Passenger car carbon emission monitoring and regulation system and method based on multi-source spatio-temporal data

Abstract

The application discloses a passenger car carbon emission monitoring and regulation system and method based on multi-source space-time data, comprising: a data fusion module for processing multi-source data and outputting space-time aligned fusion data; a carbon emission tracing and user analysis module for calculating vehicle-level carbon emission contribution and dividing user groups based on the fusion data; a multi-level regional accounting and early warning module for receiving and based on the vehicle-level carbon emission contribution and user group division results, and generating early warning signals by comparing preset targets; a regulation simulation and feedback module for simulating and generating regulation instructions according to the early warning signals, and feeding back the effect data after the instruction execution to the carbon emission tracing and user analysis module. The application realizes the tracing of massive mobile source carbon emissions from a macro region to specific vehicles, driving sections and time segments by constructing a data fusion-vehicle-level carbon emission calculation-user group clustering analysis chain, and solves the industry problem.

Description

Technical Field

[0001] This invention relates to the field of digital transportation environment technology, and in particular to a system and method for monitoring and controlling carbon emissions of passenger vehicles based on multi-source spatiotemporal data. Background Technology

[0002] As the global response to climate change deepens, the green and low-carbon transformation of the transportation sector, particularly passenger vehicles, has become a crucial link. Currently, several technological approaches have been developed within the industry for monitoring and managing carbon emissions from passenger vehicles. Mainstream practices focus on the collection and analysis of macro-level statistical data. For example, aggregated indicators such as total regional fuel consumption, charging volume, and the proportion of new energy vehicles in the vehicle fleet are used to assess overall carbon emission levels and transformation progress. Simultaneously, some systems indirectly estimate the contribution of transportation emissions by deploying monitoring equipment on key urban roads or utilizing satellite remote sensing data to invert regional air pollutant concentrations. In terms of management decision support, a common method is to set fixed annual or phased emission reduction targets and thresholds based on historical data and experience, triggering alarms when macro-level statistical indicators exceed preset red lines. These technological solutions constitute the foundational tools for current carbon emission control and trend assessment, providing certain data references for policy formulation.

[0003] However, the existing technological systems mentioned above have significant limitations and shortcomings in supporting the needs of precise, dynamic, and intelligent governance. First, at the monitoring and tracing level, existing methods rely on macro-statistics or regional concentration inversion, resulting in coarse data granularity and low spatiotemporal resolution. This makes it impossible to achieve precise carbon emission contribution tracing from specific vehicles, road sections, to specific time points, leading to a lack of precise targeting in governance measures and making it difficult to implement refined management based on the principle of "whoever emits, is responsible." Second, at the decision support level, the early warning mechanism based on static thresholds is rigid and unable to dynamically respond to complex factors such as traffic flow fluctuations, changes in grid cleanliness, and policy interventions, resulting in insufficient accuracy and foresight in early warnings. Furthermore, there is a lack of intelligent and parameterized correlation models between early warning signals and specific control measures. The process from "identifying problems" to "forming executable solutions" heavily relies on human experience, and the efficiency and scientific rigor of decision-making need improvement. Crucially, most existing systems are open-loop architectures, meaning that real-world data on policy implementation cannot be systematically fed back and used to optimize monitoring models and decision-making algorithms. System models are prone to failure over time and with changing environments, lacking adaptive and continuous learning capabilities, making it difficult to support long-term, dynamic transformation process management. These shortcomings have hindered the deep evolution of the carbon emission control system towards digitalization, intelligence, and precision. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, this invention provides a passenger vehicle carbon emission monitoring and control system and method based on multi-source spatiotemporal data, which solves the problem that the prior art cannot achieve accurate carbon emission source tracing, dynamic intelligent early warning and closed-loop adaptive optimization.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A system and method for monitoring and controlling carbon emissions from passenger vehicles based on multi-source spatiotemporal data, including:

[0007] The data fusion module is used to process multi-source data and output spatiotemporally aligned fused data;

[0008] The carbon emission traceability and user analysis module is used to calculate vehicle-level carbon emission contributions and classify user groups based on the fused data. This module includes a carbon emission traceability subunit and a user clustering subunit, configured with a driving behavior correction factor model and a clustering model, respectively, to undertake the tasks of carbon emission calculation and user segmentation. The driving behavior correction factor model is a mathematical model that corrects standard operating condition energy consumption in real time based on vehicle speed (s) and acceleration (a), and its expression is F(s,a). It is implemented through a vehicle bench test calibration coefficient table or a feedforward neural network trained based on historical driving segment data. The structure of the neural network is: input layer (2 nodes, corresponding to s,a) → hidden layer 1 (32 nodes, ReLU activation) → hidden layer 2 (16 nodes, ReLU activation) → output layer (1 node, Sigmoid activation). The core algorithm of the clustering model is K-means++, and its input feature vector is user generation, city level of residence, and average weekly commuting mileage. The features are Z-score standardized before clustering.

[0009] The multi-level regional accounting and early warning module is used to receive and aggregate multi-level regional carbon emission and low-carbon penetration rate indicators based on the vehicle-level carbon emission contribution and user group segmentation results, and generate early warning signals by comparing with preset targets.

[0010] The control simulation and feedback module is used to simulate and generate control commands based on the warning signal, and feed back the effect data after the command execution to the carbon emission tracing and user analysis module to drive model parameter optimization. Specifically, it includes: using the mini-batch gradient descent method, with the newly collected vehicle energy consumption data after command execution as the supervision signal, to update the weight parameters of the driving behavior correction factor neural network; retraining the user clustering model with the same new dataset to update the cluster centers; and retraining the time series prediction model based on the updated carbon emission data to iteratively optimize the dynamic decision threshold, thereby forming a monitoring-early warning-control closed loop with continuous learning capabilities.

