National-owned asset intelligent supervision method and system based on digital twinning

Abstract

The invention provides an intelligent supervision method and system for state-owned assets based on digital twinning, and relates to the technical field of asset management. The method comprises the steps of establishing a digital twinborn model of state-owned assets, analyzing a potential causal relationship among variables in an asset behavior evolution process according to the digital twinborn model, constructing a causal graph, constructing a reinforcement learning scheduling strategy model through the causal graph and the digital twinborn model, and outputting an optimal scheduling strategy. And a supervisor performs remote control and management operation on the state-owned assets through the intelligent terminal according to the optimal scheduling strategy, and an operation instruction is processed by the data processing center and then is fed back to the corresponding Internet of Things sensor and execution equipment, so that intelligent management and control on the state-owned assets are realized. According to the invention, real-time visual supervision and intelligent management and control of the state-owned assets are realized, the supervision efficiency and accuracy are improved, the management cost and risk are reduced, and the operation benefit and safety of the state-owned assets are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of asset management, and in particular to a method and system for intelligent supervision of state-owned assets based on digital twins. Background Art

[0002] In recent years, with the development of information technology, various regions have gradually established state-owned asset management platforms, and achieved basic records of asset life cycles through asset registration, ledger management, and transfer approval. However, the existing systems generally show the characteristics of "static ledger management", that is, they mainly rely on manual input and regular updates to maintain asset information, lacking real-time perception, behavior analysis, and intelligent decision-making capabilities. This makes it impossible to accurately capture and forward-lookingly judge issues such as asset status changes, value loss, and abuse risks during dynamic operation. In addition, traditional regulatory methods are mostly carried out in a decentralized manner by institutions, lacking a unified perspective to evaluate the overall efficiency of state-owned assets, scheduling structure, and configuration rationality.

[0003] To improve the above problems, some studies have tried to introduce the concept of "digital twin", that is, to build a virtual digital model to map and synchronously update physical assets, so as to realize real-time visualization of asset status and monitoring of operation process. However, these applications are mostly in engineering equipment or industrial manufacturing scenarios, and their implementation in the field of state-owned asset management is still in its early stages. On the one hand, state-owned asset management involves multi-dimensional and multi-departmental collaboration, and the attributes of the assets themselves are complex (such as fixed assets, office space, rental assets, etc.). Their operating behaviors are often affected by multiple factors such as policies, markets, geography, and budgets. Existing digital twin solutions are difficult to fully model these influencing factors; on the other hand, although the existing system can present the status of assets, it still lacks in-depth data understanding capabilities, especially in identifying abnormal asset use, hidden vacancy, repeated configuration, and low efficiency. It cannot provide reliable risk warnings and disposal suggestions.

[0004] More importantly, the existing management system cannot analyze the "causes of asset behavior" or perform simulation and strategy optimization based on historical experience. For example, when an asset has not been allocated for use for a long time, the system cannot determine whether its idleness is due to obsolete functions, management negligence, or delayed allocation process; for example, when multiple units frequently exchange similar assets, the system lacks a mechanism to evaluate whether there is unreasonable scheduling or resource mismatch. The existing system presents the characteristics of "information display" rather than "intelligent supervision" in data analysis and decision support, making it difficult to achieve the leap from "visible" to "well managed". Summary of the invention

[0005] In view of the deficiencies of the current state-owned asset management system in aspects such as data perception dimension, behavior analysis ability, scheduling intelligence level, and regulatory decision-making interpretability, the present invention proposes a digital twin intelligent supervision method and system for the state-owned asset management scenario. By constructing a digital twin model covering multiple dimensions such as asset status, transfer behavior, environmental information, and management strategies, this method realizes in-depth modeling of assets in the actual operating environment. Further combined with the causal analysis and strategy learning mechanism for asset behavior changes, the system can identify the potential causes of abnormal asset behaviors and generate target-oriented scheduling or configuration suggestions based on this, thereby realizing closed-loop dynamic supervision and optimization.

