Vehicle-mounted aromatherapy control method, device and equipment and readable storage medium

Abstract

The invention relates to the technical field of vehicle-mounted fragrance control, and discloses a vehicle-mounted aromatherapy control method, device and equipment and a readable storage medium, and the vehicle-mounted aromatherapy control method comprises the steps: obtaining multi-source vehicle-mounted environment data and vehicle operation data, and generating a multi-dimensional spatial-temporal feature vector representing a current vehicle-mounted environment state; clustering processing is carried out according to the feature vectors, and the in-vehicle environment state grade is obtained; determining aromatherapy control reference information based on the environmental state grade and the scene rule; constructing a fusion vector according to the multi-dimensional spatial-temporal feature vector and aromatherapy control reference information, inputting the fusion vector into an aromatherapy control model, and generating target aromatherapy control data; and receiving user preference information, and performing personalized adjustment on the target aromatherapy control data to obtain aromatherapy control data. According to the vehicle-mounted aromatherapy control method and device, efficient fusion of self-adaption, scenario and individuation of the environment inside and outside the vehicle is achieved, and the vehicle-mounted aromatherapy control accuracy and the user experience are remarkably improved.

Description

Technical Field

[0001] This application relates to the field of in-vehicle fragrance control technology, and in particular to an in-vehicle fragrance control method, device, equipment and readable storage medium. Background Technology

[0002] With the rapid development of smart cockpits and vehicle networking technologies, in-vehicle fragrance systems are gradually becoming an important feature for enhancing the user's driving experience. Current technologies typically release fragrance gases into the vehicle through preset scents and fixed concentrations to improve air quality and driving comfort. Theoretically, an ideal fragrance system should be able to dynamically and personally adjust the scent and concentration based on the driving environment, user preferences, and in-vehicle air quality. However, most existing systems rely on data from single in-vehicle sensors, such as temperature, humidity, and air quality, lacking comprehensive perception capabilities of external environmental information (such as traffic accidents, traffic congestion, and inclement weather). This results in fragrance strategies failing to effectively adapt to changing driving scenarios. Furthermore, due to the lack of cross-temporal and spatial user preference modeling and memory transfer mechanisms, fragrance control strategies are usually limited to static rules and fixed scene patterns, making it difficult to meet the increasingly diverse needs of the driving experience.

[0003] Existing in-vehicle fragrance systems still have significant technical shortcomings in areas such as environmental perception, concentration adjustment, personalized adaptation, and cross-system collaboration. First, environmental perception is limited to a single dimension, focusing only on in-vehicle air quality while neglecting the potential impact of external environmental factors (such as traffic accidents and congestion) on the driver's psychological state, resulting in insufficient adaptability of fragrance strategies. Second, fragrance adjustment methods remain relatively crude, typically relying on fixed scenario strategies such as "city mode / highway mode," lacking the ability to dynamically calculate fragrance type and concentration based on real-time data. Third, the lack of a correlation model between user preferences and geographical environment makes it difficult to achieve cross-temporal and spatial fragrance memory transfer, thus failing to provide personalized services. Finally, existing systems suffer from data silos with subsystems such as in-vehicle navigation, hindering collaborative control based on route, traffic conditions, and weather information, and lacking concentration over-limit warnings and automatic dilution mechanisms, posing health risks. Summary of the Invention

[0004] In view of this, embodiments of this application provide a method, apparatus, device and readable storage medium for controlling in-vehicle aromatherapy, which can effectively solve the problems of insufficient environmental perception, imprecise aromatherapy adjustment, lack of personalization and isolated system data in the prior art.

[0005] In a first aspect, embodiments of this application provide a method for controlling in-vehicle air fresheners, including: Acquire multi-source vehicle environment data and vehicle operation data to generate a multi-dimensional spatiotemporal feature vector characterizing the current vehicle environment; Clustering is performed on the multidimensional spatiotemporal feature vectors to obtain the current in-vehicle environment status level; Based on the in-vehicle environment status level and preset scene rules, aromatherapy control reference information is determined; Based on the multidimensional spatiotemporal feature vector and the aroma control reference information, a fusion vector is constructed and input into the aroma control model to obtain the target aroma control data. The system receives user preference information and adjusts the target aromatherapy control data to obtain aromatherapy control data.

[0006] In some embodiments, the method further includes: The vehicle-mounted air freshener is controlled to release air fragrance according to the aforementioned air fragrance control data. The aroma release operation includes: Select the target fragrance channel corresponding to the fragrance control data; Based on the concentration parameters in the aromatherapy control data, the working state of the atomizing component corresponding to the target fragrance channel is adjusted to generate the actual aromatherapy release state.

[0007] In some embodiments, the method further includes: The operation feedback data of the in-vehicle aromatherapy device is obtained, including the fragrance parameters and concentration parameters before and after user adjustment, as well as the environmental feature vector at the time of adjustment. Based on the aforementioned operational feedback data, fine-tuning samples are constructed, and a fine-tuning dataset is generated; When the number of user adjustments meets the preset triggering conditions, the aromatherapy control model is incrementally fine-tuned based on the fine-tuning dataset to obtain the updated model.

[0008] In some embodiments, acquiring multi-source in-vehicle environment data and vehicle operation data to generate a multi-dimensional spatiotemporal feature vector characterizing the current in-vehicle environment includes: The multi-source vehicle environment data is subjected to multi-channel filtering and outlier removal is performed in combination with preset threshold rules to obtain environmental data. The vehicle operation data is smoothed and normalized using a sliding window process to obtain the operation data. The environmental data and the operational data are concatenated according to the time and spatial dimensions to form a multidimensional spatiotemporal feature vector.

[0009] In some embodiments, the clustering process of the multidimensional spatiotemporal feature vector to obtain the current in-vehicle environment state level includes: The multidimensional spatiotemporal feature vectors are weighted according to preset feature weights to obtain weighted feature data; Based on the weighted feature data, fuzzy C-means clustering is performed in a high-dimensional feature space to obtain multiple candidate environmental state categories and their corresponding membership vectors. The confidence levels of each candidate environmental state category are ranked according to the membership vector to determine the in-vehicle environmental state level.

