Multi-source data driven automobile carbon emission characteristic space-time prediction method and system

Abstract

The invention relates to the technical field of carbon emission prediction, and discloses a multi-source data driven automobile carbon emission characteristic space-time prediction method and system, and the method comprises the steps: collecting multi-source data, carrying out the preprocessing, and constructing a multi-factor carbon emission data set; performing factor analysis on the data set, and extracting three principal component factors; a CNN-LSTM mixed deep learning model containing an attention mechanism is constructed to predict carbon emission; calculating carbon emission intensity of different spatial scales; based on a Gaussian diffusion theory, constructing a carbon emission diffusion model under a multi-wind-speed condition, and calibrating a diffusion coefficient by adopting measured data of an air monitoring station; modeling is carried out on the fused spatio-temporal data, and a carbon emission intensity prediction value of a grid scale in a specified time period in the future is output; performing visual display on the prediction result, and triggering early warning and generating a corresponding ecological driving control strategy based on the prediction value; the problem that an existing carbon emission prediction mode is low in prediction precision is solved.

Description

Technical Field

[0001] This invention relates to the field of carbon emission prediction technology, and in particular to a multi-source data-driven spatiotemporal prediction method and system for automobile carbon emission characteristics. Background Technology

[0002] With the increasingly severe global climate change problem, greenhouse gas emission reduction has become an international consensus. As one of the main sources of carbon emissions, vehicle exhaust emissions have received significant attention for reduction efforts. Statistics show that the global transportation sector accounts for 24% of total carbon emissions, with automobiles, as the primary mode of transportation, contributing nearly 70% of these emissions. Accelerated urbanization has driven rapid growth in vehicle ownership. This surge has not only exacerbated urban traffic congestion but also brought about serious carbon emission problems. In some first-tier cities, vehicle carbon emissions account for over 30% of total urban carbon emissions, becoming a key bottleneck restricting the improvement of the urban ecological environment and the achievement of "dual carbon" goals. Currently, the control and management of automobile carbon emissions urgently requires scientific and accurate carbon emission prediction technologies. In recent years, the rapid development of deep learning technology has provided a new path for solving the prediction problem of complex systems—deep learning models have powerful capabilities for processing massive and complex data, and the hybrid application of different deep learning models can further improve prediction accuracy and practicality, creating conditions for achieving higher accuracy in carbon emission prediction. However, existing automobile carbon emission prediction methods still have many technical limitations and are difficult to meet the application needs in complex scenarios, especially in the three core areas of data acquisition and utilization, model closed-loop optimization, and result visualization. From a technical practice perspective, existing research and patented technologies have not yet formed a complete technology chain encompassing "data acquisition - model building - evaluation and optimization - visualization." On the one hand, traditional prediction models often focus on the impact of single factors on carbon emissions, such as building prediction models based solely on isolated indicators like engine parameters, mileage, or traffic flow, neglecting the coupling effect of multiple factors within the "human-road-vehicle" system. In real-world scenarios, vehicle carbon emissions are comprehensively affected by multiple dimensions of factors, including climate change (such as temperature, precipitation, and wind speed), traffic flow (such as vehicle density, driving speed, and congestion), and driving behavior (such as rapid acceleration, rapid deceleration, and idling time). Complex nonlinear relationships exist between these factors, making it difficult for single-factor models to accurately depict the dynamic changes in carbon emissions, resulting in prediction errors generally exceeding 20%, failing to provide a reliable basis for formulating control strategies. Even when some studies introduce deep learning technology, they often remain within a one-way process of "data input - model prediction," failing to address the systematic challenge of acquiring regional carbon emission data. Existing technologies also rely on macroscopic statistical data and lack real-time data support at the microscopic level. Its data source is singular and cannot cover all dimensions of information in the "people-road-vehicle" system, resulting in insufficient completeness and timeliness of the model input data. Furthermore, while existing prediction methods have gradually transitioned from traditional statistical models to machine learning and deep learning, technical bottlenecks still exist. For example, there is insufficient multi-source data fusion capability, making it difficult to efficiently integrate heterogeneous data such as climate monitoring data, traffic checkpoint data, and vehicle OBD (On-Board Diagnostics) data; the model architecture is too simplistic, failing to fully leverage the advantages of hybrid deep learning models (such as combining CNN-LSTM, Transformer, and time-series models) in spatiotemporal feature extraction and nonlinear relationship fitting; and the lack of a closed-loop optimization mechanism prevents the use of evaluation results to adjust model parameters, leading to a long-term decline in prediction accuracy. These problems collectively make it difficult for existing technologies to meet the urban traffic management requirements for "high-precision, dynamic, and visualized" carbon emission prediction. Summary of the Invention

[0003] This invention provides a spatiotemporal prediction method and system for automobile carbon emission characteristics driven by multi-source data, in order to solve the problem of low prediction accuracy in existing carbon emission prediction methods.

