An urban electric vehicle time-sharing charging implicit carbon emission accounting method

Abstract

The present application relates to the technical field of machine learning, in particular to a kind of city electric vehicle time-sharing charging implicit carbon emission accounting method.The meteorological characteristics are fused with map fixed-point coordinate alignment, node basis vector is generated;Based on the cross-zone contact electric quantity and network frame power flow state, directed acyclic network is constructed, node basis vector is mapped and tracked along the graph topology path, and carbon intensity base is output;Combining node load parameter with external environment parameter, obtain network loss compensation value, superimposed to carbon intensity base and verified by regional carbon emission allocation, generate time-sharing carbon emission matrix;Time-sharing carbon emission matrix is packaged with multi-element power output parameter as source network carbon coupling tensor and dimension reduction aggregation, and the basic carbon emission equivalent parameter and power source correction electric quantity parameter are obtained by inversion, to generate carbon emission result.The present application constructs a coherent logical closed loop in three dimensions of time series, spatial grid and physical flow direction, avoids the phenomenon that measured value deviates from physical energy transfer process.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, specifically to a method for calculating the implicit carbon emissions of time-sharing charging of electric vehicles in cities. Background Technology

[0002] Accurately measuring the implicit carbon emissions of time-of-use charging of mobile power terminals (such as electric vehicles) in specific physical environments (such as complex urban power grids) is an important current technological application direction. Existing technologies generally adopt a measurement method that combines fixed emission parameters with physical metering of terminal electricity. That is, the electrical energy consumed by the terminal is directly collected through a physical interface and then simply multiplied and matched with the static carbon emission equivalent parameters preset by the regional power grid to estimate the scale of carbon emissions generated by the physical conversion of electrical energy. This is currently the mainstream technical means to achieve the quantification of implicit carbon emissions from terminal electricity.

[0003] However, existing technical solutions have significant shortcomings in measurement accuracy and physical traceability continuity. The physical parameter models used to quantify carbon intensity suffer from staticity and coarse accuracy. Existing measurement methods are mostly based on fixed historical averages or static topological state delineation. Their underlying parameters cannot be dynamically adjusted with real-time changes in the external natural environment, making it difficult to accurately match specific physical links requiring high-frequency state monitoring (such as the dynamic impedance and thermal dissipation of transmission lines under temperature and light fluctuations). This leads to deviations in the measurement of physical energy losses from source to end. Furthermore, "snapshot" numerical calculations are performed only at the instant of physical energy transfer, failing to achieve continuous tracking of the carbon emission material flow transfer process. This means they cannot measure the sudden changes in local carbon intensity caused by cross-regional physical energy dispatch, nor can they accurately track the specific spatial topological path of energy transmission within complex physical grid structures. The output of measurement results suffers from a lack of diversity. Terminals connected to the grid fail to adapt parameters in multiple dimensions according to the dynamic proportions of multiple energy sources or the dynamic distribution attributes of the environment in the energy physical production process. This results in carbon emission measurement values ​​deviating from the physical energy transfer process and failing to reflect the real environmental physical load of different terminals at specific spatiotemporal nodes.

[0004] To address this, a method for calculating the implicit carbon emissions from time-sharing charging of urban electric vehicles is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide a method for calculating the implicit carbon emissions of time-sharing charging of urban electric vehicles, so as to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for calculating the implicit carbon emissions of time-sharing charging of urban electric vehicles, comprising: The system acquires remote sensing images, map coordinates, inter-regional power transmission, grid power flow status, unit load rate, regional carbon emission quota, multi-source power output parameters, node load parameters including transient power values ​​and state of charge percentage, and external environmental parameters including external temperature and external wind speed values, as well as sampling step size, reference thermal time constant, and power source carbon emission value. Using a time synchronization mechanism, the system aligns and fuses the meteorological features extracted from the remote sensing images and the map coordinates to establish spatial relationships and generate node base vectors. A directed acyclic network is constructed using cross-regional interconnection power and grid power flow status, and weight parameters are configured. The node base vector is mapped to the terminal graph node of the directed acyclic network. The carbon emission value at the power source is used as the source emission factor of the root node at the source. According to the weight parameters, the tracking calculation is performed along the graph topology path to output the carbon intensity base. The equivalent charging power is obtained by applying a nonlinear charging attenuation weight to the transient power value based on the percentage of state of charge; the original temperature difference fluctuation parameter is calculated based on the external temperature values ​​of adjacent time slices, and a first-order low-pass filter is performed based on the sampling step size and the reference thermal time constant to obtain the equivalent temperature difference fluctuation parameter. The equivalent temperature difference fluctuation parameter, equivalent charging power, external wind speed value, and state of charge percentage are imported into the network loss assessment model to output the heat dissipation parameter. The network loss conversion parameter is determined based on the heat dissipation parameter. The network loss conversion parameter is multiplied by the carbon intensity base to obtain the network loss compensation value. The network loss compensation value is superimposed and merged into the carbon intensity base to obtain the corrected carbon intensity parameter. The corrected carbon intensity parameters are arranged according to the time series to generate a time-series carbon emission matrix. The time-sharing carbon emission matrix and the output parameters of multiple power sources are encapsulated into a source-grid carbon coupling tensor. The source-grid carbon coupling tensor is subjected to dimension reduction and aggregation processing to obtain a dimension reduction and aggregation vector. Feature inversion mapping is performed on the dimension reduction and aggregation vector to obtain the basic carbon emission equivalent parameter and the power source correction power parameter, thus generating the carbon emission result.

[0007] Preferably, the processing of remote sensing images, map coordinates, inter-regional power transmission, grid power flow status, regional carbon emission quotas, multi-source power output parameters, node load parameters, and external environmental parameters includes: capturing remote sensing images and map coordinates containing spatial positioning parameters by docking with remote sensing mapping nodes; collecting inter-regional power transmission and grid power flow status through remote terminals, setting the grid power flow status to include grid topology matrix and line power parameters, with the grid topology matrix including on / off status markers; retrieving regional carbon emission quotas including maximum threshold parameters and warning threshold parameters; extracting multi-source power output parameters including instantaneous output ratio values; recording node load parameters including transient power values ​​and state of charge percentages through load nodes; capturing external environmental parameters through meteorological probes, determining that external environmental parameters include external wind speed values ​​and external temperature values; and interpolating the unit load rate in a preset multi-dimensional curve of carbon emission intensity characteristics to obtain the carbon emission value at the power source.

[0008] Preferably, the processing of the node base vector includes: performing multispectral separation on the remote sensing image to extract surface temperature and light intensity parameters, and constructing meteorological features; configuring sliding time window slices, and extracting timestamp labels and spatial positioning parameters attached to the meteorological features based on a time synchronization mechanism; constructing a two-dimensional grid array according to the spatial positioning parameters, projecting the meteorological features into the corresponding coordinate intervals contained in the two-dimensional grid array, performing alignment and fusion operations, establishing spatial associations, and outputting cross-dimensional attribute values; assembling the cross-dimensional attribute values ​​into a feature set matrix; performing normalization and scaling calculations on the feature set matrix to smooth the parameter value distribution intervals, and merging to generate the node base vector.

[0009] Preferably, the processing procedure for the directed acyclic network includes: parsing the network topology matrix contained in the network power flow state using a graph computing engine; performing a loop-breaking operation on the network topology matrix by calling a power flow ratio sharing algorithm to establish a unidirectional power flow state graph; decomposing the unidirectional power flow state graph into initial root nodes and intermediate transition nodes, setting terminal graph nodes in combination with spatial positioning parameters, and establishing an initial network architecture; parsing the line power parameters contained in the network power flow state, configuring the weight parameters of the edges of the directed acyclic network graph according to the line power parameters; injecting the cross-regional interconnection power as a constraint condition into the initial network architecture, and simultaneously projecting the node base vectors into the terminal graph nodes for feature mapping according to the spatial positioning parameters and weight parameters, and outputting the directed acyclic network.

[0010] Preferably, the carbon intensity baseline processing includes: calling a breadth-first search algorithm to traverse the directed acyclic network and identify graph topology paths pointing to the terminal graph nodes; extracting the on / off state markers contained in the network topology matrix, performing connectivity verification based on the on / off state markers, eliminating graph topology paths in the off state, and retaining valid graph topology paths; using the carbon emission value at the power source as the source emission factor of the root node, and based on the weight parameters of the edges in the directed acyclic network graph, performing state transition tracking calculations along the valid graph topology paths from the root node to the terminal graph nodes to obtain the carbon intensity baseline of the terminal graph nodes.

[0011] Preferably, the processing of the network loss compensation value includes: calculating the power mean and power variance of the transient power value within the sliding time window; inputting the equivalent temperature difference fluctuation parameter, equivalent charging power, external wind speed value, state of charge percentage, power mean and power variance into the network loss assessment model to calculate the heat dissipation parameter; performing quadratic polynomial fitting processing on the heat dissipation parameter to output the network loss conversion parameter; extracting the carbon intensity base, performing floating-point multiplication operation on the network loss conversion parameter and the carbon intensity base to obtain the network loss compensation value.

[0012] Preferably, the processing of the time-sharing carbon emission matrix includes: performing a scalar addition operation on the network loss compensation value and the carbon intensity base to obtain the corrected carbon intensity parameter; extracting the highest threshold parameter and the warning threshold parameter included in the regional carbon emission quota; comparing the corrected carbon intensity parameter and the warning threshold parameter, and activating a preset over-limit marking mechanism when the corrected carbon intensity parameter exceeds the warning threshold parameter but is lower than the highest threshold parameter, and outputting the over-limit status feature bit corresponding to the corrected carbon intensity parameter; outputting an extreme value alarm parameter when the corrected carbon intensity parameter exceeds the highest threshold parameter; and arranging the corrected carbon intensity parameters corresponding to each time slice into a matrix form according to the time sequence to generate the time-sharing carbon emission matrix.

