Enterprise management resource dynamic allocation intelligent system

Abstract

The invention relates to the industrial Internet of Things crossing field, and discloses an enterprise management resource dynamic allocation intelligent system, which constructs a closed-loop system of multi-modal data fusion, hierarchical decision optimization, elastic resource allocation and edge-cloud cooperative execution: through cross-modal comparison learning, association policy texts and equipment operation data are learned, and data islands are eliminated; a Pareto frontier search and reinforcement learning collaborative architecture is adopted to realize multi-target dynamic balance; establishing a resource game allocation model based on the Shapley value contribution degree quantization and non-negative matrix factorization; and industrial-grade real-time control is achieved by combining knowledge distillation compression with a federal feedback mechanism. The problems of data dimension splitting, single decision target, rigid distribution mechanism and serious response delay in the prior art are solved, the resource matching precision is remarkably improved, the vacancy rate is reduced, and the real-time response capability in an emergency scene is guaranteed.

Description

Technical Field

[0001] The present invention relates to the cross-technical field of industrial Internet of Things, and in particular to an intelligent system for dynamically allocating enterprise management resources. Background Art

[0002] Enterprise management currently faces complex challenges in dynamic resource allocation. Existing systems suffer from a single data perception dimension, disorganized allocation mechanisms, and delayed execution responses, leading to resource misallocation, uncontrolled risks, and reduced efficiency.

[0003] Data silos lead to biased decision-making. Most systems only process structured business data (such as ERP order records) and lack the ability to parse unstructured information such as policy documents and equipment logs. This data fragmentation distorts resource portraits and makes decision-making decisions one-sided.

[0004] Single-objective optimization amplifies systemic risk. Mainstream scheduling algorithms focus on optimizing a single metric (such as cost reduction) while neglecting the dynamic balance of multiple objectives. Traditional models lack multi-objective coordination mechanisms and are unable to adapt to complex constraint scenarios.

[0005] Static allocation exacerbates resource conflicts. Fixed priority rules (such as first-come, first-served) make it difficult to quantify the dynamic value of resources. This mechanical allocation leads to resource competition between departments, reducing overall efficiency.

[0006] Centralized architectures hinder real-time response. Cloud-based decision-making models face communication latency bottlenecks, while edge devices lack autonomous decision-making capabilities. The rigid demand for real-time control in industrial scenarios is difficult to meet.

[0007] Therefore, the present invention proposes an enterprise management resource dynamic allocation intelligent system to solve the shortcomings of the existing technology Summary of the Invention

[0008] To address the shortcomings of existing technologies, a multimodal data fusion module is used to dynamically associate policy texts, device status, and market data. A hierarchical decision-making module is used to construct a collaborative mechanism of global multi-objective optimization and edge reinforcement learning. The elastic resource allocation module is used to quantify resource contributions and dynamically reconstruct priorities. The edge-cloud collaborative execution module is combined to achieve lightweight policy deployment and real-time feedback control.

[0009] To achieve the above objectives, the present invention is implemented through the following technical solutions: an intelligent system for dynamically allocating enterprise management resources, comprising:

[0010] Multimodal data fusion module, used to integrate internal and external structured data and unstructured data of the enterprise to generate a dynamic resource relationship map;

[0011] A dynamic hierarchical decision module is in communication with the multimodal data fusion module, receives the dynamic resource relationship graph, and generates a resource allocation strategy through a two-layer framework of global coordination and local execution;

[0012] A flexible resource allocation module, in communication with the dynamic hierarchical decision module, dynamically allocates resources based on the priority matrix and game theory model according to the resource allocation strategy;

[0013] The edge-cloud collaborative execution module is in communication with the elastic resource allocation module, executes resource allocation instructions in real time through edge nodes, and feeds back the execution results to the multimodal data fusion module to form a closed-loop optimization system.

[0014] Preferably, the output end of the multimodal data fusion module is connected to the input end of the dynamic hierarchical decision module through a data bus to transmit a dynamic resource relationship map; the output end of the elastic resource allocation module is connected to the input end of the edge-cloud collaborative execution module through a message queue to transmit resource allocation instructions.

[0015] Preferably, the global agent and local agents of the dynamic hierarchical decision module communicate through a distributed decision network, the global agent sends multi-objective coordination weight parameters to the local agents, and the local agents feed back department-level execution status data to the global agent.