[0004] To achieve the above objectives, the present invention employs the following technical solution: In a first aspect, the present invention provides a multi-source data-driven spatiotemporal prediction method for automobile carbon emission characteristics, comprising: S1. Acquire and preprocess multi-source data, which includes at least: real-time vehicle emission, location, and vehicle condition time-series data collected by the vehicle-mounted data acquisition unit, as well as traffic data, meteorological data, and air quality data obtained from external data sources; perform outlier removal, missing value completion, and standardization processing on the multi-source data to form a multi-factor carbon emission dataset; S2. Perform factor analysis on the original data in the multi-factor carbon emission dataset to extract multiple principal component factors; S3. Construct and train a hybrid deep learning prediction model based on CNN-LSTM. Input the time series data corresponding to the principal component factors into the model to predict automobile carbon emissions. During the model training process, an attention mechanism is introduced to assign differentiated weights to each principal component factor, and the NSGA-II multi-objective optimization algorithm is used to adaptively correct the hyperparameters of the model. The optimization algorithm has the dual optimization objectives of minimizing prediction error and minimizing model complexity. S4. Perform spatiotemporal alignment and feature fusion on the multi-source data obtained in step S1 to establish a multi-scale carbon emission calculation model and calculate the basic carbon emission data at the road segment, regional and urban scales. S5. Based on Gaussian diffusion theory and combined with diffusion coefficients under multiple wind speed levels, a carbon emission diffusion model is constructed, and parameters are calibrated using measured data from air monitoring stations to quantify the contribution of carbon emissions from different road sections to air quality. S6. A spatiotemporal prediction model is constructed using a Transformer-LSTM hybrid architecture. The multi-source data processed in step S4 is input, and the spatiotemporal prediction results of carbon emissions at the grid scale within a specified future time period are output. S7. Based on the spatiotemporal prediction results obtained in step S6, and combined with the preset early warning standards, generate carbon emission early warning information and corresponding eco-driving control strategies.

[0005] Optionally, in step S2, multiple principal component factors are extracted, including: Principal component analysis was used for factor extraction. Eigenvalues ​​and eigenvectors were obtained by solving the characteristic equation of the correlation matrix. Principal components were selected based on the criteria that the eigenvalue is ≥1 or the cumulative variance contribution rate is ≥85%. The three principal component factors, including at least traffic-road factors, driving behavior factors, and vehicle-environment factors, were extracted.

[0006] Optionally, in step S3, the adaptive adjustment of the model's hyperparameters using the NSGA-II multi-objective optimization algorithm includes: S31. Randomly generate an initial population containing multiple hyperparameter combinations; S32. Calculate the two objective function values ​​of prediction error and model complexity for each individual in the population; S33. Sort and select individuals based on non-dominance relationships and crowding levels; S34. Generate offspring populations through selection, crossover, and mutation operations, and merge them with the parent populations to select the next generation population; S35. Iterate through S32 to S34 until the termination condition is met, and output the optimal hyperparameter combination.

[0007] Optionally, in step S4, the spatiotemporal alignment includes: To address the differences in time sampling frequencies among multi-source data, linear interpolation or spline interpolation methods are used for time alignment. To address spatial coordinate differences, spatial calibration and matching are performed based on a unified coordinate system. The feature fusion adopts a weighted fusion method based on entropy weighting.

[0008] Optionally, in step S4, a multi-scale carbon emission calculation model is established to calculate the basic carbon emission data at the road segment, regional, and urban scales, including: A carbon emission calculation model is established, which satisfies the following relationship: ; In the formula, Qv represents the traffic flow of the v-th type of vehicle. is the carbon emission factor per unit mileage for Class V vehicles at speed s and acceleration a, and L is the road segment length; The study area is divided into grids, and the road segment emission data is aggregated into each grid through spatial overlay analysis to calculate the total regional emissions. Emissions data from various regions are aggregated according to administrative boundaries and corrected based on urban transportation network density and population distribution data to obtain basic carbon emission data at the city scale.

[0009] Optionally, in step S5, the carbon emission diffusion model adopts a continuous point source Gaussian diffusion model, and the formula for calculating the ground-level axial concentration is: ; In the formula, C is the diffusion concentration and u is the average wind speed (m / s). These are the diffusion coefficients in the horizontal and vertical directions, respectively. The diffusion coefficient can be determined based on the atmospheric stability classification method according to different wind speed levels, or by fitting an empirical formula for the diffusion coefficient under different wind speeds. The empirical formula for the diffusion coefficient satisfies the following relationship: ; In the formula, a, b, c, and d are empirical parameters related to wind speed; The least squares method was used to calibrate the diffusion model parameters with the goal of minimizing the error between the model predictions and the measured values ​​from air monitoring stations.

