Vehicle global working condition low-carbon power cooperative control method based on networked road condition prediction

Abstract

The application belongs to the technical field of intelligent networked vehicles, power system control and traffic carbon emission management, and specifically discloses a vehicle global working condition low-carbon power coordination control method based on networked road condition prediction, which comprises the following steps: acquiring and according to real-time multi-source data, rolling predicting a future driving working condition sequence through a pre-trained prediction model to generate a working condition characteristic vector containing vehicle speed, acceleration, slope and parking probability; taking the working condition characteristic vector as input, performing double-target optimization with carbon emission minimization as the upper target and energy consumption minimization as the lower target, and outputting power coordination control instructions; issuing the instructions, collecting energy consumption data in real time and converting the energy consumption data into carbon emissions, updating the carbon account and generating driving guidance information accordingly, and dynamically correcting the control parameters used in the double-target optimization. The application can effectively realize low-carbon coordination and carbon emission forward control of the vehicle power system, and reduce vehicle operation carbon emissions.

Description

Technical Field

[0001] This application belongs to the fields of intelligent connected vehicles, power system control and traffic carbon emission management technology, and more specifically, relates to a low-carbon power coordinated control method for vehicle operating conditions across the entire domain based on network road condition prediction. Background Technology

[0002] Current vehicle powertrain control systems primarily rely on reactive adjustments based on real-time driving conditions, lacking the ability to dynamically predict future road conditions. This results in persistently high energy consumption and carbon emissions under typical operating conditions such as slopes, traffic congestion, and traffic light intersections. While hybrid and pure electric vehicles possess the potential for multi-power source synergy, existing energy management strategies are mostly rule-based or instantaneous optimizations, failing to achieve optimal carbon emissions across all operating conditions. Furthermore, driver behavior significantly impacts carbon emissions, and real-time guidance mechanisms are lacking. In addition, commercial vehicle users, such as logistics fleets, lack carbon asset management tools, making it difficult to translate low-carbon operations into economic benefits.

[0003] Therefore, how to achieve low-carbon synergy in vehicle powertrain systems and proactive control of carbon emissions is an urgent problem that needs to be solved. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the purpose of this application is to provide a vehicle-wide low-carbon powertrain coordination control method based on networked road condition prediction, which can realize low-carbon coordination and forward-looking control of carbon emissions in the vehicle powertrain system, thereby reducing carbon emissions from vehicle operation.

[0005] To achieve the above objectives, firstly, this application provides a method for low-carbon power cooperative control of vehicles across all operating conditions based on networked road condition prediction, comprising the following steps: S10, acquire real-time multi-source data, and use a pre-built road vehicle carbon dynamic database to fuse and spatiotemporally align the real-time multi-source data to obtain the current fused data; S20, based on the current fused data and historical operating condition data, the driving operating condition sequence within a preset time period is predicted in a rolling manner through a pre-trained prediction model, generating an operating condition feature vector containing vehicle speed, acceleration, gradient, and parking probability. S30, using the aforementioned operating condition feature vector as input, performs dual-objective optimization with minimizing carbon emissions as the upper-level objective and minimizing energy consumption as the lower-level objective, and outputs power coordinated control commands; S40, issue the power coordination control command, collect the vehicle's energy consumption data in real time and convert it into carbon emissions, update the carbon account and generate driving guidance information based on the carbon emissions, and dynamically correct the control parameters used in the dual-objective optimization based on the carbon emissions.

[0006] As a further preferred embodiment, in step S10, the real-time multi-source data includes dynamic traffic information, high-precision map data, vehicle operating status data, and carbon emission factor data; the dynamic traffic information includes real-time traffic flow density, traffic light timing, accidents, and construction events; the high-precision map data includes road slope, curvature, speed limit, and number of lanes; the vehicle operating status data includes vehicle speed, acceleration, engine speed, torque, battery state of charge, and transmission gear; and the carbon emission factor data includes conversion factors from fuel consumption and electricity consumption to carbon emissions under different operating conditions.

[0007] As a further preferred embodiment, in step S20, the pre-trained prediction model adopts a multimodal prediction architecture, fusing the vehicle's own driving trajectory with the motion features of surrounding vehicles, and continuously predicts the vehicle speed sequence and operating condition type for the next 5 to 30 seconds; the pre-trained prediction model is a spatiotemporal graph convolution and Transformer hybrid architecture model, specifically including: Spatiotemporal graph convolutional layers are used to extract spatial dependencies and local temporal features between vehicles in a transportation network. The graph convolution operation formula is as follows: ,in , A For dynamically time-warped adjacency matrices, For degree matrix, For the first Layer node features For learnable weights, Activate ReLU; The Transformer encoder, used to capture long-range temporal dependencies, has the following self-attention calculation expression: ,in It is obtained by linear transformation of the output of the spatiotemporal graph convolutional layer. d k for K The dimension of a vector; The prediction output layer maps the output of the Transformer encoder to the vehicle speed, acceleration, gradient, and parking probability at future prediction time points.