[0016] Preferably, the elastic resource allocation module receives the global policy from the dynamic hierarchical decision module and generates resource preemption instructions according to the real-time priority matrix, and the resource preemption instructions are distributed to the edge-cloud collaborative execution module through a control signal.

[0017] Preferably, the cloud hub of the edge-cloud collaborative execution module periodically sends a data update request to the multimodal data fusion module, and the multimodal data fusion module responds to the request and provides an incremental data set to update the dynamic resource relationship map.

[0018] Preferably, the multimodal data fusion module includes:

[0019] The cross-modal contrastive learning unit aligns numerical data and semantic features through a contrastive loss function, which is:

[0020]

[0021] in:

[0022] represents the latent space representation of the i-th numerical data;

[0023] Represents the latent space representation of the j-th semantic data;

[0024] τ>0 is the temperature coefficient, which is used to adjust the sharpness of the similarity distribution;

[0025] is a set of matched positive sample pairs (numerical data and corresponding semantic description);

[0026] is a set of randomly sampled negative sample pairs;

[0027] Dynamic graph construction unit, which updates resource dependency weights based on event triggering mechanism.

[0028] Preferably, the global agent of the dynamic hierarchical decision module adopts the Pareto frontier dynamic search algorithm, and its multi-objective weight update formula is:

[0029]

[0030] in:

[0031] is the weight of the i-th target at time step t, satisfying

[0032] η>0 is the learning rate, which controls the weight update step size;

[0033] is the i-th objective function f i (θ t ) for the model parameters θ t gradient;

[0034] k≥2 is the total number of optimization objectives.

[0035] Preferably, the flexible resource allocation module includes:

[0036] Shapley value preemption unit, calculating resource contribution value:

[0037]

[0038] in:

[0039] represents the contribution value of resource i to the overall goal,

[0040] S is a subset of resource set N

[0041] is the utility function of the resource subset S, defined as v(S) = ∑ j∈S u j -c(S), where u j is the priority score of resource j, c(S) is the coordination cost of subset S;

[0042] Dynamic priority matrix generation unit, which assigns resource weights through non-negative matrix factorization.

[0043] Preferably, the lightweight policy network of the edge-cloud collaborative execution module compresses the global model through model distillation, and its distillation loss function is:

[0044]

[0045] in:

[0046] KL(·||·) represents the Kullback-Leibler divergence, which measures the difference between two probability distributions;

[0047] (a|s t ) is the global policy network in state s t Generate the probability distribution of action a;

[0048] π edge (a|s t ) is the edge strategy network in state s t Generate the probability distribution of action a;

[0049] s t is the system status at time step t, including resource occupancy, order urgency and equipment load.

[0050] Preferably, a method for dynamically allocating enterprise management resources comprises the following steps:

[0051] The multimodal data fusion module collects and integrates internal and external enterprise data to build a dynamic resource relationship map;

[0052] The dynamic hierarchical decision module receives the graph and generates a global resource allocation strategy and local execution instructions;

[0053] Dynamically allocate resources according to the strategy through the elastic resource allocation module and generate resource preemption instructions;

[0054] The instructions are executed by the edge-cloud collaborative execution module, and the execution results are fed back to the multimodal data fusion module to update the map.

[0055] The present invention provides an intelligent system for dynamically allocating enterprise management resources. It has the following beneficial effects:

[0056] 1. To address the resource portrait distortion problem caused by data silos in traditional systems, this invention uses cross-modal comparative learning and dynamic graph construction technology to achieve multi-dimensional semantic alignment of financial data, equipment status, and external policies. It can eliminate the semantic gap between structured data and unstructured data in management scenarios, provide a real-time and complete resource situation awareness foundation for subsequent decision-makers, and directly serve the accuracy requirements of "dynamic allocation."

[0057] 2. Aiming at the inherent contradiction between cost, efficiency and risk in enterprise resource scheduling, the present invention breaks through the limitations of single-dimensional decision-making by constructing a collaborative architecture of global Pareto frontier dynamic search and local reinforcement learning, and through a closed-loop mechanism of dynamic adjustment of multi-objective weights in the cloud and real-time optimization of the edge policy network, realizing the adaptive decision-making ability of the "intelligent system" in multi-constraint scenarios, and forming a complete mapping chain from strategic goals to execution strategies.