[0011] Preferably, the data fusion module is specifically used for: cleaning, coordinate transformation, and time synchronization of multi-source heterogeneous data from vehicle terminals, traffic monitoring equipment, and consumer surveys, uniformly mapping them to a standard spatial grid and time slices, and outputting fused data with a unified spatiotemporal code; the spatial grid uses Geohash encoding with a precision of 7 bits, corresponding to an area of ​​approximately 153 meters × 153 meters on the ground; the time slices have a fixed duration of 5 minutes; the cleaning operation includes removing GPS points that exceed the city boundary, correcting obviously abnormal vehicle speed values, and using linear interpolation to fill in short-term missing timestamp data.

[0012] Preferably, the carbon emission traceability and user analysis module includes:

[0013] The carbon emission traceability subunit is used to calculate and generate a carbon emission contribution value dataset indexed by vehicle, spatial grid, and time slice based on the fused data by associating real-time energy consumption of vehicles with dynamic grid carbon emission factors. The dynamic grid carbon emission factors are obtained from real-time power generation combination data of the regional power grid dispatch center through API interface and are updated every 15 minutes. The unit also optimizes the calculation accuracy of carbon emission contribution value by using its own configured driving behavior correction factor model.

[0014] The user clustering subunit is used to perform machine learning clustering based on user attributes and behavioral characteristics in the fused data, output user group classifications with differentiated low-carbon preferences, and perform clustering operations through its own configured clustering model. The model parameters can be iteratively optimized through feedback data.

[0015] Preferably, the carbon emission traceability subunit is specifically used for: calling the baseline energy consumption value according to the vehicle identification code, calculating the instantaneous energy consumption through the vehicle specific power model in combination with the real-time vehicle speed and acceleration, querying the grid carbon emission factor corresponding to the grid and time slice, multiplying them to obtain the carbon emission contribution value, and introducing the driving behavior correction factor model parameters configured by itself during the calculation process to dynamically correct the instantaneous energy consumption and improve the accuracy of the carbon emission contribution value; the vehicle specific power model adopts a simplified calculation formula VSP=v×(a+0.092)+0.00021×v³, where v is the vehicle speed (m / s) and a is the acceleration (m / s²), and its parameters are calibrated based on typical urban road conditions.

[0016] Preferably, the user clustering subunit is specifically used for: using user generation, place of residence city level, and commuting mileage as core feature vectors, performing clustering using the K-means++ algorithm, and determining the group division through silhouette coefficient evaluation. The K-means++ algorithm is the core algorithm of the clustering model of this subunit, and its parameters such as the number of clusters and initial centers are parameters to be optimized in the model. User age generation is divided according to birth year into: first generation (1995-2009), second generation (1980-1994), and third generation (1965-1979); the place of residence city level is determined according to preset population and economic scale thresholds; commuting mileage is taken from the weekly average driving mileage data of the vehicle terminal.

[0017] Preferably, the multi-level regional accounting and early warning module includes:

[0018] The regional accounting unit is used to aggregate carbon emission contribution and low-carbon vehicle data level by level based on the mapping relationship between spatial grid and predefined spatial hierarchy, and generate time series of carbon emission intensity and penetration rate at each level; the aggregation adopts the spatial key value grouping aggregation operation of the distributed computing framework.

[0019] The intelligent early warning unit is used to compare the time series with the regional stage carbon emission control targets, generate dynamic judgment thresholds through the time series prediction model, and output structured early warning signals when the indicators continuously exceed the thresholds. The time series prediction model is a gated recurrent unit neural network, which is trained using monthly data from the past 24 months. In addition to historical carbon emission intensity, the input features also include the average temperature of the month, the number of holidays, and the number of newly registered new energy vehicles.

[0020] Preferably, in the intelligent early warning unit, the time-series prediction model is a gated recurrent neural network; the logic for generating the early warning signal is: when the core indicator of a certain area exceeds the dynamic threshold for N consecutive cycles, where N ranges from 3 to 6 cycles, or when the exceedance reaches M%, it is triggered, where M ranges from 10% to 30%.

[0021] Preferably, the control simulation and feedback module includes:

[0022] The simulation unit is used to match and generate a set of control instructions containing control tools, target objects, and intensity parameters based on a preset rule base according to the type and attributes of the warning signal. The rule base adopts a decision tree structure, for example: IF warning type is "high carbon intensity" AND target is "high emission grid cluster" THEN control tool = "congestion charge", intensity parameter = base rate × (1 + (actual value - threshold) / threshold).

[0023] The feedback optimization unit is used to transmit new data after the command execution to update the carbon emission calculation and user clustering models, and iteratively optimize the warning threshold. Specifically, it updates the driving behavior correction factor model parameters of the carbon emission tracing subunit and the clustering model parameters of the user clustering subunit, and synchronously adapts and optimizes the warning threshold. This feedback optimization is automatically triggered after each round of control command implementation and the completion of a full calendar month of data collection. After the model is updated, the system uses an independent validation dataset to evaluate the performance. Only when the root mean square error of the driving behavior correction factor model on the validation set decreases by more than a preset first threshold, or the user clustering silhouette coefficient increases by more than a preset second threshold, is the update confirmed as effective and the online model replaced.

[0024] A preferred method for monitoring and controlling carbon emissions from passenger vehicles based on multi-source spatiotemporal data includes the following steps:

[0025] S1: Through the system's data fusion module, multi-source heterogeneous data from vehicle terminals, traffic monitoring equipment, and consumer surveys are cleaned, coordinate transformed, and time synchronized. The data is then uniformly mapped to a standard spatial grid and time slice, and fused data with unified spatiotemporal coding is output to complete the spatiotemporal fusion processing.

