Graph neural network-based ozone generation potential contribution weight dynamic regulation method and device for in-vehicle, equipment and storage medium

Abstract

The application provides a method and device for dynamically regulating ozone generation potential contribution weight based on a graph neural network, equipment and a storage medium, and relates to the technical field of environmental pollution control. The method comprises the following steps: synchronously collecting multi-source heterogeneous data in a vehicle; performing data cleaning and feature labeling, and calculating ozone generation potential; constructing a dynamic chemical graph with VOCs species as nodes and chemical reactions or synergistic effects as edges; learning the dynamic chemical graph by using a graph attention network model, and outputting real-time ozone generation potential prediction values and dynamic contribution weights of each pollution source; based on the weights and real-time environmental state, performing targeted purification, intelligent air volume circulation or refrigerant leakage collaborative management. The application realizes accurate prediction of ozone generation potential in the vehicle and intelligent analysis of pollution source contribution, and improves the pertinence, energy efficiency and foresight of air quality regulation in the vehicle.

Description

Technical Field

[0001] This application relates to the field of environmental pollution control technology, and in particular to a method, device, equipment and storage medium for dynamic control of the contribution weight of ozone generation potential in vehicles based on graph neural networks. Background Technology

[0002] Modern life, with its long commutes and long-distance road trips, poses significant health risks. In-vehicle ozone is an air health risk closely related to air conditioning systems, external pollution penetration, and the release of pollutants from interior materials. The WHO recommends an O3 limit of 50 ppb (8-hour average), the EPA's in-vehicle standard is 70 ppb (monitored at a constant speed of 40 km / h), and my country's GB / T27630-2011 standard sets a limit of 100 ppb for air conditioning recirculation for one hour when the vehicle is stationary.

[0003] In-vehicle ozone contribution weighting coefficient: 1. Ozone from the road environment penetrates into the vehicle. In road environments, the ozone permeability of outside air is approximately 0.6 in summer and 0.4 in winter. During peak urban traffic hours, the concentration of nitrogen oxides (NOx) in the air... X Higher ozone concentrations lead to increased ozone generation efficiency, with ambient ozone concentrations reaching 80-120 ppb. Especially when the air conditioner is in external air circulation mode, without an ozone filter, external O3 directly enters the vehicle.

[0004] 2. The high-voltage discharge of the drive system generates trace amounts of ozone, with an ozone contribution rate of approximately 0.1% for electric vehicles and approximately 0.05% for traditional gasoline vehicles. Corona discharge from onboard ionization air purifiers or faulty air conditioning circuits produces ozone concentrations of 10-30 ppb. The high-frequency switching arc of the onboard charger (OBC) in electric vehicles generates ozone concentrations of 30 ppb. High-voltage electrodes (±6kV) also produce 1-20 ppb of ozone when adsorbing particulate matter.

[0005] 3. The ozone contribution rate of the added high-voltage module is approximately 0.3. The artificial installation of an ozone sterilizer for vehicle interior disinfection resulted in leakage, causing unpleasant odors inside the cabin, with ozone concentrations reaching up to 200 ppb.

[0006] 4. Ozone decomposition accelerates under high temperature and humidity. The temperature and humidity inside the vehicle significantly affect the ozone decomposition process and steady-state concentration, with an impact coefficient of approximately -0.05. At high temperatures (>30℃), the increased release of VOCs from interior materials reacts with O3 to generate secondary pollutants, accelerating ozone decomposition to within 10-20 minutes compared to approximately 1 hour at normal temperature. For every 10℃ increase in temperature, the steady-state ozone concentration decreases by 15-20%. At high humidity (RH>70%), water molecules react with O3 to generate hydroxyl radicals (·OH), accelerating ozone decomposition. For every 20% increase in humidity, the ozone concentration decreases by 10-12%.

[0007] In-vehicle ozone management technology has evolved from early static models to dynamic intelligent control systems. By integrating multi-source sensor data, real-time environmental data, and vehicle data, and utilizing machine learning to dynamically analyze the contribution of each pollution source, control strategies ranging from source suppression and rapid decomposition to intelligent purification mode switching have been developed. This invention emphasizes the integration of monitoring and predictive capabilities, enabling dynamic compensation for various sources, including interior volatile organic compounds, achieving comprehensive, adaptive, and systematic management.

[0008] Refrigerant leaks from automotive air conditioning systems can enter the passenger compartment through the air vents of the evaporator. As the seals on the expansion valve / temperature sensor interface age, refrigerant can seep into the vehicle through gaps in the wiring harness. Hydrofluoroolefin (HFO) refrigerants are widely used to replace hydrofluorocarbon (HFC) refrigerants (such as R134a with a GWP of 143) due to their low global warming potential (GWP = 1 for HFO-1234yf). Although HFOs have a GWP 2-3 orders of magnitude lower, their ozone generation potential (OFP) is significantly higher than that of HFCs (OFP ≈ 0.02 g O3 / g for HFC-134a, and 4.8 g O3 / g for HFO-1234yf). Hydrofluoroolefin (HFO) refrigerants, such as HFO-1234yf (CF3CF=CH2), although containing double bonds, can absorb ultraviolet light of 200-300nm (mainly in the upper stratosphere / troposphere) and undergo direct photolysis to generate acetyl radicals, participating in chain reactions to promote the formation of O3. However, activation requires extreme conditions. Therefore, the OFP value is measured under conditions of high NOx concentration (>200ppb) + strong ultraviolet light (UV-C band, <280nm). At room temperature and pressure, HFO-1234yf is a non-reactive substance, and its OFP is close to zero. The Chinese national standard limit (GB18352.6-2016) stipulates that the leakage rate of vehicle air conditioning systems is ≤ 1.5g / yr, corresponding to an upper limit of 7.2gO3 / vehicle·year for OFP. In reality, the leakage concentration of automotive air conditioning refrigerants is usually <1ppm (far below the reaction threshold). Even if decomposition occurs, the efficiency of ozone generation is less than 0.001% of the total OFP.

