Intelligent fragrance control system and method based on olfaction perception self-adaption

Abstract

The invention belongs to the technical field of intelligent fragrance control, and discloses an intelligent fragrance control system and method based on olfaction perception self-adaption, multi-dimensional data such as exhaled gas fragrance residue and skin electroreaction are captured in real time through a multi-mode olfaction fatigue monitoring unit, and 1-5 olfaction fatigue levels are accurately judged in combination with an LSTM-attention model; and then the dynamic release control unit is matched with a layering strategy, for example, short-time pulse stimulation is started in light fatigue, fragrant cross-modulation switching and concentration zeroing are triggered in heavy fatigue, and local high-concentration pseudo fatigue is avoided synchronously through a distributed space probe. By means of the design, fragrance release is always matched with the real-time olfaction state of a user, the problem that the fragrance cannot be effectively perceived after continuous exposure is thoroughly solved, it is ensured that the user can stably perceive the fragrance in different use periods, and the use experience is remarkably improved; and a complete logic chain with multi-mode monitoring, double-target sensing and cooperative control is constructed.

Description

Technical Field

[0001] This invention belongs to the field of intelligent fragrance control technology, specifically an intelligent fragrance control system and method based on olfactory perception adaptation. Background Technology

[0002] Fragrance systems, as important devices for enhancing spatial comfort and user experience, have been widely used in homes, commercial spaces, and transportation. With the upgrading of consumer demands, users are placing higher requirements on the accuracy, comfort, and health benefits of fragrance release. Current technologies are evolving from traditional manual control to intelligent automatic control, but there is still significant room for improvement in adapting to human olfactory characteristics and dynamic environmental changes. Existing fragrance systems mostly employ fixed concentration or timed release modes, which have three significant technical shortcomings: First, the static release mode of traditional fragrance diffusers does not consider the adaptive decay law of olfactory receptors. In actual use, human sensitivity to the same scent decreases with increasing exposure time, resulting in users being unable to effectively perceive the fragrance afterward, severely impacting the user experience. Second, while some improved solutions can adjust the fragrance concentration through ambient temperature and humidity, they only use a one-way control logic of "environmental parameters → concentration adjustment," failing to establish a correlation between the user's physiological state (such as olfactory fatigue) and the release strategy, thus failing to fundamentally solve the problem of olfactory numbness. Third, continuous release at high concentrations not only results in a large waste of fragrance materials, with a consumption rate far exceeding that of reasonable release modes, but also significantly increases the health risks of excessive volatile organic compound (VOC) concentrations in the space, failing to meet indoor air quality safety standards. Summary of the Invention

[0003] The purpose of this invention is to provide an intelligent fragrance control system and method based on olfactory perception adaptation, so as to solve the problems mentioned in the background art.

[0004] To achieve the above objectives, the present invention provides the following technical solution: an intelligent fragrance control system based on olfactory perception adaptation, the system comprising: Multimodal olfactory fatigue monitoring unit: integrates dual-parameter olfactory sensor, wearable linkage component, and distributed probe. It uses LSTM-attention model to process four-dimensional data of "gas + physiology + space + behavior", outputs fatigue level 1-5 and recovery time, and builds "sensitive / tolerant" personalized model to provide accurate fatigue benchmark for subsequent units. Dynamic Environment-Health Dual-Target Sensing Unit: Added VOCs and millimeter-wave human flow sensors, adaptive diffusion model combined with prior fatigue level adjustment coefficient to generate spatial concentration report, binding environmental adaptation and health and safety, providing dual-dimensional decision-making basis for releasing control unit; Multi-strategy collaborative dynamic release control unit: stratified strategies are set according to fatigue level (different for mild / moderate / severe), embedded VOCs excess response mechanism, and the fragrance library is iterated based on user feedback to provide optimization direction for interactive units, realizing "fatigue-health-preference" collaborative control; Personalized interaction and feedback incentive unit: Supports custom fatigue thresholds and fragrance preferences, post-strategy pop-up to receive feedback, weekly health reports, unlocking customized fragrances upon completion of feedback, and supplementing the preceding fragrance library with data to provide personalized support for the central unit; Edge computing collaborative control hub unit: Adopting an architecture of "real-time edge processing + cloud iteration", it builds a health-first decision matrix, performs edge-de-identified data transmission, cloud-based federated learning, generates final instructions, supports subsequent processes, and ensures efficient system collaboration.