[0008] As a further preferred embodiment, in step S20, the input is represented as: ; The mathematical model for the bi-objective optimization is as follows: ; The power coordination control command is expressed as follows: ; In the formula, For vehicle speed; The battery is in its state of charge. For gear; This refers to engine torque; This refers to the motor torque. Fuel consumption rate; For power consumption; Carbon emission factor; For time; , These are the carbon emission weighting coefficient and the energy consumption weighting coefficient; It is a fuel with a low calorific value; For battery power; weighting coefficients are optimized using a hierarchical approach, with the upper layer based on... The Pareto front is derived with the objective of minimization; the lower layer satisfies the carbon emission constraints of the upper layer. Minimize under the premise ; For engine start / stop; This is the motor torque coefficient; This is a gear shift command; This represents the air conditioner power ratio.

[0009] As a further preferred embodiment, the dual-objective optimization employs a joint decision-making process combining model predictive control and Q-learning, specifically including: The operating condition sequence is used as an external reference trajectory; The optimal control sequence is obtained by solving a constrained nonlinear optimization problem in each control cycle. Define reinforcement learning state as , For the change in vehicle speed, The variable represents the change in battery state of charge, h represents the road gradient, and C represents the road congestion level. The reward function is a correction factor for the model's predictive control output. , This represents the change in carbon emissions. To measure the change in energy consumption, an ε-greedy strategy is used for exploration, and the Q-network is updated every 100 control steps to correct the model prediction error online. The revised control commands are output to the actuator.

[0010] As a further preferred embodiment, in step S40, before dynamically correcting the control parameters based on carbon emissions, a dynamic threshold calculation and guidance prompt trigger judgment are also included: Let the reference carbon emission intensity of the current vehicle under standard operating conditions be... Dynamic threshold Adjustments are made in real time based on road conditions ahead. ,in The change in slope The adjustment coefficient is used to reflect the impact of road slope on the carbon emission threshold. Road congestion level CThe adjustment coefficient is used to reflect the impact of traffic congestion on the carbon emission threshold. For air conditioning demand The adjustment coefficient is used to reflect the impact of vehicle air conditioning load on carbon emission thresholds; Take the average actual carbon emissions over the past 60 seconds ,like Then, the threshold is adaptively offset; Estimating cumulative carbon emissions over the next 10 seconds using predicted future driving conditions. ,like If so, an alert will be triggered, in which This is the average velocity factor.

[0011] As a further preferred embodiment, the driving guidance information generated after triggering the warning is determined by the specific type based on the predicted exceedance range and duration: when the predicted carbon emission exceedance range is no more than 10% and the duration is no less than 2 seconds, a slight lift-off prompt is generated; when the predicted carbon emission exceedance range is greater than 10% but no more than 30%, a coasting suggestion or a voice prompt to reduce the throttle is generated; when the predicted carbon emission exceedance range is greater than 30%, a mandatory suggestion to switch to pure electric mode or activate ECO mode is generated; when the predicted probability of stopping ahead is greater than 0.7, a prompt to release the throttle and coast to the intersection is generated 5 seconds in advance; if the driver does not respond and the actual carbon emissions still exceed the threshold, the air conditioning power is automatically reduced or the motor assist strategy is adjusted.

[0012] As a further preferred embodiment, the method is also configured with a communication failure and data anomaly degradation mechanism, which automatically switches to local sensor control mode when there is no network connection, map is missing, or data is abnormal.

[0013] As a further preferred embodiment, the method also includes a cloud-based collaborative optimization step: uploading the de-identified data of the vehicles in operation to a cloud-based collaborative optimization platform. The platform aggregates the de-identified operation data of the fleet and the carbon emission hotspot information of the road network based on a federated learning architecture, dynamically updates the pre-trained prediction model and the dual-objective optimization decision model, and generates a map-level carbon emission guidance service.

[0014] Secondly, this application provides a vehicle-wide low-carbon power cooperative control system based on networked road condition prediction, comprising the steps of implementing any one of the methods described above, including: The connected data acquisition and fusion module is used to acquire real-time multi-source data and use a pre-built road-vehicle carbon dynamic database to fuse and spatiotemporally align the real-time multi-source data to obtain the current fused data; The global operating condition prediction module is used to predict the driving operating condition sequence within a preset time period in the future based on the current fused data and historical operating condition data, and generate an operating condition feature vector containing vehicle speed, acceleration, slope and stopping probability. The low-carbon power coordinated control module is used to perform dual-objective optimization with the operating condition feature vector as input, which has the upper-level objective of minimizing carbon emissions and the lower-level objective of minimizing energy consumption, and outputs power coordinated control commands. The carbon emission monitoring and feedback module is used to issue the power coordination control command, collect the vehicle's energy consumption data in real time and convert it into carbon emissions, update the carbon account and generate driving guidance information based on the carbon emissions, and dynamically correct the control parameters used in the dual-objective optimization based on the carbon emissions.