[0009] Although the amount of refrigerant leaked from the vehicle's air conditioning system is small and does not directly contribute to OFP, it may synergistically interact with other pollutants in the vehicle (such as VOCs, NOx, PM2.5, CO, etc.). First, physical mixing enhances exposure risk. HFO-1234yf has a boiling point of −29°C and rapidly vaporizes after leakage, forming a gaseous layer in the enclosed vehicle compartment. This gas mixes with existing VOCs, and the cumulative concentration effect increases the overall amount of pollutants inhaled. The leaked refrigerant may be repeatedly inhaled and diffused by the air conditioning system, combining with PM2.5 in the vehicle to form aerosol carriers that penetrate deep into the respiratory tract. Second, there are toxicological synergistic effects. For example, the mild anesthetic effect of the refrigerant and the hematopoietic toxicity of benzene may synergistically damage the central nervous system and liver. Mouse experiments show that the incidence of neurobehavioral disorders increases by 30% when R134a coexists with toluene. If the refrigerant leak is accompanied by exhaust gas infiltration, the asphyxiating effects of CO and HFO-1234yf are superimposed, leading to a decrease in blood oxygen saturation. Third, material corrosion releases additional pollutants. HFO-1234yf can generate HF upon contact with water, corroding air conditioning pipes, releasing metal ions, and forming organometallic compounds with VOCs in the vehicle, irritating the respiratory tract. Refrigerant leaks lead to high temperatures and pressures in the air conditioning system, potentially accelerating the decomposition of polyurethane foam in the vehicle and releasing toxic isocyanates. Fourth, although HFOs have approximately zero OFP, they react with ·OH radicals to form a chain oxidation reaction, producing byproducts such as trifluoroacetic acid (TFA). TFA reacts with O3 to generate more toxic PANs, and with NO2 to generate CF3NO2, which has a high greenhouse effect potential. This reaction also releases highly reactive free radicals (such as ·CF3) that generate ozone, while consuming ·OH radicals. The generated CF3OO· reacts with NO, promoting the conversion of NO to NO2, leading to an intensified photochemical reaction between NOx and VOCs, increasing ozone generation.

[0010] The photochemical reaction of NOx and VOCs in a vehicle requires the simultaneous fulfillment of three conditions: UV radiation, VOCs, and NOx release. First, ultraviolet light (UV-A, 315–400nm) is the main wavelength that triggers the photolysis of VOCs, and the windshield only filters 30% of UV-A. Under the limited UVA light resistance of the windshield, external penetration, release from interior materials, and refrigerant leakage disrupt the balance of NOx / VOCs in the vehicle, resulting in potential ozone generation.

[0011] NOx sources inside a vehicle include: 1. External infiltration: When driving in congested areas or areas with a high concentration of diesel vehicles, refrigerant leaks can cause frequent start-stop cycles of the air conditioning, leading to changes in interior temperature and humidity and increased engine load. These factors can cause large amounts of NOx from outside to enter the vehicle through the air conditioning system or windows. 2. Vehicle system malfunctions causing direct exhaust gas infiltration: Faults in systems such as the three-way catalytic converter (TWC) and exhaust gas recirculation (EGR) can cause NOx from exhaust gases to directly enter the passenger compartment. 3. Trace amounts of NO generated within the vehicle itself: High-voltage arcing of onboard electrical appliances ionizing air, tobacco combustion, and the decomposition of certain cleaning agents can all produce trace amounts of NOx.

[0012] VOCs sources in vehicles include: 1. Interior materials, such as: ① materials for seats, carpets, and dashboards, such as polyurethane foam (PU), PVC, synthetic rubber, and adhesives. Typical VOCs include: formaldehyde (HCHO, IARC 1 carcinogen), benzene compounds (benzene C6H6, toluene C7H8, ethylbenzene C8H6). 10 ① Plastics and composite materials used in center console panels, door panels, and wiring harnesses, such as ABS, PP, and EPDM, often contain plasticizers (e.g., DEHP phthalates). Typical VOCs include: olefins (e.g., ethylene C2H4, propylene C3H6, released at high temperatures) and aromatic hydrocarbons (e.g., xylene, with a strong odor); ③ Windshield sealants / sheet metal sealants / sealants and coatings used in paints, with typical VOCs including: isocyanates (e.g., TDI, from polyurethane coatings, irritating to the respiratory tract) and styrene (C8H8, a component of paint thinner). The VOC release rate of interior materials such as leather and plastics fluctuates with temperature and humidity. For every 10°C increase in temperature, the VOC release rate increases by 30–50%. During summer exposure to direct sunlight (car interior >60°C), formaldehyde release can reach 5–10 times the level at room temperature (GB / T27630 limit: 0.10 mg / m³). The aging and decomposition of plastics releases olefins and oxygen-containing organic compounds (such as caprolactam C6H). 11 NO). 2. Introduced by human activities, such as: ① Car accessories such as perfumes, cleaning agents, and air fresheners, which contain aldehydes (such as citral) and camphor, masking odors but potentially generating secondary pollutants; ② Inferior floor mats / steering wheel covers, which release sulfides (such as dimethyl sulfide C2H6S) and amines (such as triethylamine C6H). 15 N); ③ Smoking and diet: Tobacco combustion releases polycyclic aromatic hydrocarbons (PAHs) (such as benzo[a]pyrene C). 20 H 12 ), Nicotine (C 10 H 14N2), hydrogen sulfide (H2S) and ammonia (NH3) are produced from the decay of food residue. 3. Infiltration from the external environment, such as: ① Traffic exhaust, benzene series compounds and aldehydes in the outside air enter the car through the air conditioning external circulation or window gaps; diesel exhaust contains acetaldehyde and acrolein (C3H4O, highly irritating); ② Gas stations / parking lots, gasoline volatilizes and releases C5-C12 alkanes (such as n-hexane C6H12H2O). 14 ), Methyl tert-butyl ether (MTBE, C5H) 12 4. Air conditioning system lubricating oil (such as PAG / POE) leaks with the refrigerant, leading to the release of aldehydes or olefins (VOCs).

[0013] Existing in-vehicle ozone control technologies suffer from three main shortcomings: First, in terms of data utilization, they primarily rely on external macroscopic environmental data or simply overlay in-vehicle sensor readings. This fails to construct a dynamic network with inherent chemical reaction relationships among various pollutants, environmental parameters, and vehicle status within the cabin, resulting in poor adaptability and significant errors in predicting ozone formation potential under complex conditions. Second, in terms of assessment methods, ozone formation potential calculations often employ static parameters based on laboratory conditions (such as fixed MIR values), failing to respond in real-time to the dynamic impacts of temperature, humidity, light intensity, and NOx / VOCs ratio changes on reactivity, particularly regarding insufficient assessment of synergistic effects caused by events such as refrigerant leaks. Third, in terms of control strategies, most rely solely on preset concentration thresholds to trigger air conditioning mode switching or activate full-effect purification, failing to differentiate the actual contributions of different pollution sources and hindering dynamic and precise intervention. This leads to simplistic control, low energy efficiency, and an inability to proactively suppress latent but rapidly increasing pollution sources. Summary of the Invention

[0014] This application provides a method, device, equipment, and storage medium for dynamic adjustment of the contribution weight of ozone generation potential in vehicles based on graph neural networks, so as to realize real-time prediction and alarm of ozone inside the car. A feature vector containing concentration, molecular characteristics, and source attributes is constructed, and the chemical reactions and synergistic effects are characterized by the connection relationships between nodes in the graph.