[0010] Optionally, in step S6, the processing procedure of the Transformer-LSTM hybrid architecture includes: The input data is reconstructed into a three-dimensional tensor of time series, grid space, and feature dimension; The computation employs a multi-head attention mechanism to capture global spatiotemporal dependencies, satisfying the following relationship: ; The temporal features are then enhanced by an LSTM layer; finally, the predicted future carbon emission intensity for each grid is output through a fully connected layer.

[0011] Secondly, embodiments of this application provide a multi-source data-driven spatiotemporal prediction system for automobile carbon emissions, comprising: The data acquisition module is used to collect actual vehicle data, traffic data, meteorological data, and air quality data, and to preprocess them to construct a multi-factor carbon emission dataset. The multi-factor carbon emission prediction module is used to perform principal component factor extraction and to build and optimize a CNN-LSTM hybrid deep learning model. The spatiotemporal prediction and diffusion analysis module is used to perform multi-scale emission calculations, carbon emission diffusion modeling and contribution analysis, and Transformer-LSTM spatiotemporal prediction. The visualization module is used to dynamically display carbon emission prediction results, spatiotemporal heat maps, and diffusion process simulations based on GIS technology. The early warning and control module is used to trigger early warnings based on predicted carbon emission intensity and preset yellow, orange, and red three-level early warning thresholds, and automatically generate eco-driving control strategies including rapid acceleration restrictions, off-peak traffic, or traffic light timing optimization.

[0012] Optionally, the data acquisition module includes: The vehicle data acquisition unit integrates a PEMS emission tester, a GPS positioning module, and a vehicle OBD diagnostic module to collect four-dimensional data of the vehicle, including "time-location-vehicle condition-emissions". The external data access unit is used to connect to the urban traffic management database, meteorological platform API, and air monitoring stations to obtain road network, traffic flow, meteorological, and air quality data.

[0013] Optionally, the multi-factor carbon emission prediction module includes: The factor analysis unit is used to perform the KMO test and Bartlett's test of sphericity, and to extract principal component factors through principal component analysis. Hybrid model unit, integrating CNN layers, LSTM layers and attention mechanism layers, is used for feature extraction and sequence prediction; The parameter optimization unit integrates the NSGA-II multi-objective optimization algorithm to adaptively adjust the model hyperparameters.

[0014] Beneficial effects: This invention provides a multi-source data-driven spatiotemporal prediction method for vehicle carbon emission characteristics. By constructing a comprehensive vehicle carbon emission prediction model, this model delves into the complex human-road-vehicle coupled system and explores the carbon emission characteristics of vehicles under the influence of multiple factors, such as climate change, traffic flow, and driving behavior. Deep learning technology is used to integrate these factors, improving the accuracy of carbon emission prediction. Furthermore, this invention aims to develop a spatiotemporal carbon emission prediction system to accurately predict carbon emissions in different urban areas and time periods. This will provide a scientific basis for urban traffic management and eco-driving control strategies, while supporting policymakers' application in environmental management decisions. The technical approach includes collecting relevant climate, traffic, and behavioral data, developing a prediction model using deep learning technology, and validating and optimizing the model using real-world urban data. This system not only helps reduce urban carbon emissions and improve air quality but also provides a solid scientific basis for urban planning and environmental policy formulation, thus generating significant positive impacts at both the social and environmental levels. Attached Figure Description

[0015] Figure 1 This is a flowchart of a multi-source data-driven spatiotemporal prediction method for automobile carbon emission characteristics, which is a preferred embodiment of the present invention. Detailed Implementation

[0016] The technical solution of the present invention will be clearly and completely described below. 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.

[0017] Unless otherwise defined, the technical or scientific terms used in this invention shall have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Similarly, the terms "an" or "a" and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms "connected" or "linked" and similar terms are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up," "down," "left," "right," etc., are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship also changes accordingly.