[0058] 3. To address the imbalance between allocation fairness and efficiency in resource competition scenarios, the present invention designs a collaborative model based on Shapley value contribution quantification and non-negative matrix decomposition. Through resource utility-cost game analysis and priority latent factor mining, a dynamic allocation mechanism that takes into account both efficiency and fairness is established to support the rapid reconstruction of complex resource relationships in "enterprise management" scenarios and ensure the supply guarantee capability of core business resources.

[0059] 4. To overcome the industrial-grade real-time shortcomings of traditional cloud-based centralized decision-making, the present invention develops knowledge distillation compression and federated aggregation feedback technology. Through the collaborative mechanism of lightweight model edge deployment and global strategy dynamic aggregation, a closed-loop control link of "dynamic resource allocation" from decision generation to physical execution is formed, meeting the technical rigidity requirements of low-latency and high-reliability execution in fields such as intelligent manufacturing and logistics scheduling. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 Provide an intelligent system architecture diagram for dynamic allocation of enterprise management resources;

[0061] Figure 2 Flowchart of the method for dynamic allocation of enterprise management resources. DETAILED DESCRIPTION

[0062] The following will clearly and completely describe the technical solution of the present invention in conjunction with the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0063] Please see the attached Figure 1The embodiment of the present invention provides an intelligent system for dynamically allocating enterprise management resources. The following describes in detail the various modules of the system of the present invention.

[0064] Multimodal data fusion module

[0065] The multimodal data fusion module of the present invention includes a data acquisition and preprocessing unit, a cross-modal comparative learning unit and a dynamic graph construction unit. The data acquisition and preprocessing unit is connected to the internal database of the enterprise and the external public opinion monitoring system through a data interface, and is used to obtain structured numerical data and unstructured text and image data in real time. The cross-modal comparative learning unit is connected to the preprocessing unit through a data bus and receives the standardized numerical matrix and semantic feature vector. The dynamic graph construction unit is connected to the cross-modal comparative learning unit through an event triggering interface, receives fusion features and generates a dynamic resource relationship map.

[0066] The data collection and preprocessing part includes structured data processing components and unstructured feature extraction components. i , order quantity y j To perform standardization:

[0067]

[0068] where μ x and σ x is the mean and standard deviation of historical data, x i , is the original value, output standardized matrix (d=10 is the numerical feature dimension). The unstructured feature extraction component uses the pre-trained BERT model to extract the semantic vector of the policy text. And extract the monitoring image spatial features through ResNet-50

[0069] The cross-modal contrastive learning unit includes a numerical tower network and a semantic tower network. The numerical tower network consists of three fully connected layers, which inputs X and outputs a latent space representation The semantic tower network uses the same structure to process the semantic vector S and outputs

[0070] The two networks are jointly trained using the contrastive loss function:

[0071]

[0072] in is a set of positive sample pairs, and the construction rule is: for each numerical data x i , randomly select a semantic data s from the associated text description set j , generate four pairs (xi ,s j ).

[0073] is a set of negative sample pairs, and the construction rule is: for each x i , randomly sample K=5 semantic data from non-related texts to generate (x i ,s k ).

[0074] is the cosine similarity function, τ=0.1 is the temperature coefficient, u,

[0075] The dynamic graph construction unit visualizes the enterprise resources as a dynamic graph G t =(V t ,E t ), node v i ∈V t Represents a department or equipment entity, edge e ij ∈E t Weight Indicates the resource dependency strength. When an external event keyword k is detected event When , the edge weights are updated through the event triggering mechanism:

[0076]

[0077] Where Δw = 0.2 is the preset impact factor, TF-IDF (k event ) is calculated as follows:

[0078]

[0079] N=10 4 is the total number of documents in the historical public opinion corpus, n k To include k event The number of documents. Node attributes Update via MLP Network:

[0080]

[0081] in is the weight matrix (input dimension d+k=10+768), is a trainable parameter.

[0082] Dynamic hierarchical decision module

[0083] The dynamic hierarchical decision-making module of the present invention comprises a global agent and local agents, which are connected to each other through a distributed decision-making network. The global agent's input is connected to the output of the multimodal data fusion module, receiving a dynamic resource relationship map; its output transmits multi-objective coordination parameters to the local agents via a control signal channel. The local agent's input is connected to the state sensors of the enterprise department's execution terminals, and its output transmits execution status data to the global agent via a feedback loop.