[0013] Preferably, the processing of the dimension reduction and aggregation vector includes: extracting the instantaneous output ratio values ​​contained in the multi-source power output parameters; setting the instantaneous output ratio values ​​as an additional feature dimension, concatenating the additional feature dimension into the time-sharing carbon emission matrix, and constructing a source-grid carbon coupling tensor; defining a calculation time window, starting a tensor orthogonal decomposition algorithm, extracting key principal component parameters, generating an approximate reconstruction tensor, subtracting the source-grid carbon coupling tensor from the approximate reconstruction tensor element by element, generating and caching an orthogonal residual matrix; and aggregating all key principal component parameters to generate a dimension reduction and aggregation vector.

[0014] Preferably, the carbon emission result processing includes: retrieving the cached orthogonal residual matrix; performing inverse decoding on the dimensionality-reduced aggregated vector and the orthogonal residual matrix, performing decoupling and splitting processing based on preset built-in transformation rules to obtain the carbon emission intensity benchmark and the power correction intensity benchmark, performing time integral conversion in combination with the transient power values ​​contained in the node load parameters to restore the basic carbon emission equivalent parameter and the power correction energy parameter; retrieving the preset unit energy emission reduction factor for the power correction energy parameter; the unit energy emission reduction factor contains emission reduction offset attributes matching the instantaneous output ratio value; and adding the product of the basic carbon emission equivalent parameter and the unit energy emission reduction factor and the power correction energy parameter to generate the carbon emission result.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. By acquiring remote sensing images, map coordinates, inter-regional power transmission, grid power flow status, regional carbon emission quotas, multi-source power output parameters, node load parameters, and external environmental parameters, a multi-dimensional physical sensing foundation was constructed. In specific implementation, a time synchronization mechanism was used to align and fuse meteorological features from remote sensing image analysis with map coordinates, establishing spatial correlation and generating node base vectors. A spatiotemporal mapping relationship was established between environmental conditions and terminal nodes, transforming complex environmental physical characteristics into underlying spatial grid parameters. This enabled multi-dimensional parameter adaptation for the dynamic distribution attributes of different geographical locations, thereby objectively characterizing the real environmental physical load of different terminals at specific spatiotemporal nodes and providing a high-resolution spatiotemporal data measurement benchmark.

[0016] 2. Based on the graph computing engine, a directed acyclic network (DAN) is constructed using cross-regional interconnection power and grid power flow status. The node base vectors are mapped to the corresponding graph nodes of the DAN, the graph topology path of the DAN is parsed, and tracing calculations are performed along the graph topology path to output the carbon intensity base. Subsequently, node load parameters and external environmental parameters are imported into the network loss assessment model to obtain network loss compensation values. The network loss compensation values ​​are superimposed and fused into the carbon intensity base, which conforms to the physical flow chain of power from the source to the end. It quantifies the specific topology and dynamic impedance fluctuations when energy is transmitted within the complex grid, avoiding the phenomenon that the measured values ​​deviate from the physical energy transfer process.

[0017] 3. By combining regional carbon emission quotas with carbon emission limit verification, a time-sharing carbon emission matrix is ​​generated. Further, the time-sharing carbon emission matrix and multi-source power output parameters are encapsulated into a source-grid carbon coupling tensor. Dimensionality reduction and aggregation are performed on the source-grid carbon coupling tensor to obtain a dimension-reduced aggregation vector. Finally, feature inversion mapping is performed on the dimension-reduced aggregation vector to obtain the basic carbon emission equivalent parameter and the power source correction power parameter. By generating carbon emission results, the associated resource physical consumption characteristics deeply related to energy physical production processes are effectively extracted and their dimensions transformed. This effectively processes heterogeneous resource consumption data, ensuring that the final output carbon emission measurement values ​​cover the multi-source collaborative emission reduction patterns and reflect the real environmental physical load borne by the terminal when connected to the grid.

[0018] 4. The system organically integrates the node base vectors generated by establishing spatial associations, the carbon intensity base calculated by tracing along the graph topology path, and the source-grid carbon coupling tensor constructed by encapsulating time-sharing carbon emission matrices and multi-source power output parameters. This breaks down the isolation between cross-domain physical data and constructs a coherent logical closed loop across three dimensions: time series, spatial grid, and physical flow direction. By unifying the dynamic meteorological environment characteristics at the front end, the micro-link transformation and attenuation in the middle, and the associated resource physical consumption at the end into a computational framework of dimensionality reduction aggregation and feature inversion mapping, the system achieves full-chain collaboration from natural environment perception to material flow equivalent analysis. This effectively smooths out the local data fragmentation caused by single-dimensional measurement and can objectively depict the energy transfer patterns under the interweaving of complex physical factors from a global perspective, thereby more rigorously restoring the real environmental physical loads of different terminals at specific spatiotemporal nodes. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a method for calculating the implicit carbon emissions from time-sharing charging of urban electric vehicles, as proposed in an embodiment of this invention. Figure 2 This is a flowchart of the node basic vector generation and spatial alignment mapping process proposed in an embodiment of this invention application; Figure 3 This is a flowchart illustrating the directed acyclic graph topology construction and network loss tracking and correction process proposed in an embodiment of this invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figures 1-3The present invention provides a method for calculating the implicit carbon emissions of time-sharing charging of urban electric vehicles, the specific steps of which are as follows: The system acquires remote sensing images, map coordinates, inter-regional power transmission, grid power flow status, unit load rate, regional carbon emission quota, multi-source power output parameters, node load parameters including transient power values ​​and state of charge percentage, and external environmental parameters including external temperature and external wind speed values, as well as sampling step size, reference thermal time constant, and power source carbon emission value. Using a time synchronization mechanism, the system aligns and fuses the meteorological features extracted from the remote sensing images and the map coordinates to establish spatial relationships and generate node base vectors. A directed acyclic network is constructed using cross-regional interconnection power and grid power flow status, and weight parameters are configured. The node base vector is mapped to the terminal graph node of the directed acyclic network. The carbon emission value at the power source is used as the source emission factor of the root node at the source. According to the weight parameters, the tracking calculation is performed along the graph topology path to output the carbon intensity base. The equivalent charging power is obtained by applying a nonlinear charging attenuation weight to the transient power value based on the percentage of state of charge; the original temperature difference fluctuation parameter is calculated based on the external temperature values ​​of adjacent time slices, and a first-order low-pass filter is performed based on the sampling step size and the reference thermal time constant to obtain the equivalent temperature difference fluctuation parameter. The equivalent temperature difference fluctuation parameter, equivalent charging power, external wind speed value, and state of charge percentage are imported into the network loss assessment model to output the heat dissipation parameter. The network loss conversion parameter is determined based on the heat dissipation parameter. The network loss conversion parameter is multiplied by the carbon intensity base to obtain the network loss compensation value. The network loss compensation value is superimposed and merged into the carbon intensity base to obtain the corrected carbon intensity parameter. The corrected carbon intensity parameters are arranged according to the time series to generate a time-series carbon emission matrix. The time-sharing carbon emission matrix and the output parameters of multiple power sources are encapsulated into a source-grid carbon coupling tensor. The source-grid carbon coupling tensor is subjected to dimension reduction and aggregation processing to obtain a dimension reduction and aggregation vector. Feature inversion mapping is performed on the dimension reduction and aggregation vector to obtain the basic carbon emission equivalent parameter and the power source correction power parameter, thus generating the carbon emission result.

[0022] The technical solution of the present invention will be further described in detail below with reference to specific embodiments.

[0023] Example 1 This application discloses a method for calculating the implicit carbon emissions from time-sharing charging of urban electric vehicles, see below. Figure 1The specific steps proposed in this invention include: S1. Acquiring remote sensing images, map coordinates, inter-regional interconnection power, grid power flow status, unit load rate, regional carbon emission quota, multi-source power output parameters, node load parameters including transient power values ​​and state of charge percentage, and external environmental parameters including external temperature and external wind speed values, as well as sampling step size, reference thermal time constant, and power source carbon emission value; using a time synchronization mechanism, aligning and fusing the meteorological features and map coordinates from the remote sensing images to establish spatial association and generate node base vectors; S2. Constructing a directed acyclic network using inter-regional interconnection power and grid power flow status and configuring weight parameters, mapping the node base vectors to the terminal graph nodes of the directed acyclic network, using the power source carbon emission value as the source emission factor of the root node, performing tracking calculations along the graph topology path according to the weight parameters, and outputting the carbon intensity base; S3. Applying non-linearity to the transient power values ​​according to the state of charge percentage. Linear charging attenuation weights are used to obtain the equivalent charging power. The original temperature difference fluctuation parameter is calculated based on the external temperature values ​​of adjacent time slices. A first-order low-pass filter is performed based on the sampling step size and the reference thermal time constant to obtain the equivalent temperature difference fluctuation parameter. The equivalent temperature difference fluctuation parameter, equivalent charging power, external wind speed value, and state of charge percentage are imported into the network loss assessment model to output the heat dissipation parameter. The network loss conversion parameter is determined based on the heat dissipation parameter. The network loss conversion parameter is multiplied by the carbon intensity base to obtain the network loss compensation value. The network loss compensation value is superimposed and fused into the carbon intensity base to obtain the corrected carbon intensity parameter. The corrected carbon intensity parameters are arranged according to the time series to generate a time-series carbon emission matrix. S4: The time-series carbon emission matrix and the multi-source power output parameters are encapsulated into a source-grid carbon coupling tensor. Dimensionality reduction and aggregation processing is performed on the source-grid carbon coupling tensor to obtain a dimension-reduced aggregation vector. Feature inversion mapping is performed on the dimension-reduced aggregation vector to obtain the basic carbon emission equivalent parameter and the power source corrected energy parameter, generating the carbon emission result.

[0024] Further, acquire remote sensing images, map coordinates, inter-regional power transmission, grid power flow status, unit load rate, regional carbon emission quota, multi-source power output parameters, node load parameters including transient power values ​​and state of charge percentage, and external environmental parameters including external temperature and external wind speed values, as well as sampling step size, reference thermal time constant, and power source carbon emission value; using a time synchronization mechanism, perform alignment and fusion of meteorological features parsed from remote sensing images and map coordinates to establish spatial correlation and generate node base vectors; corresponding to step S1 above; see [link to relevant documentation]. Figure 2 The specific implementation process includes: The system retrieves remote sensing images and map coordinates containing spatial positioning parameters from remote sensing mapping nodes; it collects cross-regional power transmission and grid power flow status via remote terminal equipment, defining the grid power flow status as including the grid topology matrix and line power parameters, with the grid topology matrix including on / off status markers; it retrieves regional carbon emission quotas including maximum threshold and warning threshold parameters; it retrieves multi-source power output parameters including instantaneous output ratios of photovoltaic, wind, and geothermal power; it records node load parameters including transient power values ​​and state of charge percentages via load nodes; it captures external environmental parameters via meteorological probes, determining that external environmental parameters include external wind speed and external temperature values; and it interpolates the unit load rate in a preset multi-dimensional carbon emission intensity characteristic curve to obtain the carbon emission value at the power source.