[0018] Please see Figure 1This application provides a multi-source data-driven spatiotemporal prediction method for automobile carbon emission characteristics, including: S1. Acquire and preprocess multi-source data, which includes at least: real-time vehicle emission, location, and vehicle condition time-series data collected by the vehicle-mounted data acquisition unit, as well as traffic data, meteorological data, and air quality data obtained from external data sources; perform outlier removal, missing value completion, and standardization processing on the multi-source data to form a multi-factor carbon emission dataset; S2. Perform factor analysis on the original data in the multi-factor carbon emission dataset to extract multiple principal component factors; S3. Construct and train a hybrid deep learning prediction model based on CNN-LSTM. Input the time series data corresponding to the principal component factors into the model to predict automobile carbon emissions. During the model training process, an attention mechanism is introduced to assign differentiated weights to each principal component factor, and the NSGA-II multi-objective optimization algorithm is used to adaptively correct the hyperparameters of the model. The optimization algorithm has the dual optimization objectives of minimizing prediction error and minimizing model complexity. S4. Perform spatiotemporal alignment and feature fusion on the multi-source data obtained in step S1 to establish a multi-scale carbon emission calculation model and calculate the basic carbon emission data at the road segment, regional and urban scales. S5. Based on Gaussian diffusion theory and combined with diffusion coefficients under multiple wind speed levels, a carbon emission diffusion model is constructed, and parameters are calibrated using measured data from air monitoring stations to quantify the contribution of carbon emissions from different road sections to air quality. S6. A spatiotemporal prediction model is constructed using a Transformer-LSTM hybrid architecture. The multi-source data processed in step S4 is input, and the spatiotemporal prediction results of carbon emissions at the grid scale within a specified future time period are output. S7. Based on the spatiotemporal prediction results obtained in step S6, and combined with the preset early warning standards, generate carbon emission early warning information and corresponding eco-driving control strategies.

[0019] Optionally, in step S2, multiple principal component factors are extracted, including: Principal component analysis was used for factor extraction. Eigenvalues ​​and eigenvectors were obtained by solving the characteristic equation of the correlation matrix. Principal components were selected based on the criteria that the eigenvalue is ≥1 or the cumulative variance contribution rate is ≥85%. The three principal component factors, including at least traffic-road factors, driving behavior factors, and vehicle-environment factors, were extracted.

[0020] Optionally, in step S3, the adaptive adjustment of the model's hyperparameters using the NSGA-II multi-objective optimization algorithm includes: S31. Randomly generate an initial population containing multiple hyperparameter combinations; S32. Calculate the two objective function values ​​of prediction error and model complexity for each individual in the population; S33. Sort and select individuals based on non-dominance relationships and crowding levels; S34. Generate offspring populations through selection, crossover, and mutation operations, and merge them with the parent populations to select the next generation population; S35. Iterate through S32 to S34 until the termination condition is met, and output the optimal hyperparameter combination.

[0021] Optionally, in step S4, the spatiotemporal alignment includes: To address the differences in time sampling frequencies among multi-source data, linear interpolation or spline interpolation methods are used for time alignment. To address spatial coordinate differences, spatial calibration and matching are performed based on a unified coordinate system. The feature fusion adopts a weighted fusion method based on entropy weighting.

[0022] Optionally, in step S4, a multi-scale carbon emission calculation model is established to calculate the basic carbon emission data at the road segment, regional, and urban scales, including: A carbon emission calculation model is established, which satisfies the following relationship: ; In the formula, Qv represents the traffic flow of the v-th type of vehicle. is the carbon emission factor per unit mileage for Class V vehicles at speed s and acceleration a, and L is the road segment length; The study area is divided into grids, and the road segment emission data is aggregated into each grid through spatial overlay analysis to calculate the total regional emissions. Emissions data from various regions are aggregated according to administrative boundaries and corrected based on urban transportation network density and population distribution data to obtain basic carbon emission data at the city scale.

[0023] Optionally, in step S5, the carbon emission diffusion model adopts a continuous point source Gaussian diffusion model, and the formula for calculating the ground-level axial concentration is: ; In the formula, C is the diffusion concentration and u is the average wind speed (m / s). These are the diffusion coefficients in the horizontal and vertical directions, respectively. The diffusion coefficient can be determined based on the atmospheric stability classification method according to different wind speed levels, or by fitting an empirical formula for the diffusion coefficient under different wind speeds. The empirical formula for the diffusion coefficient satisfies the following relationship: ; In the formula, a, b, c, and d are empirical parameters related to wind speed; The least squares method was used to calibrate the diffusion model parameters with the goal of minimizing the error between the model predictions and the measured values ​​from air monitoring stations.

[0024] Optionally, in step S6, the processing procedure of the Transformer-LSTM hybrid architecture includes: The input data is reconstructed into a three-dimensional tensor of time series, grid space, and feature dimension; The computation employs a multi-head attention mechanism to capture global spatiotemporal dependencies, satisfying the following relationship: ; The temporal features are then enhanced by an LSTM layer; finally, the predicted future carbon emission intensity for each grid is output through a fully connected layer.