[0084] The global agent adopts the Pareto frontier dynamic search algorithm, and its multi-objective weight update formula is defined as:

[0085]

[0086] in, Represents the weight coefficient of the i-th optimization objective at time step t, the initial value η=0.01 is the learning rate parameter; Denotes the objective function f i For the decision model parameter θ t The gradient of is calculated by the automatic differentiation algorithm; k=3 is the total number of preset objectives (cost, efficiency, risk). The objective function is specifically defined as:

[0087]

[0088] Among them, c i is the unit cost coefficient of resource i, obtained from the enterprise financial database; is the resource usage; t j is the standard time consumption of task j, calculated based on historical work order data; y j ∈{0,1} represents the task allocation state; r p is the loss coefficient of risk event p, determined by expert evaluation; p ∈[0,1 is the probability of risk occurrence.

[0089] The local agent adopts the double-delayed deep deterministic policy gradient (TD3) algorithm, which includes the policy network π φ and value network Q ψ The policy network input state vector (including resource occupancy rate l i ∈0,1], order urgency u j ∈[0,5]), after two hidden layers (dimension 64, ReLU activation) output action a t =π φ (s t )+∈, where is Gaussian exploration noise. The value network input state s t With action a t , after two hidden layers (dimension 128, ReLU activation) output double Q value:

[0090]

[0091]

[0092] The final Q value takes the minimum value:

[0093]

[0094] The policy network update formula is:

[0095]

[0096] in For capacity 10 6 The experience replay pool uses uniform sampling to extract batch data (batch size 256).

[0097] Adversarial training: Generate virtual emergency samples to enhance the robustness of the strategy. The adversarial state is calculated as:

[0098]

[0099] Where δ = 0.05 is the disturbance coefficient, is the randomly generated target perturbation action, is the mean square error loss function.

[0100] Flexible resource allocation module

[0101] The flexible resource allocation module of the present invention includes a dynamic priority matrix generator, a Shapley value preemption unit, and a resource constraint management unit. The dynamic priority matrix generator connects to the enterprise's real-time database via a data interface to receive resource load and task urgency data. The Shapley value preemption unit's input terminal is connected to the policy output terminal of the dynamic hierarchical decision module to receive global coordination parameters. The resource constraint management unit interacts with the enterprise's resource reservation system via a control bus to ensure that the allocation process complies with contractual constraints.

[0102] Dynamic priority matrix generation

[0103] First build a real-time priority matrix Matrix element M ij Defined as a task-resource priority score:

[0104]

[0105] Among them, u i∈[0,5] is the urgency of task i, which is calculated as (Δt i is the remaining delivery time, γ=0.1 is the attenuation coefficient); l j ∈[0,1] is the real-time load rate of resource j, which is collected by device sensors; α=1.2 and β=0.8 are exponential smoothing parameters, which are determined by fitting historical data.

[0106] Then, non-negative matrix factorization (NMF) is used to decompose M into a weight matrix With the characteristic matrix The optimization goal is:

[0107]

[0108] Among them, r = 5 is the latent factor dimension (selected by the time rule), λ1 = 0.1 and λ2 = 0.05 are regularization coefficients, ||·|| F is the Frobenius norm. The decomposition process is solved iteratively using the alternating direction multiplier method (ADMM), and the iterative termination condition is that the relative error change rate is less than 10 -4 Or the maximum number of iterations, 1000, is reached.

[0109] Shapley value preemptive calculation

[0110] Calculate the contribution value φ of resource i i , the formula is:

[0111]

[0112] Where v(S) = ∑ j∈S u j -c(S) is the utility function of resource subset S; c(S) = κ|S| + ω∑ j,k∈S d jk is the coordination cost, κ = 0.3 is the linear cost coefficient, ω = 0.05 is the conflict penalty coefficient, d jk ∈(0,1] represents the conflict intensity between resources j and k (obtained through statistics of historical work order conflict records).

[0113] Due to the high complexity of accurately calculating the number of combinations, Monte Carlo sampling approximation is used, and the number of sampling times K = 10 4 , error tolerance ∈ = 0.01, confidence level 95%. The specific sampling rules are:

[0114] Positive samples: Prioritize resource combinations associated with the current task:

[0115] Negative samples: Randomly remove at least one key resource.