[0025] Specifically, the process of acquiring remote sensing images, map coordinates, inter-regional power transmission, grid power flow status, regional carbon emission quotas, multi-source power output parameters, nodal load parameters, and external environmental parameters is as follows: Core basic parameters are captured within a fixed polling period. The polling period parameter is selected based on the data refresh frame rate of the power dispatch automation system and is set to 5 seconds. The data source and data format of each basic parameter have been calibrated. The remote sensing image serves as the data distribution interface for the high-altitude remote sensing mapping platform. It sends data requests with timestamps and regional identifiers, acquires image data streams encapsulated in multispectral matrix format, extracts high-resolution data including thermal infrared and visible light bands, and stores it in a file. The resolution of the remote sensing image is selected based on the data granularity standard of urban surface microclimate monitoring, setting it to 10 meters as the static spatial base. Simultaneously, it connects to the urban meteorological micro-station network to acquire 15-minute high-frequency sampling time series from ground micro-station sensors. Using the micro-station high-frequency series as a benchmark, Kalman filtering temporal interpolation updates are performed on the 10-meter-level remote sensing grid data.

[0026] The map's fixed-point coordinates include three-dimensional floating-point vector parameters representing longitude, latitude, and elevation. The selection of these three-dimensional floating-point vector parameters is achieved by calling a geographic information system interface, querying a database, reading the original geographic coordinate system messages of the fixed charging station mapping nodes, and extracting the values ​​through data format parsing.

[0027] The inter-regional communication power is used as a high-voltage side input constraint. The remote terminal at the high-voltage settlement gate of the substation is accessed to read the accumulated active power value of the smart meter. The selection method for the inter-regional communication power value is as follows: in the high-frequency calculation with a 5-second polling cycle, the instantaneous active power measured in real time by the remote terminal is used as a dynamic boundary constraint; while the absolute physical reading generated by the smart gate meter every 15 minutes of the settlement cycle is read only as the long-cycle energy integral adjustment calibration benchmark at the end of the time window to avoid high-frequency calculation drift.

[0028] The power flow status of the power grid is determined by extracting a data set containing the power grid topology matrix and line power parameters from the database of the power distribution management system. The power grid topology matrix embeds on / off status markers, and the values ​​of these markers are assigned based on the operating status signals fed back from the auxiliary contacts of the circuit breakers and disconnectors in the field: 0 for open status and 1 for closed status.

[0029] Regional carbon emission allowances are generated by connecting to the data platform and parsing the formatted annual allowance files. The maximum threshold parameter is selected by extracting the overall carbon emission reduction target limit of the regional power grid and multiplying it by a proportional coefficient of 1.0; the warning threshold parameter is selected by extracting the above limit value and multiplying it by a proportional coefficient of 0.8.

[0030] The multi-source power output parameters are obtained by parsing the power structure data packets pushed by the power dispatch center's data interface. The instantaneous output ratio values ​​are selected by extracting log records of the percentage of actual output power of various generator sets of photovoltaic, wind power and geothermal power to the total load power within a specific calculation period. For example, data such as 20% photovoltaic output, 15% wind power output and 5% geothermal output can be selected.

[0031] Node load parameters are obtained through continuous sampling of node load. Node load parameters include transient power values, charging mode identifiers (AC slow charging / DC fast charging), and state of charge percentage. The transient power values ​​are extracted based on real-time readings of the output active power periodically uploaded by the charging pile's built-in metering device. The charging mode identifier and state of charge percentage are extracted through the communication interface with the charging pile's built-in battery management system to characterize the unique physical charging conditions at the end of the electric vehicle.

[0032] The external environmental parameters are the received temperature and wind speed monitoring values. The external temperature value is selected by reading the Celsius reading converted from the meteorological probe. The carbon emission value at the power source is selected by initiating a query request to the energy management platform on the power plant side, retrieving the unit load rate data, and interpolating and retrieving the corresponding intensity data from the preset multidimensional carbon emission intensity characteristic curve.

[0033] The time slice step size is used to divide the calculation time window for carbon emission accounting; the sampling step size is the time interval corresponding to the participation of adjacent temperature samples in the thermal inertia hysteresis smoothing calculation, which is consistent with the time slice step size in this embodiment; the reference thermal time constant is determined based on the material thermal response characteristics of power cables and power distribution equipment.

[0034] By connecting with remote sensing mapping and various probes to build a front-end sensing network, the specific physical sources of each basic underlying parameter are defined, effectively avoiding measurement deviations caused by a single data source, and providing a high-fidelity data pool with rigorous physical meaning to support subsequent complex spatiotemporal correlations and graph topology tracing.

[0035] Multispectral separation is performed on remote sensing images to extract surface temperature and light intensity parameters, constructing meteorological features. Sliding time window slices are configured, and timestamp labels and spatial positioning parameters attached to the meteorological features are extracted using a time synchronization mechanism. A two-dimensional grid array is constructed based on the spatial positioning parameters, and the meteorological features are projected onto the corresponding coordinate intervals contained in the two-dimensional grid array. Alignment and fusion operations are performed to establish spatial relationships and output cross-dimensional attribute values. These cross-dimensional attribute values ​​are then assembled into a feature set matrix. Normalization and scaling calculations are performed on the feature set matrix to smooth the parameter value distribution intervals and merge them to generate node base vectors.

[0036] Specifically, the processing procedure for the node's fundamental vector is as follows: Meteorological features and spatial positioning are fused to generate node base vectors. First, multispectral separation is performed on the remote sensing image. The separation process extracts pixel channels of specific wavelengths, extracts surface temperature parameters from the infrared thermal imaging band, and extracts light intensity parameters from the visible and near-infrared bands. The two sets of values ​​are then merged and stitched together to form meteorological features.

[0037] A sliding time window is configured to segment the data stream into frames. The step size of the sliding time window is selected by comparing the upper limit of the remote sensing satellite data refresh cycle and the minimum common sampling cycle of the load node instruments, and the least common multiple of the two is selected and uniformly set to 15 minutes. Through this time window interception mechanism, the timestamp labels attached to meteorological features are mapped and bound to the map point coordinates within the same slice.

[0038] A two-dimensional grid array with latitude and longitude as axes is constructed based on spatial positioning parameters. The grid side length parameter is selected by setting a fixed span of 100 meters based on the spatial probability density function of urban building density layer data and distribution transformer distribution. According to the geographic information coordinate system transformation rules, the floating-point values ​​of meteorological features are projected onto the corresponding coordinate intervals contained in the two-dimensional grid array, and alignment and filling operations are performed to generate spatially coupled cross-dimensional attribute values, which are then stacked column by column within the same time window to form a feature set matrix.

[0039] In this embodiment, an outlier removal mechanism based on a local outlier factor is introduced into the feature set matrix. The local reachability density of the feature vectors within the grid is calculated, outlier nodes affected by sensor transient noise are screened out, and spatial smoothing interpolation of adjacent grids is performed using adaptive Gaussian weights. Specifically, before performing normalization scaling calculations on the feature set matrix, an outlier cleaning mechanism based on a local outlier factor is introduced. The specific process is as follows: For each grid data point in the feature set matrix, a neighborhood parameter of 5 is set, and the Manhattan distance from the target grid point to its 5th nearest normal grid data point is extracted and defined as the target grid point's neighborhood baseline distance; then, the actual Manhattan distance between the target grid point and any adjacent point is calculated; the larger value between the neighborhood baseline distance and the actual Manhattan distance is selected and set as the reachable distance. Calculate the neighborhood distance and local reachability density of the target grid point: First, calculate the neighborhood baseline distance and reachability distance of the target grid point, where the reachability distance is the larger of the neighborhood baseline distance and the actual Manhattan distance. Then, define the local reachability density of the target grid point as the reciprocal of the average reachability distance from the target grid point to all points in its neighborhood. Finally, divide the average local reachability density of the surrounding grids by the local reachability density of the target grid to obtain the local outlier factor. Set the anomaly detection threshold to 1.5. When the local outlier factor calculated from the feature value of a grid (such as extremely low light intensity caused by cloud cover) is greater than 1.5, it is marked as an anomaly. Trigger the interpolation repair mechanism: Extract the feature values ​​of normal grids in the adjacent grid regions around the anomaly point and use an adaptive Gaussian weighted algorithm for repair. The Gaussian kernel function was configured with a standard deviation of 1.2. Weights were assigned based on spatial distance: adjacent grids with a linear distance of 1 grid step were weighted at 0.6, and grids with a diagonal distance of 1.414 grid steps were weighted at 0.4. The product of each normal grid's eigenvalue and its corresponding weight was calculated and summed. This sum was then divided by the total weights involved in the assignment to generate repaired eigenvalues ​​that replaced the original outliers. The resulting feature set matrix after anomaly cleaning was output. This process effectively eliminates local feature distortions caused by transient noise from remote sensing and sensors, improving the anti-interference capability and accuracy of the underlying spatial grid parametric mapping.

[0040] To eliminate interference from different dimensions, a normalization scaling calculation is performed on the feature set matrix. The specific processing steps are as follows: The measured value of a certain original attribute in the current time slice is extracted, and the minimum value of that feature dimension in the historical sample space is subtracted to obtain the numerator of the difference; simultaneously, the maximum value of that feature dimension in the historical sample space is subtracted from the aforementioned minimum value to obtain the denominator of the range; then, a division operation is performed, dividing the obtained difference numerator by the range denominator. The numerator represents the difference of the same attribute at present, and the denominator represents the range of fluctuation of that attribute. After the division operation, the original physical dimensions are reduced to equal values, and the final output is a dimensionless smooth parameter value distributed within a closed interval of 0 to 1.

