A method for monitoring, predicting and warning carbon emissions of traffic infrastructure under the joint driving of big data and artificial intelligence

Abstract

The application discloses a traffic infrastructure carbon emission monitoring, predicting and early warning method under the joint driving of big data and artificial intelligence, which comprises the following steps: acquiring multi-source heterogeneous data, updating a digital twin containing a physical layer, an energy layer and a carbon flow layer; generating features through interlayer coupling mapping and heterogeneous graph convolution coding; combining a physical model to calculate a carbon emission benchmark, using a residual model to predict a residual to obtain a predicted value; using dynamic mutual information and an abnormal score, when the mutual information is low and the score is high, determining interlayer mapping drift based on mapping state parameters and giving an early warning. The application has the following effects: breaking through the lag of static accounting, improving the prediction accuracy through dynamic mapping, and realizing the early warning of carbon emission precursors through sensing mapping drift.

Description

Technical Field

[0001] This application relates to the field of transportation carbon emission technology, specifically to a method for monitoring, predicting, and providing early warning of carbon emissions from transportation infrastructure driven by a combination of big data and artificial intelligence. Background Technology

[0002] Carbon emissions from transportation infrastructure are characterized by strong time-varying properties and multi-factor coupling, making accurate monitoring and dynamic prediction of these emissions of significant practical importance for carbon management.

[0003] Among existing technologies, the most mainstream method is the statistical accounting method based on fixed emission factors. This method pre-determines the average emission coefficient per unit mileage for different vehicle types, and then combines it with the cumulative activity data such as traffic flow and mileage obtained from statistics of various road segments. The total carbon emissions within a certain time and space range are estimated by multiplying and summing the products of "emission factor × activity amount". This method is widely used in carbon emission inventory compilation and macro trend analysis because of its simplicity of calculation and easy data acquisition.

[0004] However, the core flaw of this method lies in the fact that the emission factor is a static average value, which cannot respond to real-time fluctuations in actual traffic flow. In real road networks, dynamic factors such as changes in vehicle speed, acceleration and deceleration frequency, and congestion levels significantly alter the instantaneous emission rate of a single vehicle, and the fixed factor cannot capture these changes, leading to a systematic deviation between the estimated results and actual emissions. Especially during traffic congestion or sudden disturbances, emission surges cannot be reflected in a timely manner, and the predicted results lag significantly behind the actual situation, thus making it difficult to support the needs of dynamic early warning and real-time control. This is the essential bottleneck faced by traditional static accounting methods in application. Summary of the Invention

[0005] To overcome the lag of static accounting, improve prediction accuracy through dynamic mapping, and achieve early warning of carbon emissions through perceptual mapping drift, this application provides a method for monitoring, predicting, and warning of carbon emissions from transportation infrastructure driven by big data and artificial intelligence.

[0006] This application provides a method for monitoring, predicting, and providing early warning of carbon emissions from transportation infrastructure driven by a combination of big data and artificial intelligence, including:

[0007] Acquire multi-source heterogeneous data on transportation infrastructure;

[0008] A pre-constructed digital twin is updated based on multi-source heterogeneous data. The digital twin includes a physical layer, an energy layer, and a carbon flow layer.

[0009] The updated physical layer traffic flow state is mapped to the energy layer load by inter-layer coupling operators, establishing inter-layer mapping relationships and generating mapping state parameters; a heterogeneous graph convolutional network is used to jointly encode the updated digital twin to obtain feature vectors.

[0010] Based on the updated physical layer traffic flow status and energy layer load, the baseline carbon emission value is calculated through a physical model; based on historical data and feature vectors in multi-source heterogeneous data, the carbon emission residual is predicted through a residual prediction model to obtain the residual prediction value; the residual prediction value is superimposed with the baseline value to obtain the carbon emission prediction value.

[0011] Based on time-series samples of carbon emission predictions and carbon emission baselines within a preset time window, a dynamic mutual information value is calculated; the carbon emission predictions are then input into an anomaly scoring network to obtain an anomaly score.

[0012] When the dynamic mutual information value is lower than the preset fluctuation threshold and the abnormal score is greater than the preset score threshold, the inter-layer mapping relationship is determined to have drifted based on the mapping state parameters and an abnormal warning signal is generated.

[0013] One of the above technical solutions has the following advantages or beneficial effects:

[0014] This application provides a method for monitoring, predicting, and warning carbon emissions from transportation infrastructure driven by big data and artificial intelligence. By constructing a digital twin comprising a physical layer, an energy layer, and a carbon flow layer, it achieves dynamic mapping and synchronization of all elements related to the physical operation status, energy supply and demand status, and carbon emission trends of transportation infrastructure. Through inter-layer coupling operators, the physical layer traffic flow status is mapped to the energy layer load, and the mapping weights are dynamically adjusted using a scenario-based mapping function, establishing an adaptive mapping relationship between the physical and energy layers, thus improving the accuracy and timeliness of cross-layer mapping. Finally, a heterogeneous graph convolutional network is used to jointly encode the digital twin, realizing multi-source heterogeneous information from the physical, energy, and carbon flow layers. Deep integration and low-dimensional unified representation: By calculating the carbon emission baseline value through a physical model and combining it with the residual prediction model to predict the carbon emission residual, data-driven and physical mechanisms are integrated, effectively correcting the systematic bias of the physical model and significantly improving the accuracy and robustness of carbon emission prediction values. By calculating the dynamic mutual information value based on time-series samples and constructing an anomaly scoring network, joint detection of carbon emission prediction anomalies and instability of inter-layer mapping relationships is achieved. When the dynamic mutual information value is lower than the preset fluctuation threshold and the anomaly score is higher than the preset scoring threshold, the drift of the inter-layer mapping relationship can be determined in a timely manner and an anomaly warning signal can be generated, providing reliable decision support for the low-carbon operation of transportation infrastructure and precise carbon emission control. Attached Figure Description

[0015] The technical solution and other beneficial effects of this application will become apparent from the following detailed description of specific embodiments in conjunction with the accompanying drawings.

[0016] Figure 1 This is a flowchart illustrating the method for monitoring, predicting, and issuing early warnings of carbon emissions from transportation infrastructure driven by the combined use of big data and artificial intelligence, as provided in this application embodiment. Detailed Implementation

[0017] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0018] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Features thus defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more unless otherwise explicitly specified.

[0019] like Figure 1 As shown in the embodiments of this application, a method for monitoring, predicting, and providing early warning of carbon emissions from transportation infrastructure driven by a combination of big data and artificial intelligence is provided, including:

[0020] S1: Acquire multi-source heterogeneous data on transportation infrastructure.

[0021] In this application embodiment, transportation infrastructure refers to a comprehensive system consisting of roads, bridges, tunnels, transportation hubs and their associated energy supply and emission monitoring facilities; multi-source heterogeneous data refers to a collection of data of various types, formats and sampling frequencies from sensing and monitoring equipment, energy metering systems, emission monitoring terminals, geographic information systems and historical operation and maintenance databases, and its data types include structured time-series data and semi-structured topological data.

[0022] The specific implementation process is as follows: Multiple types of data acquisition terminals are deployed along transportation infrastructure lines to collect traffic flow data, energy consumption data, equipment operating status data, and carbon emission monitoring data in real time according to preset sampling frequencies and preset data protocols. The collected multi-source heterogeneous data undergoes integrity verification and outlier removal to obtain multi-source heterogeneous data that can be used for subsequent digital twin updates. The preset sampling frequencies are set differently based on data types: traffic flow data is sampled every 5 to 15 minutes (5 min ≤ Δt ≤ 15 min); energy consumption data is sampled every 1 to 5 minutes (1 min ≤ Δt ≤ 5 min); equipment operating status data is sampled every 1 to 3 minutes (1 min ≤ Δt ≤ 3 min); and carbon emission monitoring data is sampled every 10 to 30 minutes (10 min ≤ Δt ≤ 30 min). The preset data protocols include at least one of MQTT, Modbus, or OPCUA protocols.

[0023] It is worth noting that the spatiotemporal resolution of multi-source heterogeneous data should meet the accuracy requirements of dynamic mapping of digital twins. When the data loss rate of a certain acquisition terminal exceeds 5%, a data supplementation mechanism is triggered.

[0024] Understandably, by acquiring multi-source heterogeneous data on transportation infrastructure, the technical problems of data silos and information fragmentation in traditional transportation management systems have been solved. This has enabled full-dimensional data perception of the operational status of transportation infrastructure, energy flow, and carbon emission trends, allowing the digital twin to be dynamically updated based on real-time data.

[0025] S2: Updates a pre-built digital twin based on multi-source heterogeneous data. The digital twin includes a physical layer, an energy layer, and a carbon flow layer.

[0026] In this embodiment, a digital twin refers to a virtual model system that performs full-element mapping and dynamic simulation of the physical entities, energy flow, and carbon emission processes of transportation infrastructure in digital space; the physical layer refers to a digital mapping layer describing the road network topology, traffic flow dynamic evolution, and vehicle operating status of transportation infrastructure; the energy layer refers to a digital mapping layer describing the connection relationships of energy equipment, energy load distribution, and energy flow paths in transportation infrastructure; the carbon flow layer refers to a digital mapping layer describing carbon emission generation nodes, carbon emission flow direction, and cumulative carbon emission distribution in transportation infrastructure; the pre-constructed digital twin refers to a basic digital twin established during the system initialization phase using historical data and benchmark parameters, and its update process refers to the process of dynamically correcting the state parameters of the basic digital twin and reconstructing the model using real-time acquired multi-source heterogeneous data; the mapping state parameters refer to a set of quantitative indicators describing the mapping accuracy, synchronization delay, and data consistency between the digital twin and the physical entities of transportation infrastructure.

[0027] It should be noted that before updating the pre-built digital twin based on multi-source heterogeneous data, the initial construction of the digital twin includes the following specific methodological steps:

[0028] S201: Obtain initial multi-source heterogeneous data, which includes static topology parameters, equipment parameters, and emission factor baseline values. Static topology parameters include road network topology information, equipment parameters include energy equipment information, and emission factor baseline values ​​include emission factor information.

[0029] In this embodiment, the initial multi-source heterogeneous data refers to the benchmark dataset collected during the initialization phase of the digital twin for constructing the basic digital twin; static topology parameters refer to structural parameters that describe the connection relationships of road network nodes, road segment lengths, number of lanes, and traffic capacity of transportation infrastructure and do not change over time; equipment parameters refer to a set of parameters describing the rated capacity, rated power, operating efficiency, and installation location of photovoltaic power generation equipment, energy storage equipment, charging piles, lighting equipment, and other energy supply and consumption equipment in transportation infrastructure; emission factor benchmark values ​​refer to benchmark coefficients describing the carbon dioxide emissions generated by various energy devices per unit of energy consumption, and their values ​​are determined according to the carbon emission accounting standards published by the industry; road network topology information refers to vector data including road node numbers, node geographic coordinates, road segment start and end nodes, road segment length, and road segment traffic capacity; energy equipment information refers to a set of parameters including equipment type, equipment rated power, equipment installation location, and equipment operating status thresholds; emission factor information refers to a set of values ​​including electricity emission factors, fuel emission factors, gas emission factors, and renewable energy emission reduction factors.

[0030] The specific implementation process is as follows: Static topology parameters, equipment parameters, and emission factor baseline values ​​are extracted from the planning database, as-built drawing database, energy equipment ledger database, and carbon emission accounting standard database of transportation infrastructure. The extracted initial multi-source heterogeneous data is then formatted and missing values ​​are filled to obtain structurally complete and field-consistent initial multi-source heterogeneous data. Among these, the road segment length is a key static topology parameter. Indicates the geometric length of the road segment, in kilometers (km); number of lanes Indicates the number of lanes for one-way traffic; traffic capacity. This indicates the maximum number of standard vehicles that can pass through a road segment per unit time, in vehicles per hour (veh / h). Rated capacity in equipment parameters represents the rated output power of the energy equipment, in kilowatts (kW); rated power represents the rated power consumption of the energy equipment, in kilowatts (kW); operating efficiency... This indicates the energy conversion efficiency of energy equipment. In the emission factor baseline values, the electricity emission factor represents the carbon dioxide emissions corresponding to a unit of electricity consumption, in kilograms of carbon dioxide per kilowatt-hour (kgCO2 / kWh); the fuel oil emission factor represents the carbon dioxide emissions corresponding to a unit of fuel oil consumption, in kilograms of carbon dioxide per liter (kgCO2 / L); and the gas oil emission factor represents the carbon dioxide emissions corresponding to a unit of gas oil consumption, in kilograms of carbon dioxide per cubic meter (kgCO2 / m³). 3 The renewable energy emission reduction factor represents the carbon dioxide emission reduction corresponding to a unit of renewable energy power generation, in kilograms of carbon dioxide per kilowatt-hour (kgCO2 / kWh).

[0031] Understandably, by acquiring initial multi-source heterogeneous data containing static topology parameters, equipment parameters, and emission factor baseline values, the technical problem of data missingness and inconsistency with baselines in the initialization stage of digital twins was solved, and the parameterized construction of the basic model of digital twins was realized, providing standardized data input for the hierarchical modeling of the physical layer, energy layer, and carbon flow layer.

[0032] S202: Perform spatiotemporal alignment and standardization on the initial multi-source heterogeneous data to obtain standardized data.

[0033] In this embodiment, spatiotemporal alignment refers to the data preprocessing process that unifies initial multi-source heterogeneous data from different acquisition terminals with different timestamps and spatial reference systems to the same time base and the same spatial coordinate system; standardization refers to the data conversion process that standardizes the dimensions, numerical ranges, and encoding formats of the initial multi-source heterogeneous data; standardized data refers to a dataset that has a unified timestamp, unified spatial reference system, unified dimensions, and unified encoding format after spatiotemporal alignment and standardization.

