A method, system, device, and medium for dynamic valuation of highway digital assets

Abstract

The present application relates to the technical field of expressway digital asset management, in particular to an expressway digital asset dynamic valuation method, system, device and medium; the method firstly obtains multi-source heterogeneous asset data to perform space-time manifold alignment and semantic fusion mechanism, obtains comprehensive asset base characteristics, generates a causal adjacency matrix, introduces an intervention operator to perform counterfactual differential evolution deduction, obtains a trajectory tensor and deduction confidence, then constructs double agents to perform double-blind game, generates a digital asset dynamic fair value under the guidance of Nash equilibrium theory, and finally constructs a Lagrange cost functional with the goal of comprehensive maximization of realized income and asset preservation, solves an optimal preventive maintenance execution strategy, and performs cryptography solidification with the comprehensive asset base characteristics, digital asset dynamic fair value and expected operation income, to obtain a digital asset ledger block, realizing high-fidelity mapping, interpretable evolution prediction and high-value realization of expressway digital assets.

Description

Technical Field

[0001] This invention relates to the field of digital asset management technology for highways, specifically to a method, system, device, and medium for dynamic valuation of digital assets for highways. Background Technology

[0002] Utilizing IoT sensing, BIM (Building Information Modeling), and GIS (Geographic Information System) technologies to construct a digital twin foundation for highways and acquire digital asset parameters such as traffic flow, meteorological environment, structural load, and disease evolution throughout the entire lifecycle has become a core means to improve intelligent operation capabilities and asset preservation and appreciation levels. With the advancement of new transportation infrastructure construction, although massive amounts of multi-source heterogeneous asset data have been accumulated, existing digital asset modeling, evolution prediction, and dynamic valuation technologies still have significant limitations when facing the complex, ultra-large-scale highway system that is cross-regional, long-term, and strongly coupled. These limitations make it difficult to meet the needs of precise maintenance decisions and refined asset governance.

[0003] Existing digital asset management technologies for highways fall short in the face of these complex scenarios, primarily due to technical bottlenecks such as evolutionary distortion caused by spatiotemporal semantic fragmentation and a lack of valuation credibility caused by conflicts in multidimensional objectives. On one hand, multi-source heterogeneous asset data exhibits severe spatiotemporal heterogeneity in virtual space, and evolutionary deduction lacks underlying physical causal constraints. Existing digital twin technologies mostly employ static spatial stacking based on BIM and simple chart mounting of IoT data. The geometric spatial topology of BIM and the high-frequency temporal spatial data of sensors belong to different mathematical expression categories, resulting in a significant semantic gap in the virtual mapping of asset status. More critically, existing asset degradation predictions (such as road rutting evolution and bridge fatigue damage) often rely on statistical regression models based on deep learning (such as LSTM sequence networks). These purely data-driven models are deeply entrenched in correlation traps and lack explicit modeling of the complex coupling mechanisms between heavy traffic dynamic loads, natural material aging, and extreme weather stresses. When faced with unknown extreme operating conditions or the need to assess specific maintenance interventions, models often collapse due to the inability to perform counterfactual causal reasoning, failing to accurately predict the true three-dimensional evolution trajectory of asset health. Furthermore, there is a serious conflict of information sources and objectives between the health valuation of digital assets and operational decisions, lacking an objective and intelligent game arbitration mechanism. Highway operation and maintenance management is essentially a multi-objective optimization problem, seeking the optimal balance between ensuring safety and compliance, reducing operational risks, controlling maintenance costs, and pursuing economic benefits. Traditional asset evaluation systems often rely on subjective expert scoring or fixed linear weight matrices for data assimilation. They tend to favor structural health monitoring systems that prioritize frequent maintenance to ensure absolute safety, which clashes with financial budget assessment systems that tend to postpone major repairs to control current costs. When decisions conflict in certain road sections, existing systems can only provide a mechanical weighted average or a single over-limit alarm. This rigid evaluation model lacks an intelligent negotiation mechanism that can simulate the perspectives of professional asset appraisers and engineering maintenance experts to dynamically determine the true fair value of assets. As a result, the final asset valuation and operation and maintenance strategy cannot eliminate systemic biases, and the economic and physical consistency and credibility of the decision-making results are seriously insufficient. Summary of the Invention

[0004] The purpose of this invention is to provide a method, system, device, and medium for dynamic valuation of digital assets of highways.

[0005] The technical solution of this invention is as follows: A dynamic valuation method for digital assets of highways includes the following operations: S1. Acquire multi-source heterogeneous asset data of highway assets, including dynamic time-series data from roadside IoT sensors, static geometric topology data from BIM, and unstructured text data from manual inspections; perform spatiotemporal manifold alignment and semantic fusion on the multi-source heterogeneous asset data to generate comprehensive asset base features; S2. Based on the comprehensive asset base characteristics, generate a causal adjacency matrix, design an interference quantifier to perform counterfactual state differential evolution deduction, and generate counterfactual healthy states; based on the causal adjacency matrix and counterfactual healthy states, obtain the deduction confidence and trajectory tensor; S3. Based on trajectory tensor and inference confidence, construct a security intelligent agent and an economic intelligent agent, conduct double-blind game, reach consensus on asset valuation under Nash equilibrium constraints, and generate dynamic fair value of digital assets. S4. Based on the dynamic fair value of digital assets, generate expected operating revenue. With the goal of maximizing the combined revenue from realization and inventory value, construct a Lagrange cost functional and solve for the optimal preventive maintenance execution strategy. Combine this with the comprehensive asset base characteristics, the dynamic fair value of digital assets, and the expected operating revenue, and cryptographically solidify it to obtain the digital asset ledger block.

[0006] A dynamic valuation system for digital assets of highways, used to implement the aforementioned dynamic valuation method for digital assets of highways, includes: The comprehensive asset base feature generation module is used to acquire multi-source heterogeneous asset data of highway assets, including dynamic time-series data of roadside IoT sensors, static geometric topology data of BIM, and unstructured text data of manual inspection; it performs spatiotemporal manifold alignment and semantic fusion on the multi-source heterogeneous asset data to generate comprehensive asset base features; The module for generating inference confidence and trajectory tensors is used to generate a causal adjacency matrix based on comprehensive asset base features, design an interference quantifier to perform counterfactual state differential evolution inference, and generate a counterfactual healthy state; based on the causal adjacency matrix and the counterfactual healthy state, the inference confidence and trajectory tensor are obtained. The digital asset dynamic fair value generation module is used to construct a secure intelligent agent and an economic intelligent agent based on trajectory tensor and inference confidence, conduct double-blind game, reach a consensus on asset valuation under Nash equilibrium constraints, and generate the dynamic fair value of digital assets. The digital asset ledger block generation module is used to generate expected operating revenue based on the dynamic fair value of digital assets. With the goal of maximizing the combined realization revenue and inventory value, it constructs a Lagrange cost functional, solves for the optimal preventive maintenance execution strategy, and cryptographically solidifies the combined asset base characteristics, the dynamic fair value of digital assets, and the expected operating revenue to obtain the digital asset ledger block.