[0025] The following are the specific implementation steps in this embodiment: I. Implementation of Multi-Source Data Acquisition and Preprocessing The implementation of the data acquisition module is the foundation of the system construction, requiring simultaneous acquisition of measured data and multi-source collaborative data integration. First, an integrated on-board experimental platform was built, comprising a PEMS emission tester, a GPS positioning module, and a vehicle OBD diagnostic module. 1.5L gasoline vehicles, 2.0L gasoline vehicles, and 2.0T diesel vehicles were selected as experimental models, and a 50km multi-level road network covering urban expressways, main roads, secondary roads, and branch roads was planned as the experimental route. During the experiment, tests were arranged throughout the day, including morning peak hours, off-peak hours, evening peak hours, and nighttime, accumulating over 400 hours of real-world emission testing. Real-time data collection included vehicle CO2 and CO emission data, driving trajectory coordinates, instantaneous speed, engine speed, throttle opening, and other vehicle condition data.

[0026] Simultaneously, multi-source collaborative data integration was initiated. Authorized access to the traffic big data platform by the urban traffic management department was obtained to acquire road network vector data (including road grade, number of lanes, intersection location, etc.) and real-time traffic data (traffic density, congestion index, average driving speed). The API interface of the local meteorological platform was connected to obtain hourly meteorological data such as temperature, wind speed, and precipitation. Air quality data such as PM2.5 and NO2 from air monitoring stations were obtained from the environmental protection department. All collected data underwent preprocessing: outliers in emission and speed data were removed using the 3σ criterion; a small amount of missing data was supplemented using linear interpolation; and all numerical indicators were standardized to the [0,1] range according to the formula "(original value - minimum value) / (maximum value - minimum value)". Finally, a multi-factor carbon emission dataset of over 1.4 million records containing four-dimensional information (time, location, vehicle condition, emissions) was integrated and stored in the system database, providing high-quality data input for subsequent modules.

[0027] II. Construction and Implementation of Multi-Factor Carbon Emission Prediction Model Based on the preprocessed multi-factor dataset, the multi-factor carbon emission prediction module was launched. The first step was to conduct factor analysis. Using SPSS, KMO and Bartley's test of sphericity were performed on indicators such as vehicle attributes, road conditions, driving behavior, and traffic conditions in the dataset to ensure that the data met the conditions for factor analysis. Then, principal component analysis was used to extract the three principal component factors: "traffic-road", "driving behavior", and "vehicle-environment". The influence weight of each factor on carbon emissions was clarified, and the coupling mechanism of multiple factors in the human-road-vehicle system was elucidated.

[0028] The second step involves building a prediction model using a CNN-LSTM hybrid deep learning architecture. Using feature data corresponding to the three principal component factors as input, the spatial correlation features of traffic flow, road grade, and other data are first extracted through the convolutional and pooling layers of the CNN. The processed feature data is then fed into the LSTM layer, where the gating units capture the dynamic changes in time-series data such as driving behavior and speed. Simultaneously, an attention mechanism is introduced between the CNN and LSTM layers, assigning differentiated weights based on the influence of each principal component factor on carbon emissions, thus strengthening the contribution of key factors to the prediction results. During model training, the NSGA-II multi-objective optimization algorithm is used, aiming to minimize prediction error and maximize model efficiency. Hyperparameters such as the number of CNN convolutional kernels and the number of LSTM hidden units are adaptively adjusted. Through iterative training, the model's prediction accuracy on the test set is improved to over 90%, completing the construction of a multi-factor carbon emission prediction model and achieving accurate prediction of carbon emissions under complex conditions for single vehicles and single road segments.

[0029] III. Implementation of the Spatiotemporal Prediction and Diffusion Analysis Module The implementation of the spatiotemporal prediction and diffusion analysis module requires a three-tiered functional approach. The first tier involves multi-source data fusion and multi-scale emission calculation. This involves processing the road network, traffic, meteorological, and air quality data acquired by the data acquisition module using a "spatiotemporal alignment-feature fusion-quality control" process: unifying the time dimension of all data using a 5-minute timestamp, matching the spatial dimension based on road network node coordinates, and using a weighted average method to fuse multi-source data within the same spatiotemporal unit (e.g., fusing floating car data and checkpoint data with average road segment speed). A sliding window method is used to eliminate abnormal fluctuations in the fused data, ensuring data consistency. Based on the fused data, a three-tiered emission calculation model of "road segment-region-city" is established: using a single road segment as a unit, the total carbon emissions of the road segment are calculated based on the results of the multi-factor prediction model; the city is divided into 1km×1km grids, and the carbon emission data of road segments within the grids are aggregated to calculate the regional carbon emission density; all regional data are overlaid and combined with the city's road network coverage to estimate the overall total carbon emissions of the city, achieving seamless output of carbon emission data at different scales.