[0010] In some embodiments, determining the aromatherapy control reference information based on the in-vehicle environment state level and preset scene rules includes: The in-vehicle environment state level is input into the scene rule base, and a set of candidate scene rules that match the in-vehicle environment state level is selected based on the condition matching strategy. Based on the correlation analysis between the candidate scene rule set and the multidimensional spatiotemporal feature vector, the target scene rule is determined. Based on the target scenario rules, obtain the fragrance selection parameters and concentration adjustment parameters, and combine the fragrance selection parameters and concentration adjustment parameters to generate aromatherapy control reference information.

[0011] In some embodiments, the step of constructing a fusion vector based on the multidimensional spatiotemporal feature vector and the aromatherapy control reference information and inputting it into the aromatherapy control model to obtain target aromatherapy control data includes: High-dimensional feature embedding encoding is performed on the multidimensional spatiotemporal feature vector to obtain the first feature representation; The aromatherapy control reference information is encoded using rule parameters to obtain a second feature representation; The first feature representation and the second feature representation are concatenated along the feature dimension to generate a fusion vector; The fusion vector is input into the feature extraction layer of the aromatherapy control model to generate a fusion feature representation; The fused feature representation is input into the prediction layer of the aromatherapy control model to generate the target aromatherapy control data.

[0012] Secondly, embodiments of this application provide a vehicle-mounted air freshener control device, comprising: The data processing module is used to acquire multi-source vehicle environment data and vehicle operation data, and generate a multi-dimensional spatiotemporal feature vector representing the current vehicle environment; The clustering module is used to perform clustering processing on the multidimensional spatiotemporal feature vectors to obtain the current in-vehicle environment status level. The retrieval module is used to determine aromatherapy control reference information based on the in-vehicle environment status level and preset scene rules; The fusion processing module is used to construct a fusion vector based on the multidimensional spatiotemporal feature vector and the aromatherapy control reference information, and input it into the aromatherapy control model to obtain the target aromatherapy control data. The adjustment module is used to receive user preference information and adjust the target aromatherapy control data to obtain aromatherapy control data.

[0013] Thirdly, embodiments of this application provide a vehicle device, the vehicle device including a processor and a memory, the memory storing a computer program, and the processor executing the computer program to implement the vehicle-mounted aromatherapy control method of the first aspect described above.

[0014] Fourthly, embodiments of this application provide a computer-readable storage medium, wherein when the computer program is executed on a processor, it implements the vehicle aromatherapy control method of the first aspect described above.

[0015] The embodiments of this application have the following beneficial effects: By acquiring multi-source in-vehicle environmental data and vehicle operation data, a multi-dimensional spatiotemporal feature vector is constructed to characterize the current in-vehicle and external environmental state; subsequently, based on this feature vector, cluster analysis is performed on the in-vehicle environmental state to classify it into different environmental state levels; further, based on the environmental state level and preset scene rules, corresponding aromatherapy control reference information is generated; the aromatherapy control reference information is fused with the spatiotemporal feature vector and input into the network model to output target aromatherapy control data; finally, personalized correction is performed by combining user preference information to generate aromatherapy control data. Through this application, deep integration of aromatherapy control in multiple stages such as environmental state recognition, rule matching, model decision-making, and personalized preference adjustment can be achieved, improving the perception and adaptability of the in-vehicle fragrance system to complex in-vehicle and external environments, enhancing the accuracy, intelligence, and personalization of aromatherapy control, significantly improving user experience, and meeting diverse fragrance needs. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart of an embodiment of the in-vehicle air freshener control method of this application is shown; Figure 2 Another flowchart of the in-vehicle air freshener control method according to an embodiment of this application is shown; Figure 3 A schematic diagram of a vehicle-mounted air freshener control method according to an embodiment of this application is shown. Detailed Implementation

[0018] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments.

[0019] The components of the embodiments of this application described and illustrated in the accompanying drawings can be arranged and designed in a variety of different configurations. Therefore, the following detailed description of the embodiments of this application provided in the drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0020] In the following text, the terms "comprising," "having," and their cognates, which may be used in various embodiments of this application, are intended only to indicate a particular feature, number, step, operation, element, component, or combination thereof, and should not be construed as primarily excluding the presence of one or more other features, numbers, steps, operations, elements, components, or combinations thereof, or adding the possibility of one or more combinations thereof. Furthermore, the terms "first," "second," "third," etc., are used only for distinguishing descriptions and should not be construed as indicating or implying relative importance.

[0021] Unless otherwise specified, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which the various embodiments of this application pertain. Terms (such as those defined in commonly used dictionaries) shall be interpreted as having the same meaning as in their contextual meaning in the relevant technical field and shall not be construed as having an idealized or overly formal meaning, unless clearly defined in the various embodiments of this application.

[0022] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0023] Considering the technical problems of existing in-vehicle aromatherapy systems, such as insufficient environmental perception, imprecise aromatherapy adjustment, lack of personalization, and isolated system data, an in-vehicle aromatherapy control method is proposed. This method acquires multi-source in-vehicle environmental data and vehicle operation data to construct a multi-dimensional spatiotemporal feature vector. It then dynamically determines aromatherapy control reference information by combining it with a scene rule base, generates target aromatherapy control data based on a deep neural network, and incorporates user preferences for personalized adjustments to achieve intelligent and adaptive control of the in-vehicle aromatherapy system.

[0024] The following examples illustrate the method for controlling in-vehicle air fresheners.

[0025] Figure 1 A flowchart of a car air freshener control method according to an embodiment of this application is shown. Exemplarily, the car air freshener control method includes the following steps: Step S100: Obtain multi-source vehicle environment data and vehicle operation data, and generate a multi-dimensional spatiotemporal feature vector representing the current vehicle environment.

[0026] The multidimensional spatiotemporal feature vector refers to a high-dimensional input vector that uniformly represents in-vehicle environmental perception data, vehicle operating parameters, and road and geographic scene information in both temporal and spatial dimensions. In this embodiment, the feature vector adopts a vector structure consisting of twelve dimensions, covering the following information: Vehicle-mounted environmental data: Temperature, humidity, air quality (AQI), noise, and light intensity are collected by deploying mini weather station modules; Vehicle operating data: vehicle speed, acceleration, engine load; Real-time traffic information: By integrating the Gaode Map Open Platform API, we obtain the road congestion index (value range 0~100) and the coordinates of construction sections; Weather and geography features: Includes rainfall probability, visibility, extreme weather warning signs, tunnel / bridge signs, and current altitude provided by the Gaode API.