[0007] A dynamic valuation device for digital assets of highways includes a processor and a memory, wherein the processor implements the above-mentioned dynamic valuation method for digital assets of highways when executing a computer program stored in the memory.

[0008] A computer-readable storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the aforementioned dynamic valuation method for digital assets of highways.

[0009] The beneficial effects of this invention are as follows: This invention provides a dynamic valuation method for digital assets of highways. First, by acquiring multi-source heterogeneous asset data and designing a spatiotemporal manifold alignment and semantic fusion mechanism, discrete BIM geometric topology, high-frequency dynamic sensing time series, and domain experience text are mapped to a unified high-dimensional feature space. This solves the semantic fragmentation and spatiotemporal heterogeneity problems of heterogeneous data in physical mapping, and constructs a high-fidelity asset base feature with physical self-consistency for subsequent causal inference. Then, based on the comprehensive asset base feature, a causal adjacency matrix is ​​generated, and a counterfactual differential evolution inference is introduced. This enables accurate inference of the evolution trajectory of asset health status under different maintenance intervention strategies in virtual space, breaking through the limitations of pure data-driven models that cannot perform causal inference and counterfactual reasoning. Furthermore, the inference confidence is generated by combining the information entropy of the causal adjacency matrix, providing a solid physical constraint and risk boundary for subsequent asset valuation. Next, based on trajectory tensors and inference confidence, a dual-agent system representing safety compliance and financial cost reduction is constructed to conduct a double-blind game. Under the guidance of Nash equilibrium theory, multiple rounds of confrontation and compromise between professional engineering appraisers and financial analysts are simulated, so that the valuation consensus can both accommodate the rigid bottom line of safety supervision and take into account the optimization demands of economic benefits, thereby generating an objective and fair dynamic fair value of digital assets. This effectively solves the fundamental problem that traditional linear weighting cannot handle the conflict of objectives. Finally, based on the dynamic fair value of digital assets, the monetization revenue of right-of-way is dynamically generated. With the goal of maximizing the combined monetization revenue and asset preservation, a Lagrange cost functional is constructed to solve the Pareto optimal maintenance strategy under the funding constraints of the entire road network. The asset base characteristics, fair value and expected revenue are cryptographically solidified to generate an immutable ledger block, realizing high-fidelity mapping, interpretable evolution prediction and high-value monetization of highway digital assets. Attached Figure Description

[0010] The solutions and advantages of this application will become clear to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of the invention.

[0011] In the attached diagram: Figure 1 This is a flowchart illustrating the method of this embodiment. Figure 2 This is a comparison chart of the errors of different methods in complex valuation tasks, as shown in the example. Figure 3 This is a comparison chart of causal consistency residuals for different methods in the embodiments; Figure 4 This is a comparison chart of consensus achievement rates for different game theory methods in the example. Figure 5 This is a comparison chart showing the increase in return on investment for different methods in the embodiments. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of the exemplary embodiments of this application clearer, the technical solutions in the exemplary embodiments of this application are described clearly and completely below. Obviously, the described exemplary embodiments are only some embodiments of this application, and not all embodiments.

[0013] This embodiment provides a method for dynamic valuation of digital assets of highways. (See also...) Figure 1 This includes the following operations: S1. Acquire multi-source heterogeneous asset data of highway assets, including dynamic time-series data from roadside IoT sensors, static geometric topology data from BIM, and unstructured text data from manual inspections; perform spatiotemporal manifold alignment and semantic fusion on the multi-source heterogeneous asset data to generate comprehensive asset base features; S2. Based on the comprehensive asset base characteristics, generate a causal adjacency matrix, design an interference quantifier to perform counterfactual state differential evolution deduction, and generate counterfactual healthy states; based on the causal adjacency matrix and counterfactual healthy states, obtain the deduction confidence and trajectory tensor; S3. Based on trajectory tensor and inference confidence, construct a security intelligent agent and an economic intelligent agent, conduct double-blind game, reach consensus on asset valuation under Nash equilibrium constraints, and generate dynamic fair value of digital assets. S4. Based on the dynamic fair value of digital assets, generate expected operating revenue. With the goal of maximizing the combined revenue from realization and inventory value, construct a Lagrange cost functional and solve for the optimal preventive maintenance execution strategy. Combine this with the comprehensive asset base characteristics, the dynamic fair value of digital assets, and expected operating revenue, and cryptographically solidify it to obtain the digital asset ledger block. The specific steps are detailed below.

[0014] S1. Acquire multi-source heterogeneous asset data of highway assets, including dynamic time-series data of roadside IoT sensors, static geometric topology data of BIM, and unstructured text data of manual inspection; perform spatiotemporal manifold alignment and semantic fusion on the multi-source heterogeneous asset data to generate comprehensive asset base features.

[0015] First, acquire multi-source heterogeneous asset data for highways, including dynamic time-series data from roadside IoT sensors, static geometric topology data from BIM, and unstructured text data from manual inspections. Highway assets include, but are not limited to, roadbeds (such as slopes and retaining walls), bridges (such as piers, continuous box girders, and expansion joints), tunnels (such as portals and linings), pavements (such as asphalt surface layers and base layers), and electromechanical facilities (such as gantry systems, lighting equipment, and surveillance cameras).

[0016] Then, spatiotemporal manifold alignment and semantic fusion are performed on multi-source heterogeneous asset data to generate comprehensive asset base features.

[0017] The spatiotemporal manifold alignment and semantic fusion operations include: performing dynamic time-series harmonic normalization on the sampling timestamps of the IoT sensor dynamic time-series data, mapping discrete observation times to continuous periodic physical space, and generating dynamic time-series harmonic feature vectors; parsing the mechanical transmission paths between assets from BIM static geometric topology data, defining the adjacency relationships between assets, and reconstructing the asset spatial topology manifold vectors through graph attention mechanisms; performing semantic parsing on unstructured text data from manual inspections using a vertical domain large language model to obtain high-order text semantic vectors; and aligning and fusing the dynamic time-series harmonic feature vectors, spatial topology manifold vectors, and high-order text semantic vectors to obtain comprehensive asset base features.

[0018] In processing the dynamic time-series data of the opposite-side IoT sensors, it is important to consider that roadside IoT sensing devices such as WIM (Dynamic Weighing Institution), piezoelectric sensors, and weather stations on highways have vastly different collection frequencies and are prone to data time-series loss. Directly inputting discrete absolute timestamps into the model would lead to the failure of the underlying time dimension representation. To eliminate this representation barrier, this embodiment abandons the traditional sequence completion interpolation method and instead designs a phase transformation approach. The sampling timestamps of the opposite-side IoT sensor dynamic time-series data are processed for dynamic time-series harmonic normalization, mapping the discrete observation time into a continuous periodic physical space. This enables the capture of the traffic tidal impact patterns of heavy trucks during the day and night, as well as the temperature gradient fatigue effect of the road surface structure during seasonal changes.