[0015] Building upon this foundation, a graph neural network is used to learn the interactions between different pollutants in real time, dynamically assess the contribution weight of each pollution source, and achieve targeted purification and intelligent regulation based on this weight. This method can not only predict and warn of ozone formation potential in real time, but also improve the accuracy and adaptability of in-vehicle air quality management through dynamic contribution weight-based pollution source ranking, material selection, and system optimization.

[0016] Firstly, this application provides a method for dynamically adjusting the contribution weight of in-vehicle ozone formation potential based on graph neural networks, including: Multi-source heterogeneous data are collected synchronously at a frequency of no less than 1Hz under various typical scenarios. The multi-source heterogeneous data includes in-vehicle VOCs concentration, in-vehicle environmental parameters, vehicle status parameters, pollutant gas concentration, external data, and refrigerant leakage data. The typical scenarios include urban congestion, high-speed cruising, high-temperature exposure, and simulated refrigerant leakage conditions. All collected data streams were timestamped and aligned. Sliding window statistical tests and interpolation algorithms were used to handle sensor failures, communication packet loss, and outliers. The total ozone formation potential was calculated for each time step. OFP total The instantaneous contribution values ​​of key VOCs species and the labels of the Top-5 species are marked; the total ozone formation potential is... OFP total The calculation formula is: In the formula, The temperature established based on the Arrhenius equation and photochemical kinetics ,humidity The dynamic correction function of UV radiation on the reaction rate For the first i Real-time concentration of volatile organic compounds inside the vehicle. For the first i The standard maximum incremental reactivity of a certain volatile organic compound. The serial number represents the volatile organic compound. A dynamic chemical graph is constructed, represented as G=(V,E), where nodes V represent each VOC species, and the node feature vector includes the standardized concentration value, the standard maximum incremental reactivity MIR value, the molecular descriptor, and the source code; edges E characterize the potential chemical reaction associations or environmental synergistic effects between species, and the initial weights of the edges are set based on known atmospheric chemical mechanisms. A graph neural network model is trained based on the constructed dynamic chemical graph and training data. The graph neural network model adopts the graph attention network GATv2 architecture and outputs real-time OFP prediction values ​​and dynamic contribution weight vectors of each VOC species. Based on real-time OFP prediction values ​​and dynamic contribution weight vectors, combined with the NOx / VOCs ratio, control measures are implemented, including targeted purification, intelligent air volume and circulation mode switching, and / or refrigerant leakage collaborative management.

[0017] In one possible design, the multi-source heterogeneous data specifically includes: The concentration of VOCs in the vehicle was monitored using proton transfer reaction mass spectrometry (PTR-MS) or a high-precision photoionization detector (PID) to detect various key VOC species, including toluene, xylene, ethylene, propylene, formaldehyde, acetaldehyde, and CF3CHO. In-vehicle environmental parameters, including temperature, relative humidity, and UV-A and UV-B radiation intensity; Vehicle status parameters, including real-time vehicle speed, air conditioning operating mode and fan speed, and air conditioning system pressure; Concentration of polluting gases, including NOx and O3 concentrations inside the vehicle; External data, including external NOx and O3 concentrations and local meteorological information, are obtained from roadside units (RSUs) via V2X communication; Refrigerant leakage data, HFO-1234yf leakage rate monitored by infrared sensors.

[0018] In one possible design, the types of edge E mentioned above include OH radical competitive edges, ozone generation cooperative edges, secondary pollutant generation edges, and environmental regulation edges. The reaction relationships corresponding to each type of edge are as follows: The OH radical competes for the oxidizing edge, connecting to a VOC species that competes for the OH· oxidant, and the reaction is VOC. i +OH·→Products i VOCs j +OH·→Products j VOC i and VOCs j Represents the i-th and j-th volatile organic compound species, Products i and Products j VOC i and VOCs j The products of the reaction with OH· are OH· radicals. Ozone generation synergistic edges connect species that share or jointly promote ozone generation chain reaction pathways, including VOCs. i +OH·→RO2·→NO→NO2+HO2·→O3、HO2·+NO→OH·+NO2; RO2· represents peroxyalkyl radical, HO2· represents peroxyhydroxyl radical; The secondary pollutant generation side connects the precursor to the secondary pollutant it generates, and the reaction formula includes CF3CHO+OH·→CF3C(O)O2·+PANs; CF3C(O)O2· represents trifluoroacetyl peroxy free radical; The environmental control edge uses environmental parameters as supernodes to connect all VOCs nodes affected by them, reflecting the impact of high temperature / UV radiation on VOCs release rate and reaction rate.

[0019] In one possible design, the architecture of the graph neural network model specifically includes: The input layer projects the node feature vectors into a 128-dimensional space. The graph attention layer consists of two GATv2 layers. The first layer has four attention heads, each with a 64-dimensional output, for a total of 256 dimensions. The second layer has one attention head, which aggregates the output information from the previous layer and outputs a 128-dimensional output. The output layer outputs the total OFP value through global average pooling and fully connected layers at the graph level; the node-level output outputs the contribution weight of each node through independent fully connected layers, which is used for Top-5 species identification.

[0020] In one possible design, the graph neural network model training process employs the following multi-task loss function: Loss = λ 1* MSE ( OFP pred , OFP true ) +λ 2 *CrossEntropy ( Importance pred , Importance true ) +λ 3 *Loss scenario In the formula, λ 1. λ 2 and λ 3 represents the weighting coefficients of each component of the loss function. MSE ( OFP pred , OFP true () is the predicted value of ozone formation potential. OFP pred Compared with the true value OFP true The mean square error between them CrossEntropy ( Importance pred , Importance true Predicted values ​​of the contribution weights of each VOC species Importance pred Compared with the true value Importance true Cross-entropy.

[0021] In one possible design, the targeted purification includes: identifying the 1-3 key VOC species with the highest contribution weight and activating the targeted purification module; if the total weight of xylene exceeds 40%, activating the molecular sieve with high adsorption rate for aromatic hydrocarbons. The intelligent airflow and circulation mode switching includes: when NOx / VOCs < 0.28 and the internal VOCs contribution weight is high, the vehicle-mounted activated carbon filter is activated and the risk of volatilization from interior materials is alerted; when NOx / VOCs > 0.35, the system switches to internal circulation mode, broadcasts NOx red zone information through V2X technology, and triggers photocatalytic or mini selective catalytic reduction (SCR) devices. The refrigerant leak collaborative management includes: when a refrigerant leak is detected and the weight of products such as CF3CHO increases, a manganese-based catalyst is activated to decompose carbonyl compounds in a targeted manner, the air conditioning operating parameters are dynamically optimized, data is recorded, and maintenance services are scheduled through the vehicle network system.