[0030] The second stage involves constructing a carbon emission diffusion model. Based on Gaussian diffusion theory, model parameters are determined using measured concentration data from air monitoring stations. The total carbon emissions from a road segment are taken as the emission source strength. Corresponding diffusion coefficients are set according to different wind speeds (1-3 m / s, 3-5 m / s, 5-10 m / s), and the height of vehicle exhaust pipes is set as the emission source height. Carbon emission diffusion models are established under different wind speeds. By comparing the pollutant concentrations calculated by the model with the measured concentrations at monitoring stations, the diffusion coefficients are calibrated using the least squares method to ensure that the relative error between the model calculations and measured values ​​is controllable. Based on the diffusion model simulation results, a circular area with a radius of 3 km centered on each air monitoring station is determined as the carbon emission diffusion concentration calculation area. The contribution of carbon emissions from each road segment within this area to the air quality at the monitoring stations is analyzed, and key emission road segments with a contribution exceeding 10% are identified.

[0031] The third level constructs a spatiotemporal prediction submodule, employing a Transformer-LSTM hybrid architecture. It uses fused multi-source data (including historical carbon emission data, traffic data, and meteorological data) as input. The Transformer's multi-head attention layer captures the correlation features of carbon emissions across different road segments and time periods. The LSTM layer enhances the time-series prediction capability. The model output is set to provide carbon emission data for the next 24 hours at a 1km × 1km grid scale. The model is trained and validated using historical data, and its hyperparameters are adjusted until the validation R² reaches 0.91, completing the spatiotemporal prediction submodule and enabling accurate prediction of the spatiotemporal distribution of urban carbon emissions.

[0032] IV. Implementation of Visualization and Early Warning Control Module The visualization module is implemented using GIS technology as its core, building a dynamic display interface. The interface development is divided into three main functional modules: the "Real-time Emission Monitoring" module matches road segment carbon emission data output from the multi-factor prediction model with the GIS road network map, using different colors to mark the carbon emission levels of each road segment, and displaying details such as total carbon emissions and major influencing factors when the mouse hovers over the segment; the "Spatiotemporal Trend Analysis" module, based on the results of the spatiotemporal prediction submodule, generates spatiotemporal trend heatmaps of carbon emissions for the past 24 hours and the past 7 days, allowing users to view changes over different time periods using a time slider; and the "Dynamic Demonstration of Diffusion Process" module loads the calculation results of the carbon emission diffusion model, dynamically simulating the diffusion range and concentration changes of carbon emissions under different wind speeds, supporting playback and pause controls. Simultaneously, the interface update frequency is synchronized with the data collection frequency to ensure the real-time nature of the displayed data.

[0033] The implementation of the early warning and control module requires first determining the early warning standards. Based on the road segment's environmental traffic capacity (calculated according to the road segment's designed capacity and carbon emission thresholds), a three-tiered standard of "yellow warning, orange warning, and red warning" is established. Early warning triggering logic is developed. The system queries real-time and predicted carbon emission data for each road segment every 5 minutes, automatically determining the warning level by comparing it with the early warning standards. When an early warning is triggered, the system sends early warning information (including the warning road segment, warning level, and current carbon emission value) to designated terminals of the traffic management department via its built-in message push function. Simultaneously, a control strategy library is established, pre-setting differentiated strategies for different warning levels and road segment types: Red warning road segments push "rapid acceleration restriction (navigation prompts to optimize driving behavior)" and "off-peak traffic (guiding vehicles to avoid peak hours)" strategies; orange warning road segments push "traffic light timing optimization (extending green light duration by 10%-20%)" strategies; and yellow warning road segments push "smooth driving guidance (traffic guidance screen prompts)" strategies. This achieves linkage between early warning and control, providing direct decision support for the traffic management department.

[0034] Through the orderly implementation of the above modules, the overall construction of the vehicle carbon emission spatiotemporal prediction system is completed, forming a complete closed loop from data input to regulatory application, which meets the needs of urban vehicle carbon emission characteristic analysis, spatiotemporal prediction and precise regulation.