[0034] The specific implementation process is as follows: A unified time base is set as Coordinated Universal Time (UTC). Time difference correction and sampling time alignment are performed on the timestamp fields of the initial multi-source heterogeneous data. The spatial reference system is set to the WGS84 geographic coordinate system. Topological integrity verification and node coordinate projection transformation are performed on the road network topology information in the static topology parameters. Field encoding standardization and unit unification conversion are performed on the energy equipment information in the equipment parameters. Numerical normalization and unit unification conversion are performed on the emission factor information in the emission factor baseline values, resulting in standardized data with spatiotemporal alignment and a standardized format. The spatiotemporal alignment sampling time window... The time slice length indicating data alignment, in minutes (min). The value range is 5 minutes to 15 minutes (5 min ≤ ≤15min); Data consistency rate This indicates the percentage of completeness of the aligned data fields. The value range is from 0.98 to 1.0 (0.98 ≤ ≤1.0). It is worth noting that when the data consistency rate... When the value is below 0.98, the missing value imputation and format unification operations are re-executed. Specifically, time alignment uses linear interpolation to interpolate data from different sampling frequencies to a unified time slice; spatial coordinate projection transformation uses the universal transverse Mercator projection, with projection zone numbers according to... The projection zone number is determined and constrained to a range of 1 to 60. If the calculation result exceeds this range, it is forcibly assigned to the nearest valid projection zone. This indicates the longitude of the road segment's center point, in degrees (°). In the above formula for projected zone designation, Used to set the value range as Degree to longitude shifted to Degree to The range of degrees The longitude width of a single projection strip in the universal transverse Mercator projection. This is a floor function used to round down the entire projection zone containing the longitude of the road segment's center point. Used to change the projection zone number from Numbering; for example, when the longitude of the center point of a certain road segment... for When, That is, this section of road is located in the first The projection zone, whose central meridian is The coordinate projection of this road segment is... The projection zone number of any road segment in the road network is uniquely determined by using the central meridian as a reference, thus avoiding spatial misalignment caused by inconsistent selection of projection zones for cross-zone road segments.

[0035] Understandably, by performing spatiotemporal alignment and standardization on the initial multi-source heterogeneous data, the technical problems of inconsistent spatiotemporal benchmarks and heterogeneous dimensional formats of multi-source data were solved, achieving spatiotemporal unification and semantic consistency at the data level, and providing a standardized data foundation for the collaborative construction of subsequent physical layer models, energy layer models and carbon flow layer models.

[0036] S203: Based on the road network topology information in standardized data, establish the topology structure of the physical layer model, and use the topology structure as input to generate dynamic traffic flow simulation results through a preset traffic flow simulation model.

[0037] In this embodiment, the topology of the physical layer model refers to a directed graph structure with road network nodes as vertices, road segments as edges, and road segment capacity as edge weights; the preset traffic flow simulation model refers to a traffic flow evolution simulation algorithm built based on a cellular transmission model or a dynamic traffic assignment model; and the traffic flow dynamic simulation results refer to time series data describing the traffic flow, traffic density, and average travel speed of each road segment at different times.

[0038] Specific implementation process: Extract the set of road nodes based on the road network topology information in the standardized data. Collection of road sections The topology of the physical layer model is constructed using road nodes as vertices, road segments as edges, and road segment capacity as edge weights. ,in Represents the set of road nodes; Represents a set of road segments; Represents the edge weight matrix. Represents a node With nodes Traffic capacity of the road segments, in vehicles per hour (veh / h). Based on the topology of the physical layer model. As input, load a preset traffic flow simulation model and set the simulation time step. For 1 minute ( (min), total simulation time For 24 hours ( h) Based on the real-time traffic demand input for each road segment, the traffic flow for each road segment at each simulation time step is calculated using a cellular transfer model. Traffic density and average travel speed This generates dynamic simulation results of traffic flow; among which, This indicates the traffic flow on a road segment, in vehicles per hour (veh / h). This indicates the traffic density of a road segment, in units of vehicles per kilometer (veh / km). This represents the average travel speed of a road segment, in kilometers per hour (km / h). In the cellular transport model, traffic flow... Traffic density Satisfying basic graph relations:

[0039] .

[0040] in, This represents the average travel speed, in kilometers per hour (km / h). Satisfy: In the basic graph relation, the left side of the equation This represents the number of vehicles passing through the cross-section per unit time. The right side represents the product of traffic density and average travel speed. Its physical meaning is that all vehicles simultaneously present on the road segment pass through the cross-section at their average travel speed. For example, when... Vehicles per kilometer At 1000 km / h, Vehicles per hour.

[0041] .

[0042] in, Indicates free-flow velocity, unit: kilometers per hour (km / h); This represents congestion density, measured in vehicles per kilometer (veh / km). It is important to note that the simulation time step should not exceed 1 minute to ensure the stability of the traffic flow dynamics simulation results. The value ranges from 60 km / h to 120 km / h, and its value is determined based on the design speed or the upper limit of the measured speed for each road section. The value ranges from 100 vehicles per kilometer to 200 vehicles per kilometer, and its value is based on the number of lanes in each road segment and the average vehicle length. The estimate is determined, among which, This indicates the average vehicle body length, in meters (m). The value range is 5 meters to 15 meters. The basic graphical parameters of the traffic flow simulation model are based on the historical traffic flow of each road segment. The measured density data were calibrated using least-squares fitting; the initial condition for the simulation was the average traffic density of each road segment during the same historical period, and the boundary condition was the measured traffic flow time series at the road network entrance and exit detection sections. This speed... In the density relationship, The maximum travel speed when the density approaches zero. The current density represents the proportion of the blocking density. The velocity reduction factor characterizes the linear suppression of velocity by density increase, for example, when kilometers per hour Vehicles per kilometer Vehicles per kilometer kilometers per hour. In the blockage density estimation formula, for The number of meters corresponding to a kilometer. For the number of lanes, For example, when the average vehicle body length is... , Rice time, Vehicles per kilometer, that is, the maximum capacity of a two-lane road per kilometer. A vehicle.

[0043] Understandably, by establishing the topology of the physical layer model and running the preset traffic flow simulation model, the technical problem of the distortion in the mapping between the physical space and digital space of traffic infrastructure is solved, and a high-fidelity simulation of the dynamic evolution process of road network traffic flow is achieved, providing dynamic traffic demand input for the energy layer and carbon flow layer.

[0044] S204: Based on the energy equipment information and traffic flow simulation results in the standardized data, establish the equipment connection relationship of the energy layer, and use the connection relationship and traffic flow simulation results as input to calculate the energy load distribution through the preset energy system simulation model.

[0045] In this embodiment, the device connection relationship of the energy layer refers to the energy supply and consumption topology relationship between photovoltaic power generation equipment, energy storage equipment, charging piles, lighting equipment and other energy equipment; the preset energy system simulation model refers to the energy load calculation algorithm constructed based on equipment operation scheduling rules and energy balance equations; and the energy load distribution refers to the spatial distribution and time series data of the net energy load power of each energy device at different times.

[0046] The specific implementation process is as follows: Based on the energy equipment information in the standardized data, the equipment type set, rated power set, and installation location set are extracted. According to the spatial deployment location and functional association of each energy device in the road network, the equipment connection relationship of the energy layer is established. Using this equipment connection relationship and traffic flow simulation results as input, a pre-set energy system simulation model is loaded. Based on the rated power, real-time operating status, and traffic flow-induced load demand of each device, the net energy load power of the energy equipment in each road segment is calculated. This generates energy load distribution. Represents net energy load power, in kilowatts (kW), satisfying:

[0047] .

[0048] in, This indicates the power consumption of electrical equipment, in kilowatts (kW). This indicates the charging power of the energy storage device, in kilowatts (kW). The value is negative when the energy storage device is discharging. This indicates the output power of fossil fuel power generation equipment, in kilowatts (kW). This represents the output power of renewable energy power generation equipment, in kilowatts (kW). In the above power balance equation, the load side (electricity consumption and energy storage charging) is positive, and the supply side (various types of power generation) is negative. Its algebraic sum represents the net load power that needs to be supplemented by the external grid; when... When, it indicates that electricity needs to be purchased from the power grid. When, it indicates power being fed into the grid. In the power balance equation, the left side represents the net load that needs to be supplemented by the external grid, and the first two terms on the right side represent the load-side power, and the last two terms represent the supply-side power. For example, when kilowatt, kilowatt, kilowatt, kilowatt, Kilowatts indicate that electricity needs to be purchased from the grid at that moment. kilowatt.

[0049] It is worth noting that the time resolution of the energy system simulation model should be consistent with that of the traffic flow simulation model to ensure data synchronization between layers. This includes the power consumption of electrical equipment. according to Conversion, Indicates the effective output power of the device, in kilowatts (kW). ), This indicates the equipment's operating efficiency; the energy storage equipment's charging and discharging scheduling strategy is based on preset peak and valley periods, with charging during valley periods and discharging during peak periods; the load demand of charging piles is determined by traffic flow guidance based on the product of the proportion of electric vehicles on that road segment and the preset average charging power per vehicle.

[0050] Understandably, by establishing the device connection relationship of the energy layer and calculating the energy load distribution, the technical problems of isolated operation of energy equipment in transportation infrastructure and the inability to correlate load demand with traffic flow status have been solved. This has enabled the dynamic coupling of energy supply and demand status with traffic flow status, providing energy load input for carbon emission tracking in the carbon flow layer.

[0051] S205: Based on emission factor information and energy load distribution in standardized data, establish emission nodes for the carbon flow stratum model, and use emission nodes and energy load distribution as inputs to calculate carbon emission flow direction through a preset carbon emission tracking algorithm.

[0052] In this embodiment of the application, the emission nodes of the carbon flow layer model refer to the carbon emission generation and transmission nodes that correspond one-to-one with the energy devices in the energy layer; the preset carbon emission tracking algorithm refers to the carbon emission source tracing and allocation algorithm constructed based on the carbon flow conservation principle and graph propagation mechanism; and the carbon emission flow direction refers to the dataset describing the entire chain transmission path and allocation of carbon emissions from the supply side to the consumption side.

[0053] The specific implementation process is as follows: Based on the emission factor information and energy load distribution in the standardized data, a set of emission nodes for the carbon flow stratum model is established. Each emission node corresponds to an energy device in the energy layer, and the energy type and corresponding emission factor baseline value of that node are recorded. Using the emission nodes and energy load distribution as input, a preset carbon emission tracking algorithm is executed to calculate the carbon emission flow direction. It should be noted that calculating the carbon emission flow direction using the preset carbon emission tracking algorithm includes the following specific steps:

[0054] S2051: Obtain the energy load distribution and emission factor baseline values ​​for each emission node, and classify each emission node into a supply side or a consumption side based on the equipment type in the energy equipment information and the energy load distribution.

[0055] In this embodiment, the supply side refers to the device nodes that output energy to other nodes, such as grid access points and distributed photovoltaic inverters; the consumption side refers to the device nodes that consume energy, such as lighting fixtures, ventilation fans, and charging piles.

[0056] The specific implementation process is as follows: Read the energy load distribution and emission factor baseline values ​​of each emission node; determine the functional role of each emission node based on the equipment type identifier in the energy equipment information. If the net energy load power of the node... If so, it is determined to be on the supply side; if If so, it is determined to be the consumer end; if If it is a neutral node, it will not participate in carbon flow propagation.

[0057] S2052: Using each emission node as a vertex, determine the energy supply relationship between devices based on energy equipment information, determine the real-time energy supply in each energy supply relationship based on energy load distribution, and establish a directed edge from the supply end to the consumption end when the supply is greater than the preset supply threshold. After traversing all energy supply relationships, a carbon flow directed graph is generated.

[0058] In the embodiments of this application, the energy supply relationship refers to the topological connection relationship between devices where there is actual energy transmission; the real-time energy supply refers to the net energy power transmitted from the supply end to the consumption end per unit time; the preset supply threshold refers to the minimum effective energy transmission power threshold value excluding line loss and measurement noise interference; the carbon flow directed graph refers to the directed graph structure with emission nodes as vertices and energy supply relationships as directed edges.

[0059] The specific implementation process is as follows: Taking each emission node as a vertex, the energy supply relationship between devices is determined based on the device connection topology in the energy device information. The real-time energy supply amount in each energy supply relationship is calculated based on the energy load distribution. (kW), when Greater than the preset supply threshold (In this embodiment, we take) When the energy supply is at a certain kW, directed edges are established from the supply side to the consumption side. After traversing all energy supply relationships, a directed graph of carbon flow is generated. ,in Represents the set of emission nodes; This represents the set of directed edges.

[0060] S2053: Based on the total energy load of each supply side and its consumption side load, the ratio of consumption side load to total supply side load is used as the load allocation coefficient, and the product of the total energy load of the supply side and the emission factor benchmark value is used as the direct carbon emissions.

[0061] In this embodiment of the application, the load allocation factor refers to the proportional weight when carbon emissions are allocated from the supply side to the consumption side; direct carbon emissions refer to the amount of carbon dioxide emissions directly generated by the supply side during the energy conversion process.

[0062] Specific implementation process: For each supply-side node Calculate its total energy load (kW) and all its downstream consumer loads (kW). From the consumer end Load and supply side The ratio of total load to load distribution factor ,satisfy:

[0063] .

[0064] in, This indicates from the supply side To the consumer end The load distribution factor; Indicating the consumer side Load, unit: kilowatt (kW); Indicating downstream consumer end Load, unit: kilowatt (kW); Indicating supply side The downstream consumer end collection. From the supply side... Total energy load Compared with emission factor baseline value The accumulation as direct carbon emissions ,satisfy:

[0065] .

[0066] in, Indicating supply side Direct carbon emissions, in kilograms of carbon dioxide per hour (kgCO2 / h). Indicating supply side Total energy load, unit: kilowatt (kW); Indicating supply side The emission factor baseline value, in kilograms of carbon dioxide per kilowatt-hour (kgCO2 / kWh). The carbon flux formula above... , and Indicates node energy load, unit: kilowatt (kW), and is similar to the symbol for segment length in the physical layer. Strict distinctions must be made to avoid multiple meanings from the same letter. In the formula for calculating direct carbon emissions, the total energy load on the supply side is the power level of energy conversion, and the emission factor baseline value is the carbon dioxide emissions per unit of energy. The product of the two is the carbon emissions per unit time on the supply side. For example, when the total energy load on the supply side is... kilowatts, emission factor is kilograms of carbon dioxide per kilowatt-hour kilograms of carbon dioxide per hour.

[0067] It is worth noting that the above load allocation coefficients satisfy the carbon flow conservation constraint, that is, for any supply side... ,have This ensures that total carbon emissions do not undergo non-physical increases or decreases during transmission. The numerator of the load allocation coefficient is the load at a specific consumption end, and the denominator is the sum of the loads at all downstream consumption ends of the supply end. The ratio represents the proportion of that consumption end in the energy output of the supply end. For example, the downstream consumption loads of a certain supply end are... kilowatt, kilowatt, kilowatts, then the corresponding load distribution factors are respectively , , The sum of the three is .

[0068] S2054: The load allocation coefficient is used as the weight of the directed edge, and the direct carbon emissions are used as the starting amount. The load is propagated from the supply side to the consumption side node by node along the carbon flow directed graph.

[0069] Specific implementation process: Load allocation coefficients Assigned as a directed edge in a carbon flow directed graph The edge weights are determined by the direct carbon emissions of each supply side. The initial propagation quantity is propagated node by node along the directed graph of carbon flow from the supply side to the consumption side.

[0070] S2055: At each node, the carbon emissions from all upstream nodes are accumulated and summed with the direct carbon emissions when that node acts as a supplier to obtain the cumulative carbon emissions.

