Specifically, a cross-domain task chain is a set of cross-departmental collaborative processes with time-series dependencies and resource reservation instructions, such as a chained task sequence of "procurement department reserving materials → production department debugging equipment → quality inspection department synchronously accepting the equipment." Its core characteristics include:

[0104] 1. Temporal Dependency: Task nodes need to be executed in logical order. Reinforcement learning is used to optimize the node order and reduce waiting gaps between departments.

[0105] 2. Resource reservation instruction: Based on the predicted status of the resource twin, resources are locked in advance to avoid conflicts during execution.

[0106] 3. Dynamic buffering mechanism: Reserves elastic resources to cope with task uncertainty and prevent chain interruptions.

[0107] S3. Inject extreme perturbations into the digital twin environment, combine reinforcement learning to dynamically adjust resource redundancy strategies, verify the robustness of the task plan, and output contingency plans.

[0108] In the description of this invention, injecting extreme perturbations into a digital twin environment, combining reinforcement learning to dynamically adjust resource redundancy strategies, verifying the robustness of the task scheme, and outputting contingency plans include:

[0109] S31. Based on historical fault data, generate a multi-dimensional disturbance scenario that conforms to the actual fault distribution.

[0110] Specifically, multidimensional disturbance scenarios refer to a set of composite disturbance events covering the physical layer, capability layer, and task layer, generated by integrating the spatiotemporal distribution characteristics of historical fault data (such as equipment downtime records and supply chain disruption events).

[0111] S32. Inject disturbances in the resource twin in layers, construct the disturbance propagation impact chain, quantify the risk of cascading failures, and dynamically adjust the amount of backup resources and real-time redistribute task paths.

[0112] Specifically, the disturbance propagation impact chain is a dynamic graph model that quantifies the risk of cascading failures. It constructs the impact topology by analyzing the cross-layer propagation path of disturbances in the resource twin (e.g., equipment failure → adjacent equipment overload → production line shutdown). The disturbance propagation impact chain addresses the problem of delayed response to implicitly related failures in traditional methods by using a graph propagation algorithm to calculate the probability of failure propagation between nodes in real time and triggering a dynamic adjustment mechanism based on the risk value.

[0113] S33. Set robustness quantitative indicators, construct a three-dimensional evaluation space in combination with cost change rate, screen the Pareto optimal solution set that balances disturbance resistance and economy, and eliminate fragile solutions.

[0114] Specifically, robustness quantification indicators are a multi-dimensional parameter system for evaluating a system's ability to withstand disturbances. The core indicators include the variance of task completion rate fluctuation (stability), the rate of change of redundant resource costs (economic efficiency), and the maximum tolerable disturbance intensity threshold (stress resistance).

[0115] S34. Analyze the behavior trajectory of the robust optimal solution, extract key response actions as emergency nodes, construct a contingency plan graph with weight constraints, and support multi-level response triggering.

[0116] Specifically, the weighted contingency plan graph is an emergency action logic network built based on a spatiotemporal graph convolutional network (ST-GCN). Nodes represent key response actions, edges represent action dependencies, and edge weights are determined by the product of trigger probability and action cost. The contingency plan graph transforms the behavioral trajectory of the robust optimal solution into hierarchical response rules, balancing response speed and the risk of over-intervention through weight thresholds.

[0117] S4. Construct a traceable task flow, monitor the cross-departmental collaboration status in real time, and trigger cross-domain resource allocation based on zero-knowledge proof when there is an abnormal status. Compare the task results with the original intent graph, locate the root cause of the deviation through causal reasoning, and drive incremental training of the intent graph and resource twin.

[0118] In the description of this invention, a traceable task flow is constructed, and the cross-departmental collaboration status is monitored in real time. When an abnormal status exists, triggering cross-domain resource reallocation based on zero-knowledge proof includes:

[0119] S41. Based on cross-domain task chains, construct a task flow graph with spatiotemporal stamps, encapsulate task nodes, resource dependencies, and execution constraints into verifiable units, and support end-to-end traceability for cross-departmental collaboration.