[0015] This application has the following beneficial effects: (1) Breaking through the limitations of traditional reactive power control, it has for the first time achieved integrated control of “networked road condition prediction - full-domain operating condition prediction - low-carbon power coordination”, and carbon emission control has been transformed from delayed response to forward intervention; (2) Innovate the dual-target optimization framework of "carbon emission target-energy consumption", reduce carbon emissions under typical operating conditions by 8%-15% while ensuring power performance, and effectively support the dual carbon targets in the transportation sector; (3) Construct a vehicle-level carbon account and a real-time driving guidance mechanism to transform macro carbon emission reduction targets into perceptible and executable driver behaviors, thereby improving the efficiency of human-machine collaboration in carbon emission control. (4) Adopting a cloud-edge collaborative federated learning framework, the model can be continuously evolved while ensuring data privacy, supporting carbon asset management of commercial vehicle fleets, and opening up the value chain of low-carbon operation and carbon trading; (5) The modular architecture covers multiple power types (gasoline, hybrid, pure electric) for commercial vehicles and passenger vehicles, and has the ability to quickly adapt to different models and scenarios, making it highly valuable for promotion. (6) The simulation comparison results of this application with existing technologies (such as traditional MPC and rule-based hybrid control) under typical urban commuting conditions are shown in [reference to the original text]. Figures 6-8 Among them, carbon emissions were reduced by 13.5% (compared to regulatory control) and 8.2% (compared to ordinary MPC), and power response delay was reduced by 22%, demonstrating significant advantages. Attached Figure Description

[0016] Figure 1 This is a system architecture diagram provided in the embodiments of this application; Figure 2 This is a schematic diagram of the model structure of the global working condition prediction module provided in the embodiments of this application; Figure 3 This is a flowchart of low-carbon power collaborative control provided in the embodiments of this application; Figure 4 This is a schematic diagram of the carbon emission monitoring and feedback module provided in the embodiments of this application; Figure 5 This is a flowchart of the MPC and Q-learning joint decision-making algorithm provided in the embodiments of this application; Figure 6 This is a bar chart comparing the carbon emission intensity of the method provided in this application embodiment with that of existing technologies (rule control, ordinary MPC) under WLTC conditions; Figure 7 This is the instantaneous carbon emission response curve (including threshold and trigger marker) of the method provided in this application embodiment on a real vehicle urban commuting route. Figure 8 This is a time-series comparison curve of the vehicle battery state of charge and engine start-stop count under conventional control provided by the embodiments of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0018] The purpose of this application is to provide a vehicle-wide low-carbon power cooperative control system and method based on network road condition prediction, so as to solve the problems of lack of road condition prediction, lagging carbon emission control, and insufficient coupling between power and carbon emissions in the existing technology, and realize low-carbon cooperation and forward-looking control of carbon emissions in the vehicle power system.

[0019] The vehicle all-domain operating condition low-carbon power collaborative control system based on network road condition prediction provided in this application includes a network data acquisition and fusion module, an all-domain operating condition prediction module, a low-carbon power collaborative control module, and a carbon emission monitoring and feedback module.

[0020] Among them, the connected data acquisition and fusion module is used to collect dynamic traffic information, high-precision map data, vehicle operation status data and carbon emission factor data. Through multi-source data fusion and spatiotemporal alignment, a dynamic database of road-vehicle-carbon is constructed.

[0021] In this application, the connected data acquisition and fusion module is configured with a communication failure and data anomaly degradation mechanism, automatically switching to local sensor control mode when there is no network, map is missing, or data is abnormal. Specifically, the dynamic traffic information of the connected data acquisition and fusion module includes real-time traffic flow density, traffic light timing, and accident / construction events; high-precision map data includes road slope, curvature, speed limit, and number of lanes; vehicle operating status data includes vehicle speed, acceleration, engine speed, torque, battery SOC, and transmission gear; and carbon emission factor data includes fuel consumption / electricity consumption-carbon emission conversion factors under different operating conditions.

[0022] The global driving condition prediction module is used to predict the sequence of vehicle driving conditions within a future time window based on the fusion of historical and real-time connected road condition information, including vehicle speed, gradient, congestion level, and probability of stopping.

[0023] In this application, the multimodal prediction architecture of the full-domain working condition prediction module integrates the vehicle's own driving trajectory with the motion characteristics of surrounding vehicles to achieve rolling prediction of vehicle speed sequence and working condition type for the next 5-30 seconds, with a working condition recognition accuracy of ≥92%. Specifically, the working condition prediction module outputs a working condition sequence including vehicle speed, gradient, congestion level, stopping probability, acceleration, and speed limit information.

[0024] The low-carbon power coordinated control module is used to make coordinated power decisions across the entire operating condition range based on the predicted operating condition sequence and real-time carbon emission feedback.

[0025] In this application, the low-carbon power cooperative control module adopts minimizing carbon emissions as the upper-level objective and minimizing energy consumption as the lower-level objective, constructing a dual-objective optimization model. Through model predictive control, it generates cooperative decisions for engine start-stop, electric motor assist / regenerative braking, gear selection, and air conditioning power allocation, reducing carbon emissions by 8%-15% compared to traditional rule-based control. Specifically, the low-carbon power cooperative control module integrates predictive cruise and battery management strategies for hybrid commercial vehicles, adaptive regenerative braking and thermal management cooperative control for pure electric passenger vehicles, and economical shifting and engine load transfer strategies for traditional fuel vehicles.

[0026] The carbon emission monitoring and feedback module is used to calculate the instantaneous and cumulative carbon emissions of vehicles in real time. Combined with carbon emission target constraints, it generates carbon emission warnings and driving behavior guidance suggestions, forming a closed loop of "prediction-control-evaluation-optimization".

[0027] In this application, the carbon emission monitoring and feedback module has a built-in vehicle-level carbon account, which calculates carbon emission intensity in grams per kilometer in real time. When the predicted carbon emissions exceed the threshold, it provides guidance prompts to the driver through the instrument panel or HUD, such as coasting suggestions, reducing throttle, and switching to pure electric mode, and generates a low-carbon driving score.