[0022] In one possible design, the training data includes three types of labeled data: a basic scenario, a refrigerant leakage scenario, and a NOx / VOCs control zone scenario; wherein: The basic scenario uses data segments with normal driving conditions, no refrigerant leakage, and NOx / VOCs ratio within the typical range, labeled as OFP calculated based on standard MIR values; The refrigerant leakage scenario selects data segments with HFO-1234yf leakage rates exceeding 1 gram / year, and adds leakage intensity features to byproduct nodes such as CF3CHO. The NOx / VOCs control zone scenario is divided into a VOCs control zone and a NOx control zone based on the NOx / VOCs ratio.

[0023] Secondly, this application provides a device for dynamic adjustment of the contribution weight of in-vehicle ozone generation potential based on graph neural networks, the device comprising: The data acquisition module is configured to synchronously collect multi-source heterogeneous data at a frequency of no less than 1Hz under various typical scenarios. The multi-source heterogeneous data includes in-vehicle VOCs concentration, in-vehicle environmental parameters, vehicle status parameters, pollutant gas concentration, external data, and refrigerant leakage data. The typical scenarios include urban congestion, high-speed cruising, high-temperature exposure, and simulated refrigerant leakage conditions. The data processing module is configured to timestamp-align all acquired data streams, employ sliding window statistical tests and interpolation algorithms to handle sensor failures, communication packet loss, and outliers, and calculate the total ozone formation potential at each time step. OFP total The instantaneous contribution values ​​of key VOCs species and the labels of the Top-5 species are marked; the total ozone formation potential is... OFP total The calculation formula is: In the formula, The temperature established based on the Arrhenius equation and photochemical kinetics ,humidity The dynamic correction function of UV radiation on the reaction rate For the first i Real-time concentration of volatile organic compounds inside the vehicle. For the first i The standard maximum incremental reactivity of a certain volatile organic compound. The serial number represents the volatile organic compound. The graph construction module is configured to construct a dynamic chemical graph, represented as G=(V,E), where nodes V represent each VOC species, and the node feature vector includes the standardized concentration value, the standard maximum incremental reactivity MIR value, the molecular descriptor, and the source code; edges E characterize the potential chemical reaction associations or environmental synergistic effects between species, and the initial weights of the edges are set based on known atmospheric chemical mechanisms. The model training module is configured to train a graph neural network model based on the constructed dynamic chemical graph and training data. The graph neural network model adopts the graph attention network GATv2 architecture and outputs real-time OFP prediction values ​​and dynamic contribution weight vectors of each VOC species. The control execution module is configured to execute control measures based on real-time OFP prediction values ​​and dynamic contribution weight vectors, combined with the NOx / VOCs ratio. These control measures include targeted purification, intelligent airflow and circulation mode switching, and / or refrigerant leakage collaborative management.

[0024] Thirdly, embodiments of this application provide an electronic device, including: at least one processor and a memory; the memory stores computer execution instructions; the at least one processor executes the computer execution instructions stored in the memory, causing the at least one processor to perform the dynamic adjustment method for the contribution weight of in-vehicle ozone generation potential based on graph neural networks as described in the first aspect and various possible designs of the first aspect.

[0025] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions. When a processor executes the computer-executable instructions, it implements the method for dynamic adjustment of in-vehicle ozone generation potential contribution weight based on graph neural networks as described in the first aspect and various possible designs of the first aspect.

[0026] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the method for dynamic adjustment of in-vehicle ozone generation potential contribution weight based on graph neural networks as described in the first aspect and various possible designs of the first aspect.

[0027] The method, apparatus, device, and storage medium for dynamic control of in-vehicle ozone generation potential contribution weights based on graph neural networks provided in this application have at least the following beneficial effects: This application demonstrates significant comprehensive advantages over existing technologies. Firstly, by constructing a dynamic chemical graph and utilizing graph neural networks for modeling, it achieves high-precision dynamic prediction of the ozone generation potential inside the vehicle, significantly reducing prediction errors compared to traditional methods. Simultaneously, the model can accurately analyze and quantify the real-time contribution of each pollution source, thereby intelligently identifying key pollutants. Furthermore, the targeted purification strategy based on dynamic contribution weights significantly improves the removal efficiency of individual pollutants and reduces the overall system energy consumption. In addition, the system possesses proactive collaborative management capabilities for complex events such as refrigerant leaks, enabling a shift from post-treatment to pre-event warning and process intervention. Ultimately, the entire system forms a complete intelligent closed loop of perception, assessment, decision-making, control, and optimization. It not only intelligently selects the optimal control strategy based on the real-time chemical environment but also provides scientific data support for material selection and health management, achieving adaptive and systematic efficient management of in-vehicle air quality. Attached Figure Description

[0028] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0029] Figure 1 A flowchart of a method for dynamically adjusting the contribution weight of in-vehicle ozone generation potential based on graph neural networks, provided in an embodiment of this application; Figure 2 A flowchart illustrating the model deployment and strategy execution provided in this application embodiment; Figure 3 This is a structural diagram of a device for dynamically adjusting the contribution weight of in-vehicle ozone generation potential based on a graph neural network, provided in an embodiment of this application.

[0030] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concepts of this application to those skilled in the art through reference to specific embodiments. Detailed Implementation

[0031] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0032] The collection, storage, use, processing, transmission, provision, and disclosure of financial data or user data involved in the technical solution of this application all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0033] It should be noted that in the embodiments of this application, certain software, components, models and other existing solutions in the industry may be mentioned. These should be regarded as exemplary and are only intended to illustrate the feasibility of implementing the technical solution of this application. However, it does not mean that the applicant has used or necessarily used the solution.

[0034] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0035] With the development of intelligent automotive cabins, the problem of ozone formation from volatile organic compounds (VOCs) and nitrogen oxides (NOx) in vehicles under photochemical reactions is becoming increasingly prominent. Traditional static control methods are difficult to adapt to complex dynamic environments, necessitating a precise control technology that can integrate multi-source data and respond to dynamic conditions in real time. Therefore, this application provides a dynamic control method for the contribution weight of ozone formation potential in vehicles based on graph neural networks. This method constructs a real-time controllable chemical graph model of various pollutants, environmental conditions, and vehicle status in the vehicle. It uses GNNs to capture the complex dynamic relationships between different VOCs and between VOCs and environmental factors such as temperature, humidity, and light, thereby accurately assessing the real-time contribution of various pollution sources to ozone formation. Based on this dynamic weight assessment, the purification module can be activated selectively, and the operating strategy can be adjusted according to the real-time environment to achieve effective prediction and precise control of ozone pollution in the vehicle. Specifically, as... Figure 1 As shown, the method for dynamic adjustment of the contribution weight of in-vehicle ozone generation potential based on graph neural network includes the following steps S10-S50.