[0041] In this operation, the original attribute measurement values ​​are selected by directly reading the sensor's sampled output values ​​within the corresponding time slice; the minimum and maximum values ​​are selected by extracting the extreme values ​​of specific dimensions from a historical sample database containing all measurement data from the past 30 days for updating and filling. The smoothing parameter values ​​of all dimensions are merged to generate the node's basic vector.

[0042] A two-dimensional grid array is constructed based on spatial positioning parameters. Meteorological features are projected into the corresponding coordinate intervals to perform alignment and fusion operations. A rigorous spatiotemporal mapping relationship is established between environmental conditions and terminal nodes. Complex external physical features are transformed into underlying spatial grid parameters, achieving high-resolution spatial alignment.

[0043] Furthermore, a directed acyclic network (DAG) is constructed using inter-regional interconnection power and grid power flow status, and weight parameters are configured. The node base vectors are mapped to the terminal graph nodes of the DAG. The carbon emission value at the power source's origin is used as the source emission factor of the root node. Following the weight parameters, tracing calculations are performed along the graph topology path to output the carbon intensity baseline; this corresponds to step S2 above; see [link to relevant documentation]. Figure 3 The specific implementation process includes: The graph computing engine analyzes the network topology matrix contained in the power flow state of the network structure; it calls the power flow ratio sharing algorithm to perform loop removal on the network topology matrix, establishes a unidirectional power flow state graph, and decomposes the unidirectional power flow state graph into the starting root node and intermediate transition nodes. Combined with spatial positioning parameters, it sets the terminal graph nodes and establishes the initial network architecture; it analyzes the line power parameters contained in the network power flow state, and configures the weight parameters of the edges of the directed acyclic network graph according to the line power parameters; it injects the inter-regional interconnection power as a constraint condition into the initial network architecture, and at the same time, according to the spatial positioning parameters and weight parameters, it projects the node base vectors into the terminal graph nodes for feature mapping, and outputs the directed acyclic network.

[0044] Specifically, the processing procedure for directed acyclic networks is as follows: Graph computing techniques are used to map the physical power grid topology into a directed acyclic network (DAG) in digital space. First, the power flow state of the network is read, and an adjacency matrix representing the device connection topology is extracted. For loop closures present in the distribution network, a power flow proportion sharing algorithm is invoked to perform a loop-breaking operation on the network topology matrix. The loop-breaking rules are as follows: Measurement data from each line branch junction is read, and the direction and absolute value of active power flow in and out of the branches are compared. Logical constraints stipulate that energy flows only from high-potential nodes to low-potential nodes within the same time slice, establishing a unidirectional power flow state diagram. After establishing the unidirectional power flow state diagram, the active power of all inflow branches to any intermediate transition node is extracted. Based on the Bialek proportion sharing principle, the power contribution ratio of a specific inflow branch to that node is calculated. Specifically, the active power of the specific inflow branch is divided by the sum of the active power of all branches flowing into that node. For any associated outflow branch, the proportion of power from the specific inflow branch it carries is the calculated proportion.

[0045] The unidirectional power flow state diagram is retrieved, and the in-degree and out-degree parameters of each independent node are scanned using a traversal algorithm. The in-degree and out-degree thresholds of the nodes are selected by directly statistically setting the number of physical ports fed back from the substation's main electrical wiring diagram. Nodes with an in-degree parameter of 0 in the scan results are marked as the starting root nodes; nodes with both in-degree and out-degree parameters greater than 0 are defined as intermediate transition nodes; simultaneously, the map coordinates are extracted, and nodes with an out-degree parameter of 0 that match the input coordinates are set as terminal nodes, completing the initial network architecture's element node setup.

[0046] The line power parameters are read, and connection weight parameters are assigned to each directed edge in the initial network architecture. The specific process of obtaining the connection weight parameters is performed using the following calculation formula: Extract the input active power value in kilowatts at the beginning of a single distribution line and the output active power value at the end. Subtract the output active power value from the input active power value to calculate the active power loss difference, which represents the physical transmission loss of the line, as the target numerator variable. At the same time, extract the input active power value at the beginning of the line as the target denominator variable. Divide the target numerator variable by the target denominator variable. Since both the numerator and denominator use kilowatt power as the unit, the dimensions are eliminated after the division operation, thus calculating the dimensionless weight parameter value representing the proportion of physical loss of the line.

[0047] In this process, the active power flow value of a single distribution line is selected by reading the power flow uploaded by the monitoring and control device installed at the line node; the active power value of all lines flowing into that node is selected by extracting the sum of the measurement data of the node's current transformers. Finally, the inter-regional interconnection power is extracted as a constraint condition for the total power of the entire network and injected into the boundary nodes of the initial network architecture. According to the spatial positioning parameters, the node's basic vector is projected and bound to the data structure inside the terminal graph node, and the constructed directed acyclic network is output.

[0048] The power flow ratio sharing algorithm is invoked to perform loop unblocking operations on the network topology matrix to establish unidirectional power flow. This effectively suppresses the cyclic deadlock paths that objectively exist in the physical power grid, and tightly maps the complex physical ring network into a directed acyclic network, laying a graph structure foundation for breadth-first search and continuous tracking of the material flow transfer process.

[0049] A breadth-first search algorithm is invoked to traverse the directed acyclic network (DAN) and identify graph topology paths pointing to the terminal graph nodes. On / off state markers are extracted from the network topology matrix, and connectivity checks are performed based on these markers. Graph topology paths in the disconnected state are eliminated, while valid paths are retained. The carbon emission value at the power source's origin is used as the source emission factor for the root node. Based on the weight parameters of the DAN graph edges, state transition tracking calculations are performed along the valid graph topology paths from the root node to the terminal node to obtain the carbon intensity baseline of the terminal graph nodes.

[0050] Specifically, the process for handling the carbon intensity base is as follows: The breadth-first search algorithm is invoked to assign terminal graph nodes as target anchors. The directed acyclic network is traversed along network connections to identify graph topology paths leading to the initial root node. Connectivity status markers are extracted from the network topology matrix, and connectivity Boolean checks are performed layer by layer: when a marker value of 0 is detected, a path culling operation is triggered to remove branches of that graph topology path; when a marker value of 1 is detected, it is retained in the set of valid graph topology paths.

[0051] Along the effective graph topology path, weighted state transition tracking calculations based on power flow ratios are performed from the initial root node to the terminal graph node. For any intermediate transition node, when multiple inflow paths exist, the line power parameters (i.e., active power) of each inflow path are extracted. Based on the proportion of each inflow path's power to the node's total inflow power, the network flow loss parameters from different paths are weighted, summed, and merged. For a single, branchless path, the path's weight parameter (i.e., the difference in line active power loss divided by the input active power) is extracted, and the loss increment of that path is calculated and incorporated into the current state. Finally, the equivalent network flow loss parameters accumulated after weighted allocation across the entire network are recorded upon reaching the terminal graph node.

[0052] From the node base vector mapped to the terminal graph node, meteorological features such as surface temperature and light intensity parameters are extracted and extracted. These are then used in the environmental impedance value derivation calculation process to extract the external temperature value with Celsius dimensions. Subtracting the standard reference temperature yields the Celsius temperature difference variable. Multiplying the Celsius temperature difference variable by the temperature resistivity coefficient (which has a reciprocal of Celsius) yields a dimensionless temperature correction. Similarly, extracting the light intensity parameter with watts per square meter and multiplying it by the thermal radiation conversion constant with watts per square meter yields a dimensionless light correction. Subsequently, the constant 1, along with the temperature and light corrections, is added together to obtain a dimensionless comprehensive damping amplification factor characterizing the environmental impact. The physical basis of this addition logic is that for conductive metal materials (such as steel-cored aluminum stranded wire), within the allowable operating temperature range, their resistivity change is usually approximated as a linear superposition mainly affected by the combined influence of ambient temperature and absorbed thermal radiation. Constant 1 represents the ideal baseline state without external environmental fluctuations; the temperature correction reflects the linear shift in resistivity caused by environmental temperature differences; and the illumination correction reflects the increase in equivalent resistivity caused by solar radiation heat absorption. Since both of these factors microscopically exacerbate the hindrance of lattice vibrations to electron movement, and are relatively small disturbances in typical urban operating environments, they are all relatively small.

[0053] Finally, a reference impedance constant with ohmic dimensions is extracted, and the reference impedance constant is multiplied by the comprehensive damping amplification factor to calculate and output the environmental impedance value with ohmic dimensions.

[0054] In this embodiment, after generating the environmental impedance value, a wind speed heat dissipation compensation model is integrated. The external wind speed value is extracted from the external environmental parameters, and the convective heat dissipation reduction constant is calculated using an empirical wind-cooling function. The environmental impedance value is multiplied by this reduction constant to obtain the wind-cooled corrected environmental impedance value. Specifically, the external wind speed value with units of meters per second is extracted from the captured external environmental parameters. The effective physical sensing range of wind speed is set to 0.5 to 15 meters per second; values ​​below 0.5 meters per second are ignored as natural convection, and values ​​above 15 meters per second are considered as surface heat exchange saturation. A heat transfer reduction function for air cooling is introduced to calculate the dimensionless convective heat loss reduction constant. When the external wind speed is between 0.5 and 15 m / s, the external wind speed is subtracted by 0.5 to obtain the difference. This difference is then square-rooted and multiplied by a fixed constant of 0.02 (square root of seconds per meter). Finally, the product is subtracted from 1 to obtain the convective heat loss reduction constant. When the external wind speed is below 0.5 m / s, the convective heat loss reduction constant is directly set to 1. When the external wind speed is above 15 m / s, the highest threshold is substituted into the above calculation process to obtain the minimum value of the constant. For example, when the local grid wind speed measured by the meteorological probe is 4.5 m / s, the constant is calculated to be 0.96. The environmental impedance value with Ohmic dimensions obtained in the previous calculation is extracted and multiplied by the dimensionless convective heat loss reduction constant using scalar multiplication. The physical significance of this step is that increased wind speed accelerates heat dissipation from the surface of overhead cables, reducing the actual operating temperature and resulting in a decrease in conductor impedance compared to when only sunlight and base temperature are considered. The multiplicative calculation result is defined as the environmental impedance value after wind cooling correction. The above process quantifies the attenuation effect of surface convection heat transfer on the dynamic impedance of the cable, further improving the physical accuracy of the microscopic thermistor impedance derivation.