[0019] In dynamic time-series harmonic normalization processing, let's assume that for any asset node... An asset node is an object that maps the physical entity of a highway asset to a virtual space with high fidelity. Its environment and load sensing data are represented by absolute timestamps of the sampling. The absolute timestamp of the sampling is The initial reference time point for system twin archiving, i.e., the asset node. The initial reference time point is Define asset nodes The fundamental periodic phase vector and digital asset nodes generated through linear transformation Dynamic time-series harmonic eigenvectors The calculation formula for dynamic time-series harmonic normalization is as follows: , , in, Pi The diurnal tidal cycle, set to 24 hours, is used to capture the peak-valley variation patterns of short-term traffic flow. The environmental seasonal cycle is set to 365 days to capture the long-period meteorological and environmental stress alternation law. and These are sine and cosine functions used to smoothly transform linear-time signals into closed-loop periodic signals. This represents the horizontal concatenation of vector elements along the channel dimension. The linear projection matrix for learnable time features maps the four-dimensional basic periodic vector to the high-dimensional hidden layer space defined by the model, thereby generating a standardized time-series vector that filters out sampling noise and contains inherent dynamic rhythms.

[0020] In processing BIM static geometric topology data, it is important to consider that highways are not isolated individual pieces, but rather a continuous mechanical transmission system. For example, minor uneven settlement of bridge abutments can increase the dynamic impact coefficient of adjacent approach road surfaces through physical structure, leading to accelerated pavement damage; poor drainage in a certain section can cause excessive moisture content in adjacent roadbeds. Traditional BIM 3D geographic coordinates can only represent static locations and cannot truly reflect the physical transmission characteristics of such loads and defects. Therefore, this embodiment first parses the actual mechanical transmission paths between assets from the BIM model to define the adjacency relationships between assets. Then, it uses a graph attention mechanism to adaptively learn the strength of this physical transmission, reconstructing a dynamic asset-related manifold to obtain the asset spatial topology manifold vector.

[0021] In the process of obtaining the asset space topology manifold vector, for asset nodes and its physical adjacent nodes Extracting asset nodes from BIM static geometric topology data, such as from BIM and asset ledgers. , The set of original geometric dimensions, structural form, and material constitutive properties of are denoted as […]. and Calculate adjacent nodes For nodes Physical effect of conduction attention coefficient Based on the transmission attention coefficient Graph attention mechanism is used to aggregate and generate asset nodes. spatial topological manifold vector The calculation formula is as follows: , , Asset nodes defined based on mechanical transmission paths The neighborhood group, This is an index of physically adjacent nodes. and They are nodes and nodes The set of original geometric dimensions, structural form, material, and constitutive property characteristics. The spatial transformation weight matrix is ​​responsible for uniformly projecting the original low-dimensional attribute features to the high-dimensional representation computation space of the model. This is a vector concatenation operator used to combine node features side-by-side. A learnable shared attention weight vector. This represents the transpose operation of a matrix or vector, used to quantify and evaluate the response intensity of structural mechanical coupling or disease propagation between adjacent nodes under specific working conditions. It is a linear rectified activation function with leakage; Represents a node For nodes The physical action of the attention coefficient For traversing the adjacency set The internal summation index variable, Represents any node in the adjacency set For nodes The physical effect conduction attention coefficient is used as a denominator to calculate the sum of the influence of all adjacent nodes; For the natural constant The calculation formula uses an exponential function with base 0.5. In this embodiment, the attention weights are globally normalized through the Softmax mechanism. A feature aggregation and summation operator for neighboring nodes. For a nonlinear activation function with Gaussian error, It is a spatial topological manifold vector that deeply contains the local structural morphology of assets in the entire road network and the complex mechanical boundary constraints.

[0022] In processing unstructured text data from manual inspections, relying solely on numerical fluctuations and spatial topology maps from IoT sensor data is insufficient to comprehensively depict the overall health of assets. To address this high-tech challenge, this step employs a large language model (LLM) within a vertical domain to semantically analyze this valuable experience from unstructured text data. Asset nodes are defined... The unstructured text data of manual inspection is This includes long historical inspection texts and maintenance record sequences, with semantic parsing used to extract high-order text semantic vectors. The calculation formula is as follows: , This is the encoding layer of a large language model fine-tuned from a deep knowledge graph in the field of highway engineering. It can identify engineering terms and extract the severity and evolution trend of potential defects. These are pre-defined, exclusive guidance word vectors for different asset types, used to force the language model's attention to focus on structural health diagnosis and operational risk assessment.

[0023] The above alignment and blending is achieved through the following calculation formula: , For asset nodes The dynamic time-series harmonic eigenvectors, For asset nodes spatial topological manifold vectors This is an operator for concatenating temporal and spatial features. This is a spatiotemporal modal adaptive mapping matrix, used to adjust the dimensions of the spatiotemporal features after concatenation. This is the text semantic alignment matrix, which is used to eliminate the scale difference between numerical physical features and symbolic semantic features; the + sign indicates the element-wise addition operation of the feature vectors; This is a layer normalization operator to eliminate the gradient divergence problem caused by excessive variance in numerical distribution when merging multi-source data; For asset nodes that deeply integrate continuous dynamic rhythms, macroscopic physical topological constraints, and microscopic disease diagnosis semantics The comprehensive asset base characteristics.

[0024] S1 in this embodiment breaks down the physical separation between dynamic temporal data from roadside IoT sensors, static geometric topology data from BIM, and unstructured textual data from manual inspections. In reality, the sensing data of highways often exhibits an extremely uneven and discrete state due to equipment power supply, network latency, or event-triggered mechanisms (such as dynamic weighing sampling only when vehicles pass by). At the same time, the spread of defects between assets has a strong physical topological dependency. S1 in this embodiment processes temporal distortion through continuous harmonic coding, transforms static BIM into a dynamic mechanical conduction manifold, and designs a large language model to deeply mine the latent degradation patterns in maintenance texts, thereby constructing a high-dimensional asset basis tensor with rigorous physical laws and rich semantic connotations for the subsequent causal inference module.

[0025] S2. Based on the comprehensive asset base characteristics, generate a causal adjacency matrix, design an interference quantifier to perform counterfactual state differential evolution deduction, and generate a counterfactual healthy state; based on the causal adjacency matrix and the counterfactual healthy state, obtain the deduction confidence and trajectory tensor.

[0026] First, to conduct reliable causal inferences, it is necessary to clearly define the true causal relationship between traffic dynamic loads, environmental stresses, and structural damage. Therefore, this embodiment constructs a structural causal model (SCM) based on comprehensive asset base features, adaptively discovers the causal network topology of asset degradation, and applies a self-attention mechanism and sparsity thresholding to the comprehensive asset base features rich in full spatiotemporal dynamics and textual semantic information to dynamically generate a causal adjacency matrix representing the strength of causal associations. The calculation formula is as follows: , For asset nodes The causal adjacency matrix represents the asset nodes. The intrinsic causal network topology in a microscopic environment For asset nodes The comprehensive asset base characteristics, This is the causal emitter feature mapping matrix, used to extract potential causal precondition features from the basis features. This is a causal receiver feature mapping matrix, used to extract potential outcome response features. This is a matrix transpose operation used to calculate the inner product of the feature matrices of the transmitter and receiver while maintaining dimension alignment. The scaling factor is the number of hidden layer dimensions for the mapped features, used to prevent the inner product result from being too large, which could lead to the vanishing gradient phenomenon in subsequent layers. It is a normalized exponential function, responsible for transforming the original correlation degree obtained from the inner product calculation into a probability distribution form with a sum of 1; The sparsity threshold function is used to cut off weak connections caused by spurious correlations or data coincidences, and to force the retention of strong causal transmission paths that conform to the laws of engineering physics, thereby ensuring that the final causal graph strictly satisfies the mathematical assumption of directed acyclicity.