[0027] The aforementioned multi-source data are constructed into a unified multi-dimensional feature structure through asynchronous sampling, spatiotemporal alignment, and format standardization.

[0028] In an optional embodiment, step S100 includes the following sub-steps: S101 performs multi-channel filtering on multi-source vehicle environment data and removes outliers by combining preset threshold rules to obtain environmental data.

[0029] Channel filtering refers to using differentiated signal smoothing methods based on data type to remove noise and sudden interference.

[0030] As an example, a moving average filter is applied to temperature and humidity; Median filtering is applied to air quality, noise, and light intensity, and threshold rejection is performed by setting upper and lower limits based on environmental physical constraints. Band-limited filtering and normalization are performed on vehicle speed, acceleration, and engine load to remove extreme points caused by abnormal data collection or malfunctions. The road congestion index is smoothed by index, and data outside the range of 0 to 100 are removed; Map projection matching is performed on the coordinates of the construction section to eliminate abnormal GPS points that do not match the actual road structure; Time series mean filtering was applied to rainfall probability and visibility, and the results were compared and corrected with forecast data from the meteorological center. Perform data consistency verification on extreme weather warning status to ensure that the alarm type matches the meteorological bureau's API; The validity of tunnel / bridge markings is confirmed via an electronic map interface; First-order difference constraints are applied to changes in altitude to filter out unreasonable fluctuations in a short period of time.

[0031] Through the above processing, a multi-source cleaned dataset with uniform format and reliable data is obtained.

[0032] S102, perform sliding window smoothing and normalization on the vehicle operation data to obtain the operation data.

[0033] Among them, the sliding window processing uses a fixed window length (such as 1 second) to update the mean or median of continuous frame data to smooth short-term jitter; the normalization processing uses the Min-Max normalization method to compress vehicle speed, acceleration, and engine load into a uniform range of 0-1 values, so that they have equal modeling weight in multi-dimensional stitching.

[0034] As an example, the range of vehicle speed values ​​is set as follows: acceleration is Engine load is Then, after normalization, the running data is obtained.

[0035] S103 concatenates environmental data and operational data according to time and space dimensions to form a multidimensional spatiotemporal feature vector.

[0036] Among them, temporal dimension stitching refers to the synchronous alignment of multi-source data such as vehicle environment, operating status, real-time traffic conditions, weather, and geographical features at the same sampling time; spatial dimension stitching refers to combining data features from different sources into a single high-dimensional vector according to the spatial location of sensors inside and outside the vehicle, road coordinates, and geographical markers. For example, five-dimensional environmental parameters such as temperature and humidity, PM2.5, VOCs, noise, and illumination, three-dimensional operating data such as vehicle speed, acceleration, and engine load, and four-dimensional external information such as congestion index, weather warnings, tunnel / bridge signs, and altitude changes are stitched together into a twelve-dimensional spatiotemporal feature vector.

[0037] Step S200: Cluster the multidimensional spatiotemporal feature vectors to obtain the current in-vehicle environment status level.

[0038] The in-vehicle environment status level refers to the environmental classification label obtained after comprehensive calculation based on twelve-dimensional multi-source features, used to characterize the graded results of different in-vehicle environmental states. For example, the environmental status level can be divided into states such as "comfortable," "slightly stuffy," "slightly cold," "polluted," and "low visibility warning." To accurately identify these states, this step sequentially performs feature weighting, fuzzy clustering, and confidence ranking.

[0039] In one alternative embodiment, such as Figure 2 As shown, step S200 includes the following sub-steps: S201, the multidimensional spatiotemporal feature vectors are weighted according to preset feature weights to obtain weighted feature data.

[0040] Among them, the preset feature weights refer to the quantification coefficients based on the importance of different features to in-vehicle comfort and aromatherapy control, including: The characteristics of the vehicle environment (e.g., temperature, humidity, air quality, noise, light intensity) have a high weight; for example, air quality and temperature can be set to 0.2. Vehicle operating characteristics (vehicle speed, acceleration, engine load) are secondary and can be set to 0.1~0.15; The weights of road condition characteristics (road congestion index, coordinates of construction sections) are moderate, such as a weight of 0.08 for the congestion index; Weather features (probability of rainfall, visibility, and extreme weather warnings) have low weights, such as visibility having a weight of 0.05; The weights of geographical features (tunnel / bridge signs, changes in altitude) can be set based on experimental statistics.

[0041] For example, assuming the feature vectors are f1, f2...f12, and the corresponding weights are w1, w2...w12, then the formula for calculating the weighted feature data is: Fi = wi•fi; i = 1, 2... 12 S202, based on weighted feature data, performs fuzzy C-means clustering in a high-dimensional feature space to obtain multiple candidate environmental state categories and their corresponding membership vectors.

[0042] Fuzzy C-Means (FCM) clustering is a soft clustering algorithm that assigns membership values ​​to multiple clusters for each data sample, rather than belonging to a single category, thus better adapting to the uncertainty of multidimensional features. For example, assuming the number of cluster centers is set to C=5 (comfortable, cold, hot, poor air quality, low visibility warning), the Euclidean distance from each sample point to the cluster center is calculated. And calculate its membership degree iteratively based on the following formula. :

[0043] in, : The membership degree of sample i to category k, with a value range of ([0,1]), representing "to what extent sample i belongs to category k" (the closer to 1, the higher the membership degree); : The distance from sample i to class k (Euclidean distance, Manhattan distance, etc. can be used to measure the distance between the sample and the class center).

[0044] : Distance from sample i to class j (j iterates through all classes) C: Total number of categories (i.e., the data should be clustered into C clusters).

[0045] m: Fuzziness index (usually (m>1)), which controls the "fuzziness" of membership: the larger m is, the more "fuzzy" the membership distribution is (the degree to which a sample can belong to multiple categories at the same time is more uniform); when (m→1), the membership is close to "hard clustering" (the sample almost belongs to only one category).