[0116] Resource preemption logic

[0117] When the resource competition index When , the preemption instruction is triggered. The preemption amount Δq is calculated as

[0118]

[0119] Among them, q reserve =100 is the minimum resource reservation amount constrained by the contract, Q total is the total amount of resources, and C is the ID set of the current competing resources.

[0120] Edge-cloud collaborative execution module

[0121] The edge-cloud collaborative execution module described in the present invention includes a model distillation unit, an edge policy execution unit, a cloud-based aggregation feedback unit, and an anomaly detection rollback unit. The input end of the model distillation unit is connected to the global policy output end of the elastic resource allocation module through a high-speed data channel to receive the global model parameters on the cloud; the output end is connected to the model loading interface of the edge policy execution unit through an encrypted channel. The edge policy execution unit is deployed on an industrial site edge computing node (such as Huawei Atlas500) and is connected to the enterprise PLC control terminal through the Modbus / TCP protocol. The input end of the cloud-based aggregation feedback unit receives the execution log data of the edge node through the 5G private network, and the output end is connected to the feedback interface of the multimodal data fusion module.

[0122] Model distillation and edge deployment

[0123] The model distillation unit uses the teacher-student distillation framework to compress the global strategy model π global , the loss function is defined as:

[0124]

[0125] Among them, α = 0.7 is the distillation weight coefficient, which is determined by grid search; is the KL divergence; is the task loss, y true It is the best action record in history; is the state vector, including:

[0126] Resource occupancy rate i ∈[0,1](real-time load of device i / maximum load);

[0127] Order urgency (Δt j The remaining time of the order (unit: hours);

[0128] Equipment health h k ∈{0,1} (0=fault, 1=normal, determined by the device self-test signal).

[0129] The edge policy execution unit deploys a lightweight model π edge , its network structure is:

[0130] Input layer: dimension 32 (same as s t Dimensional consistency);

[0131] Hidden layer: 2 fully connected layers with dimensions of 64 and 32 respectively, and activation function is ReLU;

[0132] Output layer: dimension m = 10 (action space size), activation function is Softmax. Model pruning (retaining 80% of weights, pruning threshold > 0.01) and FP16 quantization (IEEE 754 standard) compress the model size to 15MB, meeting the memory constraints of edge devices.

[0133] Cloud aggregation and abnormal rollback

[0134] The cloud-based aggregate feedback unit executes the FedAvg algorithm every 24 hours to update the global model:

[0135]

[0136] Where N=100 is the total number of edge nodes; n i is the amount of data in the node, n total =∑n i ;λ=0.01 is the differential privacy noise coefficient;

[0137] When the anomaly detection and rollback unit detects that the node disconnection rate is greater than 5% or the allocation error rate is greater than 3%, a version rollback is triggered:

[0138]

[0139] Among them, K=10 is the capacity of the historical version library; M=10 3 is the validation set size; μ = 0.1 is the entropy regularization coefficient Entropy(p) = -∑plogp to prevent strategy degradation.

[0140] Please see the attached Figure 2 The present invention also provides a method for dynamically allocating enterprise management resources. The following describes the specific implementation of each step in conjunction with the workflow of the method of the present invention.

[0141] The multimodal data fusion module collects and integrates internal and external enterprise data to build a dynamic resource relationship map;

[0142] The dynamic hierarchical decision module receives the graph and generates a global resource allocation strategy and local execution instructions;

[0143] Dynamically allocate resources according to the strategy through the elastic resource allocation module and generate resource preemption instructions;

[0144] The instructions are executed by the edge-cloud collaborative execution module, and the execution results are fed back to the multimodal data fusion module to update the map.

[0145] The device of this embodiment can be used to execute the above method embodiment, and its principles and technical effects are similar, so they will not be repeated here.

[0146] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent system for dynamic allocation of enterprise management resources, characterized by: include: Multimodal data fusion module, used to integrate internal and external structured data and unstructured data of the enterprise to generate a dynamic resource relationship map; A dynamic hierarchical decision module is in communication with the multimodal data fusion module, receives the dynamic resource relationship graph, and generates a resource allocation strategy through a two-layer framework of global coordination and local execution; A flexible resource allocation module, in communication with the dynamic hierarchical decision module, dynamically allocates resources based on the priority matrix and game theory model according to the resource allocation strategy; The edge-cloud collaborative execution module is in communication with the elastic resource allocation module, executes resource allocation instructions in real time through edge nodes, and feeds back the execution results to the multimodal data fusion module to form a closed-loop optimization system.