[0036] S10: Under various typical scenarios, synchronously collect multi-source heterogeneous data at a frequency of not less than 1Hz; the multi-source heterogeneous data includes in-vehicle VOCs concentration, in-vehicle environmental parameters, vehicle status parameters, pollutant gas concentration, external data, and refrigerant leakage data; the typical scenarios include urban congestion, high-speed cruising, high-temperature exposure, and simulated refrigerant leakage conditions.

[0037] Specifically, in step S10, under various typical scenarios (including urban congestion, high-speed cruising, high-temperature exposure, and simulated refrigerant leakage), the following multi-dimensional data are collected synchronously at a frequency of no less than 1Hz: 1) In-vehicle VOCs concentration: at least 20 key VOC species, including toluene, xylene, ethylene, propylene, formaldehyde, acetaldehyde, CF3CHO, etc., were monitored by proton transfer reaction mass spectrometry (PTR-MS) or high-precision photoionization detector (PID). 2) In-vehicle environmental parameters, including temperature (°C), relative humidity (%), and UV-A (315–400nm) and UV-B (280–315nm) radiation intensity (W / m²). 2 ); 3) Vehicle status parameters, including real-time vehicle speed (km / h), air conditioning operating mode (specifically, internal or external circulation status and fan speed), and air conditioning system pressure (kPa). 4) Concentration of polluting gases, including NOx and O3 concentrations inside the vehicle; 5. External data: Real-time acquisition of external NOx and O3 concentrations and local meteorological information from roadside units (RSUs) via V2X communication; 6) Refrigerant leakage data, HFO-1234yf leakage rate monitored by infrared sensors (in g / s or g / year).

[0038] S20: Timestamp align all collected data streams; use sliding window statistical tests and interpolation algorithms to handle sensor failures, communication packet loss, and outliers; calculate the total ozone formation potential at each time step. OFP total The instantaneous contribution value of key VOCs species and the Top-5 species labels were also marked.

[0039] Specifically, in step S20, timestamps are aligned for all collected data streams; and sliding window statistical tests and interpolation algorithms are used to handle sensor failures, communication packet loss, and outliers.

[0040] The target value calculation and labeling includes two points: first, the calculation of the true ozone formation potential (OFP), and second, the total ozone formation potential. OFP total The calculation formula is: In the formula, The temperature established based on the Arrhenius equation and photochemical kinetics ,humidity The dynamic correction function of UV radiation on the reaction rate For the first i Real-time concentration of volatile organic compounds inside the vehicle. For the first i The standard maximum incremental reactivity of a certain volatile organic compound. This refers to the serial number of the volatile organic compounds.

[0041] Second, the contribution of key species is ranked, and the instantaneous contribution value of each VOCs species is calculated. Based on this, Top-5 species labels are generated for each time step, which are then used for supervised learning of the subsequent model.

[0042] S30: Construct a dynamic chemical graph, represented as G=(V,E), where nodes V represent each VOC species, and the node feature vector includes the standardized concentration value, the standard maximum incremental reactivity MIR value, the molecular descriptor, and the source code; edges E characterize the potential chemical reaction associations or environmental synergistic effects between species, and the initial weights of the edges are set based on known atmospheric chemical mechanisms.

[0043] Specifically, in step S30, constructing the graph structure G=(V,E) is the core focus of the method in this application, whereby... Node V: Each VOC species is defined as a node vi. Its node feature vector contains multidimensional attributes such as the normalized concentration value, the standard maximum incremental reactivity (MIR) value of the species, molecular descriptors (such as the number of C=C carbon bonds, molecular weight), and source encoding (such as interior decoration / fuel / anthropogenic / refrigerant byproducts); Edge E: Represents potential chemical reaction linkages or environmental synergies between species. The initial weights of the edges are set based on known atmospheric chemical mechanisms, and the types of the main edges and their representative reactions are shown in Table 1 below.

[0044] Table 1. Edge Types, Chemical Reaction Formulas, and Physical Significance

[0045] Note: The "·" after the symbol represents a free radical (containing unpaired electrons), which has strong reactivity and is the key to driving the chain reaction; the subscripts i and j are only used to distinguish different types of VOCs and have no special chemical meaning, corresponding to specific pollutants such as toluene, xylene, and formaldehyde actually present in the vehicle.

[0046] S40: Train a graph neural network model based on the constructed dynamic chemical graph and training data. The graph neural network model adopts the graph attention network GATv2 architecture and outputs real-time OFP prediction values ​​and dynamic contribution weight vectors of each VOC species.

[0047] In this embodiment, to train the dynamic correction capability of the GNN model (Graph Neural Network model), the following three key scenarios are constructed and labeled in the training data: Basic scenario: Using data segments with normal driving conditions, no refrigerant leakage, and NOx / VOCs ratios within typical ranges, the model is labeled as using standard MIR values ​​to calculate OFP, establishing a basic understanding of ozone generation.

[0048] Refrigerant leakage scenario: Select data segments with HFO-1234yf leakage rate exceeding 1 gram / year, add leakage intensity features to byproduct nodes such as CF3CHO in the node features, strengthen the edges between them and NOx / VOCs nodes at the graph structure level, inject dynamic compensation factors, and guide the model to learn dynamic correction of MIR.

[0049] NOx / VOCs Control Zone Scenario: Based on the NOx / VOCs ratio, the data is divided into a VOCs control zone (NOx / VOCs ratio < 0.28) and a NOx control zone (NOx / VOCs ratio > 0.35), allowing the model to learn the distribution of contribution weights under different chemical mechanisms. The NOx / VOCs ratio refers to the ratio of NOx to VOCs concentrations in the target scenario (e.g., inside a vehicle).

[0050] The GNN model's specific architecture adopts Graph Attention Network v2, which has stronger dynamic expressive capabilities. It specifically includes an input layer, a graph attention layer, and an output layer.

[0051] The input layer projects the node feature vectors hi into a 128-dimensional space. The graph attention layer consists of two GATv2 layers. The first GATv2 layer comprises four attention heads, each outputting 64 dimensions, for a total of 256 dimensions. The second GATv2 layer comprises one attention head, which aggregates the output information from the previous layer's multiple heads, outputting 128 dimensions. The attention head contains the attention mechanism, the core formula of which is: α ij =softmax j (LeakyReLU(a T [W hi ||W hj ])), where a and W are learnable parameters, || denotes concatenation, α ij The update reflects the real-time correction of edge weights.