[0055] The selection methods for each variable in the impedance derivation process are as follows: the external temperature value and the light intensity parameter are directly extracted from the meteorological characteristics mapped to the terminal node of the graph; the standard reference temperature is set to a fixed 20 degrees Celsius based on the national power distribution network design specification standard; the temperature resistivity coefficient is obtained by referring to the overhead cable engineering material physical property guide manual, for example, the value of steel-cored aluminum stranded wire is 0.0039; the thermal radiation conversion constant is extracted based on the average value extracted from the environmental absorption rate test report of the insulation sheath surface and set to a value of 0.0001; the reference impedance constant is selected by reading the nominal unit impedance (ohms / km) from the line manufacturer's nameplate, and extracting the actual physical cable length (km) of the branch in combination with the graph topology path, and multiplying the two to obtain the reference impedance constant with the dimension of ohms.

[0056] Subsequently, the operating voltage and power factor parameters of the node at the beginning of the path are obtained. The input power value is divided by the product of the node operating voltage parameter, the power factor parameter, and the square root of three to deduce the current value of the line. The current value is squared and multiplied by the environmental impedance value to calculate the loss power value in kilowatts. Then, the loss power value is divided by the input power value to obtain the dimensionless static network loss rate of the path. Based on the environmental impedance value, node operating voltage parameter, power factor parameter, and line input power value, the loss power value of the corresponding line is calculated. The loss power value is divided by the line input power value to obtain the static network loss rate of the corresponding line. The weight parameters of the corresponding graph edges in the directed acyclic network are corrected according to the static network loss rate to obtain the environmentally corrected weight parameters. The static network loss rate is used to update the weight parameters of the corresponding graph edges in the current time slice and is not repeatedly accumulated with the weight parameters before correction. The environmentally corrected weight parameters are used to characterize the changes in line loss caused by the external environment during line transmission. The environmental correction weight parameter is the graph edge weight parameter after static network loss rate correction, and it serves as the input for subsequent state transition tracing calculations. The carbon emission value at the power source starting point is written into the corresponding starting root node as the source emission factor. Based on the environmental correction weight parameter corresponding to each graph edge, state transition tracing calculations are performed along the effective graph topology path. At intermediate transition nodes, the source emission factors and loss increments corresponding to different inflow paths are weighted and merged to obtain the additional loss carbon emission value accumulated when reaching the terminal graph node.

[0057] Since the carbon footprint is treated as a substance that propagates along with currents in the digital model, the additional loss carbon emissions are arithmetically added to the power source carbon emissions (measured in kilograms of carbon dioxide per kilowatt-hour), which represents the source emission factor. After the addition, the calculated output is a carbon intensity base that incorporates environmental spatiotemporal perception characteristics, while maintaining the dimensional property of kilograms of carbon dioxide per kilowatt-hour.

[0058] The calculation is performed by tracing the topology path along the effective graph, and the environmental impedance is deduced by combining meteorological characteristics to correct the grid flow loss. It fits the physical flow chain of power transfer and quantifies the response to the microscopic thermal interference generated by the external environment on the physical grid.

[0059] Furthermore, node load parameters and external environmental parameters are imported into the network loss assessment model to obtain network loss compensation values; these compensation values ​​are then superimposed and integrated into the carbon intensity baseline, and combined with regional carbon emission quotas to perform carbon emission over-limit verification, generating a time-sharing carbon emission matrix; this corresponds to step S3 above; the specific implementation process includes: Calculate the power mean and power variance of transient power values ​​within the sliding time window; input the equivalent temperature difference fluctuation parameter, equivalent charging power, external wind speed value, state of charge percentage, power mean, and power variance into the network loss assessment model to calculate the heat dissipation parameter; perform quadratic polynomial fitting on the heat dissipation parameter to output the network loss conversion parameter; extract the carbon intensity base, and perform floating-point multiplication on the network loss conversion parameter and the carbon intensity base to obtain the network loss compensation value.

[0060] Specifically, the processing procedure for network loss compensation values ​​is as follows: The aforementioned static network loss rate is used to characterize the static impact of the external environment on the power flow distribution relationship of the line; the following network loss assessment model is used to assess the dynamic additional losses caused by charging load fluctuations, charging stages and equipment thermal inertia. The calculation process of the network loss assessment model does not repeatedly include the aforementioned static network loss rate.

[0061] To compensate for dynamic loss dissipation deviations caused by nonlinear heating of equipment materials, a deep learning evaluation model is loaded and run. First, the transient power value, charging mode identifier, and state of charge (SCC) percentage stage identifier from the node load parameters, as well as the external temperature value from the external environment parameters, are extracted. Based on the SCC percentage stage identifier, a nonlinear charging attenuation weight is applied to the transient power value: when the SCC is in the constant current charging range (e.g., <80%), the weight is 1.0; when the SCC enters the constant voltage wake charging range (e.g., ≥80%), the SCC percentage minus the 80% threshold is extracted to obtain the overflow percentage difference. The natural base of the constant is then exponentially multiplied by the negative product of this overflow percentage difference and a preset attenuation coefficient (e.g., 5.0) to obtain the adjusted nonlinear charging attenuation weight. The equivalent charging power adjusted by the attenuation weight is then combined with the external temperature value. Calculate the temperature fluctuation parameter: Extract the external temperature value of the current time slice cache, subtract the external temperature value of the previous consecutive adjacent time slice cache, and perform an absolute value operation on the calculation result to obtain the temperature fluctuation parameter with Celsius unit.

[0062] In this embodiment, a thermal inertia hysteresis smoothing operator is introduced. Based on the thermal time constant of the equipment material and the sampling step size, the filtering weight is calculated, and a first-order low-pass filter is applied to the temperature difference parameter between the current and previous slices to obtain the equivalent temperature difference fluctuation parameter with time decay properties. Specifically, considering that power cables and transformers possess physical heat capacity, their temperature response exhibits an inherent hysteresis compared to environmental changes, a thermal inertia hysteresis smoothing step is added. The set sliding time window step size is extracted to 15 minutes, and the reference thermal time constant of a standard overhead conductor (taking steel-cored aluminum stranded wire as an example) is retrieved to 45 minutes. The thermal inertia smoothing coefficient is calculated. The specific calculation process is as follows: the time window step size is divided by the reference thermal time constant to obtain the ratio, the negative of this ratio is taken, and then the natural exponent is calculated, resulting in a smoothing coefficient constant of approximately 0.716. The equivalent temperature difference fluctuation parameter cached in the previous time slice and the original temperature difference fluctuation parameter calculated in the current slice are extracted. Numerical fusion calculations are performed using a first-order discrete low-pass filter formula: the equivalent temperature difference fluctuation parameter from the previous moment is multiplied by a smoothing coefficient, and then added to the product of the original temperature difference fluctuation parameter and 1 minus the smoothing coefficient, to obtain the current equivalent temperature difference fluctuation parameter. For example, if the equivalent fluctuation at the previous moment was 2 degrees Celsius, and the currently calculated original fluctuation is 5 degrees Celsius, then the current equivalent parameter is corrected to 2.852 degrees Celsius. The above process restores the true thermal hysteresis characteristics of the physical conductor under rapid power fluctuations through first-order filtering.

[0063] The equivalent temperature difference fluctuation parameter, equivalent charging power, external wind speed value, state of charge percentage stage identifier, and the power mean and power variance of transient power values ​​within the sliding time window are encapsulated and input into the network loss assessment model. Among them, the equivalent charging power is used to characterize the load level corresponding to the charging condition, the equivalent temperature difference fluctuation parameter is used to characterize the temperature change under the thermal inertia of the equipment, the external wind speed value is used to characterize the convective heat dissipation conditions, and the state of charge percentage stage identifier is used to characterize the charging stage of the electric vehicle.

[0064] The selection methods for the hyperparameters and internal architecture of the network loss assessment model are defined as follows: The input mapping layer is configured with the corresponding number of neurons based on the dimensional features of the input parameters; the hidden layers use a series of multilayer perceptrons, and the number of neurons in each hidden layer is selected by combining them through parameter tuning during the pre-training phase, aiming to minimize the mean squared error of the validation set predictions, with the numbers set to 64, 32, and 16 respectively from shallow to deep; activation functions are embedded in the connection nodes between hidden layers, and disconnected probability layers are interspersed between layers to prevent overfitting. The disconnected probability parameter is set to 0.3 after weighing the risks of overfitting suppression and underfitting; the model loss function is selected to measure the mathematical deviation between the model's forward propagation predictions and the true label values, and is set as the mean squared error function; the initial learning rate parameter of the backpropagation optimizer is set to 0.001 based on the trial-and-error convergence speed; the batch size parameter is set to 512 based on the upper limit of the processing capacity of the computing nodes; the global training epochs parameter is set to a fixed 200 epochs by continuously observing the inflection point where the loss function curve gradually approaches its flat point.

[0065] After completing the high-dimensional feature mapping through the network loss assessment model, the predicted thermal loss parameters with nonlinear correction properties are output.

[0066] Subsequently, a quadratic polynomial fitting process was performed on the dimensionless equivalent heat dissipation parameter. The fitting process for this network loss conversion parameter was performed using the following calculation formula: the equivalent heat dissipation parameter was squared, and then multiplied by the first fitting coefficient to obtain the quadratic term; at the same time, the heat dissipation parameter was directly multiplied by the second fitting coefficient to obtain the linear term; the third constant term, which characterizes the fixed no-load iron loss property of the equipment, was extracted; finally, the aforementioned quadratic term, linear term, and third constant term were summed by addition to calculate and output the network loss conversion parameter representing the global nonlinear loss ratio.