[0120] S42. Utilize resource twins to collect cross-departmental collaboration indicators in real time, and predict potential anomalies during task execution through multi-dimensional monitoring, triggering zero-knowledge proof verification.

[0121] S43. When an anomaly is detected, the responsible department is required to generate a zero-knowledge proof to verify the authenticity of the resource status without disclosing the original data. Based on the verification results, the resource allocation strategy should be dynamically adjusted to generate a minimum disturbance relocation plan, which should then be injected into the task flow for execution.

[0122] Please see Figure 2 It also provides a task allocation management system based on big data, which includes:

[0123] The cross-modal perception module 1 is used to acquire multimodal task requirement input, extract cross-domain task elements, construct an intent graph with weight constraints, and construct a resource twin containing implicit capability labels through multi-source data fusion, and update the resource heat map in real time.

[0124] Collaborative optimization module 2 is used to solve for the Pareto fronts of time, cost, and quality based on intent graph weights and resource twin states, using a multi-objective collaborative optimization model, and outputting a cross-domain task chain.

[0125] The scheduling and simulation module 3 is used to inject extreme disturbances into the digital twin environment, combine reinforcement learning to dynamically adjust resource redundancy strategies, verify the robustness of the task plan, and output contingency plans.

[0126] The execution traceability module 4 is used to build a traceable task flow, monitor the cross-departmental collaboration status in real time, and trigger cross-domain resource allocation based on zero-knowledge proof when there is an abnormal status. It also compares the task results with the original intent graph, locates the root cause of the deviation through causal reasoning, and drives incremental training of the intent graph and resource twin.

[0127] In summary, by leveraging the technical solutions described above, intelligent decision-making and dynamic adaptation for cross-domain tasks are achieved through multimodal fusion and multi-objective collaborative mechanisms. In the task planning phase, multimodal task requirements are spatiotemporally aligned and semantically mapped to generate a weighted intent graph, addressing the strategy rigidity problem caused by static weights in traditional systems. In the resource modeling phase, physical, biological, and environmental data are integrated to construct an implicit capability graph, generating a resource twin containing predictive heatmaps, overcoming the limitation of traditional monitoring focusing only on explicit indicators. In the execution phase, Pareto frontier solutions and perturbation simulations are combined to generate resilient task chains, and privacy-preserving cross-domain resource allocation is achieved through zero-knowledge proofs. Finally, incremental training of the intent graph and resource twin is driven by causal reasoning feedback, forming a closed-loop system of "perception-optimization-verification-iteration," significantly improving the overall adaptability and decision reliability of task allocation in complex environments.