[0052] The output layer is used for graph-level output. It performs global average pooling on the final embedding of all nodes, followed by a fully connected layer, and outputs the total OFP value (for regression tasks). Node-level output: An independent fully connected layer processes the final embedding of each node and outputs its importance score (contribution weight) for Top-5 species identification.

[0053] During training, the training data is divided into a training set (70%), a validation set (15%), and a test set (15%). A multi-task loss function is used. Loss = λ 1* MSE ( OFP pred , OFP true )+λ 2 *CrossEntropy ( Importance pred , Importance true ) +λ 3 *Loss scenario In the formula, λ 1. λ 2 and λ 3 represents the weighting coefficients of each component of the loss function. MSE ( OFP pred , OFP true () is the predicted value of ozone formation potential. OFP pred Compared with the true value OFP true The mean square error between them CrossEntropy ( Importance pred , Importance true Predicted values ​​of the contribution weights of each VOC species Importance pred Compared with the true value Importance true Cross-entropy.

[0054] The evaluation results of the trained model on the independent test set are shown in Table 2.

[0055] Table 2 Model Evaluation Structure

[0056] S50: Based on real-time OFP prediction values ​​and dynamic contribution weight vectors, combined with the NOx / VOCs ratio, control measures are implemented, including targeted purification, intelligent air volume and circulation mode switching, and / or refrigerant leakage collaborative management.

[0057] The purpose of step S50 is to deploy and utilize the trained GNN model. In practice, the optimized model can be lightweighted and packaged, and deployed in an automotive-grade embedded computing unit. This unit receives data streams from various in-vehicle sensors in real time via the vehicle bus and runs the model for online inference. The system collects VOCs concentration, environmental parameters, vehicle status, and refrigerant leakage signals at set intervals, dynamically constructs a chemical map and inputs it into the model, and outputs the ozone generation potential prediction value and the contribution weight vector of each pollution source in real time. Subsequently, the vehicle controller automatically adjusts the air conditioning damper circulation mode, starts and stops specific purification modules, adjusts the fan speed, and sends warnings or suggestions to the user through the in-vehicle infotainment system when necessary, based on the output results and preset strategies. At the same time, the system can upload the operating data and diagnostic results to the cloud platform via the vehicle network for model iteration optimization and fleet-level health management.

[0058] Specifically, such as Figure 2 As shown, step S50 specifically includes the following steps: S501: The trained GNN model is deployed on the vehicle-mounted embedded computing unit. The system collects sensor data in real time, constructs the dynamic chemical map at the current moment, inputs it into the model for inference, and outputs the real-time OFP prediction value and dynamic contribution weight vector.

[0059] The dynamic contribution weight vector W = [w1, w2, ..., wn], where w i Indicates the first i Normalized contribution weights of each VOC species to the current OFP.

[0060] S502: Based on the real-time contribution weight vector W and the NOx / VOCs ratio, implement the following four countermeasures as appropriate: a. Targeted purification driven by contribution weight.

[0061] The system identifies the top 1 to 3 key species contributing the most weight to W and activates targeted purification modules accordingly. For example, if the total weight of paraxylene exceeds 40%, the system will activate molecular sieves with high adsorption rates for aromatic hydrocarbons, achieving precise treatment with optimal energy efficiency.

[0062] b. Intelligent air volume and circulation based on chemical mechanisms.

[0063] When the system is in the VOCs control zone (NOx / VOCs<0.28) and the internal VOCs contribution weight is high, the system will activate the vehicle-mounted activated carbon filter and alert the system to the potential risk of volatilization from interior materials while diluting the VOCs with external fresh air. When in the NOx control zone (NOx / VOCs>0.35), the system switches to internal circulation mode to block external NOx. At the same time, it broadcasts NOx red zone information through V2X technology to form a collaborative emission reduction network with surrounding vehicles and triggers the in-vehicle photocatalytic or mini selective catalytic reduction (SCR) device to actively reduce NOx concentration.

[0064] c. Proactive and collaborative management of refrigerant leaks.

[0065] When the system detects a refrigerant leak and the weight of products such as trifluoroacetaldehyde (CF3CHO) in W increases, it enters a collaborative management mode: it activates a manganese-based catalyst to decompose carbonyl compounds in a targeted manner; it dynamically optimizes the operating parameters of the air conditioning system to try to slow down the leakage rate from the source; and it records relevant data and automatically schedules maintenance services for the vehicle through the vehicle networking system.

[0066] d. Data feedback and optimization.

[0067] The system backend continuously aggregates average contribution weight data under various scenarios to construct a dynamically updated pollution source contribution priority matrix. This data asset provides core data support for automakers' environmentally friendly material selection decisions and the generation of user-specific cabin health reports.

[0068] This embodiment also evaluates the performance of the deployed model based on three types of indicators: model prediction accuracy, dynamic control effect, and long-term robustness and reliability. The results are shown in Tables 3 to 5. The model prediction accuracy indicator is used to evaluate the accuracy of the GNN model on the core prediction task. The dynamic control effect indicator verifies the actual effect of the control strategy of this application through bench testing or real-vehicle road testing. The long-term robustness and reliability indicator is used to evaluate the stability of the system during long-term use.

[0069] Table 3 Performance Validation and Evaluation Results of Model Prediction Accuracy Indicators

[0070] Table 4 Performance Verification and Evaluation Results of Dynamic Regulation Effect Indicators

[0071] Table 5. Performance Verification and Evaluation Results of Long-Term Robustness and Reliability Indicators

[0072] This application also provides a device for dynamically adjusting the contribution weight of in-vehicle ozone generation potential based on graph neural networks, such as... Figure 3 As shown, the in-vehicle ozone generation potential contribution weight dynamic control device based on graph neural network includes: The data acquisition module 301 is configured to synchronously collect multi-source heterogeneous data at a frequency of not less than 1Hz under various typical scenarios; the multi-source heterogeneous data includes in-vehicle VOCs concentration, in-vehicle environmental parameters, vehicle status parameters, pollutant gas concentration, external data, and refrigerant leakage data; the typical scenarios include urban congestion, high-speed cruising, high-temperature exposure, and simulated refrigerant leakage conditions. Data processing module 302 is configured to timestamp-align all acquired data streams, employ sliding window statistical tests and interpolation algorithms to handle sensor failures, communication packet loss, and outliers, and calculate the total ozone formation potential at each time step. OFP total The instantaneous contribution values ​​of key VOCs species and the labels of the Top-5 species are marked; the total ozone formation potential is... OFP total The calculation formula is: In the formula, The temperature established based on the Arrhenius equation and photochemical kinetics ,humidity The dynamic correction function of UV radiation on the reaction rate For the first i Real-time concentration of volatile organic compounds inside the vehicle. For the first i The standard maximum incremental reactivity of a certain volatile organic compound. The serial number represents the volatile organic compound. Graph construction module 303 is configured to construct a dynamic chemical graph, represented as G=(V,E), where nodes V represent each VOC species, and the node feature vector includes the standardized concentration value, the standard maximum incremental reactivity MIR value, the molecular descriptor, and the source code; edges E characterize the potential chemical reaction associations or environmental synergistic effects between species, and the initial weights of the edges are set based on known atmospheric chemical mechanisms. The model training module 304 is configured to train a graph neural network model based on the constructed dynamic chemical graph and training data. The graph neural network model adopts the graph attention network GATv2 architecture and outputs real-time OFP prediction values ​​and dynamic contribution weight vectors of each VOC species. The control execution module 305 is configured to execute control measures based on real-time OFP prediction values ​​and dynamic contribution weight vectors, combined with the NOx / VOCs ratio. The control measures include targeted purification, intelligent air volume and circulation mode switching, and / or refrigerant leakage collaborative management.