[0028] Furthermore, the method provided in this application also includes a cloud-based collaborative optimization platform, which aggregates low-carbon power strategies for fleets through federated learning based on de-identified data reported by vehicles and carbon emission hotspot information of the road network, to achieve model iteration and map-level carbon emission guidance services, thereby supporting the carbon asset management of logistics companies.

[0029] Based on the same inventive concept, this application also provides a method for low-carbon power coordinated control of vehicles under all operating conditions based on the above system, including the following steps: S1: Connected data collection and fusion, which acquires dynamic road conditions, high-precision maps and vehicle operation data through V2X, vehicle GPS / IMU and cloud service platform, and stores them in the road-vehicle-carbon dynamic database after spatiotemporal alignment and anomaly removal; S2: Global driving condition prediction. Based on fused data, it predicts the future driving condition sequence and generates a condition feature vector containing vehicle speed, acceleration, gradient, and parking probability. S3: Low-carbon power collaborative decision-making, taking the predicted operating conditions as input, performs dual-objective optimization, and outputs power collaborative control commands; S4: Execution and carbon emission feedback, control commands are sent to the power domain and chassis domain, fuel / electricity consumption is collected in real time and converted into carbon emissions, carbon account is updated and driving guidance information is generated, and control parameters are dynamically corrected.

[0030] In this application, step S3 focuses on optimizing the load-adaptive energy management and anticipatory braking strategy for long downhill driving for commercial vehicles, reducing carbon emissions from long-distance transportation by 8%-12%; for passenger vehicles, it focuses on optimizing the start-stop logic and pure electric range maintenance strategy under urban congestion conditions, reducing carbon emissions from combined operating conditions by 5%-10%.

[0031] In step S4, the strategy is remotely upgraded through a cloud-edge collaboration mechanism. The decision model is continuously optimized based on user driving behavior and road network changes. The model update cycle is ≤72 hours and the carbon emission accounting error is ≤3%.

[0032] In one embodiment, the technical solution for achieving the above objective can be specifically as follows: Figure 1 As shown, the system provided in this embodiment includes a network data acquisition and fusion module, a global operating condition prediction module, a low-carbon power collaborative control module, and a carbon emission monitoring and feedback module. These modules work together to achieve a closed loop of "road condition prediction - operating condition forecasting - low-carbon control - carbon emission feedback." Connected data acquisition and fusion module: Constructs a multi-source heterogeneous data access system, supporting V2X, 5G / C-V2X, and cloud interface access to dynamic traffic information (real-time traffic flow, traffic light timing, events), high-precision maps (slope, curvature), and vehicle CAN data. Through timestamp alignment and Kalman filtering fusion, it constructs a dynamic "road-vehicle-carbon" database with a data update frequency ≥10Hz and a data effectiveness rate ≥99%.

[0033] like Figure 2 As shown, the global driving condition prediction module integrates the vehicle's historical trajectory with the motion characteristics of surrounding vehicles to predict driving condition sequences such as vehicle speed, acceleration, gradient, and parking probability for the next 5-30 seconds, with a time resolution of 1 second. The module employs a spatiotemporal graph convolution-Transformer (STGCN-Transformer) hybrid architecture. The model achieves a driving condition recognition accuracy of ≥92% in typical urban and highway scenarios, providing forward-looking input for driving decisions.

[0034] The detailed architecture and algorithm of this model are described below: Overall architecture: The model consists of three sub-modules connected in series: ① Spatiotemporal graph convolutional layer (STGCN) is used to extract spatial dependencies and local temporal features between vehicles in the traffic network; ② Transformer encoder is used to capture long-distance temporal dependencies; ③ Linear output layer is mapped to the predicted working condition sequence.

[0035] Input representation: Let the current time be... The length of the historical time window is (Taking 10 seconds), the feature vector at each time step includes: vehicle speed. acceleration Road slope Traffic flow density ahead Traffic light phase Constructing a spacetime graph , where the node set Including the vehicle itself and up to eight adjacent vehicles, the edge group Adjacency matrix weighted by the reciprocal of the distance between vehicles Dynamic updates are achieved through dynamic time warping.

[0036] Spatiotemporal graph convolution operation: The formula for graph convolution is:

[0037] in , For degree matrix, For the first Layer node features For learnable weights, ReLU activation is used. A one-dimensional convolution (kernel size 3×1) is employed in the temporal dimension to extract local temporal patterns, outputting a feature map. .

[0038] Transformer encoder: Will After adding position encoding, the data is fed into a 6-layer Transformer encoder (each layer contains 8 self-attention heads, a feedforward network, residual connections, and layer normalization). Self-attention calculation:

[0039] Depend on Obtained through linear transformation. Encoder output. .

[0040] Predictive output layer: The future is obtained after passing through two fully connected layers (with Dropout=0.2 in between). Predicted sequences in seconds (5-30 seconds):

[0041] Output includes: vehicle speed acceleration slope Parking probability The training loss function is the weighted mean squared error:

[0042] Weight The model takes ≤15ms for a single inference iteration on the NVIDIA Orin platform.