[0071] In this embodiment of the application, the cumulative carbon emissions refer to the total carbon emission responsibility borne by a node in the carbon flow network, including the carbon emissions shared from upstream nodes and the carbon emissions directly generated by the node itself.

[0072] Specific implementation process: For each node in the carbon flow directed graph Accumulated from all upstream nodes The carbon emissions, and the direct carbon emissions when the node acts as a supply end. Summing them up yields the cumulative carbon emissions. ,satisfy:

[0073] .

[0074] in, Represents a node Cumulative carbon emissions, in kilograms of carbon dioxide per hour (kgCO2 / h). Represents a node The set of upstream nodes; Indicates upstream node Cumulative carbon emissions, in kilograms of carbon dioxide per hour (kgCO2 / h). Represents a directed edge The weights; Represents a node Direct carbon emissions when acting as a supply-side node, expressed in kilograms of carbon dioxide per hour (kgCO2 / h). If the node acts only as a consumer and not a supplier, then... ,at this time In the formula for calculating cumulative carbon emissions, the summation term represents the total upstream carbon emissions allocated to that node, and the last term represents the node's own direct carbon emissions. For example, the cumulative carbon emissions of two upstream nodes of a given node are... kilograms of carbon dioxide per hour and kilograms of carbon dioxide per hour, with corresponding edge weights as follows: and If this node is purely a consumer, then kilograms of carbon dioxide per hour.

[0075] It is worth noting that, through the aforementioned weighted propagation mechanism, carbon emissions are allocated in the directed carbon flow graph according to the energy supply ratio, with clear physical meaning and satisfying the law of conservation of mass. Furthermore, the directed carbon flow graph is a directed acyclic graph, and the cumulative carbon emissions of each node are calculated node by node according to the topological sorting order of the directed carbon flow graph, in order to avoid circular dependencies caused by circular energy transfer relationships.

[0076] S2056: Generate the carbon emission flow distribution based on the cumulative carbon emissions of each node and the weight distribution of the directed edges.

[0077] In the embodiments of this application, carbon emission flow distribution refers to the complete dataset and visualization mapping results that describe the cumulative carbon emissions of each emission node, the carbon emission allocation of each directed edge, and the path of carbon emission transmission from the supply side to the consumption side.

[0078] The specific implementation process involves: summarizing the cumulative carbon emissions of all emission nodes and the carbon emission allocation of all directed edges in the directed carbon flow graph; storing this data in a structured manner according to node type and topological level; and generating a carbon emission flow distribution that includes the cumulative carbon emissions of nodes, the carbon emission allocation of edges, carbon emission transmission paths, and carbon emission intensity indicators. Among these, the carbon emission intensity indicator... This represents the cumulative carbon emissions per unit of energy load, expressed in kilograms of carbon dioxide per kilowatt-hour (kgCO2 / kWh), satisfying the following:

[0079] .

[0080] in, Represents a node Carbon emission intensity index, unit: kilograms of carbon dioxide per kilowatt-hour (kgCO2 / kWh). Represents a node Cumulative carbon emissions, in kilograms of carbon dioxide per hour (kgCO2 / h). Represents a node Net energy load power, unit: kilowatt (kW). When node A neutral node, meaning its net energy load power is zero, has a carbon emission intensity index of zero and is not included in the carbon emission intensity comparison. In the carbon emission intensity index calculation formula, the numerator is the cumulative carbon emissions borne by the node, and the denominator is the absolute value of the node's net energy load power. For example, when... kilograms of carbon dioxide per hour kilowatt-hours kilograms of carbon dioxide per kilowatt-hour.

[0081] Understandably, by establishing emission nodes in the carbon flow layer model and executing a preset carbon emission tracking algorithm, the technical problems of unclear carbon emission sources and unclear carbon emission responsibility allocation in transportation infrastructure have been solved. This has enabled quantitative tracking of carbon emissions across the entire chain, from generation at the supply side to consumption at the consumption side, allowing the carbon flow layer to accurately map the carbon emission responsibility corresponding to energy consumption.

[0082] S206: Couple the dynamic simulation results of traffic flow, energy load distribution, and carbon emission flow to generate a digital twin and mapped state parameters.

[0083] In this embodiment, coupling refers to the data integration process of associating and fusing the simulation results of the physical layer, energy layer and carbon flow layer according to the time base and spatial nodes; mapping state parameters refer to a set of quantitative indicators describing the mapping accuracy, synchronization delay and data consistency between the digital twin and the physical entity of the transportation infrastructure.

[0084] The specific implementation process is as follows: Using a unified time reference as the association key, the traffic flow, traffic density, and average travel speed of road segments in the traffic flow dynamic simulation results, the net energy load power of each device in the energy load distribution, and the cumulative carbon emissions and carbon emission intensity indicators of nodes in the carbon emission flow distribution are spatiotemporally associated according to the road node number and device number. This constructs a cross-layer coupling relationship between the physical layer, energy layer, and carbon flow layer, generating a digital twin and mapping state parameters. Among them, the mapping state parameters include time synchronization error. Unit: seconds (s) The value range is from 0 seconds to 30 seconds (0s≤ ≤30s); Data consistency rate , The value range is from 0.95 to 1.0 (0.95 ≤ ≤1.0); spatial mapping precision The threshold is lower than the data consistency rate in step S202 initialization phase. The baseline value is 0.98, because the real-time update phase allows for a higher tolerance for dynamic data loss in order to balance data freshness and integrity; Unit: meter (m) The value range is from 0.5 meters to 5 meters (0.5m ≤ ≤5m). It is worth noting that when the time synchronization error in the mapping state parameters... s or data consistency rate At this time, the calibration and update process of the digital twin is triggered. The specific operation of coupling is a database association with road node number and equipment number as the primary key: each energy device is assigned to the nearest road node according to its installation location. When a device corresponds to multiple road segment nodes, it is uniquely assigned according to the principle of nearest distance, ensuring that the three layers of data have a unique correspondence under a unified node.

[0085] Understandably, by coupling the dynamic simulation results of traffic flow, energy load distribution, and carbon emission flow, the technical problem of the physical layer, energy layer, and carbon flow layer operating independently and lacking coordination has been solved. This has achieved the organic integration and state synchronization of the three-layer architecture of the digital twin, enabling the digital twin to comprehensively reflect the physical operation status, energy supply and demand status, and carbon emission situation of the transportation infrastructure.

[0086] In summary, the steps described above for updating the pre-constructed digital twin based on multi-source heterogeneous data, by acquiring initial multi-source heterogeneous data and performing spatiotemporal alignment and standardization, establish the topology of the physical layer model, the device connection relationships of the energy layer, and the emission nodes of the carbon flow layer model, and couple the dynamic simulation results of traffic flow, energy load distribution, and carbon emission flow direction, solve the technical problem that traditional digital twin models of transportation infrastructure only focus on physical mapping and ignore the coordination of energy and carbon emissions. It realizes the hierarchical construction and dynamic coupling of the physical layer, energy layer, and carbon flow layer, enabling the digital twin to have the ability to dynamically map all elements, the entire chain, and the entire life cycle of transportation infrastructure. Moreover, the digital twin can be continuously and dynamically updated through the multi-source heterogeneous data acquired in step S1.

[0087] S3: The updated physical layer traffic flow state is mapped to the energy layer load through inter-layer coupling operators, establishing inter-layer mapping relationships and generating mapping state parameters; a heterogeneous graph convolutional network is used to jointly encode the updated digital twin to obtain feature vectors.

[0088] In this embodiment, the interlayer coupling operator refers to a parameterized operator used to establish a mathematical mapping relationship between the physical layer traffic flow state and the energy layer load, which includes mapping weights and mapping functions; the updated physical layer traffic flow state refers to the time series data of road segment traffic flow, average travel speed, and vehicle type composition ratio obtained after updating the traffic flow dynamic simulation results generated in step S203 in step S2; the energy layer load refers to the net energy load power of energy equipment in each road segment calculated in step S204; the interlayer mapping relationship refers to the corresponding mapping rules and parameter associations between the physical layer traffic flow state and the energy layer load; the mapping state parameters refer to the set of quantitative indicators describing the accuracy of the interlayer mapping relationship, the evolution state of the mapping weights, and the statistical characteristics of the mapping error; the heterogeneous graph convolutional network refers to a deep learning model that can process heterogeneous graph data composed of different types of nodes and edges, which realizes node embedding update by aggregating the feature information of heterogeneous neighbor nodes; the feature vector refers to the low-dimensional dense vector representation obtained after the digital twin is jointly encoded by the heterogeneous graph convolutional network, which is used to characterize the comprehensive state characteristics of the digital twin in the physical layer, energy layer, and carbon flow layer.

[0089] It is worth noting that the heterogeneous graph convolutional network adopts a heterogeneous graph attention network architecture, including an input layer, Each graph has an attention hidden layer and an output layer. The value range is 2 to 4. The input layer receives physical layer road segment node features (3 dimensions, including traffic flow, average travel speed, and traffic density), energy layer equipment node features (2 dimensions, including net energy load power and equipment operating efficiency), and carbon flow layer emission node features (2 dimensions, including cumulative carbon emissions and carbon emission intensity indicators), and performs type-specific linear projection transformations based on node types. Each graph attention hidden layer contains a multi-head attention mechanism, with the number of attention heads... The value ranges from 2 to 8, and the dimension for calculating attention per head is... Type-specific linear projection maps the features of nodes at each layer to a unified map. Hidden space The value ranges from 32 to 128 to ensure comparability of heterogeneous node features in attention computation. Residual connections and layer normalization are used between hidden layers to stabilize training; the output layer compresses the embeddings of each node into a single dimension using mean pooling. eigenvectors , The value range is 64 to 256. During training, a node-level semi-supervised learning framework is adopted. The loss function is the sum of the node feature reconstruction error and the domain adaptation loss. The domain adaptation loss can be the maximum mean difference (MMD) loss. The source domain is the historical training dataset, and the target domain is the currently collected validation dataset. This loss measures the distance between the node embedding distributions of the source domain training set and the target domain validation set, or it can be the adversarial loss form of gradient inversion layer (GRL) combined with the domain classifier. The optimizer is AdamW, with an initial learning rate of 0.005, a weight decay coefficient of 0.01, and at least 1000 training iterations. The training batch size is the number of physical layer nodes, and the dropout rate is set to 0.1 to 0.3. An early stopping mechanism is triggered when the validation set reconstruction error no longer decreases for 20 consecutive iterations. As a preferred embodiment, the number of layers in the heterogeneous graph convolutional network is... Pick Attention count Pick Feature vector dimension Pick The meta-relations of the heterogeneous graph include: spatial affiliation edges between physical layer road segment nodes and energy layer device nodes, representing the spatial inclusion relationship of the device installed in the road segment; emission mapping edges between energy layer device nodes and carbon flow layer emission nodes, representing the one-to-one correspondence between the device and the emission node; topological adjacency edges between adjacent physical layer road segment nodes, representing the direct connectivity relationship between the road segments; and independent attention weight matrices corresponding to different meta-relations.

[0090] It should be noted that the process of mapping the updated physical layer traffic flow state to the energy layer load through inter-layer coupling operators, establishing inter-layer mapping relationships, and generating mapping state parameters includes the following specific steps:

[0091] S301: Obtain the updated physical layer traffic flow status; the physical layer traffic flow status includes the traffic volume, average vehicle speed, and vehicle type composition ratio of each road segment.

[0092] In this embodiment of the application, traffic flow refers to the total number of standard vehicles passing through a certain road segment within a unit of time, i.e., the traffic flow mentioned above; average vehicle speed refers to the time average of the instantaneous speeds of all vehicles in a certain road segment, i.e., the average travel speed mentioned above; vehicle type composition ratio refers to the percentage distribution of different vehicle types in a certain road segment relative to the total number of vehicles.

[0093] Specific implementation process: From the physical layer of the digital twin updated in step S2, extract the traffic flow, average travel speed, and vehicle type composition ratio of each road segment within a preset update cycle to obtain the updated physical layer traffic flow status. The preset update cycle is... The time interval for updating the state of a digital twin, in minutes (min). The value range is 5 minutes to 15 minutes (5 min ≤ ≤15min); Vehicle composition ratio Indicates the first The proportion of vehicles of the same type to the total number of vehicles on the road segment, and meeting the following requirements. It is worth noting that the vehicle classification includes at least one of the following: small passenger cars, large passenger cars, small trucks, medium-sized trucks, and heavy-duty trucks.

[0094] S302: Calculate the instantaneous energy consumption power of each vehicle model based on the preset unit mileage energy consumption coefficient, average vehicle speed, and preset speed grading correction coefficient; the instantaneous energy consumption power of a single vehicle is the product of the coefficient, average vehicle speed, and correction coefficient.

[0095] In this embodiment, the preset unit mileage energy consumption coefficient refers to the standard energy consumption of a certain type of vehicle under a unit mileage; the vehicle speed grade correction coefficient refers to the dimensionless coefficient that corrects the unit mileage energy consumption coefficient according to the speed range of the average vehicle speed; the instantaneous energy consumption power of a single vehicle refers to the energy consumption rate of a single vehicle per unit time, and its physical dimension is power.

[0096] Specific implementation process: For each type of vehicle Read its preset energy consumption coefficient per unit mileage Based on the average travel speed of this road section Determine the speed gear it belongs to. Read the corresponding preset speed gear correction coefficient Calculate the instantaneous power consumption of this vehicle model. ,satisfy:

[0097] .

[0098] in, Indicates the first Instantaneous energy consumption per vehicle for this type of vehicle, unit: kilowatt per vehicle (kW / veh). Indicates the first Preset energy consumption coefficient per unit mileage for vehicle type, unit: kilowatt-hours per kilometer per vehicle (kWh / (km·veh)); This indicates the average travel speed on the road segment, in kilometers per hour (km / h). Indicates the first Preset speed gearing correction coefficient. The value range is from 0.8 to 1.2 (0.8 ≤ ≤1.2). It is worth noting that the speed range is based on the average travel speed. Classified as low speed (0km / h≤ ≤30km / h), medium speed (30km / h< ≤60km / h) and high speed (60km / h< (≤120km / h), each gear corresponds to a different preset speed correction coefficient. As a preferred example, the low gear... Values to medium speed Values High speed gear Values to Each gear The specific values ​​are determined based on the median value of historical energy consumption measurements for each road segment, calculated according to speed ranges; a preset energy consumption coefficient per unit mileage is used. The preferred value range is: small passenger cars kilowatt-hours per kilometer per vehicle kilowatt-hours per kilometer per vehicle, heavy trucks kilowatt-hours per kilometer per vehicle Kilowatt-hours per kilometer per vehicle. The formula for calculating the instantaneous energy consumption of a single vehicle is as follows: Energy consumption per unit mileage The product of the two values ​​represents the energy consumption per unit distance, calculated as the driving speed, and converts the energy consumption per unit time to the energy consumption per unit distance. For example, when the speed range is corrected by a correction factor. kilowatt-hours per kilometer per vehicle kilometers per hour hour, kilowatts per vehicle.