[0067] The coefficients within the polynomial equation are selected as follows: the first fitting coefficient, the second fitting coefficient, and the third constant term are obtained by extracting historical measured data of load and loss from different types of distribution transformers and conductors under laboratory testing conditions, and then using the least squares algorithm to perform nonlinear regression iteration to approximate the solution. For a 10 kV distribution transformer, the first fitting coefficient is selected as 0.0045, the second fitting coefficient as 0.105, and the third constant term as 0.02. Since the thermal dissipation parameter, as the independent variable, is a dimensionless parameter, the solved network loss amplification parameter is also a dimensionless loss amplification scalar.

[0068] The dimensionless network loss conversion parameter is multiplied by the carbon intensity base, and the result is defined as the network loss compensation value that includes the attenuation characteristics under heating conditions. This multiplication operation does not change the original emission dimensions, and the network loss compensation value remains completely consistent with the carbon intensity base dimension.

[0069] By calculating temperature fluctuations, deducing the micro-link attenuation coefficient, and performing quadratic polynomial fitting, the dynamic impedance and heat dissipation of transmission lines under load fluctuations are reflected, thus avoiding the phenomenon that static loss measurements deviate from the actual physical operating environment.

[0070] The network loss compensation value and the carbon intensity base are added by scalar to obtain the corrected carbon intensity parameter; the highest threshold parameter and the warning threshold parameter included in the regional carbon emission quota are extracted; the corrected carbon intensity parameter and the warning threshold parameter are compared; if the corrected carbon intensity parameter exceeds the warning threshold parameter but is lower than the highest threshold parameter, a preset over-limit marking mechanism is activated, and the over-limit status feature bit corresponding to the corrected carbon intensity parameter is output; if the corrected carbon intensity parameter exceeds the highest threshold parameter, an extreme value alarm parameter is output; the corrected carbon intensity parameters corresponding to each time slice are arranged in matrix form according to the time series to generate a time-sharing carbon emission matrix.

[0071] Specifically, the processing procedure for the time-sharing carbon emission matrix is ​​as follows: A panoramic simulation data model for carbon emissions is constructed based on time-series evolution, and over-limit verification and protection are implemented. In the merging calculation stage, the network loss compensation value and the carbon intensity baseline are retrieved. These two parameters, both with the dimension of kilograms of carbon dioxide per kilowatt-hour, are arithmetically added together to generate a corrected carbon intensity parameter reflecting the final dynamic loss. The two values ​​with the same dimension are added together, and the resulting corrected carbon intensity parameter maintains the original dimension of kilograms of carbon dioxide per kilowatt-hour.

[0072] Logical comparison is performed on the modified carbon intensity parameter. Regional carbon emission quotas are retrieved, and the highest threshold parameter and the warning threshold parameter are extracted. A judgment is executed: when the value of the modified carbon intensity parameter exceeds the lower limit defined by the warning threshold parameter but is lower than the upper limit defined by the highest threshold parameter, a preset limit-crossing marking mechanism is activated. Under this mechanism, the limit-crossing status feature bit corresponding to the modified carbon intensity parameter is output. The assignment method for the limit-crossing status feature bit is based on conditional judgment logic; when the judgment input condition is true, a 1 is written to the feature bit, and when the judgment is false, a 0 is written. The limit-crossing status feature bit and the extreme value alarm parameter are output as independent monitoring results, do not participate in the numerical calculation of the modified carbon intensity parameter, and are not included as components of the time-sharing carbon emission matrix.

[0073] If the modified carbon intensity parameter is determined to exceed the upper limit defined by the highest threshold parameter, an extreme value alarm parameter is generated and output. The trigger threshold for the extreme value alarm parameter is selected by reading the equivalent maximum carbon emission value corresponding to the rated current carrying capacity specified for different conductors in the safe operation procedure.

[0074] Continuously capture and correct carbon intensity parameters and corresponding timestamp parameters. Using time series as the first arrangement dimension axis and spatial node coordinates as the second arrangement dimension axis, arrange them into a high-order, high-dimensional matrix data format to complete the generation of the time-series carbon emission matrix.

[0075] While achieving time-sharing dynamic monitoring, it effectively prevents human tampering with the objective measurement data.

[0076] Furthermore, the time-sharing carbon emission matrix and the output parameters of the multi-source power sources are encapsulated into a source-grid carbon coupling tensor. Dimensionality reduction and aggregation are performed on the source-grid carbon coupling tensor to obtain a dimension-reduced aggregation vector. Feature inversion mapping is then performed on the dimension-reduced aggregation vector to obtain the basic carbon emission equivalent parameter and the power source corrected power parameter, generating the carbon emission result. This corresponds to step S4 above. The specific implementation process includes: Extract the instantaneous output ratio values ​​from the power parameters of the multi-source power source; set the instantaneous output ratio values ​​as an additional feature dimension, and concatenate the additional feature dimension into the time-sharing carbon emission matrix to construct the source-grid carbon coupling tensor; define the calculation time window, start the tensor orthogonal decomposition algorithm, extract key principal component parameters, generate an approximate reconstruction tensor, subtract the source-grid carbon coupling tensor from the approximate reconstruction tensor element by element, generate and cache the orthogonal residual matrix; aggregate all key principal component parameters to generate a dimension-reduced aggregation vector.

[0077] Specifically, the process of processing the dimension reduction and aggregation vector is as follows: Beyond the existing carbon flow dimension, an implicit constraint dimension of clean energy output is further embedded into the data space. Output parameters from multiple power sources are extracted, and the dimensionless instantaneous output ratios (in percentage terms) are analyzed. These instantaneous output ratios are then set as a new, independent additional feature dimension axis. The length parameter of this additional feature dimension is selected by reading the number of nodes in the corresponding time slice of the time-sharing carbon emission matrix to ensure complete alignment of dimensions in multidimensional matrix operations. Along the depth direction of the multidimensional data, this additional feature dimension is concatenated and injected into the original time-sharing carbon emission matrix, completing a step-up dimensionality increase and constructing a source-network-carbon coupling tensor with a structure encompassing time, space, carbon, and source.

[0078] A computational time window for dimensionality reduction is defined. The time window length is selected based on the start and end times of the daily load dispatch cycle of the power system, and is set to a constant of 24 hours. Once the data frames of the source-grid carbon coupling tensor fill a computational time window, the tensor orthogonal decomposition algorithm is invoked to perform feature space transformation on the cached multiple source-grid carbon coupling tensors. By recording the unrepresented difference data of key principal component parameters, and calling the orthogonal residual matrix during reverse decoding, the reconstruction bias is reduced.

[0079] The parameter settings and operational logic of the tensor orthogonal decomposition process are as follows: An asymmetric strategy is adopted to set the target rank reduction: In this three-dimensional extended structure with a time dimension of 96, a spatial dimension of 50, and a feature dimension of 4, the time dimension rank reduction is set to 8, the spatial dimension rank reduction is set to 8, while the feature dimension rank remains at 4 without compression. The extracted principal component tensors (kernel tensors) have a size of 8×8×4.

[0080] The iterative process begins. During iterative calculations, the sum of squared differences between corresponding elements in the original source-network carbon coupling tensor and the approximate reconstructed tensor is differentiated, optimized, and minimized. The loop terminates when the difference between two consecutive iterations falls within the convergence tolerance range. In this approximation process, a set of mutually orthogonal column vectors representing each major feature dimension is extracted and used as key principal component parameters. Simultaneously, the reconstructed tensor obtained by multiplying the key principal component parameters by the original tensor is subtracted element-wise, and the small deviations obtained from the subtraction are retained and cached as orthogonal residual matrices.

[0081] Extract all the independent key principal component parameters, merge and concatenate them in dimensional order, reduce their order to aggregate vectors, and output the dimensionality-reduced aggregate vector.

[0082] The instantaneous output ratio values ​​are concatenated into the matrix to construct the source-grid carbon coupling tensor. The orthogonal residual matrix is ​​generated and cached through the tensor orthogonal decomposition algorithm, which effectively processes heterogeneous power structure data and realizes high-dimensional feature space transformation and compression.

[0083] The process involves retrieving the cached orthogonal residual matrix; performing inverse decoding on the dimensionality-reduced aggregated vector and the orthogonal residual matrix; executing decoupling and splitting based on preset built-in transformation rules to obtain the carbon emission intensity benchmark and the power correction intensity benchmark; combining the transient power values ​​contained in the node load parameters to perform time integration conversion to restore the basic carbon emission equivalent parameter and the power correction energy parameter; retrieving the preset unit energy emission reduction factor for the power correction energy parameter; the unit energy emission reduction factor contains emission reduction offset attributes matching the instantaneous output ratio; and adding the product of the basic carbon emission equivalent parameter and the unit energy emission reduction factor and the power correction energy parameter to generate the carbon emission result.

[0084] Specifically, the carbon emission results are processed as follows: The dimension-reduced and compressed vector data undergoes dimension reconstruction and pricing transformation. The temporarily stored orthogonal residual matrix is ​​retrieved. Using a reverse multiplication process, following the dimensionality order recorded during tensor orthogonal decomposition, the dimension-reduced aggregated vector is restored to key principal component parameters, and an approximate reconstructed tensor is obtained based on these parameters. The orthogonal residual matrix is ​​then element-wise superimposed onto the approximate reconstructed tensor to obtain the reconstructed source-network-carbon coupling tensor. Following the preset dimensionality arrangement order during the construction of the source-network-carbon coupling tensor, decoupling and splitting processes are performed on the reconstructed source-network-carbon coupling tensor.

[0085] After decoding and generating the initial restored data, a pre-defined built-in transformation rule is loaded to perform decoupling and splitting processing. The selection method for the mask parameters within the built-in transformation rule is based on the attribute structure offset in the data definition. The specific definition of the attribute structure offset on which the built-in transformation rule is based is as follows: When constructing the source-network carbon coupling tensor, a strict dimension arrangement order is pre-defined. It is assumed that the feature state dimension of the tensor has a specific total length parameter. Among them, the slice space from index position 0 to a specific boundary index position (this boundary index position must be less than the total length parameter) is defined as the carbon emission related feature region; the slice space from the boundary index position plus 1 to the total length parameter minus 1 is defined as the power output ratio related feature region. During the decoupling and splitting processing, the mask rule is set as follows: data is extracted using a closed interval mask from 0 to the boundary index position, and the output is the carbon emission intensity benchmark; data is extracted using a closed interval mask from the boundary index position plus 1 to the total length parameter minus 1, and the output is the power correction intensity benchmark. For example, if the total length parameter of the feature dimension is 4, and indices 0 to 2 correspond to carbon emission features and index 3 corresponds to power ratio features, then the splitting is performed according to this offset.