[0026] S2: Based on the fused data, the system's carbon emission traceability and user analysis module performs the following operations in parallel:

[0027] S2a: The carbon emission traceability subunit of this module performs the calculation of vehicle carbon emission contribution, calls the vehicle's baseline energy consumption value, calculates the instantaneous energy consumption by combining real-time driving status parameters, introduces the driving behavior correction factor model parameters configured in this subunit to dynamically correct the instantaneous energy consumption, and then associates the dynamic power grid carbon emission factor to generate a carbon emission contribution value indexed by vehicle, spatial grid, and time slice. The correction factor F(s,a) is obtained by querying a pre-calibrated three-dimensional interpolation table or by performing neural network forward propagation calculation.

[0028] S2b: The user clustering subunit of this module performs user group segmentation. Based on user attributes and behavioral characteristics in the fused data, the clustering model configured by this subunit is adopted. The core algorithm is the K-means++ algorithm for machine learning clustering, which outputs user group classification with differentiated low-carbon preferences. Before clustering, the commuting mileage feature is scaled to the [0,1] interval using the max-min normalization method.

[0029] S3: Through the system's multi-level regional accounting and early warning module, based on the mapping relationship between spatial grid and administrative level, carbon emission contribution and low-carbon vehicle data are aggregated level by level to generate time series of carbon emission intensity and penetration rate at each level. The time series is compared with the regional stage carbon emission control target, and a dynamic judgment threshold is generated through the time series prediction model to generate early warning judgment results.

[0030] S4: Through the simulation unit of the system's control simulation and feedback module, based on the type and attributes of the early warning judgment result, a control instruction set containing control tools, target objects and intensity parameters is generated by matching based on a preset rule base.

[0031] S5: Through the feedback optimization unit of the control simulation and feedback module, the effect data after the command execution is sent back to step S2, driving the driving behavior correction factor model parameters of the carbon emission traceability subunit and the clustering model parameters of the user clustering subunit to be updated iteratively. The sliding window method is used to add the latest month's data to the training set and remove the earliest month's data to dynamically retrain the time series prediction model, thereby completing the online adaptive optimization of the warning threshold.

[0032] Preferably, step S1 specifically includes: accessing the vehicle CAN bus, roadside unit and survey data, and mapping them to a unified spatial grid and time slice after cleaning and alignment;

[0033] Step S2 specifically includes:

[0034] S2a: Through formula The carbon emission contribution is calculated point by point, where E is the carbon emission contribution. As the baseline energy consumption, CI represents the grid carbon intensity, v represents the vehicle identifier, g represents the spatial grid identifier, t represents the time slice identifier, and F is the core parameter of the carbon emission traceability subunit driving behavior correction factor model, which can be iteratively updated through the feedback data in step S5. For electric vehicles, The values ​​are taken from publicly available vehicle energy consumption catalogs; for gasoline vehicles, Estimated based on displacement and curb weight using empirical formulas;

[0035] S2b: Cluster analysis is performed based on user generation, region and travel characteristics. The K-means++ clustering model with user clustering sub-units is used to perform the analysis. The model parameters are optimized by the feedback data in step S5. The number of clusters K is determined by combining the elbow rule and the silhouette coefficient. The initial cluster centers are optimized and selected by the K-means++ algorithm.

[0036] Step S3 specifically includes: aggregating spatial grids to different predefined spatial levels to calculate indicators, and outputting early warnings by comparing target paths through prediction models. The early warnings are divided into three levels: Level 1 (observation, indicator exceeding the threshold ≤ 2 periods), Level 2 (warning, exceeding the threshold for 3 consecutive periods), and Level 3 (action, exceeding the threshold for 4 consecutive periods or exceeding the standard by more than 30% in a single period).

[0037] Step S4 specifically includes: matching control tools according to the warning type, and calculating intensity parameters according to the target gap. For example, for the "Level 3 - High Carbon Intensity" warning, the "Time-based and Zone-based Differentiated Parking Fee" tool is matched, and the intensity parameter (rate increase ratio) has a piecewise linear relationship with the emission overrun.

[0038] Step S5 specifically includes: updating the driving behavior correction factor model using the gradient descent algorithm and periodically retraining the warning model. During the update, the learning rate is set to 0.001, the Adam optimizer is used, and iterative training is performed until the validation set loss function converges.

[0039] The technical effects and advantages of the passenger vehicle carbon emission monitoring and control system and method based on multi-source spatiotemporal data of this invention are as follows:

[0040] 1. This invention, by constructing an analysis chain of "data fusion - vehicle-level carbon emission calculation - user group clustering," has for the first time achieved three-dimensional, refined source tracing of massive mobile source carbon emissions, from macro-regional levels to specific vehicles, road segments, and time periods, solving the industry problem of "inaccurate measurement and unclear tracing." Simultaneously, the system, through a decision-making chain of "multi-level regional accounting - intelligent early warning - control simulation," automatically matches and transforms abstract early warning signals into quantifiable and adjustable specific control instructions (such as subsidy amounts and traffic constraint strength), achieving an intelligent leap from data insight to strategy generation, transforming the decision-making process from relying on experience-based judgment to rule-based parametric simulation.