[0046] For example, typical parameter configurations are as follows: Cluster number C=4 (corresponding to "Excellent, Good, Poor, Extremely Poor") The fuzziness index m = 2.0 Maximum number of iterations = 200 Termination condition: Change in membership degree

[0047] The clustering results divided the in-vehicle environment into four levels and established a correspondence with the aromatherapy concentration control range: Excellent [0,2): Aromatherapy concentration 0.5–1.0 ml / h Good [2,5): Aromatherapy concentration 1.0–2.5 ml / h Poor [5,8): Aromatherapy concentration 2.5–4.0 ml / h Range [8,10]: Aroma concentration 4.0–5.0 ml / h S203, Rank the candidate environmental state categories by confidence based on the membership vector to determine the target environmental state level.

[0048] Here, confidence level refers to the degree of certainty that a sample point belongs to a certain category, and it is directly taken as the category corresponding to the maximum membership degree. For example, if the membership degree vector at a certain moment is 0.05, 0.12, 0.65, 0.10, 0.08, then the current vehicle environment state level is determined to be category 3 (corresponding to "poor air quality"). The final target environment state level is obtained for subsequent steps to retrieve scene rule bases and generate aromatherapy control reference information.

[0049] Step S300: Based on the in-vehicle environment status level and preset scene rules, determine the aromatherapy control reference information.

[0050] The scene rules are stored in a pre-built scene rule library containing various typical in-vehicle scenarios and corresponding aromatherapy control strategies. This library stores control information such as fragrance selection parameters and concentration adjustment parameters corresponding to different in-vehicle environmental state levels. For example, after the in-vehicle system detects the current in-vehicle environmental state level, it calls the retrieval interface in the scene rule library. By inputting the environmental state level, it triggers the scene rule matching process and progressively filters out the target scene rule that best matches the current environmental state, ensuring the personalization and accuracy of subsequent aromatherapy control.

[0051] In an optional embodiment, step S300 includes the following sub-steps: S301: Input the in-vehicle environment status level into the scene rule base, and filter the candidate scene rule set that matches the in-vehicle environment status level based on the condition matching strategy.

[0052] Among them, the condition matching strategy refers to comparing the in-vehicle environment state level with preset conditions by setting matching conditions for each rule in the scenario rule base, and filtering out a set of candidate rules that meet the conditions.

[0053] For example, when the in-vehicle environment status level is detected to correspond to the "congestion + tunnel + rain" scenario, the candidate rule that is closest to it will be extracted according to the condition matching strategy. The recommended fragrance is lavender with a concentration coefficient of 1.2. The recommended fragrances and concentration coefficients for other scenario types are shown in Table 1 below:

[0054] Table 1 S302, based on the correlation analysis of the candidate scene rule set and the multi-dimensional spatiotemporal feature vector, the target scene rule is determined.

[0055] The multidimensional spatiotemporal feature vector refers to a feature vector formed by integrating onboard environmental data and operational status data collected by vehicle sensors, including multidimensional information such as temperature, humidity, carbon dioxide concentration, real-time road conditions, weather data, and geographical features. As an example, the matching degree between candidate rules and real-time feature vectors is calculated one by one, for example: If the road congestion index is 92, the tunnel sign is 1, and the rainfall intensity is 0.8 mm / h, then the match with the rule "rainy day + congestion + tunnel" is the highest. Select this rule as the target scenario rule and extract its corresponding aroma parameters and concentration coefficients.

[0056] S303: Obtain fragrance selection parameters and concentration adjustment parameters according to the target scene rules, and combine the fragrance selection parameters and concentration adjustment parameters to generate aromatherapy control reference information.

[0057] Among them, the fragrance selection parameter refers to the preset fragrance identifier data in the target scene rules, used to indicate the fragrance to be activated in a specific scene; the concentration adjustment parameter refers to the fragrance concentration level corresponding to the target fragrance. For example, when the target scene rules recommend lavender with a concentration coefficient of 1.2, but the user preference database records that the user prefers a stronger fragrance, the actual concentration coefficient is dynamically adjusted to 1.5, and combined to form the final aromatherapy control reference information: Target fragrance: Lavender Target fragrance concentration: Base value 1.2 × Preference correction factor 1.25 = 1.5 Aromatherapy control reference information: {Fragrance type = Lavender, Concentration = 1.5ml / h} The generated aromatherapy control reference information will be transmitted to the aromatherapy release control model, driving the in-vehicle aromatherapy device to perform precise fragrance spraying, realizing adaptive fragrance control and personalized adjustment of the in-vehicle environment.

[0058] Step S400: Based on the multidimensional spatiotemporal feature vector and aromatherapy control reference information, construct a fusion vector and input it into the aromatherapy control model to obtain the target aromatherapy control data.

[0059] The aromatherapy control model refers to a deep neural network model trained using historical in-vehicle data. This model learns the mapping relationship between different in-vehicle environmental states, scene rules, and aromatherapy control parameters. Exemplarily, multi-dimensional spatiotemporal feature vectors are encoded and fused with aromatherapy control reference information to obtain a high-dimensional vector that comprehensively reflects the in-vehicle environmental state and control intent. This fused vector is input into the aromatherapy control model, and after feature extraction and prediction operations, it outputs target aromatherapy control data for controlling the in-vehicle aromatherapy device.

[0060] In an optional embodiment, step S400 includes the following sub-steps: S401, perform high-dimensional feature embedding encoding on the multidimensional spatiotemporal feature vector to obtain the first feature representation.

[0061] The multidimensional spatiotemporal feature vector includes 12 environmental features such as temperature, humidity, air quality index, road congestion index, construction section coordinates, rainfall probability, visibility, extreme weather warnings, tunnel / bridge signage, and altitude change rate. As an example, to unify different dimensions, continuous variables (temperature, humidity, etc.) are standardized to the interval 0, 10, 10, 1 using the Min-Max method, as shown in the formula:

[0062] in, A sample value in the original data (such as the temperature value at a certain moment).

[0063] : The minimum value of this set of data (such as the lowest temperature among all temperature samples).

[0064] The maximum value of this set of data (e.g., the highest temperature among all temperature samples).

[0065] : Data after Min-Max standardization (mapped to the ([0,1]) range by default).

[0066] For discrete variables (such as tunnel signs, bridge signs, etc.), 0 / 1 binary encoding is used; the congestion index is first logarithmically transformed and then standardized; for the rate of change of altitude, linearization is achieved through sigmoid compression. After the above-mentioned unified dimensions, the feature vector is input into a high-dimensional embedding network, and a dense first feature representation is obtained through multi-layer nonlinear mapping.