[0043] like Figure 3 As shown, the low-carbon powertrain coordinated control module constructs a dual-objective optimization model with minimizing carbon emissions as the upper-level objective and minimizing energy consumption as the lower-level objective. It adopts model predictive control as the main framework, introduces predicted operating conditions as reference trajectories, and combines Q-learning reinforcement learning to correct model errors. It generates coordinated commands for engine start / stop, electric motor assist / regenerative braking, transmission gear selection, and air conditioning power distribution. For commercial vehicles, it integrates predictive cruise and load adaptive strategies; for passenger vehicles, it integrates start-stop optimization for congestion conditions and pure electric range maintenance logic, reducing carbon emissions by 8%-15% compared to traditional control methods.

[0044] The specific mathematical expression of the dual-objective optimization model and the joint decision-making algorithm are as follows: Status and control inputs: System status Control input ,in For engine start / stop. This is the motor torque coefficient. This is a gear shift command. This represents the air conditioner power ratio.

[0045] Bi-objective function (prediction time domain) Control time domain ):

[0046] Fuel consumption rate (g / s) Power consumption (kW). These are carbon emission factors (g CO2 / g fuel and gCO2 / kWh). It has a low calorific value (J / g) for fuel oil. This represents battery power (positive for discharge). Weighting coefficients are optimized using a hierarchical approach: the upper layer uses... The Pareto front is derived with the objective of minimization; the lower layer satisfies the carbon emission constraints of the upper layer. Minimize under the premise In practical applications, take And adjust dynamically.

[0047] like Figure 5 As shown, the MPC-Q-learning joint decision-making process is as follows: Predictive model: using the operating condition sequence output from step S2 As an external reference trajectory.

[0048] Rolling optimization: In each control cycle, solve the above constrained nonlinear optimization problem (using Sequential Quadratic Programming, SQP) to obtain the optimal control sequence. .

[0049] Q-learning error correction: defining reinforcement learning states , For the change in vehicle speed, The change in battery state of charge, h is the road gradient, C is the road congestion level, and the action... This is a correction factor for the MPC output control quantity (discretized to {-0.1, 0, 0.1}). Reward function. The Q-network is a 3-layer fully connected network (128-64-3) that uses an ε-greedy strategy (ε=0.1) for exploration. The Q-network is updated every 100 control steps to correct model prediction errors online.

[0050] Issue and execute: Send the revised control commands Output to the actuator.

[0051] like Figure 4As shown, the carbon emission monitoring and feedback module constructs a vehicle-level carbon account based on real-time fuel / electricity consumption and carbon emission conversion factors, calculating instantaneous and cumulative carbon emission intensity (g / km). When it is predicted that carbon emissions will exceed the dynamic threshold in the next 10 seconds, it provides guidance to the driver via the instrument panel, HUD, or voice prompts, such as coasting suggestions, reducing throttle, and switching to pure electric mode. It generates a low-carbon driving score and carbon emission report, supporting carbon credit redemption or carbon asset account management.

[0052] The method for determining dynamic thresholds and the triggering logic for guidance prompts: Dynamic threshold calculation: Let the reference carbon emission intensity of the current vehicle under standard operating conditions (such as WLTC) be... (g / km). Dynamic threshold Adjustments are made in real time based on road conditions ahead:

[0053] in (15% increase for every 1% slope) We set the values ​​to 0.1 (mild congestion), 0.3 (moderate congestion), and 0.6 (severe congestion). (Air conditioning is on). A sliding window correction is also introduced: the average actual carbon emissions over the past 60 seconds are used. ,like Then, the threshold is adaptively offset.

[0054] Guided prompt triggering logic: Using a 1-second period, the S2 module predicts the vehicle speed sequence and engine / motor power demand for the next 10 seconds, estimating the cumulative carbon emissions for the next 10 seconds. (g). If If this occurs, an alert will be triggered. The specific guidance type is determined based on the magnitude and duration of the predicted exceedance: Amplitude ≤10% and duration ≥2 seconds → Slight throttle release prompt (green icon on HUD); A 10%~30% increase → Coasting suggestion or voice prompt to reduce throttle; If the amplitude is greater than 30%, it is strongly recommended to switch to pure electric mode (if it is a hybrid) or turn on ECO mode; When encountering a predicted stop ahead (probability of stop > 0.7), a prompt to release the accelerator and coast to the intersection is issued 5 seconds in advance. If the driver does not respond after the prompt and the actual carbon emissions still exceed the threshold, the system will automatically reduce the air conditioning power or adjust the motor assist strategy.

[0055] Cloud-based collaborative optimization platform: Employing a federated learning architecture, this platform aggregates anonymized fleet operation data with road network carbon emission hotspot information, dynamically updating low-carbon power strategy models and high-precision carbon emission maps. It supports logistics companies in monitoring fleet carbon emissions, scheduling low-carbon routes, and verifying carbon assets, responding to the demands of the carbon trading market.

[0056] The vehicle-wide low-carbon power coordinated control method based on the above system includes the following steps: S1: Connected Data Acquisition and Fusion: Dynamic road conditions, maps, and vehicle status data are acquired through V2X roadside units, vehicle GPS / IMU, high-precision map cloud services, and CAN bus. After spatiotemporal alignment, anomaly detection, and Kalman filtering, the data is fused and stored in the "road-vehicle-carbon" dynamic database.

[0057] S2: Global Operating Condition Prediction: Calls a pre-trained spatiotemporal graph convolutional-transformer model, takes the fused historical sequence as input, and predicts the vehicle speed, acceleration, gradient, and parking probability sequence for the next 5-30 seconds, outputting an operating condition feature vector.