[0041] 2. This invention's system can automatically feed back the actual effect data after the execution of control commands to the upstream carbon emission calculation and user analysis models, driving iterative optimization of early warning thresholds. This transforms the system from a static tool into a living entity with continuous learning and self-calibration capabilities. The system can proactively adapt to data distribution shifts caused by factors such as changes in driving behavior, policy interventions, and social development, ensuring the accuracy of long-term monitoring and the timeliness of early warnings. It fundamentally overcomes the shortcomings of traditional systems that fail due to outdated models, achieving sustainable evolution of governance capabilities. Attached Figure Description

[0042] Figure 1 This is a flowchart of the passenger vehicle carbon emission monitoring and control system and method based on multi-source spatiotemporal data proposed in this invention. Detailed Implementation

[0043] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0044] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include," "contain," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "includes..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0045] refer to Figure 1 This invention discloses a passenger vehicle carbon emission monitoring and control system and method based on multi-source spatiotemporal data. The system includes: a data fusion module, used to clean, transform, and synchronize the coordinates of multi-source heterogeneous data from vehicle terminals, traffic monitoring equipment, and consumer surveys, uniformly mapping it to a standard spatial grid and time slice, and outputting a spatiotemporally aligned fused data stream; a carbon emission tracing and user analysis module, used to calculate carbon emission contribution values ​​indexed by vehicle, spatial grid, and time slice based on the fused data, by correlating real-time vehicle energy consumption with dynamic grid carbon emission factors, and performing machine learning clustering based on user attributes and behavioral characteristics to segment user groups with differentiated low-carbon preferences; and a multi-level regional accounting and early warning module. Based on the mapping relationship between spatial grids and administrative levels, this system aggregates carbon emission and low-carbon vehicle data level by level to generate regional-level carbon emission intensity and penetration time series. A gated recurrent unit neural network prediction model then compares these series with preset targets, generating warning signals when indicators continuously exceed dynamic thresholds. The control simulation and feedback module, based on the type and attributes of the warning signals and using a pre-set rule base, matches and simulates to generate a control instruction set containing control tools, target objects, and intensity parameters. It then feeds back the effect data after instruction execution to update the carbon emission calculation and user clustering models, iteratively optimize warning thresholds, and drive adaptive optimization of system parameters. This system achieves closed-loop carbon emission management from micro-vehicles to macro-regions, and from precise monitoring to intelligent control.

[0046] Example 1

[0047] This embodiment provides a passenger vehicle carbon emission monitoring and control system and method based on multi-source spatiotemporal data, serving as an example of overall system operation and multi-module collaboration. Specific implementation details include:

[0048] Purpose of implementation:

[0049] This embodiment aims to demonstrate how the core modules of the monitoring and control system work together to complete a closed-loop process from data acquisition, detailed analysis, intelligent early warning to simulation control.

[0050] Implementation System:

[0051] This embodiment involves all the core modules of the system: data fusion module, carbon emission tracing and user analysis module, multi-level regional accounting and early warning module, and control simulation and feedback module. The system is deployed on the traffic big data platform of a certain city (hereinafter referred to as "City A").

[0052] Implementation steps:

[0053] Step one involves the data fusion module concurrently accessing real-time CAN bus data (including vehicle speed and location) from approximately 500,000 connected vehicles in City A, traffic flow data from major intersections throughout the city, and 100,000 integrated user attribute data sets. This module divides the entire City A into 500-meter-side spatial grids, using a 5-minute "time slice." After cleaning all the data, it maps it uniformly to a "grid-time slice" coordinate system, forming a spatiotemporally aligned fused data stream. The cleaning rules include: removing abnormal trajectory points with speeds exceeding 200 km / h or absolute acceleration values ​​exceeding 5 m / s²; filling in missing timestamps within 30 seconds using linear interpolation; and converting all GPS coordinates to the WGS-84 coordinate system.

[0054] Step two: The carbon emission traceability and user analysis module receives the fused data stream. On one hand, for each vehicle traveling within a specific grid and time period, a baseline energy consumption is obtained based on its vehicle type. Combining the vehicle's average speed *s* and acceleration *a* within this time slice, the correction factor is obtained by querying the local driving behavior correction coefficient table. And obtain the current carbon intensity of the grid from the power grid data interface. According to the formula The carbon emission contribution is calculated, where v represents the vehicle identifier, g represents the spatial grid identifier, and t represents the time slice identifier. On the other hand, the user clustering subunit is based on three features: user age, residential area size, and average weekly mileage. After feature standardization, the K-means++ algorithm is used to divide all users into four typical groups (e.g., "young high-mileage group," "middle-aged urban commuter group," etc.). The clustering quality is verified by calculating the silhouette score, ensuring its value is greater than 0.5.

[0055] Step three: The multi-level regional accounting and early warning module first spatially aggregates the massive vehicle-level carbon emission data obtained from the calculations according to the administrative levels of streets, districts, and cities to obtain the daily total carbon emissions and new energy vehicle penetration rate for each level of region. Subsequently, the early warning unit inputs the monthly carbon emission intensity sequence of Area B over the past year into a time-series prediction model (in this embodiment, a GRU neural network model) for training, predicting its emission trend for the following month, and dynamically generating an early warning threshold based on the area's emission reduction target for the following month. The dynamic threshold calculation formula is: Threshold = Target value × (1 + 0.05), where 0.05 is a buffer coefficient. Monitoring revealed that the carbon emission intensity of Area B has exceeded its dynamic threshold for three consecutive months, and 40% of its subordinate streets have also triggered early warnings. Therefore, the system determines and generates a Level II high-carbon scenario early warning signal.

[0056] Step four: After receiving the early warning signal, the control simulation and feedback module calculates the control intensity parameter of "increasing the parking fee rate of the core road section by 20%" based on the preset rules (the "Regional Traffic Management" tool corresponds to the Level 2 high-carbon scenario early warning) and the total excess emissions in Zone B, using a stepped response function, and generates a control instruction. One month after the instruction is executed, the system collects new vehicle trajectory and energy consumption data and feeds it back to the carbon emission tracing module, triggering the parameter update process of the driving behavior correction factor model. This allows the system to more accurately reflect changes in driving behavior after the policy implementation, thereby completing closed-loop optimization.

[0057] Implementation results:

[0058] This embodiment fully demonstrates how the system achieves a closed loop of "data-information-decision-optimization". It specifically supports the core features of the claims regarding multi-module connection to form a closed-loop system, carbon emission traceability calculation, user group segmentation, multi-level regional aggregation, dynamic early warning generation, and feedback-based model optimization, proving the feasibility and integrity of the system.