[0067] S402, the aromatherapy control reference information is encoded using rule parameters to obtain the second feature representation.

[0068] The aromatherapy control reference information includes fragrance selection parameters and concentration adjustment parameters. One-hot encoding is used to vectorize the fragrance categories, and the concentration parameters are mapped to normalized values, combining them to form a vectorized expression that can be directly input into the neural network. For example, if the target scene rule matches the "lavender" fragrance and the concentration adjustment coefficient is 1.2, the system encodes the fragrance as 0, 1, 0, 0 (corresponding to four fragrance types), and the normalized concentration parameters are concatenated to form the second feature representation.

[0069] S403 concatenates the first feature representation and the second feature representation along the feature dimension to generate a fused vector.

[0070] As an example, after acquiring the first and second feature representations, the system calls the tensor concatenation interface to concatenate the two sets of vectors element-wise along the feature dimensions. For instance, if the first feature representation has a dimension of 128 and the second feature representation has a dimension of 32, the resulting fused vector will have a dimension of 160. This concatenation operation preserves all features of the in-vehicle environment perception results and the aromatherapy control reference information, and expresses their correlation in a unified vector space, ultimately generating a fused vector.

[0071] S404: Input the fusion vector into the feature extraction layer of the aromatherapy control model to generate a fusion feature representation.

[0072] Exemplary for this purpose, the feature extraction layer employs a bidirectional LSTM (Bi-LSTM) network, with each layer containing 64 hidden units, to capture the temporal dependencies and dynamic characteristics of multidimensional features. To prevent overfitting, Dropout (p=0.2) is introduced after the LSTM layer, and higher-order features are further extracted through a multi-layer fully connected network. The final fused feature representation not only includes the comprehensive relationship between the in-vehicle environment and the aromatherapy control parameters but also retains historical trends and dynamic change information.

[0073] S405, the fused feature representation is input into the prediction layer of the aromatherapy control model to generate target aromatherapy control data.

[0074] As an example, the prediction layer is designed with two branches: Fragrance prediction branch: the target fragrance category is output using a Softmax classifier; Concentration prediction branch: the fragrance spray concentration is directly predicted through a regression layer, ranging from 0.5 ml / h to 5.0 ml / h.

[0075] In terms of training strategy, the model jointly optimizes the cross-entropy loss (fragrance type) and the mean squared error loss (concentration), and achieves adaptive optimization by dynamically adjusting the weight ratio: The initial weights are set as follows: cross-entropy = 0.7, MSE = 0.3; When the aroma type is predicted incorrectly, the cross-entropy loss weight is automatically increased by 0.05; When the concentration error exceeds 0.5 ml / h, the MSE loss weight increases by 0.03.

[0076] The model uses the AdamW optimizer (learning rate = 3e-4, weight decay = 1e-4) and adds Gaussian noise (σ = 0.2) to the historical geographic data for data augmentation.

[0077] In actual deployment, the system accelerates model inference through TensorRT, with a single decision latency of less than 50ms, ensuring real-time responsiveness in the vehicle.

[0078] Step S500: Receive user preference information, adjust the target aromatherapy control data, and obtain aromatherapy control data.

[0079] User preference information refers to personalized control habits collected through in-vehicle interactive terminals, mobile clients, or historical aromatherapy usage records, including fragrance preferences, concentration sensitivity, and aromatherapy usage strategies in different driving scenarios. For example, preset preference parameters, such as commonly used fragrance categories, optimal concentration ranges, and corresponding start / stop conditions, are read from the user account configuration file. Simultaneously, based on historical vehicle usage data, the user's aromatherapy control selection behavior under different weather conditions, road conditions, or driving times can be dynamically analyzed to obtain a user preference mapping vector across multiple scenarios.

[0080] After acquiring user preference information, the target aromatherapy control data and user preference information are input into the personalization adjustment module. The personalization adjustment module, based on a multi-dimensional weighted fusion strategy, adjusts parameters such as fragrance selection, fragrance concentration, and duration in the target aromatherapy control data. For the fragrance selection parameter, if the user has set a specific fragrance preference, the user-specified fragrance will be used first; if the user has not specified one, the fragrance selection in the target aromatherapy control data will remain unchanged. The system will adjust the fragrance concentration parameter according to the user's sensitivity to concentration. For example, if the user prefers "light" fragrance, the target concentration value will decrease by a preset ratio, while if the user prefers "strong" fragrance, the target concentration value will increase. Regarding the duration parameter, if the user prefers short-duration fragrance application, the system will shorten the default duration in the target control data; otherwise, it will extend it.

[0081] Finally, aromatherapy control data is generated based on the adjusted control parameters. This aromatherapy control data will serve as the final execution command sent to the in-vehicle aromatherapy control device, ensuring that the aromatherapy control process not only adapts to environmental conditions but also better suits the user's personal preferences, achieving a unity of user experience and intelligent environmental control.

[0082] In one optional embodiment, the in-vehicle aromatherapy control method further includes the following steps: The aromatherapy device is controlled to release aromatherapy based on aromatherapy control data.

[0083] The in-vehicle aromatherapy device refers to an automated aromatherapy release device installed inside a vehicle, comprising multiple independent fragrance channels, atomizing components, air ducts, and a drive control unit. Each fragrance channel is pre-filled with different types of essential oils, and the atomizing component converts the liquid aroma into an aerosol state, which is then released into the vehicle's interior environment through the air duct. Demonstratively, the in-vehicle aromatherapy device employs a four-channel aroma diffusion structure, corresponding to four fragrances: lavender, lemon, pine, and mint. Each fragrance channel is independently equipped with an atomizing component and an airflow control unit, supporting parallel management and precise control of multiple fragrances. Simultaneously, the device integrates a concentration adjustment module, using PWM to control the atomizing plate's operating frequency, enabling dynamic adjustment of the atomization volume with an accuracy of ±0.1 ml / h, ensuring consistent and controllable release effects for different fragrances and concentration parameters.