[0005] Preferably, the multimodal olfactory fatigue monitoring unit includes: The dual-parameter olfactory sensor for exhaled fragrance residue and nasal temperature and humidity uses the MSP430 series gas-sensitive film, with a response time of ≤1 second. It can capture temperature and humidity fluctuations of 0.1℃ in the olfactory receptor mucosa, helping to judge subtle changes in the early stage of fatigue. Wearable linkage components connect to mainstream smart bracelets via Bluetooth 5.0 protocol to collect skin conductance response (GSR, sampling rate 100Hz) and heart rate variability (HRV, time resolution 1ms) in real time. Olfactory fatigue can cause a decrease in the excitability of sympathetic nerve endings. Three to five distributed spatial fragrance detectors, employing SHT3x series temperature and humidity compensation technology, are deployed in key areas such as the living room and bedroom, achieving a detection accuracy of 0.01 mg / m³. 3 This can rule out false fatigue misjudgments caused by localized high concentrations; At the algorithm level, an LSTM-attention mechanism deep learning model is used. The training data covers 100,000 time-series data from 500 users of different ages (20-60 years old) and different olfactory sensitivities. After inputting four-dimensional data of "gas residue, physiological signal, spatial concentration difference and user behavior records (on-hook duration, adjustment frequency)", the system outputs individual olfactory fatigue levels of 1-5 levels and recovery time accurate to the minute. Establish a dynamic and personalized model. For example, the fatigue threshold for "sensitive" users (who have a strong daily reaction to floral fragrances) is set to level 1, and for "tolerant" users (who are exposed to fragrance environments for a long time) it is set to level 3. The model will automatically fine-tune the threshold based on the user's weekly interaction data, providing a fatigue benchmark that is accurate to the individual for the subsequent dynamic environment-health dual-objective perception unit. The core formula for calculating olfactory fatigue level:

[0006] In the formula: This function represents the individual's olfactory fatigue level, outputting a fatigue level of 1-5, with higher levels indicating more severe fatigue. The calculated value is rounded down by a function, with a range of 1-5. This indicates the weight of exhaled breath fragrance residue, a core fatigue indicator. Higher residue levels indicate more severe fatigue. The unit is m. 3 / mg, a fixed value after model training. =2.0, derived from training with 100,000 data points from 500 people; This indicates the amount of fragrance molecules remaining in the user's exhaled breath, directly reflecting the degree to which olfactory receptors absorb fragrance. The higher the residue, the more severe the fatigue. The unit is mg / m³. 3 Data is collected by the MSP430 series sensor. This indicates the weight of skin conductance response (GSR), a physiological indicator. The lower the GSR, the more severe the fatigue. The unit is µS. -1 Fixed values ​​after model training =0.3; GSR stands for skin conductance, reflecting the level of sympathetic nerve excitation. When the sense of smell is fatigued, the level of sympathetic nerve excitation decreases, and the GSR decreases. It is measured in µS and is collected by wearable devices. This indicates the weighting of heart rate variability (HRV), a physiological indicator. Higher HRV indicates greater fatigue, measured in milliseconds (ms). -1 Fixed values ​​after model training =0.1; HRV represents heart rate variability, reflecting autonomic balance. HRV increases during olfactory fatigue, measured in milliseconds, and is collected by wearable devices. This represents the weighting of the difference in spatial fragrance concentration, which is an environmental indicator weighting. The larger the concentration difference, the more prone one is to pseudo-fatigue. The unit is m. 3 / mg, a fixed value after model training. =1.5; This represents the difference between the highest and lowest fragrance concentrations within a region, reflecting the spatial uniformity of fragrance concentration. A large difference can easily lead to misjudgment of localized fatigue. The unit is mg / m³. 3 It is calculated by 3-5 distributed probes.

[0007] To address the unique physiological phenomenon of "olfactory fatigue" (distinct from the perception of environmental parameters such as temperature and humidity), a deep coupling of multi-dimensional hardware and dedicated algorithms is achieved: a dual-parameter olfactory sensor (MSP430 series) specifically captures residual fragrance in exhaled breath and fluctuations in nasal temperature and humidity (a subtle change of 0.1℃ in the olfactory receptor mucosa is directly related to the initial state of fatigue), rather than a general gas sensor; GSR (conductivity for skin response) and HRV (heart rate variability) data collected by wearable linkage components need to be transmitted in real time to the edge via Bluetooth 5.0 protocol, and synchronously input into the LSTM-attention model along with the spatial concentration difference data collected by the distributed spatial fragrance probe (SHT3x series)—the training data of this model (100,000 time-series data from 500 users of different ages and different olfactory sensitivities) are all specifically labeled for olfactory fatigue (e.g., level 1 fatigue corresponds to GSR ≥ 5μS, exhaled residual fragrance ≤ 0.2mg / m³). 3 The model outputs "fatigue levels 1-5 + recovery time accurate to the minute" directly for subsequent fragrance release strategy adjustments, forming a complete technology chain of "hardware acquisition (olfactory-specific data) → algorithm processing (olfactory fatigue-specific model) → decision output (service fragrance control)," which is different from the general physiological monitoring of other home devices such as smart air conditioners (such as only monitoring heart rate for sleep judgment) and does not have the problem of "simple transplantation of existing technology."