[0006] To achieve the above object, the technical solution adopted by the present invention is: to provide an intelligent supervision method for state-owned assets based on digital twins, including the following steps: S1. Map data from different sources into state vectors of a unified structure through an alignment function, and construct a digital twin model based on the time-series states of all assets; S2. Analyze the potential causal relationships between variables in the asset behavior evolution process based on the digital twin model, and construct a causal graph, which is used to extract the causal path of abnormal behavior of each asset at a certain moment; S3. Construct a reinforcement learning scheduling policy model according to the obtained asset status and abnormal causal path in the digital twin model, and output the optimal scheduling policy; S4. Apply the optimal scheduling policy to the supervision system, and collect actual execution feedback while performing the asset scheduling decision action.

[0007] Preferably, in S1, the data from different sources include ledger data, transfer and usage logs, sensor data, geographical data, and policy data, and form a time-series expression according to the current status and historical behavior, which is used to construct a system-sensitive twin state.

[0008] Preferably, in S2, the causal graph constructs its local abnormal causal path for each asset at a specific time to obtain a path chain, which is used to represent the causal source chain of the current abnormal behavior.

[0009] Preferably, in S3, by introducing a policy learning objective function, the form of the policy learning objective function is as follows: ; Among them, represents the optimal policy function, which will be deployed to the asset supervision platform for scheduling execution; represents the scheduling policy model, which maps from the state to the action ; represents the reward function constructed by double constraints; represents the total duration of the scheduling period; represents under the policy expectation; represents the current scheduling period.

[0010] More preferably, a monitoring interface is developed based on the optimal policy function and visualization technology, combining the scheduling policy model with real-time data, and intuitively displaying the key state parameters and operation metrics of physical assets on the interface.

[0011] Preferably, in S4, according to the output of the optimal scheduling policy for scheduling, the state-owned assets are remotely controlled and managed through an operation terminal, and the operation instructions are processed by the data processing center and then fed back to the corresponding Internet of Things sensors and execution devices.

[0012] Preferably, in S4, by introducing a feedback writing function, the real-world feedback is written back to the asset state of the digital twin model, and the edge weight confidence in the causal graph is adjusted.

[0013] In a second aspect of the present invention, there is also provided an intelligent supervision system for state-owned assets based on digital twins, including: A digital twin model construction module for mapping data from different sources into state vectors of a unified structure through an alignment function and constructing a digital twin model; A causal graph construction module for analyzing the potential causal relationships between variables in the process of asset behavior evolution based on the digital twin model, constructing a causal graph, and extracting the causal path of abnormal behavior of each asset at a certain moment; A reinforcement learning scheduling policy model construction module for constructing a reinforcement learning scheduling policy model according to the obtained asset state and abnormal causal path in the digital twin model; An execution and feedback module for executing the optimal scheduling policy output by the reinforcement learning scheduling policy model in the supervision system, writing back the real-world feedback to the asset state of the digital twin model through a feedback writing function, and adjusting the edge weight confidence in the causal graph.

[0014] The beneficial effects of the present invention are as follows: Compared with traditional technologies, the intelligent supervision method and system for state-owned assets based on digital twins provided by the present invention no longer stay at the static visualization of asset status, but penetrate into the entire life cycle of asset use, and integrate an "interpretable anomaly recognition mechanism" and an "adaptive strategy optimization ability" on the basis of digital twins. The former enhances the logic and accountability of supervision behaviors by mining the driving factors behind asset behaviors; the latter enables the system to have the ability to respond to complex management requirements in real time through simulation feedback and strategy evolution, so as to effectively address practical problems such as cross-regional and cross-institutional asset scheduling optimization and asset precipitation risk prevention and control. This method is adapted to state-owned asset scenarios of different types and scales, has strong generalization ability and deployability, and can significantly improve the intelligent level and resource allocation efficiency of the current state-owned asset supervision system. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 is a flowchart of the intelligent supervision method for state-owned assets based on digital twins of the present invention.