[0027] Then, based on the causal adjacency matrix, the designed intervention predictor is used to perform counterfactual state differential evolution deduction, generating a counterfactual health state. Specifically, after successfully obtaining the causal network skeleton reflecting physical laws, this embodiment breaks through the traditional simple temporal extrapolation and performs high-order counterfactual deduction, that is, deducing how the health evolution state of the asset will change if a specific maintenance intervention is implemented at a specific time in the future. Therefore, this embodiment designs the intervention predictor. By combining differential equations with state integration, a time-varying vector of human physical intervention actions, i.e., asset nodes, is defined. Human physical intervention action vector To deduce asset nodes Future target time Counterfactual health status The calculation formula is as follows: , For asset nodes The comprehensive asset base characteristics, It is a definite integral operator used to accumulate state changes over time. This is the starting time of the simulation, i.e., the current observation time. For continuous-time integral infinitesimal variables; This is a nonlinear asset natural decay function, parameterized by a deep neural network, specifically designed to simulate the natural fatigue process of materials without external intervention. For asset nodes At the micro-element moment The instantaneous health state tensor is not directly obtained through objective data acquisition from external sensors, but rather derived from the initial state by a numerical integration algorithm when solving the counterfactual differential evolution equation. With time microelement Intermediate inference variables are dynamically generated through stepwise accumulation of natural decay rate and causal transmission term of physical intervention; For asset nodes The causal adjacency matrix, This is the intervention effect mapping function, responsible for mathematically transforming specific engineering maintenance actions into a repair gain vector in a high-dimensional feature space. For asset nodes At the micro-element moment Injected virtual intervention actions, For causal repair transmission terms, ensure that repair benefits are transmitted to relevant latent variables strictly following the causal network topology.

[0028] Finally, based on the causal adjacency matrix and the counterfactual health state, the inference confidence and trajectory tensor are obtained. Since any long-term inference about the future inevitably involves uncertainty, especially after designing complex counterfactual interventions, the reliability of the prediction results directly affects the subsequent financial valuation of digital assets. Therefore, this embodiment quantifies the inference confidence by calculating the information entropy of the causal adjacency matrix and integrates it with the counterfactual state to generate a trajectory tensor with a risk boundary. The calculation formula is as follows: , , For asset nodes Quantify the confidence level of this counterfactual causal inference. It is an exponential function. This is a penalty adjustment coefficient used to control the extent to which uncertainty reduces the confidence level; An information entropy calculation operator used to measure asset nodes. Causal adjacency matrix The higher the entropy value of the matrix, the more ambiguous the causal relationship and the lower the confidence level of the inference; For asset nodes The trajectory tensor contains information about the health evolution state and the predicted risk boundary. This is the channel alignment matrix. To construct the concatenation operator, the high-dimensional health state tensor and the scalar confidence are connected along the feature dimension. For layer normalization operation; For asset nodes The counterfactual health status; the final output of the calculation formula It not only provides accurate asset evolution trends, but also comes with a credibility label, serving as the core trump card for the bidding game between the security agent and the economic agent in the subsequent "risk-utility double-blind game".

[0029] Existing pure data-driven models can only discover statistical correlations between data, but cannot understand the underlying causal logic. When faced with unknown extreme weather conditions or when it is necessary to evaluate a preventive maintenance strategy that has never been implemented before, traditional models lacking physical mechanism constraints are prone to prediction failure. To solve this technical problem, S2 in this embodiment adaptively discovers the underlying causal network topology of asset degradation from high-dimensional basis features. Then, by designing the interference quantifier in causal calculus, it realizes the counterfactual differential evolution deduction of specific physical intervention actions, calculates the causal confidence of the counterfactual deduction and generates a trajectory tensor with uncertainty boundaries, thereby accurately predicting the asset's full life cycle decay trajectory under different maintenance strategies.

[0030] S3. Based on trajectory tensor and inference confidence, construct a safe intelligent agent and an economic intelligent agent, conduct a double-blind game, reach a consensus on asset valuation under Nash equilibrium constraints, and generate the dynamic fair value of digital assets.

[0031] In the actual operation of highways, the valuation and maintenance decisions of digital assets are often entangled in conflicts of interest among multiple parties: the management perspective representing safety supervision tends to frequently carry out major repairs in order to pursue absolute structural safety and extremely low risk; while the management perspective representing financial operations tends to delay maintenance in order to reduce current cost expenditures and pursue maximum economic benefits. The traditional weighted average method cannot resolve this fundamental conflict of objectives. To solve this technical problem, the technical details of this embodiment are as follows.

[0032] First, based on the trajectory tensor and the inference confidence, a safety agent and an economic agent are constructed. Since the primary premise of game theory is establishing the core demands of both agents in negotiation, i.e., their utility functions, the utility of the safety agent depends on the degree to which the predicted asset health is significantly higher than the safety threshold, while the utility of the economic agent depends on the degree to which maintenance intervention costs are significantly lower than the budget ceiling. To construct the safety and economic agents, this embodiment decouples the health score and the inference confidence from the trajectory tensor, constructing an asymmetric utility evaluation system for both parties, and calculating the utility values ​​of the safety agent and the economic agent using the following formulas: , , For asset nodes The utility value of the safety agent is the satisfaction utility value of the safety agent with the current predicted trajectory, and its value range is [range missing]. , It is a sigmoid activation function used to smoothly map linear interpolation to the probability utility space. This is a safety sensitivity adjustment coefficient used to control the agent's tolerance to potential diseases. For asset nodes The confidence level of the projection, used here as a utility adjustment lever, means that the more reliable the prediction, the more certain the utility of the safety agent's evaluation. A nonlinear health dimensionality reduction scoring function is used to reduce the dimensionality of high-dimensional trajectory tensor asset nodes. trajectory tensor Transformed into an intuitive percentage-based health metric. These are the legally mandated structural hazard thresholds in highway safety regulations, such as the minimum bearing capacity of bridges. For asset nodes The utility value of the economic intelligent agent For the natural constant An exponential function with base 0. This is a cost sensitivity adjustment coefficient; This is a function for calculating project cost, used to calculate asset milestones. Virtual intervention actions The actual financial expenses incurred This represents the upper limit of the maintenance budget allocated to this asset node within the current financial cycle.