[0059] Example 2

[0060] This embodiment provides a passenger vehicle carbon emission monitoring and control system and method based on multi-source spatiotemporal data, used for the implementation of spatiotemporal data fusion. Specific implementation details include:

[0061] Purpose of implementation:

[0062] This embodiment aims to explain in detail how to achieve accurate spatiotemporal alignment of multi-source heterogeneous data, providing a unified benchmark for subsequent analysis.

[0063] Implementation System:

[0064] This embodiment mainly involves the specific implementation of the data fusion module.

[0065] Implementation steps:

[0066] Step 1: Spatial Alignment. The system uses the Geohash encoding algorithm to recursively binary divide the longitude interval [-180, 180] and the latitude interval [-90, 90], using an encoding string of length 7. This encoding corresponds to a grid of approximately 153 meters × 153 meters on the ground. For each vehicle GPS point... The algorithm library function is called to calculate its 7-bit Geohash grid code, where, Longitude For dimension.

[0067] Step 2, Time Alignment. The system defines time slices starting at 00:00 each day, with a length of 5 minutes (300 seconds). For any data record with timestamp t, its corresponding time slice number is... Through formula Confirmed, among which This is the timestamp at 00:00 on the current day, and ⌊⋅⌋ represents the floor function.

[0068] Step 3: Data Association and Output. Vehicle trajectory points and intersection traffic statistics are processed, and all data are assigned a "Geohash7 code" and a "time slice number". "Two labels. Data from different sources are correlated and fused using these spatiotemporal labels to output a structured fused data frame."

[0069] Implementation results:

[0070] This embodiment clarifies the specific technical means of data fusion (Geohash encoding, fixed-duration slicing, calculation formula), supports the feature of "outputting spatiotemporally aligned fused data" in the claims, and provides operable technical details to ensure that subsequent analysis is carried out on a unified spatiotemporal reference.

[0071] Example 3

[0072] This embodiment provides a passenger vehicle carbon emission monitoring and control system and method based on multi-source spatiotemporal data, used for constructing driving behavior correction factors and accurately calculating carbon emissions. Specific implementation details include:

[0073] Purpose of implementation:

[0074] This embodiment aims to reveal how to correct standard operating condition energy consumption to actual road energy consumption, i.e., the specific construction and application method of the "driving behavior correction factor", so as to achieve accurate vehicle-level carbon emission traceability.

[0075] Implementation System:

[0076] This embodiment relates to the carbon emission traceability subunit in the carbon emission traceability and user analysis module.

[0077] Implementation steps:

[0078] Step 1: Construction of the Correction Factor Model. This system employs a machine learning-based neural network model as the core implementation method for the driving behavior correction factor F(s,a) to overcome the shortcomings of the lookup table method in terms of data smoothness and generalization ability. The specific construction process is as follows:

[0079] Data preparation: More than 1 million valid "driving segments" were extracted from massive connected vehicle data, each segment lasting 5 minutes. The characteristics of each segment are average vehicle speed s (km / h) and average acceleration a (m / s²), and the label is the ratio of the actual calculated energy consumption of the segment to the vehicle's standard operating condition energy consumption.

[0080] Model Architecture: A lightweight, fully connected feedforward neural network is constructed. The input layer has two nodes, receiving (s, a). This is followed by two hidden layers with 32 and 16 nodes respectively, both using the ReLU activation function to introduce non-linearity. The output layer has one node, using the Sigmoid activation function to constrain the output within a reasonable range (e.g., 0.5 to 2.0), representing a correction factor. This network structure significantly differs from simple linear regression or lookup tables, enabling it to capture the complex non-linear relationship between driving behavior and energy consumption.

[0081] Model Training: The dataset was divided into training, validation, and test sets in an 8:1:1 ratio. Using the training set, the Adam optimizer was employed with mean squared error (MSE) as the loss function and an initial learning rate of 0.001. Early stopping was used to prevent overfitting. On the independent test set, the model's predicted values ​​achieved a determination coefficient (R²) of over 0.92, demonstrating its high accuracy and generalization ability. A lookup table method was used to calibrate energy consumption correction coefficients for different combinations of vehicle speeds (s) (0-120 km / h) and accelerations (a) (-2.5 to 2.5 m / s²) through vehicle bench tests. This table is a pre-calibrated static lookup table, representing an application of existing technology. A neural network method was used to construct a fully connected network with s and a as inputs and correction factors as outputs, and it was trained using a large amount of real-world road data.

[0082] Step 2, Real-time Carbon Emission Calculation. For vehicle v in grid g within time slice, the system executes the following calculation chain:

[0083] 1) Obtain its standard operating condition energy consumption based on the vehicle identification number. ;

[0084] 2) Based on the average vehicle speed *s* and acceleration *a* within the time slice, obtain the correction factor *F(s,a)* from the above model;

[0085] 3) Input (s,a) into the pre-trained neural network model for driving behavior correction factors, perform one forward propagation, and output the correction factor F(s,a) in real time. This process is completed in milliseconds, meeting the requirements of real-time computing;

[0086] 4) Query the real-time grid carbon intensity CI(g,t) of grid g in time slice t via API;

[0087] 5) Substitute into the core formula The carbon emission contribution of the vehicle in that spatiotemporal unit was calculated.

[0088] Implementation results:

[0089] This embodiment not only discloses the core formula for carbon emission calculation, but also delves into the construction method of the most crucial and innovative "driving behavior correction factor." By employing a neural network model with a specific structure and elaborating on the entire process from data preparation, model design, training to integration, it powerfully demonstrates that this invention does not simply apply existing models, but rather solves the industry problem of accurately estimating actual energy consumption by combining an innovative model architecture with specific data and methods in the field of transportation carbon emissions. This provides fully disclosed and robust technical support for "calculating vehicle-level carbon emission contributions."