2. The intelligent system for dynamic allocation of enterprise management resources according to claim 1, characterized in that: The output end of the multimodal data fusion module is connected to the input end of the dynamic hierarchical decision module through a data bus to transmit a dynamic resource relationship map; the output end of the elastic resource allocation module is connected to the input end of the edge-cloud collaborative execution module through a message queue to transmit resource allocation instructions.

3. The intelligent system for dynamic allocation of enterprise management resources according to claim 1, characterized in that: The global agent and local agents of the dynamic hierarchical decision module communicate through a distributed decision network. The global agent sends multi-objective coordination weight parameters to the local agents, and the local agents feed back department-level execution status data to the global agent.

4. The intelligent system for dynamic allocation of enterprise management resources according to claim 1, characterized in that: The elastic resource allocation module receives the global policy from the dynamic hierarchical decision module and generates resource preemption instructions according to the real-time priority matrix. The resource preemption instructions are distributed to the edge-cloud collaborative execution module through control signals.

5. The intelligent system for dynamic allocation of enterprise management resources according to claim 1, characterized in that: The cloud hub of the edge-cloud collaborative execution module periodically sends a data update request to the multimodal data fusion module, and the multimodal data fusion module responds to the request and provides an incremental data set to update the dynamic resource relationship map.

6. The intelligent system for dynamic allocation of enterprise management resources according to claim 1, characterized in that: The multimodal data fusion module includes: The cross-modal contrastive learning unit aligns numerical data and semantic features through a contrastive loss function, which is: in: represents the latent space representation of the i-th numerical data; Represents the latent space representation of the j-th semantic data; τ>0 is the temperature coefficient, which is used to adjust the sharpness of the similarity distribution; is a set of matched positive sample pairs (numerical data and corresponding semantic description); is a set of randomly sampled negative sample pairs; Dynamic graph construction unit, which updates resource dependency weights based on event triggering mechanism.

7. The intelligent system for dynamic allocation of enterprise management resources according to claim 1, characterized in that: The global agent of the dynamic hierarchical decision module adopts the Pareto frontier dynamic search algorithm, and its multi-objective weight update formula is: in: is the weight of the i-th target at time step t, satisfying η>0 is the learning rate, which controls the weight update step size; is the i-th objective function f i (θ t ) for the model parameters θ t gradient; k≥2 is the total number of optimization objectives.

8. The intelligent system for dynamic allocation of enterprise management resources according to claim 1, characterized in that: The flexible resource allocation module includes: Shapley value preemption unit, calculating resource contribution value: in: represents the contribution value of resource i to the overall goal, S is a subset of resource set N is the utility function of the resource subset S, defined as v(S) = ∑ j∈S u j -c(S), where u j is the priority score of resource j, c(S) is the coordination cost of subset S; Dynamic priority matrix generation unit, which assigns resource weights through non-negative matrix factorization.

9. The intelligent system for dynamic allocation of enterprise management resources according to claim 1, characterized in that: The lightweight policy network of the edge-cloud collaborative execution module compresses the global model through model distillation, and its distillation loss function is: in: KL(·||·) represents the Kullback-Leibler divergence, which measures the difference between two probability distributions; (a|s t ) is the global policy network in state s t Generate the probability distribution of action a; π edge (a|s t ) is the edge strategy network in state s t Generate the probability distribution of action a; s t is the system status at time step t, including resource occupancy, order urgency and equipment load.

10. A method for dynamically allocating enterprise management resources, applied to the system according to any one of claims 1 to 9, characterized in that: The following steps are involved: The multimodal data fusion module collects and integrates internal and external enterprise data to build a dynamic resource relationship map; The dynamic hierarchical decision module receives the graph and generates a global resource allocation strategy and local execution instructions; Dynamically allocate resources according to the strategy through the elastic resource allocation module and generate resource preemption instructions; The instructions are executed by the edge-cloud collaborative execution module, and the execution results are fed back to the multimodal data fusion module to update the map.

Images