[0016] Figure 2 is a block diagram of the intelligent supervision system for state-owned assets based on digital twins of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0017] Please refer to Figure 1 as shown, the intelligent supervision method for state-owned assets based on digital twins of the present invention includes the following steps: S1. Map data from different sources into state vectors with a unified structure through an alignment function, and build a digital twin model based on the time-series states of all assets; The purpose of this step is to establish a multi-dimensional and dynamic digital twin model for state-owned assets to support subsequent causal modeling (S2), scheduling strategy generation (S3), and state feedback update (S4). Its core task is to transform the data with scattered sources, heterogeneous complexity in reality into a twin state sequence with a unified structure, continuous time series, and institutional perception ability.

[0018] It includes various Internet of Things sensors deployed on state-owned assets (such as the above-mentioned production equipment), including temperature sensors, pressure sensors, vibration sensors, current sensors, voltage sensors, etc. These sensors collect the operation data and environmental data of assets in real time, including ledger data, transfer and usage logs, sensor data, geographical data, and policy data, form a time-series expression according to the current state and historical behaviors, and are used to construct an institution-sensitive twin state. And transmit the data to the data processing center of the system through wireless communication technology for constructing a digital twin model.

[0019] Define the structure alignment function , and map data from different sources into state vectors with a unified structure : ; Among them, represents the current asset status, represents the ledger data, such as asset type, purchase age, and affiliated institution; represents the transfer and usage logs; represents the sensor data, such as operating status and whether it is online; represents the geographical data, such as longitude and latitude, and regional classification; represents the policy data, such as transfer restrictions and budget priorities; represents the multi-modal structure aligner defined in the present invention, which is used to integrate the above five types of data and output a state vector in a unified format ; represents a vector space containing d real number dimensions.

[0020] Subsequently, construct the "time-aware state fusion function" , which is used to synthesize the current state and historical behaviors to form a temporal expression : ; Among them, represents the current asset status; : the state at the th moment forward; K is the total historical length; represents the decay weight of the historical state, is the decay rate and ; , represents controlling the weight balance between the current state and the historical state, with a default of ; represents the twin state expression after time fusion.

[0021] Introduce the institution vector and the embedding mapping to construct the institution-sensitive twin state : ; Among them, represents the institution factor vector, such as policy priority, constraint level, and weight of the affiliated unit, etc.; represents the policy embedding mapping function, which converts the institution factor into an embedding vector; represents the institution influence coefficient, indicating the degree of intervention of the institution on the behavior state (such as represents medium intervention); represents the final institution-enhanced twin state, which is used as the input for downstream tasks.

[0022] Finally, generate the final digital twin model set based on the above data : Organize the time series status of all assets into the following structure: ; Among them, represents the asset number; represents the th asset's institutional enhancement status at time ; represents the total time length for which asset is tracked; represents the total number of assets; represents the digital twin model set, which is used for causal modeling (S2), policy training (S3), and feedback update (S4).

[0023] This structure provides a standardized, dynamic, and institution-aware status representation for the entire patent system, with temporal consistency, structural consistency, and semantic stability, and is a necessary basis for realizing "digital twin + intelligent supervision".

[0024] S2. Analyze the potential causal relationships between variables during the evolution of asset behavior based on the digital twin model, and construct a causal graph, which is used to extract the causal path of behavior anomalies for each asset at a certain moment; The goal of this step S2 is to analyze the potential causal relationships between variables during the evolution of asset behavior based on the digital twin model constructed in Step 1 , construct a causal graph , and accordingly extract the behavior anomaly path for each asset at a certain moment

[0025] . This structural result will be used as the input for the subsequent policy generation stage.

[0026] First, the state sequence data output in Step 1 needs to be formatted into the time window form required for graph construction to provide a data basis for subsequent causal relationship learning. Using the obtained in S1 as the input, select a fixed time window length , and construct a time window sequence for each asset ; Among them, represents the unique identifier of the th asset; represents the institutional enhanced twin state of asset at time (from S1); represents asset Total tracking time length; Denotes the window length for local data sampling in structure learning; Denotes the asset At time State time window sequence.