[0073] This application provides an electronic device. The electronic device may include a processor and a memory, wherein the processor and the memory can communicate; exemplarily, the processor and the memory communicate via a communication bus.

[0074] The processor executes computer execution instructions stored in memory, causing the processor to perform the scheme in the above embodiments. The processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0075] The communication bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The system bus can be divided into address bus, data bus, control bus, etc. Transceivers are used to enable communication between database access devices and other computers (e.g., clients, read-write libraries, and read-only libraries). Memory may include random access memory (RAM) and may also include non-volatile memory.

[0076] The electronic device provided in this application embodiment can be the terminal device described in the above embodiments.

[0077] This application also provides a computer-readable storage medium storing computer instructions. When the computer instructions are executed on a computer, the computer performs the technical solution of the above-described embodiment of the dynamic control method for the contribution weight of in-vehicle ozone generation potential based on graph neural networks.

[0078] This application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When the at least one processor executes the computer program, it can implement the technical solution of the method for dynamic adjustment of the contribution weight of in-vehicle ozone generation potential based on graph neural networks in the above embodiments.

[0079] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be indirect coupling or communication connection through some interfaces, devices, or modules, and may be electrical, mechanical, or other forms.

[0080] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to implement the solution of this embodiment according to actual needs.

[0081] Furthermore, the functional modules in the various embodiments of this application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The unit composed of the above modules can be implemented in hardware or in the form of hardware plus software functional units.

[0082] The integrated modules described above, implemented as software functional modules, can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods of the various embodiments of this application.

[0083] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly manifested as execution by a hardware processor, or execution by a combination of hardware and software modules within the processor.

[0084] The memory may include high-speed RAM, and may also include non-volatile storage (NVM), such as at least one disk storage device, and may also be a USB flash drive, external hard drive, read-only memory, disk or optical disc, etc.

[0085] Buses can be Industry Standard Architecture (ISA) buses, Peripheral Component Interconnect (PCI) buses, or Extended Industry Standard Architecture (EISA) buses, etc. Buses can be categorized into address buses, data buses, control buses, etc.

[0086] The aforementioned storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The storage medium can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0087] An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Alternatively, the storage medium can be an integral part of the processor. The processor and storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and storage medium can exist as discrete components in an electronic control unit or main control device.

[0088] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0089] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A method for dynamically adjusting the contribution weight of in-vehicle ozone formation potential based on graph neural networks, characterized in that, The method includes: Multi-source heterogeneous data are collected synchronously at a frequency of no less than 1Hz under various typical scenarios. The multi-source heterogeneous data includes in-vehicle VOCs concentration, in-vehicle environmental parameters, vehicle status parameters, pollutant gas concentration, external data, and refrigerant leakage data. The typical scenarios include urban congestion, high-speed cruising, high-temperature exposure, and simulated refrigerant leakage conditions. All collected data streams were timestamped and aligned. Sliding window statistical tests and interpolation algorithms were used to handle sensor failures, communication packet loss, and outliers. The total ozone formation potential was calculated for each time step. OFP total The instantaneous contribution values ​​of key VOCs species and the labels of the Top-5 species are marked; the total ozone formation potential is... OFP total The calculation formula is: In the formula, The temperature established based on the Arrhenius equation and photochemical kinetics ,humidity The dynamic correction function of UV radiation on the reaction rate For the first i Real-time concentration of volatile organic compounds inside the vehicle. For the first i The standard maximum incremental reactivity of a certain volatile organic compound. The serial number represents the volatile organic compound. A dynamic chemical graph is constructed, represented as G=(V,E), where nodes V represent each VOC species, and the node feature vector includes the standardized concentration value, the standard maximum incremental reactivity MIR value, the molecular descriptor, and the source code; edges E characterize the potential chemical reaction associations or environmental synergistic effects between species, and the initial weights of the edges are set based on known atmospheric chemical mechanisms. A graph neural network model is trained based on the constructed dynamic chemical graph and training data. The graph neural network model adopts the graph attention network GATv2 architecture and outputs real-time OFP prediction values ​​and dynamic contribution weight vectors of each VOC species. Based on real-time OFP prediction values ​​and dynamic contribution weight vectors, combined with the NOx / VOCs ratio, control measures are implemented, including targeted purification, intelligent air volume and circulation mode switching, and / or refrigerant leakage collaborative management.

2. The method for dynamic adjustment of the contribution weight of in-vehicle ozone generation potential based on graph neural networks according to claim 1, characterized in that, The multi-source heterogeneous data specifically includes: The concentration of VOCs in the vehicle was monitored using proton transfer reaction mass spectrometry (PTR-MS) or a high-precision photoionization detector (PID) to detect various key VOC species, including toluene, xylene, ethylene, propylene, formaldehyde, acetaldehyde, and CF3CHO. In-vehicle environmental parameters, including temperature, relative humidity, and UV-A and UV-B radiation intensity; Vehicle status parameters, including real-time vehicle speed, air conditioning operating mode and fan speed, and air conditioning system pressure; Concentration of polluting gases, including NOx and O3 concentrations inside the vehicle; External data, including external NOx and O3 concentrations and local meteorological information, are obtained from roadside units (RSUs) via V2X communication; Refrigerant leakage data, HFO-1234yf leakage rate monitored by infrared sensors.