[0090] Example 4

[0091] This embodiment provides a passenger vehicle carbon emission monitoring and control system and method based on multi-source spatiotemporal data, used for a multi-level regional early warning mechanism with dynamic thresholds. Specific implementation details include:

[0092] Purpose of implementation:

[0093] This embodiment aims to illustrate in detail how to aggregate micro-level carbon emission data to generate macro-level regional indicators, and how to use machine learning models to generate dynamic early warning thresholds and trigger tiered early warnings.

[0094] Implementation System:

[0095] This embodiment involves a multi-level regional accounting and early warning module.

[0096] Implementation steps:

[0097] Step 1: Multi-level regional indicator aggregation. For a district / county k, the system aggregates the carbon emission contributions of all vehicles in all its subordinate spatial grids within a day d, obtaining the total daily carbon emissions for the region. The daily low-carbon vehicle penetration rate in the district / county was calculated by combining vehicle registration data. The calculation formula is: (carbon emission intensity, (where k is the area of ​​the district / county). (penetration rate, Let k be the total number of vehicles in district / county k. (This refers to the number of new energy vehicles). A time-series sequence is generated daily.

[0098] Step two: Dynamic early warning threshold generation and triggering. The early warning unit uses the monthly carbon emission intensity series of the district / county over the past 12 months to train a time-series prediction model (in this embodiment, a GRU neural network model) to predict the trend value for the next month. The dynamic threshold is set as follows: ,in The target for the next month is defined by δ, which is a fixed buffer value, and σ is the standard deviation of the model prediction error. Early warning triggering uses a tiered logic: a Level 1 warning is triggered when the actual value exceeds the dynamic threshold for N consecutive months; if a lower-level sub-region exceeding a certain proportion (e.g., 30%) in the same region also triggers a warning, the warning is upgraded to a Level 2 warning.

[0099] Implementation results:

[0100] This embodiment clearly illustrates the implementation process of "aggregating and generating regional-level indicators" and "generating early warning signals by comparing with preset targets" through specific aggregation formulas and dynamic threshold generation algorithms. The introduction of a time-series prediction model to generate dynamic thresholds and the logic of tiered triggering effectively supports the specific technical structure of intelligent early warning as described in the claims, distinguishing it from static threshold early warning.

[0101] Example 5

[0102] This embodiment provides a passenger vehicle carbon emission monitoring and control system and method based on multi-source spatiotemporal data, used for control rule simulation and feedback optimization closed loop, and the specific implementation content includes:

[0103] Purpose of implementation:

[0104] This embodiment aims to demonstrate how the system transforms early warning signals into specific control parameters, and how it utilizes feedback data after policy implementation to achieve model self-optimization.

[0105] Implementation System:

[0106] This embodiment involves a control simulation and feedback module.

[0107] Implementation steps

[0108] Step 1: Simulation generation of control commands. The system has pre-set mapping rules between early warning types, targets, and control tools. For example, when receiving an "early warning of delayed transformation" and the target is "youth group", At that time, the mapping tool was "purchase subsidy". The system is based on the current penetration rate of this group. With target value gap The subsidy amount S is dynamically calculated using the linear response function S=A+B×G, generating executable control instructions. Here, A is the benchmark subsidy amount, and B is the subsidy intensity coefficient.

[0109] Step two: Feedback-driven model iterative optimization. New vehicle operation data is collected in the new monitoring cycle following the execution of the control command. The feedback optimization unit uses this new data as a benchmark to define a loss function. Where W is the weight vector of the driving behavior correction factor neural network, and M is the number of new data samples. The weights are updated using the gradient descent algorithm: , where η is the learning rate, and ∇L(W) is the gradient of the loss function with respect to the weights W. The prediction error is minimized through iteration, thereby completing the calibration of the core model and the re-optimization of the warning threshold.

[0110] Implementation results:

[0111] This embodiment empirically demonstrates the two key features of "simulation-generated control commands" and "feedback-driven model parameter optimization" through specific mapping examples, parameter calculation formulas, and gradient descent optimization processes. It reveals how the system constitutes an intelligent closed loop capable of self-learning and self-adjustment, rather than merely a static monitoring tool, strongly supporting the inventiveness of the claims.

[0112] Comparative Example 1

[0113] This comparative example provides a traditional static monitoring and early warning solution, the specific contents of which include:

[0114] Overview of the comparison system:

[0115] Traditional solutions typically consist of data acquisition, static databases, report statistics, and fixed threshold early warning modules. The data mainly comes from macro-level statistics (such as citywide fuel consumption and new vehicle registrations), lacking real-time vehicle-level trajectory and energy consumption data.

[0116] Workflow and defects:

[0117] The system generates macroeconomic statistical reports periodically (e.g., quarterly). The early warning module relies on pre-set absolute thresholds (e.g., total carbon emissions must not exceed...). When the report data exceeds this fixed threshold, a general over-limit alarm is triggered.

[0118] Its main drawback is:

[0119] First, it is impossible to trace carbon emissions at the vehicle or grid level, resulting in a lack of precise targeting in governance;

[0120] Second, static thresholds cannot adapt to changes in traffic flow and dynamic factors such as policy interventions, resulting in poor accuracy of early warnings.

[0121] Third, there is a lack of rule-based quantitative correlation between early warning and specific control measures, resulting in weak decision support;

[0122] Fourth, the system is completely open-loop, making it impossible to optimize its own model using execution performance data, resulting in diminishing performance over the long term.