[0027] By designing a structural causal graph Learning mechanism: By extracting the causal structure from the asset state variables, it provides a structured prior for downstream policy generation, rather than relying on black-box prediction.

[0028] The present invention proposes a cross-asset joint graph construction method , based on covariance dependence and structural stability constraints, constructs a causal structure graph of learning variables : ; Wherein, , both denote any two variables in the set of asset behavior variables; Denotes the complete set of asset behavior variables, including the asset state feature dimensions used to construct all nodes in the causal graph; Denotes the covariance calculated in the window sample, used to measure the strength of causal dependence; Denotes the structural stability (standard deviation) of this causal edge in different units or regions; Denotes the adjustment coefficient of the institutional disturbance factor, used to control the importance of the stability constraint (recommended range ); Denotes the structure graph learning objective function, used to select the optimal causal edge set.

[0029] Extracting the abnormal causal path is to use the causal structure for the interpretation of single assets and subsequent action guidance, ensuring that the generated policy has a clear behavior traceability path.

[0030] Based on the causal graph , for each asset At time Construct its local abnormal causal path : First, detect Whether there is a variable Showing extreme behavior (such as exceeding the quantile threshold); From Trace back upwards The causal path in until the root variable; Get the path chain , indicating the causal source chain of this abnormal behavior.

[0031] The final output structural result will be directly used as the core input for the next step of policy training, ensuring that the scheduling policy operates based on structural interpretability. The final output of this step S2 is the following structure: : Causal structure diagram of asset behavior (globally shared), provided to the subsequent policy training stage; : Asset Abnormal behavior causal path at time , providing a structural prior for subsequent policy generation.

[0032] S3. Construct a reinforcement learning scheduling policy model based on the obtained asset status and abnormal causal path in the digital twin model, and output the optimal scheduling policy; The goal of this step S3 is: based on the obtained asset status (from in S1) and the abnormal causal path (from in S2), construct a reinforcement learning scheduling policy model to achieve dynamic optimization allocation, shared management, idle release, and risk avoidance of assets.

[0033] Carry forward the digital twin state of step one , and directly use it as the state input of reinforcement learning, ensuring the consistency between the state space and the actual asset operation characteristics, which is the basic premise for the operability of policy training.

[0034] By directly using the twin state vector in S1 as the state space of reinforcement learning , this state has integrated historical behaviors and institutional characteristics, avoiding repeated processing.

[0035] Represents the institutional enhanced state of the asset at time , with a dimension of ; : Reinforcement learning state space.

[0036] By defining the action space , which includes executable scheduling operations for state-owned asset management, the scheduling behavior set that can be executed in this invention is defined in this step, and it is also the action range output by the policy model, which must conform to the business process and authority boundary of state-owned asset management. Including: : Allocate to unit ; : Mark as shared and available; : Release as idle; : Recycling and warehousing; : Restricting usage permissions, etc.

[0037] Causal path-guided action screening mechanism: Based on the abnormal path output in Step 2 , design a structured action screening method to exclude non-compliant or risky actions during policy search, improve training efficiency, and strengthen causal consistency.

[0038] The present invention designs a structure-guided action screening function , used to exclude high-risk actions that conflict with abnormal path variables. The calculation formula of the action screening function is as follows: ; Among them, represents the asset at time 's abnormal causal path; represents each variable in the path; represents the action 's institutional risk conflict score for variable , the higher the value, the more non-compliant; represents the action institutional compliance threshold (such as 1.0); represents the final legal action subset, serving as the action space of the policy model.

[0039] The present invention hereby proposes a reward function mechanism that simultaneously considers causal path perturbation and institutional compliance, which is the core control point for strengthening the rationality of the reinforcement learning policy, ensuring that the policy is not purely profit-oriented, but emphasizes both behavioral structure and institutional goals.