3. The method for dynamic adjustment of the contribution weight of in-vehicle ozone generation potential based on graph neural networks according to claim 1, characterized in that, The types of edges E mentioned in the training include OH radical competitive edges, ozone generation cooperative edges, secondary pollutant generation edges, and environmental regulation edges. The reaction relationships corresponding to each type of edge are as follows: The OH radical competes for the oxidizing edge, connecting to a VOC species that competes for the OH· oxidant, and the reaction is VOC. i +OH·→Products i VOCs j +OH·→Products j VOCs i and VOCs j Represents the i-th and j-th volatile organic compound species, Products i and Products j VOC i and VOCs j The products of the reaction with OH· are OH· radicals. Ozone generation synergistic edges connect species that share or jointly promote ozone generation chain reaction pathways, including VOCs. i +OH·→RO2·→NO→NO2+HO2·→O3、HO2·+NO→OH·+NO2; RO2· represents peroxyalkyl radical, HO2· represents peroxyhydroxyl radical; The secondary pollutant generation side connects the precursor to the secondary pollutant it generates, and the reaction formula includes CF3CHO+OH·→CF3C(O)O2·+PANs; CF3C(O)O2· represents trifluoroacetyl peroxy free radical; The environmental control edge uses environmental parameters as supernodes to connect all VOCs nodes affected by them, reflecting the impact of high temperature / UV radiation on VOCs release rate and reaction rate.

4. The method for dynamic adjustment of the contribution weight of in-vehicle ozone generation potential based on graph neural networks according to claim 1, characterized in that, The architecture of the graph neural network model specifically includes: The input layer projects the node feature vectors into a 128-dimensional space. The graph attention layer consists of two GATv2 layers. The first layer has four attention heads, each with a 64-dimensional output, for a total of 256 dimensions. The second layer has one attention head, which aggregates the output information from the previous layer and outputs a 128-dimensional output. The output layer outputs the total OFP value through global average pooling and fully connected layers at the graph level; the node-level output outputs the contribution weight of each node through independent fully connected layers, which is used for Top-5 species identification.

5. The method for dynamic adjustment of the contribution weight of in-vehicle ozone generation potential based on graph neural networks according to claim 1, characterized in that, The graph neural network model training process employs the following multi-task loss function: Loss = λ 1* MSE ( OFP pred OFP true ) +λ 2 *CrossEntropy ( Importance pred Importance true ) +λ 3 * Loss scenario In the formula, λ 1. λ 2 and λ 3 represents the weighting coefficients of each component of the loss function. MSE ( OFP pred OFP true () is the predicted value of ozone formation potential. OFP pred Compared with the true value OFP true The mean square error between them CrossEntropy ( Importance pred , Importance true Predicted values ​​of the contribution weights of each VOC species Importance pred Compared with the true value Importance true cross-entropy, Loss scenario Loss due to scene perception.

6. The method for dynamic adjustment of the contribution weight of in-vehicle ozone generation potential based on graph neural networks according to claim 1, characterized in that, The targeted purification includes: identifying the 1-3 key VOC species with the highest contribution weight and activating the targeted purification module; if the total weight of xylene exceeds 40%, activating the molecular sieve with high adsorption rate for aromatic hydrocarbons. The intelligent airflow and circulation mode switching includes: when NOx / VOCs < 0.28 and the internal VOCs contribution weight is high, the vehicle-mounted activated carbon filter is activated and the risk of volatilization from interior materials is alerted; when NOx / VOCs > 0.35, the system switches to internal circulation mode, broadcasts NOx red zone information through V2X technology, and triggers photocatalytic or mini selective catalytic reduction (SCR) devices. The refrigerant leak collaborative management includes: when a refrigerant leak is detected and the weight of products such as CF3CHO increases, a manganese-based catalyst is activated to decompose carbonyl compounds in a targeted manner, the air conditioning operating parameters are dynamically optimized, data is recorded, and maintenance services are scheduled through the vehicle network system.

7. The method for dynamic adjustment of the contribution weight of in-vehicle ozone generation potential based on graph neural networks according to claim 1, characterized in that, The training data includes three types of labeled data: basic scenarios, refrigerant leakage scenarios, and NOx / VOCs control zone scenarios; among which: The basic scenario uses data segments with normal driving conditions, no refrigerant leakage, and NOx / VOCs ratio within the typical range, labeled as OFP calculated based on standard MIR values; The refrigerant leakage scenario selects data segments with HFO-1234yf leakage rates exceeding 1 gram / year, and adds leakage intensity features to byproduct nodes such as CF3CHO. The NOx / VOCs control zone scenario is divided into a VOCs control zone and a NOx control zone based on the NOx / VOCs ratio.

8. A device for dynamic adjustment of the contribution weight of in-vehicle ozone generation potential based on graph neural networks, characterized in that, The device includes: The data acquisition module is configured to synchronously collect multi-source heterogeneous data at a frequency of no less than 1Hz under various typical scenarios. The multi-source heterogeneous data includes in-vehicle VOCs concentration, in-vehicle environmental parameters, vehicle status parameters, pollutant gas concentration, external data, and refrigerant leakage data. The typical scenarios include urban congestion, high-speed cruising, high-temperature exposure, and simulated refrigerant leakage conditions. The data processing module is configured to timestamp-align all acquired data streams, employ sliding window statistical tests and interpolation algorithms to handle sensor failures, communication packet loss, and outliers, and calculate the total ozone formation potential at each time step. OFP total The instantaneous contribution values ​​of key VOCs species and the labels of the Top-5 species are marked; the total ozone formation potential is... OFP total The calculation formula is: In the formula, The temperature established based on the Arrhenius equation and photochemical kinetics ,humidity The dynamic correction function of UV radiation on the reaction rate For the first i Real-time concentration of volatile organic compounds inside the vehicle. For the first i The standard maximum incremental reactivity of a certain volatile organic compound. The serial number represents the volatile organic compound. The graph construction module is configured to construct a dynamic chemical graph, represented as G=(V,E), where nodes V represent each VOC species, and the node feature vector includes the standardized concentration value, the standard maximum incremental reactivity MIR value, the molecular descriptor, and the source code; edges E characterize the potential chemical reaction associations or environmental synergistic effects between species, and the initial weights of the edges are set based on known atmospheric chemical mechanisms. The model training module is configured to train a graph neural network model based on the constructed dynamic chemical graph and training data. The graph neural network model adopts the graph attention network GATv2 architecture and outputs real-time OFP prediction values ​​and dynamic contribution weight vectors of each VOC species. The control execution module is configured to execute control measures based on real-time OFP prediction values ​​and dynamic contribution weight vectors, combined with the NOx / VOCs ratio. These control measures include targeted purification, intelligent airflow and circulation mode switching, and / or refrigerant leakage collaborative management.

9. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes the computer execution instructions stored in the memory to implement the method for dynamic adjustment of the contribution weight of in-vehicle ozone generation potential based on graph neural networks as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method for dynamic adjustment of the contribution weight of in-vehicle ozone generation potential based on graph neural networks as described in any one of claims 1-7.

Images