[0099] S303: Based on the vehicle type composition ratio and traffic flow, determine the number of vehicles of each type in each road segment, multiply the instantaneous energy consumption power of each vehicle type by the corresponding number of vehicles, and then sum them up to obtain the traffic flow energy consumption power of that road segment.

[0100] In this embodiment of the application, the number of vehicles refers to the equivalent number of vehicles of a certain type in the traffic flow within a certain road segment; the traffic flow energy consumption power refers to the total energy consumption rate of all vehicles in a certain road segment due to driving.

[0101] Specific implementation process: For each road segment, based on the proportion of vehicle types... Traffic flow on road sections Road segment length and average travel speed Calculate the equivalent number of vehicles for each vehicle type. ,satisfy:

[0102] .

[0103] in, Indicates the first The equivalent number of vehicles for a given vehicle type, in units of vehicles (veh). This indicates the traffic flow on a road segment, in vehicles per hour (veh / h). Indicates the first The proportion of different vehicle types in a given category; Indicates the length of the road segment, in kilometers (km). This indicates the average travel speed on the road segment, in kilometers per hour (km / h). It also shows the instantaneous power consumption per vehicle for each vehicle type. Corresponding equivalent number of vehicles Multiply and sum them to obtain the traffic flow energy consumption power of that road segment. ,satisfy:

[0104] .

[0105] in, This indicates the energy consumption of traffic flow on a road segment, in kilowatts (kW). This represents the total number of vehicle types. It's worth noting that traffic flow energy consumption power characterizes the energy consumption demand power corresponding to the traffic flow state at the physical layer, and is a key intermediate variable in the inter-layer mapping. In the formula for calculating the equivalent number of vehicles, For the traffic volume of this model, The product of the time required for a vehicle to pass through a road segment and the number of vehicles of that type simultaneously on the road segment is the total number of vehicles of that type present. For example, when... Vehicles per hour , km, At 1000 km / h, Vehicles; traffic flow energy consumption power is the sum of the product of the number of vehicles of each type and the energy consumption power per vehicle. For example, the energy consumption power per vehicle of this type is... This model contributes kilowatts per vehicle hour. kilowatt.

[0106] S304: Based on the inter-layer coupling operator, the energy consumption power of traffic flow is mapped to the energy layer load; the inter-layer coupling operator contains mapping weights that are dynamically adjusted according to the traffic flow state.

[0107] In this embodiment, the interlayer coupling operator refers to the mathematical mapping operator that converts the energy consumption power of the physical layer traffic flow into the net energy load power of the energy layer; the mapping weight refers to the adjustable parameter in the interlayer coupling operator used to quantify the proportion of traffic flow energy consumption power converted into energy layer load.

[0108] Specific implementation process: Read the initial mapping weights from the interlayer coupling operator. ,in Indicates the initial mapping weights. The value range is from 0.5 to 1.5 (0.5 ≤ ≤1.5), traffic flow energy consumption power Multiply by the initial mapping weight The energy layer load was initially mapped. ,satisfy:

[0109] .

[0110] in, This represents the initial mapped energy layer load, in kilowatts (kW). It should be noted that the mapping weights in the inter-layer coupling operator need to be dynamically adjusted according to the current traffic flow state. The dynamic adjustment of the mapping weights includes the following specific methods and steps:

[0111] S3041: Obtain the traffic flow and average speed of each road segment during the current time period.

[0112] In this embodiment of the application, the current time period refers to the smallest time unit in which the inter-layer mapping weights are dynamically updated.

[0113] Specific implementation process: Extract the current time period from the updated physical layer traffic flow state. Traffic flow on various road sections and average travel speed ,in This represents the index for the current time period. Indicates the current time period Traffic flow on the road segment, in vehicles per hour (veh / h); Indicates the current time period Average travel speed on the road segment, in kilometers per hour (km / h).

[0114] S3042: Using traffic flow and average vehicle speed as independent variables, a preset scenario-based mapping function is called to calculate the mapping weight for the current time period; the scenario-based mapping function is obtained by fitting historical traffic flow status and corresponding energy layer load data.

[0115] In this embodiment, the scenario-based mapping function refers to a nonlinear mapping function obtained by training and fitting historical traffic flow, average travel speed and measured energy layer load using a machine learning regression method; the mapping weight refers to the inter-layer mapping coefficient for the current time period output by the scenario-based mapping function.

[0116] Specific implementation process: based on the traffic flow of the current time period. and average travel speed As the independent variable, call the preset scenario-based mapping function. Calculate the mapping weight for the current time period. , satisfy:

[0117] .

[0118] in, Indicates the current time period Mapping weights; This represents a predefined scenario-based mapping function; Indicates the current time period Traffic flow on the road segment, in vehicles per hour (veh / h); Indicates the current time period Average travel speed on the road segment, in kilometers per hour (km / h).

[0119] It is worth noting that the preset scenario-based mapping function employs either a multivariate nonlinear regression model or a neural network regression model. When the number of historical training samples is less than 50,000, a multivariate nonlinear regression model is used; when the number of historical training samples is at least 50,000, a neural network regression model is used. Both models are trained and fitted based on historical traffic flow states and corresponding energy layer load data. The corresponding network structure includes: an input layer with 2 neurons, receiving the normalized road segment traffic flow. With average travel speed The first hidden layer has 128 neurons, the second hidden layer has 64 neurons, and the ReLU activation function is used for both. The output layer has 1 neuron, and the output is mapped to weights. The output layer activation function uses a linear activation function. During training, the Adam optimizer is used with an initial learning rate of 0.001. The learning rate decay strategy is 50% every 100 epochs. The training batch size is 256, and the number of training iterations is no less than 5000 epochs. The loss function is the mean squared error function, and an early stopping mechanism is triggered when the validation set loss no longer decreases for 50 consecutive epochs. The training dataset is divided into training and validation sets in an 8:2 ratio, and the input features are Min-Max standardized to the [0,1] interval. As a preferred embodiment, the scene mapping function uses quadratic polynomial regression: ,in and These are the normalized traffic flow and average travel speed, respectively. to The fitting coefficients are determined by least-squares fitting of historical samples; the output of the scenario-based mapping function. The regression output is truncated and constrained to the interval [0.3, 2.0]: if the regression output is below 0.3, it is set to 0.3; if it is above 2.0, it is set to 2.0, to avoid extreme inputs causing non-physical mapping weights. The fitted label data is the ratio of the measured load power of traffic-related equipment recorded by the energy metering system to the energy consumption power of traffic flow during the same period. The measured load power comes from the meter readings of the electricity meters deployed in the energy distribution circuits of each road segment. The goodness of fit of the scenario-based mapping function is... Not less than 0.85. Traffic-related equipment refers to energy-consuming equipment physically related to traffic flow, such as charging piles, tunnel ventilation fans, variable message signs, and traffic signal control equipment; the measured load power excludes non-traffic-related base load components such as lighting, office, and backup equipment before training. The base load is calculated based on historical periods without vehicles (traffic flow). The mean of veh / h is determined and subtracted. In the quadratic polynomial, Item and The terms respectively characterize the nonlinear saturation effect of flow rate and velocity on the mapping weights. The interaction term describes the interaction between flow rate and velocity, the linear term describes the linear main effect, and the constant term... As the baseline weight, for example when , , , , , ,and , hour, .

[0120] S3043: Update the mapping weights in the inter-layer coupling operator with the mapping weights of the current time period, and map the traffic flow energy consumption power to the dynamically adjusted energy layer load according to the updated mapping weights.

[0121] In this embodiment of the application, the dynamically adjusted energy layer load refers to the more accurate energy layer load mapping value obtained after dynamic updating of the mapping weight.

[0122] Specific implementation process: The mapping weights in the interlayer coupling operator are changed from the initial mapping weights. Updated to the mapping weights for the current time period. Based on the updated mapping weights, traffic flow energy consumption power is mapped to dynamically adjusted energy layer load. ,satisfy:

[0123] .

[0124] in, This indicates the dynamically adjusted energy layer load, in kilowatts (kW). Indicates the current time period Mapping weights; This represents the energy consumption power of traffic flow on a road segment, in kilowatts (kW). In the dynamic mapping calculation formula, the traffic flow energy consumption power is the energy demand calculated from the physical layer, and the mapping weight is the conversion ratio of this demand to the actual load of the energy layer. For example, when... kilowatt, hour, kilowatt.

[0125] It is worth noting that the dynamically adjusted energy layer load The segment-level energy consumption demand, which is characterized by mapping the physical layer traffic flow state to the energy layer, is then processed in step S4. According to the rated capacity ratio of each energy device in step S204, the power is allocated to each device, and the net energy load power of each device is adjusted. Corrected net energy load power of equipment satisfy ,in This indicates the consumer-end equipment within this road section ( >0) and energy storage charging equipment ( The sum of the absolute values ​​of net energy load power (>0), in kilowatts (kW); supply-side equipment such as photovoltaic power generation and gas-fired power generation ( <0) is not included in the summation, and its corrected load remains the original value. Unchanged; when When the power load is less than the preset power stability threshold (0.01 kW in this embodiment), the load of each device remains unchanged from its original value. This value will be used as input for subsequent carbon flow layer calculations and carbon emission baseline calculations; if step S304 is not executed or If missing, the calculation in step S204 is used directly. Equipment overload maintenance is underway. and The ratio is equal to the ratio of the dynamically mapped load to the total load of the road segment. Essentially, it scales the load of each device by the same proportion to make the sum of the device loads consistent with the dynamically mapped load. For example, when... kilowatt, kilowatt, kilowatt-hours kilowatt.

[0126] S3044: Calculate the difference between the mapping weight of the current time period and the mapping weight of the previous time period to obtain the weight change, and store the weight change in association with the mapping state parameter.

[0127] In the embodiments of this application, the previous time period refers to the previous adjacent time unit of the current time period; the weight change refers to the difference in the mapping weights of two adjacent time periods, which is used to characterize the evolution trend of the inter-layer mapping relationship.

[0128] Specific implementation process: Read the previous time period Mapping weights Calculate the mapping weight for the current time period. Mapping weights with the previous time period The difference yields the change in weight. , satisfy:

[0129] .

[0130] in, Indicates the amount of weight change; Indicates the current time period Mapping weights; Indicates the previous time period Mapping weights. Weight changes. It is stored in association with the mapping state parameters and serves as a component of the mapping relationship evolution. The weight change... This refers to the evolution of the mapping relationship mentioned in the subsequent calculation of carbon emission baselines, used to characterize the temporal drift of the inter-layer mapping relationship. The weight change is the difference in mapping weights between adjacent time periods; for example, the weight of the previous time period was... The current time period is hour, The larger the absolute value, the more drastic the change in the interlayer mapping relationship.

[0131] Understandably, by dynamically adjusting the mapping weights, the technical problem that fixed mapping weights cannot adapt to the dynamic changes in traffic flow states is solved, the adaptive updating of inter-layer mapping relationships is realized, and the accuracy and timeliness of mapping physical layer traffic flow states to energy layer loads are improved.

[0132] S305: Associate and store the energy layer load with the traffic flow and average vehicle speed of the corresponding road segment to generate mapping state parameters; the mapping state parameters include load-flow mapping pairs and load-speed mapping pairs.

[0133] In the embodiments of this application, the load-flow mapping pair refers to a binary data set consisting of energy layer load and corresponding road segment traffic flow; the load-speed mapping pair refers to a binary data set consisting of energy layer load and corresponding road segment average travel speed.

[0134] Specific implementation process: The dynamically adjusted energy layer load obtained in step S3043 is... Traffic flow of the corresponding road segment Associated storage, generating load-flow mapping pairs The dynamically adjusted energy layer load Average travel speed of the corresponding road segment Associated storage, generating load-speed mapping pairs Integrate the load-flow mapping pairs and the load-speed mapping pairs into a single mapping state parameter. , This represents the set of mapped state parameters. Among them, the mapped state parameters... It includes a set of load-flow mapping pairs and a set of load-speed mapping pairs, and records the corresponding time period identifiers. Road signs ,in This indicates the road segment identifier. It's worth noting the mapping status parameters. Used for determining the coupling overlap region and correcting the carbon emission superposition coefficient in subsequent carbon emission benchmark calculations.

[0135] Understandably, by establishing inter-layer mapping relationships and generating mapping state parameters, the technical problems of data isolation between the physical layer and the energy layer and the untraceability of mapping relationships have been solved. This has enabled the accurate correlation and state tracking between the traffic flow state of the physical layer and the load of the energy layer, providing a data foundation for the coupled calculation and anomaly detection of the carbon flow layer.

[0136] It should be noted that the specific steps of using a heterogeneous graph convolutional network to jointly encode the updated digital twin and obtain the feature vector include: constructing a heterogeneous graph structure, where the nodes of the heterogeneous graph structure include road segment nodes in the physical layer, equipment nodes in the energy layer, and emission nodes in the carbon flow layer, and the edges of the heterogeneous graph structure include intra-layer connection edges and inter-layer mapping edges; assigning initial node features to each node of the heterogeneous graph structure, where the initial node features of the road segment nodes in the physical layer include traffic flow, average travel speed, and traffic density, the initial node features of the equipment nodes in the energy layer include net energy load power and equipment operating efficiency, and the initial node features of the emission nodes in the carbon flow layer include cumulative carbon emissions and carbon emission intensity indicators; aggregating the feature information of heterogeneous neighbor nodes through the message passing mechanism of the heterogeneous graph convolutional network, performing feature transformation and dimensionality reduction through a preset number of graph convolutional layers, and obtaining low-dimensional embedded representations of each node; concatenating or pooling the embedded nodes of all nodes in the physical layer, energy layer, and carbon flow layer to obtain the joint feature vector of the digital twin. , Represents a feature vector with dimension . Dimension, among which Represents the dimension of the feature vector. The value range is 64 to 256 (64 ≤ ≤256). Where, the number of graph convolutional layers... This indicates the number of hidden layers in a heterogeneous graph convolutional network. The value range is 2 to 4 (2≤ ≤4).

[0137] It is worth noting that heterogeneous graph convolutional networks adopt relational graph attention networks or heterogeneous graph neural network architectures, with different types of edges corresponding to different attention weight matrices.

[0138] Understandably, by employing heterogeneous graph convolutional networks to jointly encode the updated digital twin, the technical problem of isolated features and difficulty in unified representation of the physical layer, energy layer, and carbon flow layer is solved. This achieves deep fusion and low-dimensional representation of multi-source heterogeneous information of the digital twin, providing a unified feature input for subsequent residual prediction models and anomaly scoring networks.