[0086] The initial data was separated to extract two independent intensity benchmark parameters: the first parameter was defined as the carbon emission intensity benchmark, with the physical dimension being kilograms per kilowatt-hour; the second parameter was defined as the power supply correction intensity benchmark, with the physical dimension being a dimensionless percentage parameter. Subsequently, energy integration and conversion were performed: the transient power values ​​from the node load parameters were retrieved, and the transient power values ​​were multiplied and time-series integrated with the corresponding time slice step size to obtain the total terminal charging energy in kilowatt-hours; the carbon emission intensity benchmark was multiplied by the total terminal charging energy to restore the basic carbon emission equivalent parameter with the physical dimension of kilograms of carbon dioxide equivalent; the power supply correction intensity benchmark was multiplied by the total terminal charging energy to restore the power supply correction energy parameter with the physical dimension of kilowatt-hours.

[0087] The unit energy emission reduction factor for power generation parameters is retrieved synchronously. The parameter selection method for the unit energy emission reduction factor is as follows: statistical values ​​of the unit energy emission reduction factor for various clean energy sources (such as photovoltaic, wind power, and geothermal power) compared to the benchmark thermal power are retrieved from the power grid carbon emission management report, and weighted calculations are performed in conjunction with the instantaneous output ratio. Through physical conversion and deduction, a value of -0.8 kg of carbon dioxide equivalent per kilowatt-hour was selected as the unit energy emission reduction factor. This negative value represents the emission reduction offset attribute derived from the conversion when the proportion of clean energy increases.

[0088] The basic carbon emission equivalent parameter with kilogram carbon dioxide equivalent dimensions is extracted. Simultaneously, the unit electricity emission reduction factor with kilogram carbon dioxide equivalent dimensions per kilowatt-hour and the power supply correction electricity parameter with kilowatt-hour dimensions are extracted, and both are multiplied. During the calculation, the kilowatt-hour dimensions in the denominator and multiplier cancel each other out, and the resulting product is defined as the indirect emission reduction offset value, its dimension converted to kilogram carbon dioxide equivalent. Finally, the calculated basic carbon emission equivalent parameter and the indirect emission reduction offset value are extracted, and the two values ​​are added together. Since the two added items have completely identical quantitative dimensions (the indirect emission reduction offset value has a negative sign to achieve carbon reduction verification), the final carbon emission result is output after the aggregation calculation.

[0089] The orthogonal residual matrix is ​​retrieved to perform reverse decoding, restore the independent dimensional parameters, and rely on the summation and multiplication of the unit power emission reduction factor to construct a coherent closed loop at the mathematical operation level.

[0090] The specific architecture, dataset construction, and training process of the network loss assessment model in this embodiment are supplemented as follows: The final deployment structure of the network loss assessment model is a multilayer perceptron deep neural network. Its input layer contains 6 neurons, which receive parameters such as equivalent temperature difference fluctuation, equivalent charging power, ambient wind speed, state of charge percentage, and the mean and variance of power within a sliding time window. The network contains three hidden layers, with the number of neurons in each layer set to 64, 32, and 16 respectively, from shallowest to deepest. Each hidden layer is followed by a ReLU activation function to introduce non-linear expressive power, and a random deactivation layer with a dropout probability of 0.3 is embedded between layers to suppress overfitting during training. The output layer contains 1 neuron, which directly outputs the model's predicted heat loss value after forward propagation mapping.

[0091] The dataset used is derived from offline physical simulation and field measurements. By simulating the heating operation of a standard overhead conductor (taking steel-cored aluminum stranded wire as an example) under various external temperature gradients (-20 degrees Celsius to 60 degrees Celsius, with a step size of 1 degree Celsius) and various transient input power loads (0 kW to rated current carrying capacity, with a step size of 5 kW) in a temperature-controlled physical sandbox, the corresponding instantaneous temperature difference of the conductor, input power, and physically measured heat dissipation values ​​are measured and recorded at high frequency. Under each set of operating conditions, the equivalent temperature difference fluctuation parameter, equivalent charging power, external wind speed, state of charge percentage, and the power mean and power variance within the corresponding sliding time window are recorded simultaneously. The physically measured heat dissipation values ​​are used as supervised training labels to form training samples corresponding to six-dimensional input features and a single output label.

[0092] Before training, the dataset was randomly divided into training, validation, and test sets in an 8:1:1 ratio. The training set was used to calculate the loss gradient during backpropagation and iteratively optimize the connection weights and bias parameters of each network layer. The validation set was used for cross-validation during training, evaluating the model's fit at the end of each training epoch, and serving as a basis for dynamically adjusting the learning rate, initiating early termination of training, and optimizing network hyperparameters. The test set did not participate in any training or hyperparameter tuning process; it was only used after model training was completed to perform a final objective accuracy assessment of its generalization error, calculating the absolute average error between the output equivalent parameters and the actual physical heat generation.

[0093] Before training, the input data was preprocessed using extreme value scaling for normalization. During training, the mean squared error function (MSE) was used as the loss function to quantify the mathematical deviation between the network's predicted values ​​and the numerical labels of measured thermal loss in laboratory physics. The Adam algorithm was selected as the optimizer, with an initial learning rate of 0.001. Mini-batch gradient descent was used for training, with a mini-batch size of 512 samples and a maximum global training epoch of 200 epochs. At the end of each training epoch, the loss value on the validation set was monitored. If the loss value on the validation set did not decrease significantly within 15 consecutive training epochs, an early termination mechanism was triggered, and the set of network weights and bias parameters with the lowest MSE on the validation set was saved as the final deployment parameters for the network loss evaluation model.

[0094] This invention provides a method for calculating the implicit carbon emissions of time-sharing charging of urban electric vehicles. By acquiring multi-dimensional data such as remote sensing images and cross-regional power consumption, meteorological features and map coordinates are aligned and fused to generate node base vectors. A directed acyclic network is constructed based on a graph computing engine to perform tracking calculations. Subsequently, network loss compensation values ​​are superimposed and fused into the carbon intensity base. Finally, the encapsulated source-network carbon coupling tensor is subjected to dimensionality reduction aggregation and feature inversion mapping to generate carbon emission results. This method breaks the isolation between cross-domain physical data and constructs a coherent closed loop in three dimensions: time series, spatial grid, and physical flow direction. It handles the heterogeneous associated resource physical consumption characteristics and realizes multi-dimensional parameter adaptation for dynamic distribution attributes of different geographical locations and environments. Thus, it can effectively reflect the real environmental physical load of different terminals at specific spatiotemporal nodes.

[0095] Example 2 This embodiment takes an example of applying a method for calculating the implicit carbon emissions of time-sharing charging of urban electric vehicles to a scenario of centralized charging of electric truck fleets at night in a logistics park.

[0096] First, multi-dimensional parameter acquisition is performed. Within a fixed polling cycle, multispectral remote sensing images are acquired by connecting to the mapping platform to extract regional infrared and visible light features; the latitude, longitude, and elevation three-dimensional coordinates of the logistics park charging station are obtained by calling the geographic information interface; the cross-regional interconnection power is extracted through the substation gate; the grid topology matrix and line power parameters containing on / off status markers are obtained by connecting to the power distribution management system; the regional carbon emission quota is analyzed to obtain the maximum threshold parameter and the warning threshold parameter; the output parameters of multiple power sources, including the instantaneous output ratio values ​​of photovoltaic, wind power, and geothermal power, are retrieved; transient power values, charging mode identifiers, and state of charge percentage stage identifiers are obtained through the built-in metering device and battery management system of the charging pile; external temperature and external wind speed values ​​are obtained through meteorological probes; the real-time load rate of the unit is queried from the energy management platform on the power plant side, and interpolation is performed in the pre-set carbon emission intensity characteristic multi-dimensional curve mapping table to obtain the carbon emission value at the power source.

[0097] Subsequently, spatiotemporal data alignment was performed. Surface temperature and light intensity were separated from the remote sensing images, and bound to spatial coordinates using a sliding time window mechanism. A two-dimensional spatial grid was established, and the aforementioned meteorological features were projected onto the corresponding grid intervals. Historical ranges were calculated, and normalized division scaling was performed to eliminate the influence of dimensions, merging to generate dimensionless node basis vectors.

[0098] Next, a directed acyclic network (DAG) is constructed. The network topology matrix is ​​extracted, the power flow direction of each branch is compared, and the reverse power flow is cut off to resolve the loop. By calculating the in-degree and out-degree, the power plant is set as the starting node, and the logistics park charging station is set as the ending node. The weight parameters of each connecting edge are configured according to the ratio of the power of a single line to the total inflow power, and the node base vectors are projected to complete the network construction.

[0099] Next, impedance correction and topology tracing are performed. The environmental damping amplification factor is calculated by combining the temperature and illumination characteristics in the node's base vector, and the reference impedance is corrected. Based on the corrected impedance, line power parameters, node operating voltage parameters, and power factor parameters, the line loss power value and static network loss rate are calculated. The weight parameters of the corresponding graph edges in the directed acyclic network are corrected based on the static network loss rate. Then, the process is reversed from the terminal graph node to the starting root node, verifying the on / off state markers included in each graph topology path and eliminating disconnected paths. The carbon emission value at the power source's starting point is used as the source emission factor of the starting root node. Based on the corrected weight parameters, state transition tracing is performed along the effective graph topology path to obtain the additional loss carbon emission value, which is then added to the carbon emission value at the power source's starting point to obtain the carbon intensity base of the terminal graph node.

[0100] Furthermore, dynamic network loss compensation is evaluated. Temperature fluctuations and transient power from adjacent time slices are extracted and input into a pre-trained multilayer neural network evaluation model, which outputs a thermal dissipation parameter. A polynomial fitting involving quadratic, linear, and constant terms is performed on this parameter to obtain a network loss calculation parameter. This parameter is then multiplied by the carbon intensity base to obtain the network loss compensation value.