[0084] After acquiring the aromatherapy control data, the system selects the corresponding target aromatherapy channel from multiple aromatherapy channels in the car air freshener based on the aroma selection parameters in the data. For example, when the aromatherapy control data specifies the aroma as "woody," the corresponding aromatherapy channel will be activated while the other channels remain closed to avoid fragrance interference.

[0085] Subsequently, based on the concentration parameters in the aromatherapy control data, the operating status of the atomizing component in the target fragrance channel is dynamically adjusted. The atomizing component employs piezoelectric ultrasonic atomization to atomize the aromatherapy liquid into tiny particles and release them into the vehicle's air. The concentration parameters directly correspond to the atomizing component's driving voltage, operating frequency, or on-time duration. When the concentration parameter is set to a low level, the atomizing component operates intermittently at a low frequency, resulting in a smaller atomization volume. When the concentration parameter is set to medium, the atomizing component operates continuously at a medium frequency, producing a moderate amount of atomization. When the concentration parameter is set to high, the atomizing component operates continuously at a high frequency, maximizing the atomization volume.

[0086] Finally, the real-time operating status of the atomizing component is compared with the aromatherapy control data to generate the actual aromatherapy release status. This actual release status includes information such as the target fragrance type, spray duration, and atomization intensity level, and is updated within the vehicle's central control unit for subsequent closed-loop monitoring and status synchronization. This method ensures that the aromatherapy release process not only meets user preferences but also closely matches the current in-vehicle environment, achieving personalized and intelligent aromatherapy control.

[0087] In one optional embodiment, the in-vehicle aromatherapy control method further includes the following steps: Acquire operational feedback data from the in-vehicle aromatherapy device. Construct fine-tuning samples based on this feedback data and generate a fine-tuning dataset. When the number of user adjustments meets a preset trigger condition, incrementally fine-tune the aromatherapy control model based on the fine-tuning dataset to obtain an updated model.

[0088] The operational feedback data refers to the interactive data recorded when the in-vehicle aromatherapy device manually adjusts the fragrance and concentration parameters generated by the system during the aromatherapy release process. This includes the fragrance and concentration parameters before and after the user's adjustment, as well as the corresponding environmental feature vectors. For example, during an aromatherapy release process where the device automatically selects a "woody" fragrance and a "medium" concentration, and the user manually adjusts it to a "floral" fragrance and a "low" concentration, triggering the recording mechanism, the collected operational feedback data includes: Original fragrance parameters: woody Original concentration parameters: Medium Adjusted fragrance parameters: Floral Adjusted concentration parameters: Low setting Adjustment time environmental feature vector: a vector composed of twelve features such as in-vehicle temperature and humidity, real-time road conditions, weather information, and geographical features.

[0089] The feedback data is transformed into standardized fine-tuning samples. Each sample set includes "input feature vector + original output label + user-corrected label," with the structure: [X_env, Y_model, Y_user], where X_env is the twelve-dimensional spatiotemporal environment feature vector, Y_model is the original model output (fragrance type and concentration), and Y_user is the user-adjusted target label. The system continuously accumulates multiple sets of fine-tuning samples and constructs a fine-tuning dataset according to the set sample organization rules to ensure data dimensionality consistency and label quality reliability.

[0090] Furthermore, the system incorporates a pre-defined trigger condition judgment mechanism to determine whether to perform model fine-tuning. The trigger condition is met when a single user makes N consecutive adjustments (e.g., N=3) or the cumulative number of feedback samples reaches M (e.g., M=20). Once this trigger condition is detected, incremental fine-tuning is performed on the original aromatherapy control model based on the current fine-tuning dataset. Specifically, by preserving the pre-trained parameters, the original model structure remains unchanged, and only a small number of gradient updates are performed on some weight parameters to avoid damaging the existing model performance. The model uses a multilayer perceptron based on the Transformer architecture. During fine-tuning, a small number of training epochs (e.g., 5 epochs) are used, with the learning rate set to 1 / 10 of the original value, to achieve gradual adaptation of the model to user preferences.

[0091] Furthermore, a model performance verification mechanism is introduced. After fine-tuning, the predictive performance of the updated model is verified on historical user preference data. If the accuracy improvement in aroma or concentration prediction is less than 5%, the model is rolled back to the version before fine-tuning to ensure the effectiveness and stability of the model update. All model versions and rollback operations are recorded in the version control log to ensure the traceability of the model evolution process.

[0092] To improve the efficiency of adaptive adjustment, a reinforcement learning mechanism based on Deep Deterministic Policy Gradient (DDPG) is introduced. The designed reward function is:

[0093] Where w1 represents user satisfaction; w2 represents aromatherapy consumption rate; and w3 represents mode switching frequency. The optimal weight configuration was determined through extensive real-vehicle testing. These correspond to user questionnaire scores, 30-day consumable statistics, and interaction log analysis results, respectively. During the optimization process, a multi-armed slot machine algorithm was used to perform online parameter tuning on two weeks of interaction data from 200 users, dynamically balancing user experience and resource consumption.

[0094] At the user interaction level, explicit feedback can be provided: users can directly influence the reward function in reinforcement learning through a rating system (1-5 stars); Implicit feedback can also be provided: if a user stays in a certain fragrance mode for more than 5 minutes, the system considers it positive feedback; if the user quickly switches fragrances within 30 seconds, a negative penalty is triggered. Personalized memory: Constructing feature vectors for each user This is used to characterize the user's personalized features, and a user bias term is introduced into the model output layer. This enables personalized customization of prediction results. Among them, Basic prediction results (general predictions given by the model without considering user personalization, such as the basic matching score for product recommendations and the basic score for content recommendations). : Weight vector of user features (and Dimension matching is used to measure the degree of influence of each dimension of user features on the final prediction. Dot product operation: Calculates the adjustment amount corresponding to "user personalized features" (i.e., the correction value that needs to be made to the basic prediction due to user differences). The final personalized prediction result.

[0095] Ultimately, through incremental fine-tuning driven by runtime feedback, reinforcement learning optimization, and interactive closed-loop design, the system achieves continuous evolution and personalized adaptation of the in-vehicle aromatherapy control model.

[0096] In a specific application scenario, aromatherapy control is performed in an in-vehicle environment.