[0058] S3: Low-carbon power collaborative decision-making: The predicted operating conditions are input into the model predictive controller. With the goal of minimizing carbon emissions and energy consumption, the optimal control sequence is solved by combining reinforcement learning correction terms, and the output commands such as engine / motor torque, gear position, and braking feedback intensity are output.

[0059] S4: Execution and Carbon Emission Feedback: Commands are issued to the powertrain and chassis domains via the vehicle controller, real-time fuel / electricity consumption data is collected and converted into carbon emissions, carbon accounts are updated, and driving guidance information is generated. Anonymized operational data is periodically uploaded to the cloud, and model parameters are optimized through federated learning to achieve continuous strategy iteration.

[0060] The following are specific implementation examples of this application: Example 1: Implementation Process of Low-Carbon Power Co-control for Heavy-Duty Commercial Vehicles (Long-Distance Freight) 1. Applicable scenario: This is designed for long-haul diesel heavy-duty trucks operated by a logistics company, with a load capacity of 25 tons, and routes that include long downhill sections, mountainous areas, and highways. It aims to solve the problems of high fuel consumption and carbon emissions, as well as the strong reliance on driver experience.

[0061] 2. Implementation steps: S1: Connected Data Acquisition and Fusion: Obtain traffic event and traffic light information 5km ahead via V2X, extract road slope sequence (sampling interval 50m) by combining high-precision map, and fuse vehicle CAN data (vehicle speed, engine torque, load estimation) to build a road-vehicle-carbon database.

[0062] S2: Global Driving Condition Prediction: The spatiotemporal graph convolutional-Transformer model uses 10 seconds of historical data to predict the vehicle speed and gradient sequence for the next 30 seconds. The model identifies a continuous 4% uphill slope 2km ahead and a long 6% downhill slope 1.2km ahead.

[0063] S3: Low-carbon power collaborative decision-making: Model predictive control is based on predicted operating conditions and aims to minimize carbon emissions. It reduces the cruising speed in advance, adjusts the engine MAP operating point, and prompts appropriate braking recovery (if it is a hybrid) or engine braking intervention before long downhill sections to avoid unnecessary fuel consumption.

[0064] S4: Execution and Feedback: Control commands are sent to the engine ECU and transmission TCU to achieve predictive deceleration and gear optimization. The carbon emission monitoring module calculates in real time that the carbon emissions for this trip are 11.3% lower than regular driving, and the driver receives a low-carbon driving score and corresponding carbon credit rewards.

[0065] 3. Implementation results: Fuel consumption per vehicle per 100 kilometers is reduced by 2.8L, carbon emissions are reduced by about 8.9kg / 100km, the fleet can reduce carbon emissions by about 280 tons per year, and reduce fuel costs by about 350,000 yuan per year.

[0066] Example 2: Implementation Process of Low-Carbon Collaborative Control for Urban Commuting of Plug-in Hybrid Passenger Vehicles 1. Applicable scenario: A user drives a PHEV passenger vehicle and his daily commute route includes congested city roads, traffic light intersections, and urban expressways. The user often causes the engine to start and stop frequently and fuel consumption to be high due to improper battery management.

[0067] 2. Implementation steps: S1: Connected Data Acquisition and Fusion: Real-time traffic flow data and traffic light phase information are obtained through mobile phone interconnection, combined with vehicle GPS and gyroscope, and integrated with battery SOC, vehicle speed, and accelerator pedal signals.

[0068] S2: Global operating condition prediction: The model predicts the probability of three consecutive red lights and queue lengths at intersections within 500m ahead, and predicts that the driving conditions will be deceleration-idle-start sequence within the next 15 seconds.

[0069] S3: Low-carbon powertrain collaborative decision-making: Model predictive control increases electric motor assistance in advance, delays engine start, and optimizes braking feedback intensity to recover more energy. Based on red light countdown information, it provides suggestions for coasting by releasing the accelerator.

[0070] S4: Execution and Feedback: In actual driving, the number of engine start-stop cycles decreased from 9 to 3, and the proportion of electric drive increased to 78%. The carbon emission monitoring module showed a 13.2% reduction in carbon emissions per trip, and the driver's HUD displayed the emission reduction results in real time.

[0071] 3. Implementation results: Combined fuel consumption decreased from 5.6L / 100km to 4.8L / 100km, carbon emissions decreased to about 110g / km, and annual CO2 emissions were reduced by about 0.36 tons in urban commuting scenarios, while improving the smoothness of pure electric driving and the economic benefits for users.

[0072] Example 3: Cloud-based collaborative low-carbon scheduling and carbon asset management for commercial buses (urban public transport) 1. Applicable scenario: A bus company operates 20 hybrid buses with fixed routes, but fuel consumption fluctuates greatly due to morning and evening rush hours. The company wants to quantify carbon emission reductions and participate in carbon trading.

[0073] 2. Implementation steps: S1: Data Acquisition and Fusion: Each vehicle communicates with the roadside unit via V2X to obtain traffic light priority request information and station passenger flow prediction, and integrates vehicle load, SOC, and engine data.

[0074] S2: Overall Traffic Condition Prediction: The cloud-based model integrates all vehicle data to predict the traffic situation of each route in the next 5 minutes and generates prompts such as "Congestion Ahead - Suggest Switching to Pure Electric Vehicle in Advance".