[0123] Comparison results:

[0124] In the same market scenario as Example 1 (Market A), the traditional solution can only detect overall emissions "exceeding standards" in Zone B at the end of the quarter, but it cannot identify trends months in advance, nor can it pinpoint internal hotspot streets. Its early warnings cannot be linked to specific measures such as "increasing parking fees," and it cannot automatically evaluate the effects and optimize the model after the measures are implemented. Simulation analysis shows that the present invention can identify risks more than 40% earlier than the traditional solution, and improve the accuracy of predicting the effects of measures by more than 20%. This comparison significantly demonstrates the substantial progress of the present invention in both technical approach and practical effectiveness.

[0125] Comparing Examples 1-5 with Comparative Example 1, this invention, through a systematic comparison of Examples 1-5 with Comparative Example 1, clearly reveals the technological leap and substantial progress achieved by this invention compared to traditional static monitoring and early warning schemes. The fundamental flaw of the traditional scheme represented by Comparative Example 1 lies in its being an open-loop statistical system based on macroscopic, lagging, and isolated data: the data relies on annual or quarterly administrative summary reports, unable to break through the district / county level spatially, and suffers from severe temporal lag; the early warning mechanism relies on pre-set, unchanging absolute thresholds throughout the year, unable to respond to dynamic changes brought about by seasonal fluctuations in traffic flow, sudden economic activities, or policy interventions; its function is limited to issuing general "exceeding" alerts, with a huge "decision-making gap" between early warning and specific governance measures, relying entirely on the experience and judgment of managers, and the system itself lacks the ability to learn from the effects of policy implementation. Once the model and thresholds are set, they remain fixed, causing its effectiveness and reference value to rapidly diminish over time.

[0126] In contrast, the embodiments of this invention construct a dynamic closed-loop system that ranges from microscopic real-time perception to macroscopic intelligent decision-making and possesses self-optimization capabilities. At the data level, embodiments 2 and 3 achieve a qualitative leap from "macroscopic statistics" to "vehicle-grid-time period" level full-time and spatiotemporal fine-grained tracing. Through multi-source heterogeneous data fusion and a high-precision driving behavior correction vehicle power ratio model, the accuracy of carbon emission measurement is improved by several orders of magnitude. At the early warning level, embodiment 4 upgrades static thresholds to dynamic decision thresholds by introducing a time-series prediction model that integrates external features (such as Prophet+GRU) and quantile regression, enabling the early warning system to proactively identify risk trends and intelligently distinguish between reasonable fluctuations and abnormal exceedances. At the decision support level, embodiments 1 and 5, through a rule engine and parameterized simulation model, achieve an automated leap from "problem discovery" to "generating quantitative control schemes," capable of recommending executable instructions containing specific tools, objects, and intensity parameters for specific regions and groups. The most crucial breakthrough lies in the feedback optimization and continuous learning (MLOps) pipeline constructed in Example 5. This pipeline transforms the system into a self-evolving organism, automatically updating the core analytical model and early warning parameters using real-world data on the effects of policy implementation. This ensures the system's long-term effectiveness and completely resolves the fundamental problems of rigid models and diminishing returns in traditional systems. This series of synergistic technological improvements collectively achieves a fundamental paradigm shift in monitoring and early warning, moving from "post-event descriptive statistics" to "pre-event simulation and in-event adaptive control."

[0127] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of protection of the claims.

[0128] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A passenger vehicle carbon emission monitoring and control system based on multi-source spatiotemporal data, characterized in that, include: The data fusion module is used to process multi-source data and output spatiotemporally aligned fused data; The carbon emission traceability and user analysis module is used to calculate vehicle-level carbon emission contribution and classify user groups based on the fused data. This module includes a carbon emission traceability subunit and a user clustering subunit, which are respectively configured with a driving behavior correction factor model and a clustering model, and are responsible for carbon emission calculation and user classification tasks. The multi-level regional accounting and early warning module is used to receive and aggregate multi-level regional carbon emission and low-carbon penetration rate indicators based on the vehicle-level carbon emission contribution and user group segmentation results, and generate early warning signals by comparing with preset targets. The control simulation and feedback module is used to simulate and generate control commands based on the warning signals, and feed back the effect data after the command execution to the carbon emission traceability and user analysis module to drive model parameter optimization. Specifically, it includes the driving behavior correction factor model parameters of the carbon emission traceability subunit and the clustering model parameters of the user clustering subunit under the carbon emission traceability and user analysis module. The carbon emission calculation accuracy and user clustering rationality are improved through parameter optimization.

2. The passenger vehicle carbon emission monitoring and control system based on multi-source spatiotemporal data as described in claim 1, characterized in that, The data fusion module is specifically used to: clean, transform, and synchronize the coordinates of multi-source heterogeneous data from vehicle terminals, traffic monitoring equipment, and consumer surveys, and uniformly map them to a standard spatial grid and time slice, outputting fused data with unified spatiotemporal coding.

3. The passenger vehicle carbon emission monitoring and control system based on multi-source spatiotemporal data as described in claim 1 or 2, characterized in that, The carbon emission traceability and user analysis module includes: The carbon emission traceability subunit is used to calculate and generate a carbon emission contribution value dataset indexed by vehicle, spatial grid, and time slice based on the fused data by associating real-time energy consumption of vehicles with dynamic power grid carbon emission factors, and optimize the calculation accuracy of carbon emission contribution value by means of correction factors through its own configured driving behavior correction factor model. The user clustering subunit is used to perform machine learning clustering based on user attributes and behavioral characteristics in the fused data, output user group classifications with differentiated low-carbon preferences, and perform clustering operations through its own configured clustering model. The model parameters are iteratively optimized through feedback data.

4. The passenger vehicle carbon emission monitoring and control system based on multi-source spatiotemporal data as described in claim 3, characterized in that, The carbon emission traceability subunit is specifically used to: retrieve the baseline energy consumption value based on the vehicle identification code, calculate the instantaneous energy consumption by combining the real-time vehicle speed and acceleration with the vehicle specific power model, query the grid carbon emission factor corresponding to the grid and time slice, multiply them to obtain the carbon emission contribution value, and introduce the driving behavior correction factor model parameters configured by itself during the calculation process to dynamically correct the instantaneous energy consumption and improve the accuracy of the carbon emission contribution value.