[0139] S4: Based on the updated physical layer traffic flow status and energy layer load, calculate the carbon emission baseline value through the physical model; based on historical data and feature vectors in multi-source heterogeneous data, predict the carbon emission residual through the residual prediction model to obtain the residual prediction value; superimpose the residual prediction value with the baseline value to obtain the carbon emission prediction value.

[0140] In this embodiment, the physical model refers to a carbon emission calculation model established based on the physical mechanism of carbon emission factors and energy load; the carbon emission baseline value refers to the theoretical carbon emission calculated by the physical model without considering abnormal fluctuations and inter-layer mapping drift; the residual prediction model refers to an artificial intelligence model that predicts the calculation error of the physical model based on historical data and feature vectors using machine learning methods; the carbon emission residual refers to the difference between the carbon emission prediction value and the carbon emission baseline value, used to characterize the carbon emission fluctuation components that the physical model failed to capture; the residual prediction value refers to the carbon emission residual estimate output by the residual prediction model; and the carbon emission prediction value refers to the final carbon emission estimate obtained by superimposing the residual prediction value and the carbon emission baseline value.

[0141] It is worth noting that, based on historical data and feature vectors from multi-source heterogeneous data, the residual prediction model predicts carbon emission residuals. The process involves constructing a training dataset, which includes historical multi-source heterogeneous data, historical carbon emission baseline values, and historical measured carbon emission values. The difference between the historical measured carbon emission value and the historical carbon emission baseline value serves as the label for the historical carbon emission residual. The residual prediction model employs a two-layer Long Short-Term Memory (LSTM) network or a Temporal Convolutional Network (TCN). LSTM is used when the number of historical training samples is less than 50,000; otherwise, TCN is used. Taking LSTM as an example, its network structure includes: an input layer with a receiving dimension of... eigenvectors and historical carbon emission baseline series, series length The value ranges from 12 to 48; the first LSTM hidden layer contains 128 LSTM units, and the second LSTM hidden layer contains 64 LSTM units. The activation function for both is tanh, and the forget gate bias is initialized to 1.0 to prevent gradient vanishing; the output layer is a fully connected layer containing 32 neurons, with a linear activation function, and outputs the residual prediction value. During training, the Adam optimizer was used with an initial learning rate of 0.001, a batch size of 128, and at least 500 training iterations. The mean squared error loss function was used, and an early stopping mechanism was triggered when the root mean square error on the validation set stopped decreasing after 30 consecutive iterations. The training dataset was divided into training and validation sets in an 8:2 ratio, and the input features were Z-score standardized. The current feature vector was input into the trained residual prediction model, which outputs the residual prediction value. , This represents the predicted residual value, in kilograms of carbon dioxide per hour (kgCO2 / h).

[0142] It should be noted that calculating the baseline carbon emission value using a physical model includes the following specific methods and steps:

[0143] S401: Obtain the updated physical layer traffic flow status and energy layer load; the physical layer traffic flow status includes the traffic volume and average speed of each vehicle type on each road segment, and the energy layer load includes the energy consumption of energy equipment on each road segment.

[0144] In this embodiment of the application, the traffic flow by vehicle type refers to the traffic flow corresponding to different vehicle types in each road segment; the equipment energy consumption refers to the actual active power consumption of the energy equipment in each road segment during operation.

[0145] Specific implementation process: Extract traffic flow data by vehicle type for each road segment from the physical layer of the digital twin updated in step S2. and average travel speed Extract the energy consumption of energy equipment in each section from the energy layer of the updated digital twin in step S2. For road sections that have undergone interlayer coupling correction in step S3043, their equipment energy consumption is based on the corrected net energy load power. It is determined that the same device will not be repeatedly assigned to multiple road segments; if the device is located at the intersection of road segments, it will be uniquely assigned according to the nearest neighbor principle, ensuring spatial alignment of fixed source energy consumption and mobile source energy consumption at the road segment level, and obtaining the input data required for physical model calculation. Indicates the first Traffic flow for each vehicle type, in vehicles per hour (veh / h). Indicates the first Average speed of vehicle type, unit: kilometers per hour (km / h); This indicates the energy consumption of the equipment, in kilowatts (kW).

[0146] S402: Calculate the instantaneous carbon emission rate of each vehicle based on the preset unit mileage carbon emission factor and average vehicle speed of each vehicle type; multiply the instantaneous carbon emission rate of each vehicle type by the number of vehicles on the corresponding road segment to obtain the mobile source carbon emission component.

[0147] In this embodiment of the application, the preset unit mileage carbon emission factor refers to the standard carbon dioxide emission of a certain type of vehicle under a unit mileage; the single vehicle instantaneous carbon emission rate refers to the carbon dioxide emission of a single vehicle in a unit time; and the mobile source carbon emission component refers to the total carbon dioxide emission generated by vehicles traveling on the road.

[0148] Specific implementation process: For each type of vehicle Read its preset carbon emission factor per unit mileage Based on average travel speed Calculate the instantaneous carbon emission rate per vehicle for this model. , satisfy:

[0149] .

[0150] in, Indicates the first Instantaneous carbon emission rate per vehicle for this type of vehicle, unit: kilograms of carbon dioxide per vehicle per hour (kgCO2 / (h·veh)); Indicates the first Preset carbon emission factor per unit mileage for vehicle type, unit: kgCO2 / (km·veh) per vehicle; Indicates the first Average speed of vehicle type, in kilometers per hour (km / h). Based on vehicle type composition ratio. Traffic flow on road sections Road segment length and average travel speed Calculate the equivalent number of vehicles for each vehicle type. The instantaneous carbon emission rate of each vehicle model Corresponding equivalent number of vehicles Multiply and then sum to obtain the carbon emission component from mobile sources. , satisfy:

[0151] .

[0152] in, This indicates the carbon emission component from mobile sources, in kilograms of carbon dioxide per hour (kgCO2 / h). This indicates the total number of vehicle models. It's worth noting the preset carbon emission factor per unit mileage. The emission standards for motor vehicles are determined according to national or industry regulations, including those for small passenger vehicles. The value range is from 0.1 kg CO2 per kilometer to 0.3 kg CO2 per kilometer (0.1 kg CO2 / km ≤ ≤0.3kgCO2 / km), heavy trucks The value range is from 0.8 kg CO2 / km to 2.0 kg CO2 / km (0.8 kg CO2 / km ≤ ≤2.0kgCO2 / km). Among them, the first Traffic flow by vehicle type satisfy Equivalent number of vehicles Instantaneous carbon emission rate per vehicle The product of the two Since speed variables cancel each other out in the product, the combined form of carbon emission components from mobile sources is strictly equivalent to the vehicle-type-specific calculation form. (Preset carbon emission factor per unit mileage) Search by preset speed gear; speed gear selection and steps. The vehicle speed is divided into gears, with different speed gears corresponding to different speeds. Value, low gear The value is set higher than the mid-speed range to ensure consistency between speed corrections and calculations of mobile source carbon emissions and energy consumption. In the formula for calculating the instantaneous carbon emission rate of a single vehicle, Carbon emissions per unit mileage For driving speed, the product of the two converts emissions per unit distance to emissions per unit time. For example, when... kilograms of carbon dioxide per kilometer At 1000 km / h, Kilograms of carbon dioxide per hour per vehicle; the carbon emission component from mobile sources is the sum of the products of the number of vehicles of each type and their emission rate per vehicle, for example, when When the vehicle was in use, this model contributed kilograms of carbon dioxide per hour.

[0153] S403: Calculate the direct carbon emissions of each energy device based on the preset energy carbon emission factor and energy consumption of each device; sum up the direct carbon emissions of each device to obtain the carbon emission component from the stationary source.

[0154] In this application embodiment, the preset energy carbon emission factor refers to the carbon dioxide emission of a certain type of energy equipment per unit of energy consumption; direct carbon emission of equipment refers to the carbon dioxide emission of a single energy equipment per unit time; and stationary source carbon emission component refers to the total carbon dioxide emission generated by energy equipment permanently installed in transportation infrastructure.

[0155] Specific implementation process: For each energy device Read its preset energy carbon emission factor and equipment energy consumption When the device has performed the interlayer coupling correction in step S3043, Take the absolute value of the corrected net energy load power. Calculate the direct carbon emissions of the equipment. , satisfy:

[0156] .

[0157] in, Indicates device Direct carbon emissions from equipment, unit: kilograms of carbon dioxide per hour (kgCO2 / h). This indicates the preset energy carbon emission factor, in units of kilograms of carbon dioxide per kilowatt-hour (kgCO2 / kWh). Indicates device Equipment energy consumption, unit: kilowatt (kW). Direct carbon emissions from each piece of equipment. By summing them up, we obtain the carbon emission components from stationary sources. , satisfy: .

[0158] in, This indicates the carbon emission component from stationary sources, in kilograms of carbon dioxide per hour (kgCO2 / h). Indicates the device index; This indicates the total number of devices. It is worth noting the preset energy carbon emission factor. In China, power equipment The value range is from 0.5 kg CO2 / kWh to 0.8 kg CO2 / kWh (0.5 kg CO2 / kWh ≤ ≤0.8kgCO2 / kWh), gas-fired equipment The value ranges from 1.8 kg CO2 per cubic meter to 2.2 kg CO2 per cubic meter (1.8 kg CO2 / m³). 3 ≤ ≤2.2kgCO2 / m 3 If the equipment energy consumption When measuring by volumetric flow rate, the calorific value must first be converted to power units before calculation. The specific formula for calorific value conversion is: ,in, This is expressed as kilograms of carbon dioxide per kilowatt-hour (kgCO2). Energy carbon emission factor in units of 1000; It is expressed in kilograms of carbon dioxide per cubic meter (kgCO2) Carbon emission factor of fuel gas, expressed in units of 100%. This indicates the lower calorific value of the gas, in megajoules per cubic meter (MJ / m³). ), natural gas The preferred value is Megajoules per cubic meter; This is the unit conversion factor between megajoules and kilowatt-hours (1 kWh = 3.6 MJ). In the formula for calculating direct carbon emissions from equipment, the energy carbon emission factor is the carbon dioxide emission per unit of energy consumption, and the equipment energy consumption is the equipment operating power. The product of the two is the carbon emission of the equipment per unit time. For example, when... kilograms of carbon dioxide per kilowatt-hour kilowatt-hours Kilograms of carbon dioxide per hour. In the calorific value conversion formula, Carbon emissions per cubic meter of natural gas. Energy per cubic meter of gas, Convert the volumetric basis to the energy basis and multiply by Complete the unit conversion between megajoules and kilowatt-hours, for example when kilograms of carbon dioxide per cubic meter megajoules per cubic meter, kilograms of carbon dioxide per kilowatt-hour.

[0159] S404: Determine the coupling overlap area between mobile and fixed sources at the road segment level based on the mapping state parameters, and correct the carbon emission superposition coefficient within the coupling overlap area using the mapping relationship evolution in the mapping state parameters; retain the preset benchmark superposition coefficient for road segments in non-coupling overlap areas.

[0160] In this embodiment, the coupled overlapping region refers to a spatial grid region at the road segment level where both significant mobile source carbon emissions and stationary source carbon emissions exist simultaneously and are energy-interactingly related; the carbon emission superposition coefficient refers to a correction coefficient used to quantify the superposition effect of mobile source carbon emission components and stationary source carbon emission components within the coupled overlapping region; and the mapping relationship evolution refers to the change in weights in the mapping state parameters. The degree of dynamic change in the interlayer mapping relationship is represented; the preset benchmark superposition coefficient refers to the standard weighting coefficient of the carbon emission components of mobile sources and stationary sources in the non-coupled overlapping area.

[0161] Specific implementation process: Reading mapping state parameters The load-flow mapping pairs and load-speed mapping pairs are extracted, and the coupling overlap area between mobile and stationary sources at the road segment level is determined based on the mapping state parameters. It should be noted that determining the coupling overlap area between mobile and stationary sources at the road segment level based on the mapping state parameters includes the following specific method steps:

[0162] S4041: Retrieve the load-flow mapping pair from the mapping status parameters.

[0163] In this embodiment of the application, the load-flow mapping pair refers to a binary data set consisting of dynamically adjusted energy layer load and corresponding road segment traffic flow.

[0164] Specific implementation process: From mapping state parameters In the process, extract the load-flow mapping pair set for each road segment within a preset time window. ,in, This indicates the dynamically adjusted energy layer load, in kilowatts (kW). This indicates the traffic flow rate of a road segment, measured in vehicles per hour (veh / h).

[0165] S4042: Determine the average energy consumption per vehicle for each road segment based on the ratio of load value to flow value in the load-flow mapping pair.

[0166] In this embodiment, the average energy consumption per vehicle refers to the energy layer load consumption corresponding to a unit of traffic flow, used to characterize the energy efficiency of a road segment, and its physical dimension is kilowatt-hour per vehicle (kWh / veh). Its physical meaning is the energy layer power load distributed per unit hour of traffic flow, used to reflect the coupling strength between traffic flow and energy load in a road segment.

[0167] Specific implementation process: For each road segment, calculate the load value. With flow value The ratio gives the average energy consumption per vehicle. , satisfy:

[0168] .

[0169] in, This indicates the average energy consumption of a single vehicle, in kilowatt-hours per vehicle (kWh / veh). This indicates the dynamically adjusted energy layer load, in kilowatts (kW). This indicates the traffic flow rate of a road segment, measured in vehicles per hour (veh / h). It's important to note that when the traffic flow rate of a road segment... When it is 0, the average energy consumption per vehicle Take the calculated value from the previous valid time period or the preset default value, which is 0.5 kWh per vehicle. (kWh / veh). In the formula for calculating the average energy consumption per vehicle, the numerator is the dynamically mapped energy layer load of the road segment, the denominator is the traffic flow of the road segment, and the ratio represents the power load allocated to the traffic flow per unit hour. For example, when... kilowatt, Vehicles per hour kilowatt-hours per vehicle.

[0170] S4043: Divide the carbon emission components from mobile sources and stationary sources by the preset road segment length to obtain the carbon emission per unit length of mobile sources and the carbon emission per unit length of stationary sources for each road segment.

[0171] In this embodiment of the application, the preset road segment length refers to the geometric length benchmark value of each road segment; the carbon emission per unit length of mobile source refers to the carbon emission intensity of mobile source per unit road segment length; and the carbon emission per unit length of stationary source refers to the carbon emission intensity of stationary source per unit road segment length.