[0033] Then, after establishing the utility evaluation criteria for the security agent and the economic agent, the two agents need to make initial bids for the current financial value of the digital asset back-to-back. The security agent will significantly lower the asset valuation due to minor defects, while the economic agent will ignore minor hidden dangers to maintain a high asset valuation. In this embodiment, a neural network is used to generate the independent bids of both parties, and a game adversarial loss function is constructed to quantify the negotiation differences at this time. The formulas for calculating the bid tensors of both parties and the game loss are as follows: , , , For asset nodes The security agent's bid tensor is an asset financial valuation bid vector generated by the security agent based on risk-averse logic. A policy generation network for security agents, whose inputs include asset nodes. The counterfactual trajectory tensor and its own utility perception. For asset nodes The trajectory tensor, For asset nodes The utility value of the secure intelligent agent. For asset nodes The bid tensor of an economic agent is an asset financial valuation bid vector generated by the economic agent based on profit-seeking logic. A strategy generation network for economic intelligent agents; For asset nodes The utility value of the economic intelligent agent For asset nodes The game loss scalar is the adversarial loss scalar that guides the evolution of double-blind games. This represents calculating the square of the L2 norm over the vector difference, which quantifies the Euclidean distance between the bids of both parties, forcing them to narrow their differences in subsequent iterations. This is a joint utility balance coefficient used to adjust the ratio of compromise to confrontation. It is a natural logarithmic function; the part with the negative sign has its core mathematical meaning derived from the Nash negotiation solution, which aims to maximize the product of the combined utility of both parties, thereby ensuring that the game process does not degenerate into a unilateral extreme compromise by one party when seeking consensus, but rather approaches the Pareto optimal state in economics.

[0034] Finally, the security agent and the economic agent engage in a double-blind game, reaching a consensus on asset valuation under Nash equilibrium constraints, thus generating a dynamic fair value for the digital asset. After calculating the game loss, the agent's policy network is iteratively updated through multiple rounds using a gradient descent algorithm. When the loss function converges, it means that both parties have reached a Nash equilibrium state under the given physical evolution constraints. At this point, the final bids of both parties no longer fluctuate drastically. This embodiment will integrate the converged bids of both parties and superimpose compliance penalties from the macroeconomic policy level to generate the final dynamic fair value of the digital asset used for financial accounting or decision-making transactions. The calculation formula is as follows: , For asset nodes The dynamic fair value of digital assets. , Asset nodes The optimal utility values ​​of the security agent and the optimal utility values ​​of the economic agent represent the optimal utility values ​​finally locked by the security agent and the economic agent after multiple rounds of game iterations and convergence. , Asset nodes The optimal safety agent bidding tensor and the optimal economic agent bidding tensor represent the asset valuation bids determined by the final compromise between the two parties when the game reaches Nash equilibrium. The main part within the parentheses of the formula uses an adaptive weighted average based on utility proportions, meaning that in equilibrium, the valuation is more skewed towards whoever has stronger utility. If causal inference shows extreme danger, then safety utility will dominate. The Hadamard product of vectors, i.e., the element-wise multiplication operator. For asset nodes The compliance penalty deduction vector is a veto mechanism. When it is predicted that the asset's condition will inevitably reach a fatal safety threshold in a very short period of time and the current intervention strategy fails, the value of this vector approaches 1, forcibly resulting in a precipitous reduction in the final valuation of the asset. It not only profoundly reflects the current physical properties of the asset, but also encompasses the risks of future repair costs, thus achieving a unity of engineering value and financial value.

[0035] S4. Based on the dynamic fair value of digital assets, a dynamic right-of-way monetization revenue matrix is ​​generated. Combined with global budget constraints, the optimal maintenance strategy is obtained by optimizing the preventive maintenance path at the road network level. After anti-tampering processing, the digital asset ledger block is obtained.

[0036] First, because traditional toll pricing uses static standards, it fails to reflect the real-time asset depreciation of the road surface during actual use. To address this technical issue, this embodiment designs a pricing and revenue calculation mechanism based on the dynamic fair value of digital assets. Combined with real-time traffic flow data, it calculates the marginal value loss of digital assets caused by single-vehicle load, thereby dynamically generating the expected operating revenue of asset nodes within the current financial cycle. This not only reflects direct toll revenue but also includes the implicit financial changes brought about by asset depreciation. The calculation formula is as follows: , For asset nodes The expected operating income represents the asset's current financial liquidity under a dynamic pricing strategy. For asset nodes within the current statistical period The equivalent axle load cumulative traffic flow vector, For element-wise multiplication of vectors, This is the vector of the basic statutory toll rates for highways. The hyperbolic tangent activation function is used to smooth the nonlinear mapping of the value depreciation rate to a reasonable price adjustment range. It is a learnable rate elasticity mapping weight matrix, responsible for transforming pure asset differences into pricing economic indicators. For asset nodes The absolute value of the initial filing. For asset nodes The dynamic fair value of digital assets. It is a very small positive constant, specifically used to prevent division-by-zero overflow exceptions caused by a denominator of zero. This serves as a macro-level constraint vector for transportation pricing policies, acting as a mask constraint to ensure that the final generated price adjustment coefficient remains strictly within the compliant range.

[0037] Then, due to the strict upper limit on preventive maintenance funds for the entire road network, the system must coordinate maintenance resources globally at the road network level. Therefore, this embodiment deeply couples expected operating revenue with the dynamic fair value of digital assets, aiming to maximize both realization revenue and inventory value. A Lagrange cost functional is constructed to solve for the optimal preventive maintenance execution strategy allocated to each asset node. The calculation formula is as follows: , For asset nodes The optimal preventative maintenance strategy, To find the mathematical optimization operator for the set of independent variables that maximizes the objective functional, A candidate set of compliant engineering intervention actions, This represents the total number of physical nodes in the road network. This is a summation operator for all asset nodes in the entire road network. This is a liquidity preference adjustment coefficient, ranging from 0 to 1. A larger value indicates that the system prioritizes current cash flow returns such as toll fees, while a smaller value indicates that the system prioritizes the long-term value preservation rate of the underlying digital assets. In order to attempt to allocate assets to nodes during the optimization process The candidate preventive maintenance intervention actions are the independent variables of the objective optimization function. For asset nodes The expected operating revenue, at this point, appears as a function dependent on intervention actions. For asset nodes The dynamic fair value of digital assets. This is the budget overrun penalty multiplier, used to control the cost of exceeding the budget limit and disrupting the overall optimization objective. The activation function is a linear rectification function, ensuring that the penalty term is triggered only when the actual total cost exceeds the limit. To accurately calculate the actual engineering cost function for a specific intervention action; Given the current fiscal year's overall preventative maintenance budget limit, the solution is as follows: Pareto optimal allocation at the road network level under funding constraints was achieved.