5. The passenger vehicle carbon emission monitoring and control system based on multi-source spatiotemporal data as described in claim 3, characterized in that, The user clustering subunit is specifically used to: use user generation, city level of residence and commuting distance as core feature vectors, perform clustering using the K-means++ algorithm, and determine the group division by silhouette coefficient evaluation. The K-means++ algorithm is the core algorithm of the clustering model of this subunit, and its number of clusters and initial center parameters are the parameters to be optimized in the model.

6. The passenger vehicle carbon emission monitoring and control system based on multi-source spatiotemporal data as described in claim 1, characterized in that, The multi-level regional accounting and early warning module includes: Regional accounting units are used to aggregate carbon emission contributions and low-carbon vehicle data level by level based on the mapping relationship between spatial grids and administrative levels, and generate time series of carbon emission intensity and penetration rate at each level; The intelligent early warning unit is used to compare the time series with the regional stage carbon emission control targets, generate dynamic judgment thresholds through a time series prediction model, and output structured early warning signals when the indicators continuously exceed the thresholds.

7. The passenger vehicle carbon emission monitoring and control system based on multi-source spatiotemporal data as described in claim 6, characterized in that, In the intelligent early warning unit, the time-series prediction model is a gated recurrent neural network; The logic for generating early warning signals is as follows: when a core indicator in a certain area exceeds the dynamic threshold for N consecutive cycles, where N ranges from 3 to 6 cycles, or when the exceedance reaches M%, it is triggered, where M ranges from 10% to 30%.

8. The passenger vehicle carbon emission monitoring and control system based on multi-source spatiotemporal data as described in claim 1, characterized in that, The control simulation and feedback module includes: The simulation unit is used to match and generate a set of control instructions containing control tools, target objects and intensity parameters based on a preset rule base according to the type and attributes of the warning signal. The feedback optimization unit is used to send back new data after the instruction is executed to update the carbon emission calculation and user clustering models, and to iteratively optimize the warning threshold. Specifically, it updates the driving behavior correction factor model parameters of the carbon emission tracing subunit and the clustering model parameters of the user clustering subunit, and synchronously adapts and optimizes the warning threshold.

9. A method for monitoring and controlling carbon emissions of passenger vehicles based on multi-source spatiotemporal data, characterized in that, Includes the following steps: S1: Through the system's data fusion module, multi-source heterogeneous data from vehicle terminals, traffic monitoring equipment, and consumer surveys are cleaned, coordinate transformed, and time synchronized. The data is then uniformly mapped to a standard spatial grid and time slice, and fused data with unified spatiotemporal coding is output to complete the spatiotemporal fusion processing. S2: Based on the fused data, the system's carbon emission traceability and user analysis module performs the following operations in parallel: S2a: The carbon emission traceability subunit of this module performs the calculation of vehicle carbon emission contribution, calls the vehicle's baseline energy consumption value, calculates the instantaneous energy consumption by combining real-time driving status parameters, introduces the driving behavior correction factor model parameters configured in this subunit to dynamically correct the instantaneous energy consumption, and then associates the dynamic power grid carbon emission factor to generate a carbon emission contribution value indexed by vehicle, spatial grid, and time slice. S2b: The user clustering subunit of this module performs user group segmentation. Based on user attributes and behavioral characteristics in the fused data, it adopts the clustering model configured in this subunit. The core algorithm is the K-means++ algorithm for machine learning clustering, and outputs user group classification with differentiated low-carbon preferences. S3: Through the system's multi-level regional accounting and early warning module, based on the mapping relationship between spatial grid and administrative level, carbon emission contribution and low-carbon vehicle data are aggregated level by level to generate time series of carbon emission intensity and penetration rate at each level. The time series is compared with the regional stage carbon emission control target, and a dynamic judgment threshold is generated through the time series prediction model to generate early warning judgment results. S4: Through the simulation unit of the system's control simulation and feedback module, based on the type and attributes of the early warning judgment result, a control instruction set containing control tools, target objects and intensity parameters is generated by matching based on a preset rule base. S5: Through the feedback optimization unit of the control simulation and feedback module, the effect data after the command is executed is sent back to step S2, driving the driving behavior correction factor model parameters of the carbon emission traceability subunit and the clustering model parameters of the user clustering subunit to be updated iteratively, and simultaneously adapting and optimizing the warning threshold.

10. The method for monitoring and controlling carbon emissions of passenger vehicles based on multi-source spatiotemporal data as described in claim 9, characterized in that, Step S1 specifically includes: accessing the vehicle CAN bus, roadside unit and survey data, and mapping them to a unified spatial grid and time slice after cleaning and alignment; Step S2 specifically includes: S2a: Through formula The carbon emission contribution is calculated point by point, where E is the carbon emission contribution. The baseline energy consumption is CI, the grid carbon intensity is v, the vehicle identifier is g, the spatial grid identifier is t, the time slice identifier is F, and the core parameters of the carbon emission traceability subunit driving behavior correction factor model are updated iteratively through the feedback data in step S5. S2b: Cluster analysis is performed based on user generation, region and travel characteristics. The K-means++ clustering model configured with user clustering sub-units is used to perform the analysis. The model parameters are optimized by feedback data in step S5. Step S3 specifically includes: aggregating spatial grids into accounting indicators at different administrative levels, and outputting early warnings by comparing target paths through prediction models; Step S4 specifically includes: matching control tools according to the early warning type, and calculating intensity parameters based on the target gap; Step S5 specifically includes: updating the driving behavior correction factor model using the gradient descent algorithm, and periodically retraining the early warning model.

Images