[0040] Also designed a reward function jointly driven by structure + institutional constraints , and the calculation formula of the reward function is as follows: ; Among them, represents the basic benefit function, such as the improvement of asset utilization rate (such as ); represents the abnormal causal path at the current moment; represents the causal edge in the path; represents variable 's change value after executing action ; represents the path perturbation penalty coefficient (such as ); represents the degree of institutional constraint conflict triggered by executing action under the given asset state . Its calculation method is: The system will take the action The affected institutional fields (such as transfer destination unit type, budget level, asset category) are compared with the institutional constraints (such as non-transfer across regions is not allowed, transfer to non-shared units is not allowed) in the current asset status and one basic penalty point is accumulated for each violation condition met. For example, if the asset status indicates that "transfer is only allowed to units in the same region", while the actual transfer target is across regions, then ; if the "asset category restriction" is also violated at the same time, then it is , and so on, forming a positive integer value used to punish non-compliant situations in the policy behavior. If the current action violates the policy system, this item is a positive value; represents the regular weight of institutional consistency (such as ); represents the final reward value used to train the intelligent policy.

[0041] By defining the policy optimization objective of reinforcement learning, the finally trained policy model can form a stable, reasonable, and deployable scheduling function under the joint guidance of structural constraints and institutional penalties.

[0042] By introducing the policy learning objective function, the form of the policy learning objective function is as follows: ; where, represents the optimal policy function, which will be deployed to the asset supervision platform for scheduling execution; represents the scheduling policy model, which maps from the state to the action ; represents the reward function constructed by double constraints; represents the total duration of the scheduling period; represents the expectation under the policy ; represents the current scheduling period.

[0043] The training process of this policy model will adopt a constrained action space and a customized reward function to ensure that the generated scheduling policy is not only effective, but also compliant, interpretable, and in line with causal logic and institutional goals.

[0044] Based on the optimal policy function and visualization technology, a monitoring interface is developed. The scheduling policy model is combined with real-time data to intuitively display the key state parameters and operation indicators of physical assets on the interface.

[0045] S4. Apply the optimal scheduling policy to the supervision system, and collect actual execution feedback while performing the asset scheduling decision action.

[0046] The responsibility of this step S4 is to apply the reinforcement learning policy output by S3 to the real - world supervision system. While executing the asset scheduling decision action, collect the actual execution feedback, write the feedback back to the asset status in the digital twin model and fine - tune the edge - weight confidence in the causal structure diagram . This process does not involve any modeling behavior, but is limited to state update and parameter writing, which is the last link in building the system's closed - loop regulation ability.

[0047] Schedule according to the output of the optimal scheduling policy, remotely control and manage state - owned assets through the operation terminal. The operation instructions are processed by the data processing center and then fed back to the corresponding Internet of Things sensors and execution devices.

[0048] For each asset status , the policy model outputs an execution action: ; Among them, is the optimal policy function output from the policy model in S3; is the current asset status, from the twin model of S1 ; represents the execution action, such as transfer, recovery, sharing, restriction, etc.; a new state is generated after execution, that is, the result state in reality.

[0049] By introducing a feedback writing function, write the real - world feedback back to the asset status of the digital twin model, and adjust the edge - weight confidence in the causal graph. The mathematical expression is as follows: ; Among them, is the state update function, used to describe how to derive the next state from the current state and observation value, represents the feedback fusion weight, controlling the update amplitude, generally taking the value ; represents the final updated state written into , without changing its structure, only changing the value; Use the response of a certain variable after policy execution to adjust the weights of relevant edges in the causal graph (instead of structure learning): ; Among them, represents the confidence of the edge in the causal graph ; represents the response after the execution of this variable; represents the learning rate coefficient, generally taking the value of 。

[0050] This step S4 does not involve model training, learning, or reconstruction. It only updates the results into the model to ensure that the model remains up-to-date.

[0051] This step S4 does not modify the edge set of the causal graph. It only fine-tunes the weights of the existing edges to ensure evolvability and dynamics.