[0097] Data acquisition is completed through the vehicle's OBD interface, including basic operating parameters such as vehicle speed and altitude change rate. The vehicle speed collected is 80 km / h, and the altitude rise rate is 5 m / s. Simultaneously, the system implements data transmission based on a communication protocol stack, which includes: an application layer of ISO 15031-5 (DAQ mode), a transport layer of CAN 2.0B (11-bit ID), and a physical layer of ISO15765-4 (500 kbps), ensuring timing synchronization and data consistency across different sampling channels.

[0098] Furthermore, a periodic sampling and preprocessing mechanism is established based on multi-source heterogeneous data. Basic parameters (such as vehicle speed) are sampled at 100ms intervals, while high-energy-consuming parameters (such as altitude) are sampled at 1s intervals. The state vector is then processed using a Kalman filter. Dynamic smoothing is performed; at the same time, outlier detection is performed using the 3σ principle, and mean smoothing is performed based on a sliding window with a window size of 5 to obtain high-quality sampled data.

[0099] Regarding external environmental data, the system calls the Gaode API interface to obtain real-time road type (such as highway) and weather information (such as sunny day). At the same time, it obtains environmental status information such as VOCs concentration (0.8ppm) and in-vehicle temperature (25℃) through in-vehicle sensors, and processes them to obtain a twelve-dimensional spatiotemporal environmental vector: [25,50,0.8,0.1,3,0,0,0,5,1,0,0]. Based on this environmental vector, the environmental quality level is calculated. For example, the comprehensive score is 4.2, corresponding to the "good" level.

[0100] Then, the multidimensional environmental vector and aromatherapy control reference information are input into the aromatherapy control model for inference to obtain the target aromatherapy control data. For example, the model predicts that the optimal fragrance in the current scenario is lemon with a probability of 0.82, corresponding to a concentration output of 2.1 ml / h. Based on the model output, the control actuator activates the lemon fragrance channel and sets the PWM duty cycle of the atomizing component to 42%, thus forming the target aromatherapy release state.

[0101] In the aromatherapy release control process, a nonlinear calibration formula is used to compensate for the actual atomization amount. The formula is as follows:

[0102] Where Q represents the actual output concentration (ml / h), and D represents the PWM duty cycle (%). , The coefficients of the quadratic and linear terms (obtained through fitting experimental data, used to characterize the nonlinear relationship between D and Q, reflecting the influence of the "quadratic effect of the input variable" on the atomization amount). For example... This is the factory calibration coefficient for the lemon-scented flavor.

[0103] In addition, it supports multi-level calibration mechanisms, including: Factory calibration: Concentration was measured at 9 PWM duty cycle calibration points (10%-90%) at 25℃; Online calibration: Closed-loop control is performed based on the feedback results of the liquid level sensor, and zero-point calibration is automatically completed every 24 hours; Temperature compensation: When the ambient temperature T deviates, it is compensated by... Dynamically adjust atomization output to ensure release accuracy. Among these features, : Current ambient temperature (usually in °C); 25: Reference temperature (default "normal operating temperature", the reference for temperature compensation); : Control parameters related to atomization (such as valve opening, control signal strength of the atomizing device, etc., representing the "basic adjustment amount" of atomization); Correction value for atomization amount (the amount of atomization that needs to be increased or decreased when the temperature deviates from the baseline); : Proportional coefficient (controls the magnitude of atomization correction for every 1°C change in temperature).

[0104] This application scenario enables high-precision adjustment of aromatherapy control under multi-dimensional dynamic conditions such as real-time road conditions, weather, and environment.

[0105] In another exemplary application scenario, aromatherapy control is performed in a congested tunnel environment: Based on multi-source real-time data, a current environmental feature vector is constructed. For example, the collected data shows a road congestion index of 92, a tunnel identifier of 1, and a rainfall intensity of 0.8 mm / h. The system performs scenario matching based on the environmental feature vector and a preset set of rules. Upon detecting that the triggering conditions of "rainy day + congestion + tunnel" are met, the corresponding aromatherapy control strategy is initiated.

[0106] Furthermore, based on the prediction results of the user preference database and the aromatherapy control model, personalized aromatherapy adjustment optimization is performed. When the user selects "Relaxation Mode," the target aromatherapy concentration parameter output by the model is adjusted to 1.5 times the original value to enhance the soothing effect. After the aromatherapy release operation is completed, the system updates the user preference database based on the operational feedback data and writes the adjusted parameters as reinforcement learning samples into the Q-table, realizing dynamic optimization and continuous iteration of the aromatherapy control strategy.

[0107] In another exemplary application scenario, aromatherapy control and air purification are performed in tandem under extreme weather conditions: The system obtains real-time red rainstorm warnings pushed by Gaode Maps via an external weather data interface. Upon detecting a rainstorm warning trigger signal, the system automatically switches to the target fragrance channel matching the current scene based on environmental feature vectors and aromatherapy control reference information. Based on the target aromatherapy control data output by the aromatherapy control model, the system selects the mint fragrance and adjusts the target concentration parameter to 4.5 ml / h to enhance the driver's alertness.

[0108] Simultaneously, the air purification mode is activated, and the atomization component is switched to its maximum atomization output based on the atomization parameters in the aromatherapy control data to improve the air quality inside the vehicle. While performing these operations, a prompt message is displayed on the user interface: "Rainstorm warning detected, refreshment protection mode activated," achieving multi-dimensional linkage between environmental status perception, aromatherapy control, air purification, and user prompts.

[0109] Figure 3 A schematic diagram of a vehicle-mounted air freshener control device according to an embodiment of this application is shown. Exemplarily, the vehicle-mounted air freshener control device 100 includes: Data processing module 110 is used to acquire multi-source vehicle environment data and vehicle operation data, and generate a multi-dimensional spatiotemporal feature vector characterizing the current vehicle environment; Clustering module 120 is used to perform clustering processing on the multidimensional spatiotemporal feature vector to obtain the current in-vehicle environment status level; The retrieval module 130 is used to determine aromatherapy control reference information based on the in-vehicle environment status level and preset scene rules; The fusion processing module 140 is used to construct a fusion vector based on the multidimensional spatiotemporal feature vector and the aroma control reference information, and input it into the aroma control model to obtain the target aroma control data; The adjustment module 150 is used to receive user preference information and adjust the target aromatherapy control data to obtain aromatherapy control data.