[0038] Finally, to meet the financial compliance requirements for subsequent highway asset audits, traffic accident liability tracing, and asset securitization transactions of infrastructure real estate investment trusts, the core derivation data of the entire decision-making chain, namely the optimal preventive maintenance execution strategy, must be cryptographically solidified to achieve tamper-proof processing. Therefore, this embodiment designs a hash package of the optimal preventive maintenance execution strategy, comprehensive asset base characteristics, dynamic fair value of digital assets, and expected operating income to generate a digital asset ledger block with a spatiotemporal stamp. The calculation formula is as follows: , For asset nodes The digital asset ledger block can be directly linked to the Transportation Consortium blockchain network as proof. This is a high-strength secure hash algorithm function used to map heterogeneous input tensors to fixed-length unique digital fingerprints. , , , Asset nodes The comprehensive asset base characteristics, dynamic fair value of digital assets, expected operating income, and optimal preventive maintenance execution strategy; This is the splicing operator for the feature channel dimension. It is this operator that seamlessly stitches together the evolution results of all preceding steps, ensuring the integrity of the causal logic chain. For absolute physical system timestamps; It is an XOR logic cryptographic operator; This is a digital signature vector for the management department's private key, used to confirm the legitimate ownership of the ledger.

[0039] The method in this embodiment overcomes the problem of spatiotemporal semantic fragmentation of multimodal asset data in virtual space, and realizes a high-fidelity twin with high physical self-consistency.

[0040] To verify the effectiveness of the dynamic valuation method in this embodiment, an experiment was conducted. Two datasets with different characteristics were constructed and compared: a public benchmark traffic flow and road surface dataset, and a self-built full lifecycle multidimensional asset twin dataset.

[0041] Dataset Description. Trans-Public (Public Benchmark Dataset): Collected from the internationally available Federal Highway Administration LTPP (Long-Term Road Performance) database and public traffic flow network data. This dataset mainly includes a single-modal Pavement Damage Index (PCI), International Roughness Index (IRI), and basic traffic flow time series. It features broad coverage and regular time series, primarily used to verify the model's basic generalization ability in handling standard spatiotemporal road performance prediction tasks. Since this dataset does not include BIM geometric topology, unstructured maintenance text, or detailed engineering financial costs, and does not involve multi-party intelligent agent game decision-making, some high-order indicators (such as CF-RMSE and Consensus-Rate) are not applicable to this dataset (N / A). Highway-Asset-Twin (Self-built Multidimensional Asset Twin Dataset): This is a high-difficulty, real-world measured dataset built by this research based on the actual highway network of a provincial transportation control group. It simulates and measures the entire "physical-digital-financial" process data covering continuous long downhill slopes, cross-river bridges, and heavy-load passages. This dataset contains multi-source heterogeneous data: high-frequency WIM (Dynamic Weighing) and weather station time-series data, BIM static topology and constitutive parameters accurate to the component level, and unstructured manual inspection and maintenance logs spanning up to 5 years. Historical preventative maintenance project settlement data was also included as a financial game counterpart. Fine-grained physical causality and compliance annotations were performed by experts in both transportation engineering and asset appraisal, aiming to verify the core advantages of this invention in handling counterfactual inference, multimodal alignment, and risk-cost double-blind game theory.

[0042] Experimental Setup. To comprehensively evaluate performance, four existing advanced methods were selected for comparison with the method in this embodiment. The existing methods are as follows: CNN-LSTM method: Traffic flow and road surface detection indicators are input into convolutional layers in chronological order to extract features, and then LSTM is used for temporal extrapolation prediction, representing a pure data-driven deep learning benchmark that ignores physical topology and causal logic; LCCA (Lifecycle Cost Analysis) method: A basic engineering economics evaluation method that uses a fixed Markov transition matrix to predict state decay and perform financial discounted valuation, representing the traditional static experience-based asset management approach; ST-GNN (Spatiotemporal Graph Neural Network) method: Uses a standard spatiotemporal graph convolutional network to process road grid point data, representing a strong AI competitor in the current traffic prediction field, but it does not combine structural causal models (SCM) and game constraints; MADDPG (Multi-Agent Deep Deterministic Policy Gradient): Capable of handling multi-agent reinforcement learning problems, but lacks explicit mathematical joint utility constraints on physical conservation and Nash bargaining mechanisms. Finally, the CAG-Net (Causal Game Model) method proposed in this embodiment is presented.

[0043] Evaluation Metrics. The experiment uses five core metrics for evaluation. **Estimated Mean Absolute Percentage Error (↓):** Measures the degree to which the model's financial valuation matches the actual market inventory value; a lower value indicates a more accurate valuation. **Counterfactual Root Mean Square Error (↓):** Specifically measures the accuracy of projecting future asset health under specific virtual maintenance interventions (e.g., simulating a micro-surfacing treatment); a lower value indicates a more reliable projection. **Causal Consistency Residual (↓):** Measures whether the predicted asset degradation trajectory violates underlying physics and the laws of material aging (e.g., pseudo-physical phenomena where health rebounds without intervention); a lower value indicates better physical consistency. **Consensus Achievement Rate (↑):** Measures the frequency of strategic agreement between the security agent and the economic agent after game interaction; a higher value indicates a stronger conflict resolution capability. **Increase in Return on Investment (↑):** Measures the increased financial and economic benefits of the model's global preventative maintenance strategy compared to traditional empirical strategies; a higher value is better.

[0044] The statistical results of the experimental data are shown in Table 1, and the corresponding comparisons are shown in Table 2. Figure 2 , Figure 3 , Figure 4 and Figure 5 By conducting an in-depth comparative analysis of the experimental data of each model in Table 1 on the dual dataset, the technical advantages of this embodiment in handling the dynamic valuation and complex strategy optimization tasks of highway digital assets can be clearly revealed. The specific analysis is as follows.

[0045] Table 1. Performance comparison of different methods on the Trans-Public and Highway-Asset-Twin datasets.

[0046] It should be noted that in the Trans-Public basic dataset in Table 1, all models show N / A (Not Applicable) for both the "Counterfactual Root Mean Square Error" and "Consensus Achievement Rate" higher-order metrics. This is because the public benchmark dataset only contains traditional road damage indicators and traffic flow time series, lacking the engineering intervention variables necessary to support counterfactual inference, such as maintenance action labels. Furthermore, it lacks real financial cost details and safety assessment red lines, making it impossible to trigger a risk and cost double-blind game mechanism among multiple agents. This data gap itself also indirectly confirms that traditional "pure perception" monitoring data and prediction tasks are far from meeting the high-order digital asset management needs of modern highways with "physical-financial" dual causal mapping, thus further highlighting the necessity of constructing a complex multidimensional twin foundation in this embodiment.

[0047] Secondly, compared with CNN-LSTM and traditional LCCA benchmark models, this embodiment demonstrates an overwhelming advantage in handling counterfactual evolution inference. Experimental data shows that on the complex Highway-Asset-Twin dataset with multiple intervention variables, the counterfactual root mean square error of CNN-LSTM is as high as 0.85, and its estimated mean absolute percentage error even deteriorates to 24.6%. This indicates that the pure data-driven model is deeply trapped in the "correlation trap" and almost completely loses its ability to predict the aftereffects of maintenance strategies under complex road conditions. Although traditional LCCA has good physical regularity (causal consistency residual of 0.12), its Markov chain is extremely rigid and has very poor adaptability to dynamic variables. In contrast, this embodiment, through the "digital asset counterfactual evolution inference module based on structural causal model," reconstructs the asset decay differential equation using intervention variables, significantly reducing the counterfactual root mean square error to 0.18. This proves that this embodiment can not only understand past data but also accurately answer complex engineering questions such as what would happen if the strategy were changed.