[0052] As Figure 2 shown, in the second aspect, the present invention also provides an intelligent supervision system for state-owned assets based on digital twins, including: A digital twin model construction module for mapping data from different sources into state vectors of a unified structure through an alignment function and constructing a digital twin model; A causal graph construction module for analyzing the potential causal relationships between variables during the asset behavior evolution process based on the digital twin model, constructing a causal graph, and extracting the causal path of abnormal behavior of each asset at a certain moment; A reinforcement learning scheduling policy model construction module for constructing a reinforcement learning scheduling policy model according to the obtained asset states and abnormal causal paths in the digital twin model; An execution and feedback module for executing the optimal scheduling policy output by the reinforcement learning scheduling policy model in the supervision system, writing the real-world feedback back to the asset states of the digital twin model through a feedback writing function, and adjusting the edge weight confidence in the causal graph.

[0053] The above embodiments are only descriptions of the preferred embodiments of the present invention and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solutions of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. An intelligent supervision method for state-owned assets based on digital twins, characterized in that, It includes the following steps: S1. Map data from different sources into state vectors of a unified structure through an alignment function, and construct a digital twin model based on the time-series states of all assets; S2. Analyze the potential causal relationships among variables during the asset behavior evolution process based on the digital twin model, construct a causal graph, and the causal graph is used to extract the causal path of abnormal behavior of each asset at a certain moment; S3. Construct a reinforcement learning scheduling policy model according to the obtained asset states and abnormal causal paths in the digital twin model, and output the optimal scheduling policy; S4. Apply the optimal scheduling policy to the supervision system, and collect actual execution feedback while performing asset scheduling decision-making actions.

2. The intelligent supervision method for state-owned assets based on digital twin according to claim 1, characterized in that, In S1, the data from different sources include ledger data, transfer and usage logs, sensor data, geographical data, and policy data, and form a time-series expression according to the current state and historical behavior, which is used to construct a system-sensitive twin state.

3. The intelligent supervision method for state-owned assets based on digital twin according to claim 1, wherein, In S2, the causal graph constructs its local abnormal causal path for each asset at a specific time to obtain a path chain, and the path chain is used to represent the causal source chain of the current abnormal behavior.

4. The intelligent supervision method for state-owned assets based on digital twin according to claim 1, wherein, In S3, by introducing a policy learning objective function, the form of the policy learning objective function is as follows: ； Among them, represents the optimal policy function, which will be deployed to the asset supervision platform for scheduling and execution; represents the scheduling policy model, mapping from the state to the action ; represents the reward function constructed by double constraints; represents the total duration of the scheduling period; represents the expectation under the policy ; represents the current scheduling period.

5. The intelligent supervision method of state-owned assets based on digital twin according to claim 4, wherein Develop a monitoring interface based on the optimal policy function and visualization technology, combine the scheduling policy model with real-time data, and intuitively display the key state parameters and operation indicators of physical assets on the interface.

6. The intelligent supervision method for state-owned assets based on digital twin according to claim 1, wherein In S4, according to the output of the optimal scheduling policy, remotely control and manage state-owned assets through an operation terminal, and the operation instructions are processed by the data processing center and then fed back to the corresponding Internet of Things sensors and execution devices.

7. The intelligent supervision method of state-owned assets based on digital twin according to claim 1, characterized in that In S4, introduce a feedback writing function to write the real-world feedback back into the asset state of the digital twin model, and adjust the edge weight confidence in the causal graph.

8. The intelligent supervision system for state-owned assets based on digital twins is characterized in that, It includes: A digital twin model construction module, which is used to map data from different sources into state vectors of a unified structure through an alignment function and construct a digital twin model; A causal graph construction module, which is used to analyze the potential causal relationships among variables during the asset behavior evolution process based on the digital twin model, construct a causal graph, and extract the causal path of abnormal behavior of each asset at a certain moment; A reinforcement learning scheduling policy model construction module, which is used to construct a reinforcement learning scheduling policy model according to the obtained asset states and abnormal causal paths in the digital twin model; An execution and feedback module, which executes the optimal scheduling policy output by the reinforcement learning scheduling policy model in the supervision system, writes the real-world feedback back into the asset state of the digital twin model through a feedback writing function, and adjusts the edge weight confidence in the causal graph.

Images