[0128] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

Claims

1. A task allocation and management method based on big data, characterized in that, include: S1. Synchronously capture multimodal task requirements input, generate spatiotemporally aligned raw data streams through timestamp synchronization and spatial coordinate mapping, extract high-order features from the aligned multimodal data, and map them to a unified semantic space to achieve cross-modal semantic alignment. The system associates task elements extracted across modalities, dynamically generates weight constraints based on business rules and context, integrates the associated intent elements and weights to generate a constrained intent graph, and verifies feasibility through logical consistency. Simultaneously, it collects real-time, multi-dimensional status data of physical resources, merges spatiotemporally aligned physical, biological, and environmental data streams, extracts high-order features, and generates enhanced labels through feature cross-referencing. It constructs a dynamically evolving latent capability graph, integrates the collected real-time data and latent capability labels to build an interactive resource twin, and generates a predictive dynamic heatmap. Through real-time comparison between the resource twin and real data, it identifies latent capability degradation or resource conflicts, triggering self-correction of the dynamic heatmap parameters. S2. Based on intent graph weights and resource twin states, global multi-objective optimization is decomposed into distributively solvable subproblems, generating a set of departmental subproblems. A dual-population co-evolutionary framework is adopted to achieve privacy protection. Based on the real-time state of the resource twin, an adaptive weight adjustment strategy is used to correct the objective function weights, solving for a diverse Pareto solution set and maintaining a uniform distribution of the solution set in the objective-resource dual space. Non-dominated solutions in the Pareto solution set are analyzed to generate a cross-domain task chain with temporal dependencies and resource reservation instructions. The task node order is optimized through reinforcement learning to minimize inter-departmental waiting time and reserve dynamic buffer resources to cope with task uncertainty. S3. Based on historical fault data, generate multi-dimensional disturbance scenarios that conform to the actual fault distribution; inject disturbances in layers into the resource twin, construct the disturbance propagation impact chain, quantify the risk of cascading failures, and dynamically adjust the amount of backup resources and real-time reallocation task paths. We set robustness quantification indicators, construct a three-dimensional evaluation space by combining cost change rate, screen Pareto optimal solution set that balances disturbance resistance and economy, and eliminate vulnerable solutions; analyze the behavior trajectory of robust optimal solution, extract key response actions as emergency nodes, construct a contingency plan map with weight constraints, and support multi-level response triggering. S4. Based on cross-domain task chains, construct a task flow graph with spatiotemporal stamps, encapsulating task nodes, resource dependencies, and execution constraints into verifiable units to support end-to-end traceability for cross-departmental collaboration; utilize resource twins to collect cross-departmental collaboration indicators in real time, and predict potential anomalies during task execution through multi-dimensional monitoring, triggering zero-knowledge proof verification; when an anomaly is detected, the responsible department is required to generate a zero-knowledge proof to verify the authenticity of the resource status without leaking the original data, and based on the verification results, dynamically adjust the resource allocation strategy, generate a minimum disturbance adjustment scheme, and inject the adjustment scheme into the task flow execution; compare the task results with the original intent graph, locate the root cause of the deviation through causal reasoning, and drive incremental training of the intent graph and resource twins.

2. A task allocation and management method based on big data according to claim 1, characterized in that, The task elements extracted across modalities are associated, and weight constraints are dynamically generated based on business rules and context. The associated intent elements and weights are then integrated to generate a constrained intent graph, including: Task elements, including text event elements and visual event elements, are extracted from multimodal data respectively. A cross-modal link matrix is ​​constructed, and text event element features, visual event element features, and cross-modal link features are fused to generate a unified event element embedding vector. Based on a pre-defined business rule base and real-time context, dynamic weights are assigned to different task elements. The aligned task elements are used as nodes, and the dynamic weights are used as edge attributes. The graph nodes are filled based on the intent quintuple to construct a constrained intent graph.

3. A task allocation and management method based on big data according to claim 1, characterized in that, Constructing a dynamically evolving latent capability map, integrating collected real-time data with latent capability tags, building an interactive resource twin, and generating a dynamic heatmap with predictive capabilities includes: Based on enhanced labels, a latent capability graph is constructed with task-performing individuals as nodes, quantifying unstructured behaviors, capturing behavioral data in real time, and dynamically adjusting the weights of latent capabilities through reinforcement learning. By combining real-time data of physical resources with the embedding vectors of implicit capability graph nodes, an interactive resource twin is constructed, and a real-time message channel is established to trigger the update of resource twin attributes. Based on the current state and historical behavior data of the resource twin, a spatiotemporal graph convolutional network is used to predict the future resource load distribution and generate a dynamic heatmap with a time dimension.