[0048] Furthermore, compared to the mainstream spatiotemporal graph network ST-GNN, this embodiment achieves significant superiority in spatiotemporal physical topology consistency and estimation accuracy. While ST-GNN performs reasonably well on simple trans-public datasets (estimation error 10.8%), on complex datasets, due to the lack of cross-modal semantic alignment, its causal consistency residual rises to 0.36, indicating that the model generates non-physical artifacts such as "increased vehicle traffic but road surface cracks automatically heal." This embodiment, through the "spatiotemporal manifold alignment and semantic fusion module for multi-source heterogeneous asset data," not only reshapes the non-Euclidean manifold of mechanical transmission using graph attention networks, but also designs a large model to accurately extract the latent disease patterns in maintenance texts. Even on complex datasets, it still controls the causal consistency residual at an excellent 0.08, thereby stabilizing the estimation error of digital assets at 9.2%, achieving true high-fidelity twins.

[0049] Finally, compared with the multi-agent reinforcement learning model MADDPG, this embodiment achieves a breakthrough improvement in resolving the conflict between the goals of "safety compliance" and "financial cost reduction". MADDPG is prone to getting stuck in local optima or policy oscillations when faced with massive data conflicts in real-world engineering projects, with a consensus rate of only 71.5%, meaning the system cannot make a clear decision nearly 30% of the time. This embodiment, through a "risk-utility double-blind game-based dynamic asset valuation and strategy consensus module", uses the Nash negotiation solution as a hard constraint on the loss function, forcing both parties to compromise with the goal of maximizing the joint utility product. This underlying mathematical mechanism effectively eliminates the extreme bias of a single perspective, not only pushing the consensus rate in complex scenarios to 96.8%, but also achieving Pareto optimal allocation of all road network resources under the hard constraint of global budget, resulting in a return on investment improvement of up to 12.5%, nearly double that of the suboptimal model.

[0050] In summary, this embodiment not only demonstrates robustness in general traffic timing tasks, but also surpasses mainstream benchmark solutions in highly challenging and complex scenarios such as multimodal causal alignment of highway digital assets, counterfactual evolutionary inference, multi-party utility intelligent game theory, and global funding constraint optimization. Particularly under the impact of complex data streams, this method exhibits excellent noise resistance and physical reliability, proving the practical value of this interdisciplinary technical approach in intelligent transportation asset management.

[0051] Existing techniques such as the Spatiotemporal Graph Neural Network (ST-GNN) often lack deep constraints on the physical and mechanical transmission mechanisms when constructing digital twins, leading to frequent artifacts in the generated prediction results that violate physical common sense. This embodiment designs a spatiotemporal manifold alignment and semantic fusion mechanism for multi-source heterogeneous asset data, combining graph attention networks and a large language model to map discrete geometric topologies, dynamic temporal sequences, and empirical texts to a unified high-dimensional feature space. Experiments show that even on the complex and heterogeneous Highway-Asset-Twin dataset, the causal consistency residual of this embodiment remains as low as 0.08, a reduction of nearly 77.8% compared to ST-GNN (0.36), eliminating non-physical artifacts and ensuring the rigor of the virtual asset mapping.

[0052] Secondly, this embodiment successfully overcomes the "correlation trap" of purely data-driven models, endowing digital asset models with powerful counterfactual evolutionary deduction capabilities. Traditional deep learning models such as CNN-LSTM cannot distinguish between causality and coincidence in the asset degradation process, and are prone to prediction collapse when faced with unknown intervention strategy evaluations. This embodiment innovatively designs a structural causal model (SCM) and an intervention quantifier to construct the counterfactual differential evolution equation for asset degradation. Experimental data show that under complex road network conditions with multiple intervention variables, the counterfactual root mean square error of this embodiment for a specific maintenance strategy is as low as 0.18, while that of CNN-LSTM is as high as 0.85. This achieves a leap from passively fitting history to actively inferring the future, providing a solid physical inference foundation for high-precision asset valuation.

[0053] Furthermore, this embodiment effectively resolves the fundamental conflict between safety supervision and financial cost reduction, achieving an objective and fair consensus on asset valuation. Addressing the challenge of multi-party interest games in highway operations, existing reinforcement learning techniques such as MADDPG are prone to policy oscillations or local deadlocks when data conflicts are intense. This embodiment constructs a double-blind game mechanism for both safety and economics, and creatively uses the Nash negotiation solution as a hard loss constraint in the game evolution, forcing agents to seek maximum joint utility. Experimental results demonstrate that in complex decision-making scenarios, this embodiment achieves a consensus rate as high as 96.8%, far exceeding MADDPG's 71.5%. This not only significantly reduces decision-making deadlock time but also stabilizes the average absolute percentage error of the final generated digital asset valuation at 9.2%, achieving a perfect balance between engineering safety baselines and financial assessment value.

[0054] Finally, this embodiment achieves a leap from individual asset valuation to a comprehensive, network-level preventative maintenance approach, improving the return on investment for highway assets. Existing traditional methods such as Life Cycle Cost Analysis (LCCA) often only provide static, localized assessments. This embodiment deeply couples dynamic asset valuation, expected right-of-way monetization revenue, and global financial budgeting, constructing a network-level Lagrange cost functional. Through intelligent optimization, it generates a tamper-proof digital asset decision ledger. Experiments demonstrate that the Pareto-optimal maintenance scheduling strategy output by this embodiment can bring highway operators up to a 12.5% ​​increase in return on investment compared to traditional experience-based management models, promoting the high-level value realization and sustainable preservation and appreciation of highway digital assets.

[0055] This embodiment also provides a dynamic valuation system for digital assets of highways, used to implement the above-mentioned dynamic valuation method for digital assets of highways, including: The comprehensive asset base feature generation module is used to acquire multi-source heterogeneous asset data of highway assets, including dynamic time-series data of roadside IoT sensors, static geometric topology data of BIM, and unstructured text data of manual inspection; it performs spatiotemporal manifold alignment and semantic fusion on the multi-source heterogeneous asset data to generate comprehensive asset base features; The module for generating inference confidence and trajectory tensors is used to generate a causal adjacency matrix based on comprehensive asset base features, design an interference quantifier to perform counterfactual state differential evolution inference, and generate a counterfactual healthy state; based on the causal adjacency matrix and the counterfactual healthy state, the inference confidence and trajectory tensor are obtained. The digital asset dynamic fair value generation module is used to construct a secure intelligent agent and an economic intelligent agent based on trajectory tensor and inference confidence, conduct double-blind game, reach a consensus on asset valuation under Nash equilibrium constraints, and generate the dynamic fair value of digital assets. The digital asset ledger block generation module is used to generate expected operating revenue based on the dynamic fair value of digital assets. With the goal of maximizing the combined realization revenue and inventory value, it constructs a Lagrange cost functional, solves for the optimal preventive maintenance execution strategy, and cryptographically solidifies the combined asset base characteristics, the dynamic fair value of digital assets, and the expected operating revenue to obtain the digital asset ledger block.