[0101] Subsequently, a time-sharing carbon emission matrix is ​​generated and verified. Network loss compensation is superimposed on the carbon intensity baseline to obtain a corrected carbon intensity. This is compared with the quota warning and maximum threshold; if the limit is exceeded, an over-limit status feature flag is written and output. The data is encapsulated into a high-order time-sharing carbon emission matrix according to time and spatial dimensions, and the over-limit status feature is output separately.

[0102] Finally, a multi-source power dimension is introduced and the results are calculated. The instantaneous power output ratio is added as a new dimension and incorporated into the time-sharing carbon emission matrix to form a source-grid carbon coupling tensor. Within the complete daily scheduling window, tensor orthogonal decomposition is performed to separate the dimension-reduced aggregation vector and the orthogonal residual matrix. Using inverse decoding and mask splitting, the basic carbon emission equivalent and the power source correction energy parameter are restored. The unit energy emission reduction factor is determined, and the unit energy emission reduction factor is multiplied by the power source correction energy parameter, and then added to the basic carbon emission equivalent parameter to obtain the comprehensive carbon emission result for the charging of the fleet in the logistics park.

[0103] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for calculating the implicit carbon emissions of time-sharing charging of urban electric vehicles, characterized in that, include: Acquire remote sensing images, map coordinates, cross-regional power supply, power flow status of the grid, output parameters of multiple power sources, nodal load parameters including transient power values ​​and state of charge percentage, external environmental parameters including external temperature and external wind speed values, as well as sampling step size, reference thermal time constant, and carbon emission value at the power source. Using a time synchronization mechanism, meteorological features extracted from remote sensing images and map point coordinates are aligned and fused to establish spatial relationships and generate node base vectors. A directed acyclic network is constructed using cross-regional interconnection power and grid power flow status, and weight parameters are configured. The node base vector is mapped to the terminal graph node of the directed acyclic network. The carbon emission value at the power source is used as the source emission factor of the root node at the source. According to the weight parameters, the tracking calculation is performed along the graph topology path to output the carbon intensity base. The equivalent charging power is obtained by applying a nonlinear charging attenuation weight to the transient power value based on the percentage of state of charge; the original temperature difference fluctuation parameter is calculated based on the external temperature values ​​of adjacent time slices, and a first-order low-pass filter is performed based on the sampling step size and the reference thermal time constant to obtain the equivalent temperature difference fluctuation parameter. The equivalent temperature difference fluctuation parameter, equivalent charging power, external wind speed value, and state of charge percentage are imported into the network loss assessment model to output the heat dissipation parameter. The network loss conversion parameter is determined based on the heat dissipation parameter. The network loss conversion parameter is multiplied by the carbon intensity base to obtain the network loss compensation value. The network loss compensation value is superimposed and merged into the carbon intensity base to obtain the corrected carbon intensity parameter. The corrected carbon intensity parameters are arranged according to the time series to generate a time-series carbon emission matrix. The time-sharing carbon emission matrix and the output parameters of multiple power sources are encapsulated into a source-grid carbon coupling tensor. The source-grid carbon coupling tensor is subjected to dimension reduction and aggregation processing to obtain a dimension reduction and aggregation vector. Feature inversion mapping is performed on the dimension reduction and aggregation vector to obtain the basic carbon emission equivalent parameter and the power source correction power parameter, thus generating the carbon emission result.

2. The method for calculating the implicit carbon emissions of time-sharing charging of urban electric vehicles according to claim 1, characterized in that, The processing of remote sensing images, map coordinates, inter-regional power transmission, grid power flow status, regional carbon emission quotas, multi-source power output parameters, node load parameters, and external environmental parameters includes: capturing remote sensing images and map coordinates containing spatial positioning parameters by connecting to remote sensing mapping nodes; collecting inter-regional power transmission and grid power flow status through remote terminals, setting the grid power flow status to include the grid topology matrix and line power parameters, with the grid topology matrix including on / off status markers; retrieving regional carbon emission quotas including maximum threshold parameters and warning threshold parameters; retrieving multi-source power output parameters including instantaneous output ratio values, the multi-source power output parameters including the instantaneous output ratios of photovoltaic, wind, and geothermal power; recording node load parameters including transient power values ​​and state of charge percentages through load nodes; capturing external environmental parameters through meteorological probes, determining that external environmental parameters include external wind speed and external temperature values; obtaining the unit load rate, interpolating the unit load rate in a preset multi-dimensional carbon emission intensity characteristic curve to obtain the carbon emission value at the power source.

3. The method for calculating the implicit carbon emissions of time-sharing charging of urban electric vehicles according to claim 2, characterized in that, The processing of the node base vector includes: performing multispectral separation on the remote sensing image to extract surface temperature and light intensity parameters and construct meteorological features; configuring sliding time window slices and extracting timestamp labels and spatial positioning parameters attached to the meteorological features based on the time synchronization mechanism; constructing a two-dimensional grid array according to the spatial positioning parameters, projecting the meteorological features into the corresponding coordinate intervals contained in the two-dimensional grid array, performing alignment and fusion operations, establishing spatial correlation, and outputting cross-dimensional attribute values; and assembling the cross-dimensional attribute values ​​into a feature set matrix. Perform normalization scaling on the feature set matrix to smooth the distribution range of parameter values ​​and merge them to generate node basic vectors.

4. The method for calculating the implicit carbon emissions of time-sharing charging of urban electric vehicles according to claim 1, characterized in that, The processing steps for the directed acyclic network (DAN) include: parsing the network topology matrix contained in the network power flow state using a graph computing engine; performing a loop-breaking operation on the network topology matrix by calling the power flow ratio sharing algorithm to establish a unidirectional power flow state graph; decomposing the unidirectional power flow state graph into initial root nodes and intermediate transition nodes, setting terminal graph nodes in conjunction with spatial positioning parameters, and establishing an initial network architecture; parsing the line power parameters contained in the network power flow state, configuring the weight parameters of the edges of the DAN network graph based on the line power parameters; injecting the cross-regional interconnection power as a constraint condition into the initial network architecture, and simultaneously projecting the node base vectors into the terminal graph nodes for feature mapping according to the spatial positioning parameters and weight parameters, outputting the DAN network.

5. The method for calculating the implicit carbon emissions of time-sharing charging of urban electric vehicles according to claim 4, characterized in that, The carbon intensity baseline processing includes: calling a breadth-first search algorithm to traverse the directed acyclic network and identify graph topology paths pointing to the terminal graph nodes; extracting the on / off state markers contained in the network topology matrix, performing connectivity checks based on the on / off state markers, eliminating disconnected graph topology paths, and retaining valid graph topology paths; using the carbon emission value at the power source as the source emission factor of the root node, and based on the weight parameters of the edges in the directed acyclic network graph, performing state transition tracking calculations along the valid graph topology paths from the root node to the terminal graph nodes to obtain the carbon intensity baseline of the terminal graph nodes.

6. The method for calculating the implicit carbon emissions of time-sharing charging of urban electric vehicles according to claim 1, characterized in that, The processing steps for the network loss compensation values ​​include: calculating the power mean and power variance of the transient power values ​​within the sliding time window; inputting the equivalent temperature difference fluctuation parameter, equivalent charging power, external wind speed value, state of charge percentage, power mean, and power variance into the network loss assessment model to calculate the heat dissipation parameter; performing quadratic polynomial fitting processing on the heat dissipation parameter to output the network loss conversion parameter; extracting the carbon intensity base value, and performing floating-point multiplication operation on the network loss conversion parameter and the carbon intensity base value to obtain the network loss compensation value.

7. The method for calculating the implicit carbon emissions of time-sharing charging of urban electric vehicles according to claim 1, characterized in that, The processing steps of the time-sharing carbon emission matrix include: performing a scalar addition operation on the network loss compensation value and the carbon intensity base to obtain the corrected carbon intensity parameter; obtaining the highest threshold parameter and the warning threshold parameter included in the regional carbon emission quota; comparing the corrected carbon intensity parameter and the warning threshold parameter, and activating a preset over-limit marking mechanism when the corrected carbon intensity parameter exceeds the warning threshold parameter but is lower than the highest threshold parameter, and outputting the over-limit status feature bit corresponding to the corrected carbon intensity parameter; outputting an extreme value alarm parameter when the corrected carbon intensity parameter exceeds the highest threshold parameter; and arranging the corrected carbon intensity parameters corresponding to each time slice into a matrix form according to the time sequence to generate the time-sharing carbon emission matrix.

8. The method for calculating the implicit carbon emissions of time-sharing charging of urban electric vehicles according to claim 7, characterized in that, The process of processing the dimension reduction and aggregation vector includes: extracting the instantaneous output ratio values ​​contained in the multi-source power output parameters; setting the instantaneous output ratio values ​​as an additional feature dimension, concatenating the additional feature dimension into the time-sharing carbon emission matrix to construct a source-grid carbon coupling tensor; defining a calculation time window, starting the tensor orthogonal decomposition algorithm, extracting key principal component parameters, generating an approximate reconstruction tensor, subtracting the source-grid carbon coupling tensor from the approximate reconstruction tensor element by element to generate and cache an orthogonal residual matrix; and aggregating all key principal component parameters to generate a dimension reduction and aggregation vector.

9. The method for calculating the implicit carbon emissions of time-sharing charging of urban electric vehicles according to claim 8, characterized in that, The carbon emission result processing includes: retrieving the cached orthogonal residual matrix; performing inverse decoding on the dimension-reduced aggregated vector and the orthogonal residual matrix, performing decoupling and splitting processing based on preset built-in transformation rules to obtain the carbon emission intensity benchmark and the power correction intensity benchmark, performing time integral conversion in combination with the transient power values ​​contained in the node load parameters to restore the basic carbon emission equivalent parameters and the power correction energy parameters; retrieving the preset unit energy emission reduction factor for the power correction energy parameters; the unit energy emission reduction factor contains emission reduction offset attributes that match the instantaneous output ratio values; The carbon emission result is generated by adding the product of the basic carbon emission equivalent parameter, the emission reduction factor per unit of electricity, and the power source correction electricity parameter.

Images