[0075] S3: Low-carbon powertrain collaborative decision-making: The vehicle controller adopts a pure electric queuing start-stop strategy based on predictions to reduce idling emissions. The cloud platform aggregates data based on federated learning and pushes differentiated charging strategies to the fleet.

[0076] S4: Carbon Emission Feedback and Asset Management: The system generates monthly carbon emission reports for the bus fleet, verifying carbon emission reductions. The bus company will sell the verified emission reductions (approximately 420 tons per year) through a carbon trading platform to generate revenue.

[0077] 3. Implementation Results: The overall carbon emissions of the bus fleet decreased by 10.8%, and the average daily fuel cost savings per vehicle were approximately 45 yuan. The annual carbon trading revenue was approximately 25,000 yuan, forming a positive cycle of "low-carbon operation - cost savings - carbon asset returns".

[0078] To further demonstrate the technical effectiveness of this application, models of a heavy-duty commercial vehicle (parameters same as in Example 1) and a plug-in hybrid passenger vehicle (parameters same as in Example 2) were built based on the MATLAB / Simulink and CarSim co-simulation platform. WLTC, CLTC-P, and actual urban congestion scenarios were then run. The system proposed in this application (hereinafter referred to as "this system") is compared with the following two existing technologies: Comparison Method 1: Rule-based energy management strategy (power consumption - power maintenance mode); Comparison Method 2: Traditional Model Predictive Control (excluding road condition prediction, based only on instantaneous operating conditions).

[0079] Figure 6A comparison of carbon emission intensity under WLTC conditions is presented for the three methods. The carbon emission of this system in commercial vehicles is 892 g / km, a reduction of 8.8% compared to Comparative Method 1 (978 g / km) and 5.3% compared to Comparative Method 2 (942 g / km); in passenger vehicles, the carbon emission is 112 g / km, a reduction of 12.5% ​​compared to Comparative Method 1 (128 g / km).

[0080] Figure 7 The instantaneous carbon emission curves of passenger vehicles on a real-world urban commuting route (including 6 traffic lights and 3 congestion sections) are displayed, along with the dynamic threshold (dashed line) and system-triggered coasting suggestions (blue area) and throttle reduction (red area). It can be seen that this system can predict the peak carbon emission 5-10 seconds in advance and provide guidance, resulting in an actual peak carbon emission reduction of approximately 23% compared to Comparative Method 2.

[0081] Figure 8 The changes in State of Charge (SOC) and the number of engine start-stop cycles in hybrid vehicles under this system and those under traditional MPC were compared. This system, through operating condition prediction, actively recovers electrical energy before red lights, resulting in a more stable SOC and reducing the number of engine start-stop cycles from 9 times / 100 km to 3 times / 100 km, effectively avoiding inefficient power generation.

[0082] The simulation results above fully verify the effectiveness of the mathematical model and algorithm supplemented in this application, and achieve the expected low-carbon power synergy effect.

[0083] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for low-carbon power coordinated control of vehicles across all operating conditions based on networked road condition prediction, characterized in that, Includes the following steps: S10, acquire real-time multi-source data, and use a pre-built road vehicle carbon dynamic database to fuse and spatiotemporally align the real-time multi-source data to obtain the current fused data; S20, based on the current fused data and historical operating condition data, the driving operating condition sequence within a preset time period is predicted in a rolling manner through a pre-trained prediction model, generating an operating condition feature vector containing vehicle speed, acceleration, gradient, and parking probability. S30, taking the operating condition feature vector as input, performs dual-objective optimization with minimizing carbon emissions as the upper-level objective and minimizing energy consumption as the lower-level objective, and outputs power coordinated control commands; S40, issue the power coordination control command, collect the vehicle's energy consumption data in real time and convert it into carbon emissions, update the carbon account and generate driving guidance information based on the carbon emissions, and dynamically correct the control parameters used in the dual-objective optimization based on the carbon emissions.

2. The vehicle-wide low-carbon power collaborative control method based on networked road condition prediction as described in claim 1, characterized in that, In step S10, the real-time multi-source data includes dynamic traffic information, high-precision map data, vehicle operating status data, and carbon emission factor data; the dynamic traffic information includes real-time traffic flow density, traffic light timing, accidents, and construction events; the high-precision map data includes road slope, curvature, speed limit, and number of lanes; the vehicle operating status data includes vehicle speed, acceleration, engine speed, torque, battery state of charge, and transmission gear; and the carbon emission factor data includes conversion factors from fuel consumption and electricity consumption to carbon emissions under different operating conditions.

3. The vehicle-wide low-carbon power collaborative control method based on networked road condition prediction as described in claim 1, characterized in that, In step S20, the pre-trained prediction model adopts a multimodal prediction architecture, fusing the vehicle's own driving trajectory with the motion features of surrounding vehicles, and continuously predicts the vehicle speed sequence and operating condition type for the next 5 to 30 seconds; the pre-trained prediction model is a spatiotemporal graph convolution and Transformer hybrid architecture model, specifically including: Spatiotemporal graph convolutional layers are used to extract spatial dependencies and local temporal features between vehicles in a transportation network. The graph convolution operation formula is as follows: ,in , A For dynamically time-warped adjacency matrices, For degree matrix, For the first Layer node features For learnable weights, Activate ReLU; The Transformer encoder, used to capture long-range temporal dependencies, has the following self-attention calculation expression: ,in It is obtained by linear transformation of the output of the spatiotemporal graph convolutional layer. d k for K The dimension of a vector; The prediction output layer maps the output of the Transformer encoder to the vehicle speed, acceleration, gradient, and parking probability at future prediction time points.