[0110] It is understood that the apparatus of this embodiment corresponds to the method of the above embodiments, and the options in the above embodiments are also applicable to this embodiment, so they will not be described again here.

[0111] This application also provides a vehicle device, exemplary of which includes a processor and a memory, wherein the memory stores a computer program, and the processor executes the computer program to enable the vehicle device to perform the functions of the various modules in the above-described method or apparatus.

[0112] The processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, including at least one of a Central Processing Unit (CPU), Graphics Processing Unit (GPU), Network Processor (NP), Digital Signal Processor (DSP), Application-Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. The general-purpose processor can be a microprocessor or any conventional processor, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application.

[0113] The memory can be, but is not limited to, Random Access Memory (RAM), Read Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM), Electrically Erasable Programmable Read-Only Memory (EEPROM), etc. The memory is used to store computer programs, and the processor can execute the computer programs accordingly after receiving execution instructions.

[0114] This application also provides a computer-readable storage medium for storing the computer program used in the aforementioned vehicle equipment. For example, the computer-readable storage medium may include, but is not limited to, various media capable of storing program code, such as a USB flash drive, a portable hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0115] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the architecture, functionality, and operation of possible implementations of apparatus, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that, in alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0116] In addition, the functional modules or units in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0117] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a smartphone, personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application.

[0118] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A vehicle-mounted incense control method, characterized in that, The method comprises: acquiring multi-source vehicle-mounted environment data and vehicle operation data to generate a multi-dimensional space-time feature vector representing a current vehicle environment; performing clustering processing on the multi-dimensional space-time feature vector to obtain a current in-vehicle environment state level; determining a reference information for incense control based on the in-vehicle environment state level and a preset scene rule; constructing a fusion vector according to the multi-dimensional space-time feature vector and the reference information for incense control, and inputting the fusion vector into an incense control model to obtain target incense control data; receiving user preference information and adjusting the target incense control data to obtain incense control data. 2.The car-mounted incense control method according to claim 1, characterized in that, The method further comprises: controlling a vehicle-mounted incense device to perform an incense release operation according to the incense control data; wherein the incense release operation comprises: selecting a target fragrance channel corresponding to the incense control data; adjusting a working state of an atomization component corresponding to the target fragrance channel based on a concentration parameter in the incense control data to generate an actual incense release state. 3.The car-mounted incense control method according to claim 1, characterized in that, The method further comprises: acquiring operation feedback data of the vehicle-mounted incense device, the operation feedback data comprising fragrance parameters and concentration parameters before and after user adjustment, and an environment feature vector during adjustment; constructing a fine-tuning sample based on the operation feedback data and generating a fine-tuning data set; when detecting that a user adjustment frequency meets a preset triggering condition, performing incremental fine-tuning on the incense control model based on the fine-tuning data set to obtain an updated model. 4.The vehicle-mounted incense control method according to claim 1, characterized in that, The acquiring of the multi-source vehicle-mounted environment data and the vehicle operation data to generate a multi-dimensional space-time feature vector representing a current vehicle environment comprises: performing multi-channel filtering processing on the multi-source vehicle-mounted environment data, and performing outlier rejection based on a preset threshold rule to obtain environment data; performing sliding window smoothing processing and normalization processing on the vehicle operation data to obtain operation data; performing feature concatenation on the environment data and the operation data according to time and space dimensions to form a multi-dimensional space-time feature vector. 5.The vehicle-mounted incense control method according to claim 1, characterized in that, The clustering processing on the multi-dimensional space-time feature vector to obtain a current in-vehicle environment state level comprises: performing weighted processing on the multi-dimensional space-time feature vector according to a preset feature weight to obtain weighted feature data; performing fuzzy C-means clustering processing on the weighted feature data in a high-dimensional feature space to obtain a plurality of candidate environment state categories and corresponding membership vectors; performing confidence sorting on each candidate environment state category according to the membership vectors to determine an in-vehicle environment state level. 6.The vehicle-mounted incense control method according to claim 1, characterized in that, The determination of the reference information for incense control based on the in-vehicle environment state level and a preset scene rule comprises: inputting the in-vehicle environment state level into a scene rule library, and screening a candidate scene rule set meeting the in-vehicle environment state level based on a condition matching strategy; performing correlation analysis on the candidate scene rule set and the multi-dimensional space-time feature vector to determine a target scene rule; obtaining fragrance selection parameters and concentration adjustment parameters according to the target scene rule, and combining the fragrance selection parameters and the concentration adjustment parameters to generate the reference information for incense control. 7.The car-mounted incense control method according to claim 1, characterized in that, The fusion vector is constructed according to the multi-dimensional space-time feature vector and the aromatherapy control reference information, and input into an aromatherapy control model to obtain target aromatherapy control data, including: Performing high-dimensional feature embedding coding on the multi-dimensional space-time feature vector to obtain a first feature representation; Performing rule parameter coding on the aromatherapy control reference information to obtain a second feature representation; Splicing the first feature representation and the second feature representation in the feature dimension to generate a fusion vector; Inputting the fusion vector into a feature extraction layer of the aromatherapy control model to generate a fusion feature representation; Inputting the fusion feature representation into a prediction layer of the aromatherapy control model to generate the target aromatherapy control data.

8. An in-vehicle incense control device, characterized by, The data processing module is configured to obtain multi-source vehicle-mounted environment data and vehicle operation data, and generate a multi-dimensional space-time feature vector representing a current vehicle-mounted environment; The clustering module is configured to perform clustering processing on the multi-dimensional space-time feature vector to obtain a current in-vehicle environment state level; The retrieval module is configured to determine aromatherapy control reference information based on the in-vehicle environment state level and a preset scene rule; The fusion processing module is configured to construct a fusion vector according to the multi-dimensional space-time feature vector and the aromatherapy control reference information, and input the fusion vector into an aromatherapy control model to obtain target aromatherapy control data; The adjustment module is configured to receive user preference information, and adjust the target aromatherapy control data to obtain aromatherapy control data. The vehicle device includes a processor and a memory, and the memory stores a computer program, and the processor is configured to execute the computer program to implement the vehicle-mounted aromatherapy control method of any one of claims 1-7.

9. A vehicle apparatus characterized by comprising: The computer program is stored in the memory and executed on the processor to implement the vehicle-mounted aromatherapy control method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, ​

Images