[0035] This application also provides a multi-source data-driven spatiotemporal prediction system for automobile carbon emissions, including: The data acquisition module is used to collect actual vehicle data, traffic data, meteorological data, and air quality data, and to preprocess them to construct a multi-factor carbon emission dataset. The multi-factor carbon emission prediction module is used to perform principal component factor extraction and to build and optimize a CNN-LSTM hybrid deep learning model. The spatiotemporal prediction and diffusion analysis module is used to perform multi-scale emission calculations, carbon emission diffusion modeling and contribution analysis, and Transformer-LSTM spatiotemporal prediction. The visualization module is used to dynamically display carbon emission prediction results, spatiotemporal heat maps, and diffusion process simulations based on GIS technology. The early warning and control module is used to trigger early warnings based on predicted carbon emission intensity and preset yellow, orange, and red three-level early warning thresholds, and automatically generate eco-driving control strategies including rapid acceleration restrictions, off-peak traffic, or traffic light timing optimization.

[0036] Optionally, the data acquisition module includes: The vehicle data acquisition unit integrates a PEMS emission tester, a GPS positioning module, and a vehicle OBD diagnostic module to collect four-dimensional data of the vehicle, including "time-location-vehicle condition-emissions". The external data access unit is used to connect to the urban traffic management database, meteorological platform API, and air monitoring stations to obtain road network, traffic flow, meteorological, and air quality data.

[0037] Optionally, the multi-factor carbon emission prediction module includes: The factor analysis unit is used to perform the KMO test and Bartlett's test of sphericity, and to extract principal component factors through principal component analysis. Hybrid model unit, integrating CNN layers, LSTM layers and attention mechanism layers, is used for feature extraction and sequence prediction; The parameter optimization unit integrates the NSGA-II multi-objective optimization algorithm to adaptively adjust the model hyperparameters.

[0038] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A multi-source data-driven spatiotemporal prediction method for automobile carbon emission characteristics, characterized in that, include: S1. Acquire and preprocess multi-source data, including: vehicle emission, location, and vehicle condition time-series data collected in real time by the vehicle-mounted data acquisition unit, as well as traffic data, meteorological data, and air quality data obtained from external data sources. Preprocessing includes: removing outliers, filling in missing values, and standardizing the multi-source data to form a multi-factor carbon emission dataset. S2. Perform factor analysis on the raw data in the multi-factor carbon emission dataset to extract multiple principal component factors; S3. Construct and train a hybrid deep learning prediction model based on CNN-LSTM. Input the time series data corresponding to the principal component factors into the model to predict automobile carbon emissions. During the model training process, an attention mechanism is introduced to assign differentiated weights to each principal component factor, and the NSGA-II multi-objective optimization algorithm is used to adaptively correct the hyperparameters of the model. The optimization algorithm has the dual optimization objectives of minimizing prediction error and minimizing model complexity. S4. Perform spatiotemporal alignment and feature fusion on the multi-source data obtained in step S1 to establish a multi-scale carbon emission calculation model and calculate the basic carbon emission data at the road segment, regional and urban scales. S5. Based on Gaussian diffusion theory and combined with diffusion coefficients under multiple wind speed levels, a carbon emission diffusion model is constructed, and parameters are calibrated using measured data from air monitoring stations to quantify the contribution of carbon emissions from different road sections to air quality. S6. A spatiotemporal prediction model is constructed using a Transformer-LSTM hybrid architecture. The multi-source data processed in step S4 is input, and the spatiotemporal prediction results of carbon emissions at the grid scale within a specified future time period are output. S7. Based on the spatiotemporal prediction results obtained in step S6, and combined with the preset early warning standards, generate carbon emission early warning information and corresponding eco-driving control strategies.

2. The multi-source data-driven spatiotemporal prediction method for automobile carbon emission characteristics according to claim 1, characterized in that, In step S2, multiple principal component factors are extracted, including: Principal component analysis was used for factor extraction. Eigenvalues ​​and eigenvectors were obtained by solving the characteristic equation of the correlation matrix. Principal components were selected based on the criteria that the eigenvalue is ≥1 or the cumulative variance contribution rate is ≥85%. The three principal component factors, including at least traffic-road factors, driving behavior factors, and vehicle-environment factors, were extracted.

3. The multi-source data-driven spatiotemporal prediction method for automobile carbon emission characteristics according to claim 1, characterized in that, In step S3, the adaptive adjustment of the model's hyperparameters using the NSGA-II multi-objective optimization algorithm includes: S31. Randomly generate an initial population containing multiple hyperparameter combinations; S32. Calculate the two objective function values ​​of prediction error and model complexity for each individual in the population; S33. Sort and select individuals based on non-dominance relationships and crowding levels; S34. Generate offspring populations through selection, crossover, and mutation operations, and merge them with the parent populations to select the next generation population; S35. Iterate through S32 to S34 until the termination condition is met, and output the optimal hyperparameter combination.