[0172] Specific implementation process: Read the preset road segment length for each road segment. The carbon emissions from mobile sources Divide by the preset road segment length The carbon emissions per unit length of mobile sources were obtained. ; Components of carbon emissions from stationary sources Divide by the preset road segment length The carbon emissions per unit length from stationary sources were obtained. .in, This indicates carbon emissions per unit length from mobile sources, expressed in kilograms of carbon dioxide per hour per kilometer (kgCO2 / (h·km)). This indicates carbon emissions per unit length from stationary sources, expressed in kilograms of carbon dioxide per hour per kilometer (kgCO2 / (h·km)). This represents the preset road segment length, in kilometers (km). The calculation formula satisfies the following: In the formula for calculating carbon emissions per unit length, the numerator is the carbon emission component of the road segment, the denominator is the road segment length, and the ratio converts the total amount to the intensity per unit mileage to eliminate the influence of differences in road segment length. For example, when... kilograms of carbon dioxide per hour kilometer kilograms of carbon dioxide per hour per kilometer.

[0173] .

[0174] .

[0175] S4044: Divide each road segment into grids using a preset spatiotemporal resolution. When the average energy consumption of a single vehicle in the same grid is greater than the preset energy consumption coupling threshold, and the sum of the carbon emissions per unit length of the mobile source and the carbon emissions per unit length of the fixed source is greater than the preset carbon emission coupling threshold, the grid is determined to be a coupling overlap area.

[0176] In this embodiment, the preset spatiotemporal resolution refers to the temporal and spatial granularity of spatial discretization of the road segment; grid division refers to dividing the road segment into several continuous spatial units according to the preset spatiotemporal resolution; the preset energy consumption coupling threshold refers to the critical value of average energy consumption per vehicle for judging whether there is a significant coupling relationship between traffic flow and energy load within the grid; and the preset carbon emission coupling threshold refers to the critical value of carbon emission intensity for judging whether the carbon emissions from mobile sources and stationary sources within the grid reach the superposition judgment standard.

[0177] Specific implementation process: Setting a preset spatiotemporal resolution The distance is between 100 meters and 500 meters (100m ≤ ≤500m), with a preset spatiotemporal resolution Divide each road segment into a one-dimensional grid. If the road segment length is not... If the length is an integer multiple of the given length, the remaining length is treated as an independent grid cell, resulting in several grid cells. For each grid cell, the average energy consumption per vehicle within that grid is read. Carbon emissions per unit length from mobile sources and carbon emissions per unit length from stationary sources When the average energy consumption of a single vehicle Greater than the preset energy consumption coupling threshold And the carbon emissions per unit length of mobile sources carbon emissions per unit length from stationary sources The sum is greater than the preset carbon emission coupling threshold. When this occurs, the mesh is determined to be a coupled overlapping region. Among them, Indicates the preset spatiotemporal resolution, in meters (m). This indicates the preset energy consumption coupling threshold, in kilowatt-hours per vehicle (kWh / veh). The value ranges from 0.3 kWh per vehicle to 0.8 kWh per vehicle (0.3 kWh / veh ≤ ≤0.8kWh / veh); This indicates the preset carbon emission coupling threshold, in kilograms of carbon dioxide per hour per kilometer (kgCO2 / (h·km)). The value range is from 0.5 kg CO2 / (h·km) to 2.0 kg CO2 / (h·km) ≤ ≤2.0kgCO2 / (h·km)). The grid is divided into a one-dimensional grid along the road segment mileage direction, and the grid coordinates are represented by mileage markers; the average energy consumption per vehicle in each grid is taken as the average energy consumption per vehicle of the road segment to which that grid belongs; a preset energy consumption coupling threshold is used. Coupled with preset carbon emission threshold Based on historical data statistical quantile calibration Take the historical average energy consumption sequence of individual vehicles for this road section. to Quantiles Take the historical sequence of carbon emissions per unit length to Quantiles.

[0178] S4045: Associate and store the grid coordinates of the coupled overlapping area with the corresponding road segment identifier to generate the coupled overlapping area identifier.

[0179] In this embodiment of the application, the coupling overlap region identifier refers to the encoded information used to uniquely mark the spatial location of the coupling overlap region and the road segment to which it belongs.

[0180] Specific implementation process: For each grid cell identified as a coupled overlapping area, record the starting mileage marker of that grid cell within the road segment. Termination mileage marker and the road section signs Generate a set of identifiers for coupled overlapping regions. , This represents the set of identifiers for coupled overlapping regions. Indicates the starting position of the grid cell on the road segment, in meters (m). This indicates the termination position of the grid cell on the road segment, in meters (m). It is worth noting the set of coupled overlapping area identifiers. Used for the differential weighted calculation of subsequent carbon emission superposition coefficients.

[0181] Understandably, by identifying the coupling overlap region, the technical problem of repeated calculations or omissions caused by the simple spatial superposition of carbon emissions from mobile and stationary sources has been solved. This has enabled the accurate identification of the spatial coupling relationship of carbon emissions and provided spatial constraints for the refined calculation of carbon emission benchmark values.

[0182] S405: The carbon emission components of mobile sources and stationary sources in the coupled overlapping region are weighted and summed using the corrected carbon emission superposition coefficient, and the components in the non-coupled overlapping region are weighted and summed using the preset benchmark superposition coefficient, and the carbon emission benchmark value is obtained by accumulating the summation.

[0183] In this embodiment, the modified carbon emission superposition coefficient refers to the carbon emission weighting coefficient within the coupled overlapping region after being corrected by the mapping relationship evolution; the preset benchmark superposition coefficient refers to the standard weighting coefficient of the carbon emission components of mobile and stationary sources within the non-coupled overlapping region.

[0184] Specific implementation process: Read the evolution of the mapping relationship in the mapping state parameters, i.e., the change in weights. Calculate the corrected carbon emission superposition factor , satisfy:

[0185] .

[0186] in, This represents the corrected carbon emission superposition factor; Indicates the preset benchmark superposition coefficient. The value range is from 0.8 to 1.2 (0.8 ≤ ≤1.2); The correction gain coefficient represents the evolution of the mapping relationship. The value range is from 0.1 to 0.5 (0.1 ≤ 0.5). ≤0.5); This represents the change in weight. It assigns the carbon emission components from mobile sources at the road segment level. Compared with stationary source carbon emissions By grid length Distribute evenly to each grid, grid The mobile source component is Fixed source component is The carbon emission components of the coupled and uncoupled overlapping regions are the sum of the grid components of all coupled and uncoupled grids, respectively; for the coupled overlapping region identifier set The identified grid is based on the corrected carbon emission overlay factor. carbon emissions from mobile sources Compared with stationary source carbon emissions The carbon emission components of the coupled overlapping region are obtained by weighted summation of the grid components. For meshes in non-coupled overlapping regions, a preset reference stacking coefficient is used. carbon emissions from mobile sources Compared with stationary source carbon emissions The carbon emission components of the non-coupling overlapping region are obtained by weighted summation and accumulation of the grid components. .in, This indicates the carbon emission component of the coupled overlapping region, in kilograms of carbon dioxide per hour (kgCO2 / h). Satisfying: The superposition coefficient is being revised. As the baseline, the superimposed level, Here, is the mapping drift correction factor, where To ensure that the superposition coefficients do not degenerate when the mapping is stable, For drift sensitivity gain, For the drift amplitude, for example when , , hour, .

[0187] .

[0188] This represents the carbon emission component of the non-coupling overlapping region, in kilograms of carbon dioxide per hour (kgCO2 / h). satisfy:

[0189] .

[0190] Carbon emission components of the coupled overlapping region Carbon emission components of the non-coupling overlap region By summing them up, we obtain the baseline value for carbon emissions. , satisfy:

[0191] .

[0192] in, This represents the baseline value for carbon emissions, in kilograms of carbon dioxide per hour (kgCO2 / h). The coupled and non-coupled overlapping regions are determined by mutually exclusive grid partitioning; both cover the entire road segment without overlap or omission. It is worth noting that when the weight changes... When the absolute value is greater than 0.3, the corrected carbon emission superposition factor Trigger upper limit truncation, The maximum value does not exceed 1.5. The linear correction relationship between the corrected carbon emission superposition coefficient and the absolute value of the weight change is a linear sensitivity relationship between inter-layer mapping drift and inter-source coupling effects obtained based on historical calibration data fitting. This relationship is used to increase the superposition weights of mobile and stationary carbon emission components within the coupling overlap region as the degree of mapping relationship evolution increases. The reason for using the absolute value of the weight change is that regardless of whether the mapping weight drifts positively or negatively, it indicates that the inter-layer mapping relationship deviates from the steady state. Its influence on the coupling strength of mobile and stationary carbon emissions is the same in degree but independent of direction. Therefore, it is more effective in... Take the absolute value and perform symmetry processing; The cutoff trigger value is taken from the historical normal operating conditions. sequential % quantile. The carbon emission baseline is the sum of the carbon emission components in the coupled and uncoupled overlapping regions. For example, within the coupled and uncoupled overlapping regions... kilograms of carbon dioxide per hour hour, kilograms of carbon dioxide per hour, if the non-coupling overlapping region component is zero, then Kilograms of carbon dioxide per hour. This example corresponds to the case where the entire road segment's grid is determined to be a coupled overlapping region.

[0193] Understandably, calculating carbon emission baseline values ​​through physical models solves the technical problem that independent calculations of single emission sources cannot reflect the coupling effect between sources. It enables coupled perception and superimposed correction of carbon emissions from mobile and stationary sources, allowing carbon emission baseline values ​​to accurately reflect the comprehensive carbon emission level at the road segment level.

[0194] It should be noted that, based on historical data and feature vectors from multi-source heterogeneous data, the carbon emission residual is predicted using a residual prediction model to obtain the residual prediction value. This process includes: constructing a training dataset, which includes historical multi-source heterogeneous data, historical carbon emission baseline values, and historical measured carbon emission values. The difference between the historical measured carbon emission value and the historical carbon emission baseline value is used as the label for the historical carbon emission residual; training a residual prediction model based on a Long Short-Term Memory network or a temporal convolutional network using the feature vector as input and the historical carbon emission residual label as output; and inputting the current feature vector into the trained residual prediction model to output the residual prediction value. , This represents the residual prediction value, in kilograms of carbon dioxide per hour (kgCO2 / h). The residual prediction value... Compared with carbon emission benchmarks By superimposing the values, we obtain the predicted carbon emissions. , satisfy:

[0195] .

[0196] in, This represents the predicted carbon emissions, in kilograms of carbon dioxide per hour (kgCO2 / h). It's important to note that the residual prediction model undergoes at least 500 training iterations, and the mean squared error function is used as the loss function. The time alignment for residual prediction is as follows: using the historical carbon emission baseline sequence within a sliding time window as the time series input, with a window length of Tseq, predict the carbon emission residual at the current moment; training samples are constructed using a sliding window, sliding one sampling interval at a time; feature vectors... As static features, the carbon emission baseline value at each time step is concatenated along the feature dimension and then input into the residual prediction model. The mechanistic baseline given for the physical model, This is the compensation amount for the systematic bias of the physical model by the data-driven model. The superposition of the two achieves the fusion of physical mechanism and data-driven approach. For example, when kilograms of carbon dioxide per hour kilograms of carbon dioxide per hour kilograms of carbon dioxide per hour.

[0197] Understandably, by predicting carbon emission residuals using a residual prediction model and superimposing them on a benchmark value, the systematic bias caused by simplification assumptions and time-varying parameters in the physical model is resolved. This achieves a fusion of data-driven and physical mechanism-based prediction, improving the accuracy and robustness of carbon emission predictions.

[0198] S5: Calculate the dynamic mutual information value based on the time series samples of carbon emission prediction values ​​and carbon emission benchmark values ​​within a preset time window; input the carbon emission prediction values ​​into the anomaly scoring network to obtain an anomaly score; when the dynamic mutual information value is lower than the preset fluctuation threshold and the anomaly score is greater than the preset scoring threshold, determine the drift of the inter-layer mapping relationship based on the mapping state parameters and generate an anomaly warning signal.

[0199] In this embodiment, the preset time window refers to a fixed time interval for extracting time-series samples; the time-series sample refers to the sequence of carbon emission prediction values ​​and carbon emission benchmark values ​​collected at fixed sampling intervals within the preset time window; the dynamic mutual information value refers to an information theory index based on the concept of mutual information in information theory, used to quantify the statistical dependence between the carbon emission residual time-series sequence and the mapping state parameter time-series update sequence, and the lower the value, the stronger the random fluctuation of the inter-layer mapping relationship; the anomaly scoring network refers to a deep learning model based on graph neural networks and attention mechanisms, used to output a quantitative score of the anomaly degree of carbon emission prediction values; the preset fluctuation threshold refers to the critical value for judging whether the dynamic mutual information value has decreased significantly; the preset scoring threshold refers to the critical value for judging whether the anomaly score has reached the anomaly judgment standard; the anomaly warning signal refers to the alarm information generated when the inter-layer mapping relationship drifts.

[0200] It is worth noting that the anomaly scoring network adopts a graph attention network architecture, including an input layer, three graph attention hidden layers, a fully connected layer, and an output layer. The input layer receives the spatiotemporal graph structure. Node features With edge weight The hidden layer configuration for the graph attention mechanism is similar to that of the heterogeneous graph convolutional network in step S3, but it only handles homogeneous node relationships. Each attention head has a dimension of 16, and the number of attention heads is 4. The fully connected layer configuration is as described in step S515. During training, a semi-supervised anomaly detection framework is used, with the reconstruction error of normal samples and the classification error of abnormal samples jointly used as the loss function. The ratio of normal samples to abnormal samples is approximately 10:1. The optimizer used is Adam, with an initial learning rate of 0.0005. The number of training iterations is no less than 1000 epochs, the batch size is 64, and a cosine annealing strategy with a learning rate is adopted, with a dropout rate set to 0.2. The training dataset includes historical normal carbon emission prediction sequences and artificially injected anomaly samples (including mutation anomalies and drift anomalies), with the injected anomaly ratio being 10%.

[0201] It should be noted that the calculation method for dynamic mutual information values ​​includes the following specific steps:

[0202] S501: Obtain the carbon emission prediction value within a preset time window and generate a time series of the carbon emission prediction value; obtain the carbon emission benchmark value within the same preset time window and generate a time series of the carbon emission benchmark value.

[0203] In the embodiments of this application, a time series refers to a set of discrete data points arranged in chronological order.

[0204] Specific implementation process: Setting a preset time window Its value should guarantee ,in and The units must be converted before calculation, preferably... The sampling interval is 7.5 hours to 24 hours. 5 to 15 minutes (5 min ≤ ≤15min); if Then expand or shorten To meet the minimum sample size requirement, within the preset time window Within, according to sampling interval Collect carbon emission forecasts and generate a time series of carbon emission forecasts. ,in Indicates a time-series index; within the same preset time window Within, according to the same sampling interval Collect carbon emission baseline values ​​and generate a time series of carbon emission baseline values. .in, Indicates time The predicted carbon emissions, in kilograms of carbon dioxide per hour (kgCO2 / h). Indicates time The carbon emission baseline, in kilograms of carbon dioxide per hour (kgCO2 / h).