[0056] This embodiment also provides a dynamic valuation device for digital assets of highways, including a processor and a memory, wherein the processor executes a computer program stored in the memory to implement the above-mentioned dynamic valuation method for digital assets of highways.

[0057] This embodiment also provides a computer-readable storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the above-described dynamic valuation method for digital assets of highways.

[0058] While exemplary embodiments of the invention have been described herein, many other variations or modifications conforming to the principles of the invention can be directly determined or derived from the disclosure of this invention without departing from its spirit and scope. Therefore, the scope of the invention should be understood and recognized to cover all such other variations or modifications.

Claims

1. A dynamic valuation method for digital assets of highways, characterized in that, This includes the following operations: S1. Acquire multi-source heterogeneous asset data of highway assets, including dynamic time-series data from roadside IoT sensors, static geometric topology data from BIM, and unstructured text data from manual inspections; perform spatiotemporal manifold alignment and semantic fusion on the multi-source heterogeneous asset data to generate comprehensive asset base features; S2. Based on the comprehensive asset base characteristics, generate a causal adjacency matrix, design an interference quantifier to perform counterfactual state differential evolution deduction, and generate counterfactual health states; Based on the causal adjacency matrix and counterfactual health status, the inference confidence and trajectory tensor are obtained; S3. Based on trajectory tensor and inference confidence, construct a security intelligent agent and an economic intelligent agent, conduct double-blind game, reach consensus on asset valuation under Nash equilibrium constraints, and generate dynamic fair value of digital assets. S4. Based on the dynamic fair value of digital assets, generate expected operating revenue. With the goal of maximizing the combined revenue from realization and inventory value, construct a Lagrange cost functional and solve for the optimal preventive maintenance execution strategy. Combine this with the comprehensive asset base characteristics, the dynamic fair value of digital assets, and the expected operating revenue, and cryptographically solidify it to obtain the digital asset ledger block.

2. The dynamic valuation method for digital assets of highways according to claim 1, characterized in that, The spatiotemporal manifold alignment and semantic fusion operations in S1 include: The sampling timestamps of the IoT sensor dynamic time series data are processed by dynamic time series harmonic normalization, which maps discrete observation time to continuous periodic physical space and generates dynamic time series harmonic feature vectors. The mechanical transmission paths between assets are extracted from the static geometric topology data of BIM, the adjacency relationships between assets are defined, and the asset space topology manifold vector is reconstructed through graph attention mechanism. Semantic parsing of unstructured text data from manual inspection is performed using a large language model in a vertical domain to obtain high-order text semantic vectors; By aligning and fusing dynamic temporal harmonic feature vectors, spatial topological manifold vectors, and high-order textual semantic vectors, a comprehensive asset-based feature is obtained.

3. The dynamic valuation method for digital assets of highways according to claim 1, characterized in that, The causal adjacency matrix in S2 is obtained by applying a self-attention mechanism and a sparsity threshold to the comprehensive asset base features.

4. The dynamic valuation method for digital assets of highways according to claim 1, characterized in that, In S2, the counterfactual evolutionary deduction operation is implemented through the following formula: ， For asset nodes Future target time Counterfactual health conditions For asset nodes The comprehensive asset base characteristics, For definite integral operators, This is the starting moment of the simulation. For continuous-time integral differential variables, For nonlinear asset natural decay function, For asset nodes At the micro-element moment The instantaneous health state tensor, For asset nodes The causal adjacency matrix, Let be the intervention effect mapping function. For asset nodes At the micro-element moment Injected virtual intervention actions.

5. The dynamic valuation method for digital assets of highways according to claim 1, characterized in that, The formulas for calculating the inference confidence and trajectory tensor in S2 are as follows: ， ， For asset nodes The confidence level of the inference, It is an exponential function. The penalty adjustment coefficient, Operators for calculating information entropy. For asset nodes The causal adjacency matrix, For asset nodes The trajectory tensor, This is the channel alignment matrix. To concatenate operators, For layer normalization operation, For asset nodes The counterfactual state of health.

6. The dynamic valuation method for digital assets of highways according to claim 1, characterized in that, In the S3 double-blind game, the adversarial loss function is calculated as follows: ， ， ， For asset nodes The scalar of game loss, This represents calculating the square of the L2 norm for the vector difference. The joint utility balance coefficient, It is the natural logarithm function. For asset nodes The secure intelligent agent bidding tensor. For asset nodes The economic agent's bid tensor For asset nodes The utility value of the secure intelligent agent. For asset nodes The utility value of the economic intelligent agent A policy generation network for secure intelligent agents. For asset nodes The trajectory tensor, A strategy generation network for economic intelligent agents.

7. The dynamic valuation method for digital assets of highways according to claim 1, characterized in that, In S4, the expected operating revenue is calculated using the following formula: ， For asset nodes The dynamic fair value of digital assets. For asset nodes Expected operating revenue, For asset nodes within the current statistical period The equivalent axle load cumulative traffic flow vector, For element-wise multiplication of vectors, This is the vector of the basic statutory toll rates for highways. The hyperbolic tangent activation function is used. For learnable rate elasticity mapping weight matrix, For asset nodes The absolute value of the initial filing. It is a positive constant. This is a vector for traffic pricing policy restrictions.

8. A dynamic valuation system for digital assets of highways, used to implement the dynamic valuation method for digital assets of highways as described in claim 1, characterized in that, include: The comprehensive asset base feature generation module is used to acquire multi-source heterogeneous asset data of highway assets, including dynamic time-series data of roadside IoT sensors, static geometric topology data of BIM, and unstructured text data of manual inspection; it performs spatiotemporal manifold alignment and semantic fusion on the multi-source heterogeneous asset data to generate comprehensive asset base features; The inference confidence and trajectory tensor generation module is used to generate a causal adjacency matrix based on comprehensive asset base features, design interference quantifiers to perform counterfactual state differential evolution inference, and generate counterfactual health states; Based on the causal adjacency matrix and counterfactual health status, the inference confidence and trajectory tensor are obtained; The digital asset dynamic fair value generation module is used to construct a secure intelligent agent and an economic intelligent agent based on trajectory tensor and inference confidence, conduct double-blind game, reach a consensus on asset valuation under Nash equilibrium constraints, and generate the dynamic fair value of digital assets. The digital asset ledger block generation module is used to generate expected operating revenue based on the dynamic fair value of digital assets. With the goal of maximizing the combined realization revenue and inventory value, it constructs a Lagrange cost functional, solves for the optimal preventive maintenance execution strategy, and cryptographically solidifies the combined asset base characteristics, the dynamic fair value of digital assets, and the expected operating revenue to obtain the digital asset ledger block.

9. A dynamic valuation device for digital assets of highways, characterized in that, It includes a processor and a memory, wherein the processor implements the dynamic valuation method for digital assets of highways as described in any one of claims 1-7 when executing a computer program stored in the memory.

10. A computer-readable storage medium, characterized in that, Used to store a computer program, wherein the computer program, when executed by a processor, implements the dynamic valuation method for digital assets of highways as described in any one of claims 1-7.

Images