4. The multi-source data-driven spatiotemporal prediction method for automobile carbon emission characteristics according to claim 1, characterized in that, In step S4, the spatiotemporal alignment includes: To address the differences in time sampling frequencies among multi-source data, linear interpolation or spline interpolation methods are used for time alignment. To address spatial coordinate differences, spatial calibration and matching are performed based on a unified coordinate system. The feature fusion adopts a weighted fusion method based on entropy weighting.

5. The multi-source data-driven spatiotemporal prediction method for automobile carbon emission characteristics according to claim 1, characterized in that, In step S4, a multi-scale carbon emission calculation model is established to calculate the basic carbon emission data at the road segment, regional, and urban scales, including: A carbon emission calculation model is established, which satisfies the following relationship: ； In the formula, Qv represents the traffic flow of the v-th type of vehicle. Let L be the carbon emission factor per unit mileage for vehicle type v at speed s and acceleration a, where L is the road segment length and E is the carbon emission data. The study area is divided into grids, and the road segment emission data is aggregated into each grid through spatial overlay analysis to calculate the total regional emissions. Emissions data from various regions are aggregated according to administrative boundaries and corrected based on urban transportation network density and population distribution data to obtain basic carbon emission data at the city scale.

6. The multi-source data-driven spatiotemporal prediction method for automobile carbon emission characteristics according to claim 1, characterized in that, In step S5, the carbon emission diffusion model adopts a continuous point source Gaussian diffusion model, and the formula for calculating the ground-level axial concentration is: ； In the formula, C is the diffusion concentration, and u is the average wind speed. Here, represents the horizontal and vertical diffusion coefficients, Q represents the pollution source intensity, and H represents the pollution source height; The diffusion coefficient can be determined based on the atmospheric stability classification method according to different wind speed levels, or by fitting an empirical formula for the diffusion coefficient under different wind speeds. The empirical formula for the diffusion coefficient satisfies the following relationship: ； In the formula, a, b, c, and d are empirical parameters related to wind speed; The least squares method was used to calibrate the diffusion model parameters with the goal of minimizing the error between the model predictions and the measured values ​​from air monitoring stations.

7. The multi-source data-driven spatiotemporal prediction method for automobile carbon emission characteristics according to claim 1, characterized in that, In step S6, the processing procedure of the Transformer-LSTM hybrid architecture includes: The input data is reconstructed into a three-dimensional tensor of time series, grid space, and feature dimension; The computation employs a multi-head attention mechanism to capture global spatiotemporal dependencies, satisfying the following relationship: ； The temporal features are then enhanced by an LSTM layer; finally, the predicted future carbon emission intensity for each grid is output through a fully connected layer.

8. A multi-source data-driven spatiotemporal prediction system for automobile carbon emission characteristics, characterized in that, include: The data acquisition module is used to collect actual vehicle data, traffic data, meteorological data, and air quality data, and to preprocess them to construct a multi-factor carbon emission dataset. The multi-factor carbon emission prediction module is used to perform principal component factor extraction and to build and optimize a CNN-LSTM hybrid deep learning model. The spatiotemporal prediction and diffusion analysis module is used to perform multi-scale emission calculations, carbon emission diffusion modeling and contribution analysis, and Transformer-LSTM spatiotemporal prediction. The visualization module is used to dynamically display carbon emission prediction results, spatiotemporal heat maps, and diffusion process simulations based on GIS technology. The early warning and control module is used to trigger early warnings based on predicted carbon emission intensity and preset yellow, orange, and red three-level early warning thresholds, and automatically generate eco-driving control strategies including rapid acceleration restrictions, off-peak traffic, or traffic light timing optimization.

9. The multi-source data-driven spatiotemporal prediction system for automobile carbon emission characteristics according to claim 8, characterized in that, The data acquisition module includes: The on-board data acquisition unit integrates a PEMS emission tester, a GPS positioning module, and a vehicle OBD diagnostic module to collect four-dimensional data on vehicle time, location, vehicle condition, and emissions. The external data access unit is used to connect to the urban traffic management database, meteorological platform API, and air monitoring stations to obtain road network, traffic flow, meteorological, and air quality data.

10. The multi-source data-driven spatiotemporal prediction system for automobile carbon emission characteristics according to claim 8, characterized in that, The multi-factor carbon emission prediction module includes: The factor analysis unit is used to perform the KMO test and Bartlett's test of sphericity, and to extract principal component factors through principal component analysis. Hybrid model unit, integrating CNN layers, LSTM layers and attention mechanism layers, is used for feature extraction and sequence prediction; The parameter optimization unit integrates the NSGA-II multi-objective optimization algorithm to adaptively adjust the model hyperparameters.

Images