[0205] S502: Calculate the point-by-point difference between the time series of predicted carbon emissions and the time series of the baseline carbon emissions to obtain the time series of carbon emission residuals.

[0206] In the embodiments of this application, the point-by-point difference refers to the algebraic difference between the predicted carbon emission value and the carbon emission benchmark value at the same time.

[0207] Specific implementation process: For each time step Calculate carbon emission forecasts Compared with carbon emission benchmarks The difference is used to obtain the carbon emission residual. , satisfy:

[0208] .

[0209] in, Indicates time Carbon emission residuals, in kilograms of carbon dioxide per hour (kgCO2 / h). Iterate through the preset time windows. Generate a time series of carbon emission residuals from all sampling times within the time frame. In the formula for calculating point-by-point difference, (t) is carbon emission forecasts at time E base(t) represents the baseline carbon emissions at that moment, and the difference represents the residual component that the physical model could not explain at that moment. For example, when Ê(t) = 326.8, E base When ϵ(t) = 321.8 kg of carbon dioxide per hour, ϵ(t) = 5 kg of carbon dioxide per hour.

[0210] S503: Obtain the time-series update sequence of the mapped state parameters within a preset time window.

[0211] In this embodiment of the application, the time-series update sequence refers to the set of update records of the mapped state parameters arranged according to the sampling time within a preset time window.

[0212] Specific implementation process: within the preset time window Within, according to sampling interval Read mapping status parameters The update records are used to extract the mapping weights at each time step. and weight change Generate a time-series update sequence for the mapped state parameters. ,in Indicates time The mapping state parameters.

[0213] S504: Using a preset kernel density estimation method, estimate the probability distribution of the time-series update sequence of the mapped state parameters to obtain the first probability distribution; estimate the probability distribution of the time-series sequence of carbon emission residuals to obtain the second probability distribution; and estimate the joint probability distribution of the time-series update sequence of the mapped state parameters and the time-series sequence of carbon emission residuals.

[0214] In this embodiment, the kernel density estimation method refers to a statistical method for nonparametric estimation of the probability density of a random variable based on a kernel function; the first probability distribution refers to the marginal probability distribution of the time-series update sequence of the mapped state parameters; the second probability distribution refers to the marginal probability distribution of the time-series sequence of carbon emission residuals; and the joint probability distribution refers to the joint probability distribution of the time-series update sequence of the mapped state parameters and the time-series sequence of carbon emission residuals.

[0215] Specific implementation process: The Gaussian kernel function is used as the kernel function of the preset kernel density estimation method, and the kernel bandwidth is set. The value is between 0.1 and 0.5 (0.1 ≤ ≤0.5), the time-series update sequence of the mapped state parameters Kernel density estimation is performed to obtain the first probability distribution. Time series of carbon emission residuals Kernel density estimation is performed to obtain the second probability distribution. Kernel density estimation is performed on a two-dimensional joint sample of the time-series update sequence of the mapped state parameters and the time-series sequence of the carbon emission residuals to obtain the joint probability distribution. .in, Indicates kernel bandwidth; Represents the first probability distribution; This represents the second probability distribution; This represents the joint probability distribution. It is worth noting that the kernel density estimation uses the Parzen-Rosenblatt window method, and the kernel function... Satisfies the standard Gaussian kernel function form:

[0216] .

[0217] in, This represents the independent variable of the kernel function. Kernel bandwidth. Determined adaptively according to Silverman's empirical rules: ,in, Indicates the sample standard deviation. This represents the interquartile range of the sample. This represents the sample size, and the calculation results are constrained by... to Within the range of values. In the Gaussian kernel function, The normalization constant makes the kernel function integral equal to 1, and the exponential term makes the weight of the sample points vary with distance. The increase of decreases exponentially, for example when ... hour For the peak value, when hour That is, the closer a sample is to the estimated point, the greater its contribution to the probability density. In Silverman's empirical rule, and The smaller of the two is a robust measure of data dispersion. of The power of 1 is the bandwidth factor that shrinks as the sample size increases, for example, when , , hour, , in to Within the constraints. Before calculating mutual information, the weight changes are considered. carbon emission residuals Z-score standardization was performed separately to eliminate the influence of differences in dimensions and magnitudes on kernel density estimation.

[0218] S505: Calculate the dynamic mutual information value between the time-series update sequence of the mapped state parameters and the time-series sequence of the carbon emission residuals based on the first probability distribution, the second probability distribution, and the joint probability distribution.

[0219] In the embodiments of this application, the dynamic mutual information value refers to a quantitative index based on the concept of mutual information in information theory, used to measure the degree of nonlinear statistical dependence between two time-series variables.

[0220] Specific implementation process: Based on the first probability distribution Second probability distribution and joint probability distribution Calculate dynamic mutual information value , satisfy:

[0221] .

[0222] in, Represents dynamic mutual information value, unit: bit; Represents the joint probability distribution; Represents the first probability distribution; This represents the second probability distribution. It's worth noting the dynamic mutual information value. The value range is from 0 bits to positive infinity. bit), when A value of 0 bits indicates that the mapped state parameters are independent of the carbon emission residuals. A larger value indicates a stronger statistical dependence between the two. In this application, the temporal update sequence of the mapped state parameters is taken as the weight change. The variable used in mutual information calculation is a scalar time series. The double integral of mutual information is approximated by a two-dimensional uniform grid numerical integral, which represents the change in weights. carbon emission residuals The range of values ​​is evenly divided into: There are 1 grid intervals, totaling 1 For each grid cell, the product of the integrand value and the grid area is accumulated using the rectangular quadrature method to obtain a numerical approximation of the dynamic mutual information value. In the mutual information integral formula, Let be the joint probability distribution of the two variables. and Given their respective marginal probability distributions, the logarithmic terms measure the degree to which the joint distribution deviates from the independence hypothesis. The double integral accumulates this deviation over the entire value space; for example, when the two variables are completely independent, the integrand is always zero. Bit; In this application, the mapped state parameters under normal operating conditions have a stable correlation with the carbon emission residual. Maintaining a high level, when inter-layer mapping drift leads to weakened correlation. decline.

[0223] Understandably, by calculating the dynamic mutual information value, the technical problem that traditional correlation coefficients cannot capture nonlinear dependencies has been solved, and a quantitative measurement of the degree of information coupling between inter-layer mapping relationship and carbon emission residual has been achieved, providing an information theory basis for determining inter-layer mapping drift.

[0224] It should be noted that inputting the carbon emission prediction values ​​into the anomaly scoring network to obtain the anomaly score involves the following specific methods and steps:

[0225] S511: Obtain carbon emission predictions, feature vectors, and dynamic mutual information values.

[0226] Specific implementation process: Read the carbon emission prediction value obtained in step S4. Read the feature vector obtained in step S3 Read the dynamic mutual information value obtained in step S505. , as input data for the anomaly scoring network.

[0227] S512: Obtain the time sequence of dynamic mutual information values ​​within a preset time window, calculate the difference of dynamic mutual information values ​​between adjacent time points according to a preset difference rule, and obtain the rate of change of dynamic mutual information values.

[0228] In this embodiment, the rate of change of dynamic mutual information value refers to the difference ratio of dynamic mutual information values ​​at adjacent sampling times, which is used to characterize the evolution rate of the degree of information coupling of inter-layer mapping relationship; the preset difference rule refers to the numerical difference method used to calculate the difference of dynamic mutual information values ​​at adjacent time points.

[0229] Specific implementation process: within the preset time window Within, obtain the time series sequence of dynamic mutual information values. Calculate adjacent time points according to the preset difference rules and The rate of change of the dynamic mutual information value is obtained by differentiating the dynamic mutual information value. , satisfy:

[0230] .

[0231] in, Represents the rate of change of dynamic mutual information value, in bits per hour (bit / h). Indicates time The dynamic mutual information value, in bits. Indicates time The dynamic mutual information value, in bits. The sampling interval is represented by the unit "hours" from step S501, which is then converted to "hours" for calculation. It is worth noting that the preset difference rule uses a first-order backward difference scheme. A negative value indicates a decreasing trend in dynamic mutual information, suggesting a weakening of information coupling in inter-layer mapping relationships. To suppress the amplification effect of kernel density estimation noise on the difference, [the following is omitted as it is not relevant to the preceding text]. The sequence is first smoothed using a moving average (window width of 3 to 5 sampling points), and then the rate of change of difference is calculated. In the formula for calculating the rate of change of dynamic mutual information, the numerator is the difference in mutual information values ​​between adjacent time points, and the denominator is the sampling interval. For example, when... Bit, Bit, Hour( (minutes) Bits per hour; a negative value indicates that the degree of information coupling is weakening.

[0232] S513: The predicted carbon emission values ​​and feature vectors are used as the node features of each node, and the normalized dynamic mutual information values ​​are used as the weights of the edges between nodes. The spatiotemporal graph structure is constructed from the node features and the weights of the edges between nodes.

[0233] In this embodiment of the application, node features refer to vectors describing the attribute information of each node in the spatiotemporal graph structure; the inter-node association strength refers to the degree of information coupling between nodes based on dynamic mutual information value quantification; and the spatiotemporal graph structure refers to a graph data structure that simultaneously contains spatial topological relationships and temporal evolution relationships.

[0234] Specific implementation process: Using each road segment as a spatial node, and based on the predicted carbon emissions... and eigenvectors The concatenated vector serves as the node feature of each node. , Representing node features, with dimension 1 Dimension, among which Represents the dimension of the feature vector. Based on dynamic mutual information values. After Sigmoid normalization, the weights are used as prior weights for edges connecting adjacent road segment nodes. , This represents the prior weight of the edges between nodes. satisfy:

[0235] .

[0236] Based on node features and prior weights of edges between nodes Constructing a spacetime graph structure ,in The set of nodes representing the spatiotemporal graph structure; The set of edges representing the structure of a spacetime graph; This represents the edge weight matrix. It's important to note that edges in the spatiotemporal graph structure are only established between spatially adjacent road segments; the adjacency criterion is either the topological adjacency relationship of the road segments or a geographical distance of less than 500 meters. Here, the global dynamic mutual information value is used. After normalization, it serves as the shared prior weight for all edges, and its physical meaning is: when the inter-layer mapping is generally stable ( (Within the normal range), the baseline values ​​of the correlation strength between nodes are relatively high; when a global drift occurs in the mapping, the baseline values ​​of the correlation strength are adjusted accordingly. Preferably, to further characterize the differentiated correlations at the node pair level and avoid local anomalies being overwhelmed by global indicators, the mutual information between the local mapping state parameters of each node and the carbon emission residual of that node is calculated as the edge-specific weight, in which case the edge weights are prior. The mutual information between the node's local mapping parameters and the node's residual is calculated to satisfy... ,in Indicates based on road segment and road section The local mapping state parameters and the corresponding residuals are used to calculate the node-pair level dynamic mutual information values. In the prior calculation formula for edge weights, the Sigmoid function monotonically maps non-negative mutual information values ​​to... to The interval, for example when Bit time, The more stable the mapping relationship, the higher the prior edge weight.

[0237] S514: Based on the preset graph attention mechanism, the node features are aggregated and updated using the correlation strength between nodes as the prior weight of the attention coefficient to obtain the updated node embedding.

[0238] In this embodiment, the graph attention mechanism refers to a graph neural network mechanism that achieves adaptive aggregation of neighbor features by calculating the attention coefficients between nodes; the attention coefficient refers to the normalized weight that quantifies the importance of neighbor nodes to the central node; and the node embedding refers to the low-dimensional vector representation of the node after aggregation and updating by the graph attention mechanism.

[0239] Specific implementation process: For the spatiotemporal graph structure For each node in the array, read its node characteristics. and the node characteristics of neighboring nodes, based on the strength of the association between nodes. As prior weights for the attention coefficient, the attention coefficient is calculated using a pre-defined graph attention mechanism. , satisfy:

[0240] .

[0241] in, Indicates the attention coefficient; This represents the attention weight vector, with dimension . Dimension, among which Indicates the intermediate feature dimension; Represents a learnable linear transformation matrix with dimension . ; Indicates the characteristics of the central node; Represent the characteristics of neighboring nodes; Indicates the first Characteristics of each neighboring node; Represents the set of neighboring nodes of the central node; `<vector>` represents the vector concatenation operation; `LeakyReLU` represents the linear unit activation function with leakage correction. The attention coefficients are used as the basis for this operation. The updated node embedding is obtained by weighted aggregation of the features of neighboring nodes. , satisfy:

[0242] .

[0243] in, This represents the updated node embedding, with dimension . dimension; This represents a non-linear activation function. It's worth noting that the preset graph attention mechanism employs a multi-head attention structure, with the number of attention heads... The value range is 2 to 8 (2≤ ≤8). In the formula for calculating the attention coefficient, and The linear projection of the features of the center node and its neighboring nodes is concatenated and then processed by the attention weight vector. The correlation score is calculated using LeakyReLU. Summing the denominator with the summation of the neighboring fractions achieves normalization among the neighbors. The scores are modulated using prior weights, while the node embedding update uses attention coefficients to weighted sums of neighbor features and outputs after non-linear activation. For example, if a node has two neighbors, their normalized attention coefficients are... and If the node embedding is mainly contributed by the features of its more strongly associated neighbors, then its node embedding is primarily contributed by the features of its neighbors.

[0244] S515: Embed the node into the input fully connected layer and map it to the initial anomaly score.

[0245] In this embodiment, a fully connected layer refers to a linear transformation layer in a neural network where each input node is connected to each output node; the initial anomaly score refers to the quantified value of the anomaly degree of the rate of change of the unfused dynamic mutual information value obtained after mapping through the fully connected layer.

[0246] Specific implementation process: Embed the updated node Input fully connected layer, set the number of hidden layers of the fully connected layer. Floors 1 to 3 (1≤ ≤3), number of neurons per layer 32 to 128 (32 ≤ ≤128), through linear transformation of the fully connected layer and mapping with the activation function, the initial anomaly score is obtained. , satisfy: .

[0247] in, Indicates the initial anomaly score; This represents the weight matrix of the first fully connected layer; This represents the weight matrix of the second fully connected layer; This represents the bias vector of the first fully connected layer; represents the bias vector of the second fully connected layer; ReLU represents the modified linear unit activation function. σ represents the sigmoid activation function. It is worth noting the initial anomaly score. The value range is from 0 to 1 (0 ≤ The value ≤1 indicates a higher degree of anomaly; the closer to 1, the higher the anomaly score. Specifically, the output layer of the fully connected layer uses a Sigmoid activation function to map the linear output, constraining the initial anomaly score to ≤1. to Within the interval. In the initial anomaly score calculation formula, the first fully connected layer performs a linear transformation on the node embedding and then... Introducing nonlinearity, the second fully connected layer compresses the hidden layer representation into a scalar score, and the output sigmoid layer constrains the score to... to For example, when the node embedding is a linear output after two layers of transformation, hour, .