4. A task allocation and management method based on big data according to claim 1, characterized in that, Global multi-objective optimization is decomposed into distributively solvable subproblems, generating department-level subproblem sets. A dual-population co-evolutionary framework is employed to ensure privacy, including: constructing local objective functions for each department based on intent graph node weight constraints, dynamically prioritizing objective functions, and generating department-level subproblem sets; utilizing homomorphic encryption, each department locally computes the ciphertext form of constraints, and a central node aggregates the ciphertext constraints to generate the global feasible region boundary without decrypting the original data; defining a primary population and an auxiliary population, with the primary population initializing a set of solutions locally for each department and solving the objective function, while the auxiliary population uses the central node to generate relaxation constraint problems and search for potential feasible regions; and employing a federated averaging framework, each department locally performs multi-objective optimization, uploads the gradient mean of the solution set, and the central node aggregates noisy gradients, updates the global reference solution, and broadcasts it to the primary or auxiliary population to guide the search direction.

5. A task allocation management system based on big data, used to implement the task allocation management method based on big data according to any one of claims 1-4, characterized in that, include: The cross-modal perception module is used to synchronously capture the input requirements of multimodal tasks. It generates a spatiotemporally aligned raw data stream by synchronizing timestamps and mapping spatial coordinates. It extracts high-order features from the aligned multimodal data and maps them to a unified semantic space to achieve cross-modal semantic alignment. The system associates task elements extracted across modalities, dynamically generates weight constraints based on business rules and context, integrates the associated intent elements and weights to generate a constrained intent graph, and verifies feasibility through logical consistency. Simultaneously, it collects real-time, multi-dimensional status data of physical resources, merges spatiotemporally aligned physical, biological, and environmental data streams, extracts high-order features, and generates enhanced labels through feature cross-referencing. It constructs a dynamically evolving latent capability graph, integrates the collected real-time data and latent capability labels to build an interactive resource twin, and generates a predictive dynamic heatmap. Through real-time comparison between the resource twin and real data, it identifies latent capability degradation or resource conflicts, triggering self-correction of the dynamic heatmap parameters. The collaborative optimization module decomposes global multi-objective optimization into distributively solvable subproblems based on intent graph weights and resource twin states, generating a set of departmental subproblems. It employs a dual-population co-evolutionary framework to protect privacy. Based on the real-time state of the resource twins, it uses an adaptive weight adjustment strategy to correct the objective function weights, solving for a diverse Pareto solution set while maintaining a uniform distribution of the solution set across the objective-resource dual space. It analyzes the non-dominated solutions in the Pareto solution set, generating a cross-domain task chain with temporal dependencies and resource reservation instructions. Through reinforcement learning, it optimizes the task node order, minimizes inter-departmental waiting time, and reserves dynamic buffer resources to address task uncertainty. The scheduling simulation module is used to generate multi-dimensional disturbance scenarios that conform to the actual fault distribution based on historical fault data; Perturbations are injected layer by layer into the resource twin to construct the perturbation propagation impact chain, quantify the risk of cascading failures, and dynamically adjust the amount of backup resources and real-time redistribution task paths. We set robustness quantification indicators, construct a three-dimensional evaluation space by combining cost change rate, screen Pareto optimal solution set that balances disturbance resistance and economy, and eliminate vulnerable solutions; analyze the behavior trajectory of robust optimal solution, extract key response actions as emergency nodes, construct a contingency plan map with weight constraints, and support multi-level response triggering. The execution tracing module is used to construct a task flow graph with spatiotemporal stamps based on cross-domain task chains. It encapsulates task nodes, resource dependencies, and execution constraints into verifiable units, supporting end-to-end tracing for cross-departmental collaboration. It uses a resource twin to collect cross-departmental collaboration indicators in real time, and through multi-dimensional monitoring, it predicts potential anomalies in the task execution process and triggers zero-knowledge proof verification. When an anomaly is detected, the responsible department is required to generate a zero-knowledge proof to verify the authenticity of the resource status without disclosing the original data. Based on the verification results, the resource allocation strategy is dynamically adjusted to generate a minimum disturbance adjustment plan, which is then injected into the task flow execution. By comparing the task results with the original intent graph, the root cause of the deviation is located through causal reasoning, driving incremental training of the intent graph and the resource twin.

Images