4. The vehicle-wide low-carbon power collaborative control method based on networked road condition prediction as described in claim 1, characterized in that, In step S20, the input is represented as: ; The mathematical model for the bi-objective optimization is as follows: ； The power coordination control command is expressed as follows: ; In the formula, For vehicle speed; The battery is in its state of charge. For gear; This refers to engine torque; This refers to the motor torque. Fuel consumption rate; For power consumption; Carbon emission factor; For time; , These are the carbon emission weighting coefficient and the energy consumption weighting coefficient; It is a fuel with a low calorific value; Battery power; The weighting coefficients are optimized using a hierarchical approach, with the upper layer using... Find the Pareto front with the goal of minimization; The lower layer meets the carbon emission constraints of the upper layer. Minimize under the premise ; For engine start / stop; This is the motor torque coefficient; This is a gear shift command; This represents the air conditioner power ratio.

5. The vehicle-wide low-carbon power coordinated control method based on networked road condition prediction as described in claim 4, characterized in that, The dual-objective optimization employs a joint decision-making process combining model predictive control and Q-learning, specifically including: The operating condition sequence is used as an external reference trajectory; The optimal control sequence is obtained by solving a constrained nonlinear optimization problem in each control cycle. Define reinforcement learning state as , For the change in vehicle speed, The variable represents the change in battery state of charge, h represents the road gradient, and C represents the road congestion level. The reward function is a correction factor for the model's predictive control output. , This represents the change in carbon emissions. To measure the change in energy consumption, an ε-greedy strategy is used for exploration, and the Q-network is updated every 100 control steps to correct the model prediction error online. The revised control commands are output to the actuator.

6. The vehicle-wide low-carbon power collaborative control method based on networked road condition prediction as described in claim 1, characterized in that, In step S40, before dynamically adjusting the control parameters based on carbon emissions, a dynamic threshold calculation and guidance prompt trigger judgment are also included: Let the reference carbon emission intensity of the current vehicle under standard operating conditions be... Dynamic threshold Adjustments are made in real time based on road conditions ahead. ,in The change in slope The adjustment coefficient, Road congestion level C The adjustment coefficient, For air conditioning demand The adjustment coefficient; Take the average actual carbon emissions over the past 60 seconds ,like Then, the threshold is adaptively offset; Estimating cumulative carbon emissions over the next 10 seconds using predicted future driving conditions. ,like If so, an alert will be triggered, in which This is the average velocity factor.

7. The vehicle all-domain low-carbon power cooperative control method based on networked road condition prediction as described in claim 6, characterized in that, The driving guidance information generated after triggering the warning is determined by the predicted exceedance range and duration: when the predicted carbon emission exceedance range is no more than 10% and the duration is no less than 2 seconds, a slight lift-off prompt is generated; when the predicted carbon emission exceedance range is greater than 10% but no more than 30%, a coasting suggestion or a voice prompt to reduce the throttle is generated; when the predicted carbon emission exceedance range is greater than 30%, a mandatory suggestion to switch to pure electric mode or activate ECO mode is generated; when the predicted probability of stopping ahead is greater than 0.7, a prompt to release the throttle and coast to the intersection is generated 5 seconds in advance; if the driver does not respond and the actual carbon emission still exceeds the threshold, the air conditioning power is automatically reduced or the motor assist strategy is adjusted.

8. The vehicle-wide low-carbon power collaborative control method based on networked road condition prediction as described in claim 1, characterized in that, The method is also equipped with a communication failure and data anomaly degradation mechanism, which automatically switches to local sensor control mode when there is no network connection, map is missing or data is abnormal.

9. The vehicle-wide low-carbon power cooperative control method based on networked road condition prediction as described in claim 1, characterized in that, The method also includes a cloud-based collaborative optimization step: uploading anonymized vehicle operation data to a cloud-based collaborative optimization platform. The platform aggregates anonymized fleet operation data and road network carbon emission hotspot information based on a federated learning architecture, dynamically updates the pre-trained prediction model and the dual-objective optimization decision model, and generates a map-level carbon emission guidance service.

10. A vehicle-wide low-carbon power cooperative control system based on networked road condition prediction, characterized in that, The steps for implementing the method according to any one of claims 1 to 9 include: The connected data acquisition and fusion module is used to acquire real-time multi-source data and use a pre-built road-vehicle carbon dynamic database to fuse and spatiotemporally align the real-time multi-source data to obtain the current fused data; The global operating condition prediction module is used to predict the driving operating condition sequence within a preset time period in the future based on the current fused data and historical operating condition data, and generate an operating condition feature vector containing vehicle speed, acceleration, slope and stopping probability. The low-carbon power coordinated control module is used to perform dual-objective optimization with the operating condition feature vector as input, which has the upper-level objective of minimizing carbon emissions and the lower-level objective of minimizing energy consumption, and outputs power coordinated control commands. The carbon emission monitoring and feedback module is used to issue the power coordination control command, collect the vehicle's energy consumption data in real time and convert it into carbon emissions, update the carbon account and generate driving guidance information based on the carbon emissions, and dynamically correct the control parameters used in the dual-objective optimization based on the carbon emissions.

Images