[0248] S516: Using the rate of change of dynamic mutual information value as input, calculate the fusion weight through a preset weight mapping function; the fusion weight is positively correlated with the rate of change of dynamic mutual information value; the initial anomaly score and the dynamic mutual information value are weighted and fused using the fusion weight to obtain the anomaly score.

[0249] In this embodiment, the fusion weight refers to the adaptive weight coefficient used to fuse the initial anomaly score and the dynamic mutual information value; the weight mapping function refers to the monotonically increasing function that maps the rate of change of the dynamic mutual information value to the fusion weight.

[0250] Specific implementation process: using the rate of change of dynamic mutual information values As input, through a pre-defined weight mapping function Calculate fusion weights , satisfy:

[0251] .

[0252] in, Indicates the fusion weights; This represents the preset weight mapping function; This represents the slope coefficient of the weight mapping function. The value range is 1 to 5 (1≤ ≤5); This represents the offset of the weight mapping function, in bits per hour (bit / h). The value range is from -0.5 bits per hour to 0.5 bits per hour (-0.5 bits / h ≤ ≤0.5bit / h); The reference base value represents the rate of change of dynamic mutual information value, in bits per hour (bit / h), and its value is based on historical normal operating conditions. Sample standard deviation Determined, that is This makes the normalized input a dimensionless standard score. Sample standard deviation. Calculated based on the historical dynamic mutual information value change rate sequence under normal operating conditions with a time window of equal length to the preset time window. In the weight mapping function, The deviation of the rate of change relative to the offset, divided by Normalize it to a standard score. By controlling the steepness of the mapping curve, the Sigmoid maps the result to... to For example, when , , Bits per hour Bits per hour At this point, the anomaly score is primarily determined by the inverse normalization term of the dynamic mutual information value. (Using fusion weights...) Initial anomaly scoring With dynamic mutual information value Weighted fusion is performed to obtain anomaly scores. , satisfy:

[0253] .

[0254] in, Indicates abnormal rating; This represents the minimum value of dynamic mutual information within a preset time window, in bits. This represents the maximum value of the dynamic mutual information value within a preset time window, in bits. =0.01bit is the numerical stability constant. The anomaly score fusion formula is a convex combination of the initial anomaly score and the inversely normalized mutual information value, and the sum of the two coefficients is For example, when , , Bit, Bit, When the bit is inverted, the normalized value is , greater than the preset scoring threshold to The lower limit, and Bits below preset fluctuation threshold A bit satisfies both criteria and triggers an anomaly warning.

[0255] It is worth noting that when the rate of change of the dynamic mutual information value When the value is negative and the absolute value is large, the fusion weight Approaching 0, anomaly score More dependent on the inverse normalized value of the dynamic mutual information value, in this case, the lower the dynamic mutual information value, the closer the inverse normalized value is to the value of the inverse normalized value. The higher the anomaly score, the more positively correlated the anomaly score is with the degree of mapping drift; when When it is positive, the fusion weight Approaching 1, anomaly score More dependent on initial anomaly score When the dynamic mutual information value Below the preset fluctuation threshold And abnormal scoring Greater than the preset scoring threshold At that time, based on the mapping state parameters, it is determined that the inter-layer mapping relationship has drifted and an anomaly warning signal is generated, in which a preset fluctuation threshold is set. The value range is from 0.1 bit to 0.5 bit (0.1 bit ≤ 0.5 bit). ≤0.5bit), preset scoring threshold The value range is from 0.7 to 0.9 (0.7 ≤ ≤0.9).

[0256] Understandably, by constructing an anomaly scoring network and integrating the rate of change of dynamic mutual information values, the technical problem that a single indicator cannot simultaneously capture carbon emission anomalies and inter-layer mapping drift is solved. This enables joint detection of carbon emission prediction anomalies and instability in inter-layer mapping relationships. When the dynamic mutual information value is lower than the preset fluctuation threshold and the anomaly score is greater than the preset scoring threshold, it can promptly determine that the inter-layer mapping relationship has drifted and generate an anomaly warning signal, providing decision support for the operation, maintenance, and carbon emission control of digital twins of transportation infrastructure.

[0257] The embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the technical solutions and core ideas of this application. Those skilled in the art should understand that they can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. These modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for monitoring, predicting, and providing early warning of carbon emissions from transportation infrastructure driven by a combination of big data and artificial intelligence, characterized in that... include: Acquire multi-source heterogeneous data on transportation infrastructure; A pre-constructed digital twin is updated based on multi-source heterogeneous data. The digital twin includes a physical layer, an energy layer, and a carbon flow layer. The updated physical layer traffic flow state is mapped to the energy layer load by inter-layer coupling operators, establishing inter-layer mapping relationships and generating mapping state parameters; a heterogeneous graph convolutional network is used to jointly encode the updated digital twin to obtain feature vectors. Based on the updated physical layer traffic flow status and energy layer load, the carbon emission baseline value is calculated through a physical model; based on historical data and feature vectors in multi-source heterogeneous data, the carbon emission residual is predicted through a residual prediction model to obtain the residual prediction value; the residual prediction value is superimposed with the baseline value to obtain the carbon emission prediction value. Based on time-series samples of carbon emission predictions and carbon emission baselines within a preset time window, a dynamic mutual information value is calculated; the carbon emission predictions are then input into an anomaly scoring network to obtain an anomaly score. When the dynamic mutual information value is lower than the preset fluctuation threshold and the abnormal score is greater than the preset score threshold, the inter-layer mapping relationship is determined to have drifted based on the mapping state parameter and an abnormal warning signal is generated. Calculating baseline carbon emissions using physical models includes: Obtain the updated physical layer traffic flow status and energy layer load; the physical layer traffic flow status includes the traffic volume and average speed of each vehicle type on each road segment, and the energy layer load includes the equipment energy consumption of each road segment's energy equipment; Based on the preset unit mileage carbon emission factor and average vehicle speed of each vehicle type, calculate the instantaneous carbon emission rate of each vehicle type; multiply the instantaneous carbon emission rate of each vehicle type by the number of vehicles in the corresponding road segment to obtain the mobile source carbon emission component. Based on the preset energy carbon emission factor and equipment energy consumption of each energy device, calculate the direct carbon emission of the equipment; sum up the direct carbon emission of each device to obtain the carbon emission component from the stationary source. The coupling overlap area between mobile and fixed sources at the road segment level is determined based on the mapping state parameters, and the carbon emission superposition coefficient within the coupling overlap area is corrected by the mapping relationship evolution in the mapping state parameters; the preset benchmark superposition coefficient is retained for road segments in non-coupling overlap areas. The carbon emission components of mobile sources and stationary sources in the coupled overlapping region are weighted and summed using the corrected carbon emission superposition coefficient, and the components in the non-coupled overlapping region are weighted and summed using the preset benchmark superposition coefficient. The carbon emission benchmark value is obtained by accumulating these values. Calculating dynamic mutual information values ​​includes: Obtain the predicted carbon emissions within a preset time window and generate a time series of the predicted carbon emissions; obtain the baseline carbon emissions within the same preset time window and generate a time series of the baseline carbon emissions. The time series of carbon emission predictions is calculated to obtain the time series of carbon emission residuals by calculating the point-by-point difference between the time series of carbon emission predictions and the time series of carbon emission baselines. Obtain the time-series update sequence of the mapped state parameters within a preset time window; Using a pre-defined kernel density estimation method, the probability distribution of the time-series update sequence of the mapped state parameters is estimated to obtain the first probability distribution; the probability distribution of the time-series sequence of carbon emission residuals is estimated to obtain the second probability distribution; and the joint probability distribution of the time-series update sequence of the mapped state parameters and the time-series sequence of carbon emission residuals is estimated. Based on the first probability distribution, the second probability distribution, and the joint probability distribution, calculate the dynamic mutual information value between the time-series update sequence of the mapped state parameters and the time-series sequence of the carbon emission residual.

2. The method according to claim 1, characterized in that, The initial construction of a digital twin includes: Acquire initial multi-source heterogeneous data, which includes static topology parameters, equipment parameters, and emission factor baseline values. Static topology parameters include road network topology information, equipment parameters include energy equipment information, and emission factor baseline values ​​include emission factor information. Spatiotemporal alignment and standardization are performed on the initial multi-source heterogeneous data to obtain standardized data; Based on the road network topology information in the standardized data, the topology structure of the physical layer model is established, and the traffic flow dynamic simulation results are generated by using the topology structure as input and a preset traffic flow simulation model. Based on energy equipment information and traffic flow simulation results from standardized data, the equipment connection relationship of the energy layer is established. Using this connection relationship and traffic flow simulation results as input, the energy load distribution is calculated through a preset energy system simulation model. Based on emission factor information and energy load distribution in standardized data, emission nodes of the carbon flow stratum model are established, and carbon emission flow direction is calculated by using emission nodes and energy load distribution as inputs and a preset carbon emission tracking algorithm. The dynamic simulation results of traffic flow, energy load distribution, and carbon emission flow are coupled to generate a digital twin and mapped state parameters.

3. The method according to claim 2, characterized in that, The carbon emission flow is calculated using a pre-defined carbon emission tracking algorithm, including: Obtain the energy load distribution and emission factor baseline values ​​for each emission node, and classify each emission node into a supply side or a consumption side based on the equipment type in the energy equipment information and the energy load distribution. Using each emission node as a vertex, the energy supply relationship between the equipment is determined based on the energy equipment information. The real-time energy supply in each energy supply relationship is determined based on the energy load distribution. When the supply exceeds the preset supply threshold, a directed edge is established from the supply end to the consumption end. After traversing all energy supply relationships, a carbon flow directed graph is generated. Based on the total energy load of each supply side and its consumption side load, the ratio of consumption side load to total supply side load is used as the load allocation coefficient, and the product of the total energy load of the supply side and the emission factor benchmark value is used as the direct carbon emissions. The load allocation coefficient is used as the weight of the directed edge, and the direct carbon emissions are used as the starting amount. The load is then propagated from the supply side to the consumption side node by node along the carbon flow directed graph. At each node, the carbon emissions from all upstream nodes are accumulated, and then summed with the direct carbon emissions when that node acts as a supplier to obtain the cumulative carbon emissions. The carbon emission flow distribution is generated based on the cumulative carbon emissions of each node and the weight distribution of the directed edges.

4. The method according to claim 1, characterized in that, The updated physical layer traffic flow state is mapped to the energy layer load using inter-layer coupling operators. This establishes the inter-layer mapping relationship and generates mapping state parameters, including: Obtain the updated physical layer traffic flow status; the physical layer traffic flow status includes the traffic volume, average vehicle speed, and vehicle type composition ratio of each road segment; Based on the preset unit mileage energy consumption coefficient, average vehicle speed, and preset speed grading correction coefficient for each vehicle model, calculate the instantaneous energy consumption power per vehicle; the instantaneous energy consumption power per vehicle is the product of the coefficient, average vehicle speed, and correction coefficient. Based on the vehicle type composition ratio and traffic flow, determine the number of vehicles of each type in each road segment. Multiply the instantaneous energy consumption power of each vehicle type by the corresponding number of vehicles and then sum them up to obtain the traffic flow energy consumption power of that road segment. Based on the inter-layer coupling operator, the energy consumption power of traffic flow is mapped to the energy layer load; the inter-layer coupling operator contains mapping weights that are dynamically adjusted according to the traffic flow state; The energy layer load is associated with and stored in relation to the traffic flow and average vehicle speed of the corresponding road segment, generating mapping state parameters; the mapping state parameters include load-flow mapping pairs and load-speed mapping pairs.

5. The method according to claim 4, characterized in that, Dynamic adjustment of mapping weights includes: Obtain the traffic flow and average speed of each road segment during the current time period; Using traffic flow and average vehicle speed as independent variables, a preset scenario-based mapping function is invoked to calculate the mapping weight for the current time period; the scenario-based mapping function is obtained by fitting historical traffic flow status and corresponding energy layer load data. The mapping weights in the interlayer coupling operator are updated with the mapping weights of the current time period, and the traffic flow energy consumption power is mapped to the dynamically adjusted energy layer load according to the updated mapping weights. The difference between the mapping weight of the current time period and the mapping weight of the previous time period is calculated to obtain the weight change, and the weight change is associated with and stored with the mapping state parameters.

6. The method according to claim 1, characterized in that, The coupling overlap region between the mobile source and the fixed source at the road segment level, determined based on the mapping state parameters, includes: Retrieve the load-flow mapping pairs from the mapping status parameters; The energy consumption per unit flow of each road segment is determined based on the ratio of load value to flow value in the load-flow mapping pair. Divide the carbon emission components from mobile sources and stationary sources by the preset road segment length to obtain the carbon emission per unit length from mobile sources and the carbon emission per unit length from stationary sources for each road segment. Each road segment is divided into grids with a preset spatiotemporal resolution. When the average energy consumption of a single vehicle in the same grid is greater than the preset energy consumption coupling threshold, and the sum of the carbon emissions per unit length of the mobile source and the carbon emissions per unit length of the fixed source is greater than the preset carbon emission coupling threshold, the grid is determined to be a coupling overlap area. The grid coordinates of the coupled overlapping area are associated with the corresponding road segment identifier and stored to generate the coupled overlapping area identifier.

7. The method according to claim 1, characterized in that, The carbon emission predictions are input into the anomaly scoring network to obtain anomaly scores, including: Obtain carbon emission predictions, feature vectors, and dynamic mutual information values; Obtain the time series of dynamic mutual information values ​​within a preset time window, calculate the difference of dynamic mutual information values ​​between adjacent time points according to a preset difference rule, and obtain the rate of change of dynamic mutual information values. The predicted carbon emissions and feature vectors are used as the node features of each node, and the normalized dynamic mutual information value is used as the weight prior of the edge between nodes. The spatiotemporal graph structure is constructed from the node features and the weight of the edge between nodes. Based on the preset graph attention mechanism, the node features are aggregated and updated using the strength of the association between nodes as the prior weight of the attention coefficient, and the updated node embedding is obtained. Embed the nodes into the input fully connected layer and map them to the initial anomaly scores; Using the rate of change of dynamic mutual information value as input, the fusion weight is calculated through a preset weight mapping function; the fusion weight is positively correlated with the rate of change of dynamic mutual information value; the initial anomaly score and the dynamic mutual information value are weighted and fused using the fusion weight to obtain the anomaly score.

Images