[0008] Preferably, the dynamic environment-health dual-objective perception unit includes: VOCs concentration sensor, TGS2600 series, detection range 0-10 mg / m³ 3 With a response time of ≤5 seconds, it monitors the volatile organic compounds emitted by the fragrance in real time, strictly adhering to the standard of 0.6 mg / m³ in GB / T18883-2022. 3 The indoor air quality limit is set at 0.5 mg / m³. 3 A yellow warning is triggered, and a local buzzer sounds; the concentration reaches 0.6 mg / m³. 3 A strong red alert is triggered, and a notification is sent to the mobile app. The RF600 series millimeter-wave spatial crowd sensor has a detection range of 0.5-5m and an angle of 120°. It can penetrate thin curtains, glass and other obstructions, accurately count 1-10 people, and dynamically adjust the supply matching degree based on the formula for calculating the average fragrance consumption demand per person. Formula for calculating per capita fragrance consumption demand:

[0009] In the formula: This represents the fragrance consumption demand of an individual per unit of time, reflecting the real-time intensity of an individual's fragrance demand, and is expressed in mg / (person·h). This indicates the total release of the fragrance diffuser per unit time, reflecting its basic output capacity. It can be adjusted by the valve opening and is measured in mg / h. This represents the fatigue demand correction coefficient, which adjusts the average demand based on the user's fatigue level. The more severe the fatigue, the higher the demand. Levels 1-2 are 0.8, levels 3-4 are 1.0, and level 5 is 1.2. This indicates the duration of fragrance release, referring to the effective release time within the calculation period, in hours (h), with a minimum unit of 0.1h. This represents the real-time number of people in the area, as shown in the adaptive fragrance diffusion coefficient correction formula. Consistent, collected by RF600 series millimeter-wave crowd flow sensor, with values ​​ranging from 1 to 10 people; An adaptive modified fragrance diffusion model is adopted, which is optimized based on the Navier-Stokes equations of fluid dynamics. It can adjust the diffusion coefficient in real time according to the fatigue level. The coefficient increases by 2% for every 1°C increase in temperature, 5% for every 0.1m / s increase in airflow, and 15% for every additional person in the flow of people. It also generates a visual spatial concentration balance report. This unit deeply integrates environmental adaptation with health and safety, avoiding concentration imbalance of the fixed model in scenarios such as open ventilation and large gatherings, and providing a dual-dimensional decision-making basis of environmental parameters and health thresholds for the multi-strategy collaborative dynamic release control unit. Adaptive fragrance diffusion coefficient correction formula:

[0010] This represents the final diffusion coefficient of the fragrance molecules after correction, reflecting the actual diffusion capacity of the fragrance in the current environment. The larger the coefficient, the faster the diffusion. This represents the fundamental diffusion coefficient of fragrance molecules, a fixed reference value under standard environmental conditions (temperature 25℃, airflow velocity 0.1m / s, no people). =1.0; This represents the temperature correction factor, reflecting the weight of the effect of temperature changes on the diffusion coefficient, in units of... The scheme specifies the value. That is, for every 1°C increase in temperature, the diffusion coefficient increases by 2%; This represents the difference between the actual ambient temperature and the standard temperature. ,unit The temperature and humidity data are collected by the SHT4x series temperature and humidity sensor. This represents the airflow velocity correction factor, reflecting the weight of the effect of changes in airflow velocity on the diffusion coefficient, and is expressed in m / s. -1 The scheme specifies the value. =0.5, meaning that for every 0.1 m / s increase in airflow speed, the diffusion coefficient increases by 5%; This represents the difference between the actual airflow velocity and the standard airflow velocity. The unit is m / s, and the data is collected by the FS3000 series airflow sensor. This represents the personnel quantity correction factor, reflecting the weight of the influence of the number of people in the area on the diffusion coefficient. The value is specified in the plan. =0.15, meaning that for every additional person, the diffusion coefficient increases by 15%; This indicates the real-time number of people in the area, reflecting the user base with fragrance demand. It is collected by the RF600 series millimeter-wave people flow sensor, with a value range of 1 to 10 people. This represents the olfactory fatigue level correction factor, which adjusts the diffusion coefficient according to the user's fatigue level. The more severe the fatigue, the slower the diffusion. It is output by the multimodal olfactory fatigue monitoring unit: 1.05 for level 1-2, 1.0 for level 3-4, and 0.95 for level 5.

[0011] This unit does not simply monitor environmental parameters, but rather establishes a dedicated collaborative logic for "environmental parameters - olfactory fatigue - health and safety": the VOCs concentration sensor (TGS2600 series) monitors the volatile organic compounds emitted by fragrances, and the safety threshold needs to be adjusted in conjunction with the fatigue level output by the multimodal unit (e.g., in cases of severe fatigue, fragrance release decreases, and the VOCs safety threshold is simultaneously corrected according to the formula to avoid health risks caused by decreased user tolerance to irritating gases under fatigue); the per capita fragrance consumption demand calculated by the millimeter-wave crowd sensor (RF600 series) needs to be correlated with the fatigue level correction coefficient. (0.8 for levels 1-2, 1.0 for levels 3-4, and 1.2 for level 5) to ensure that the fragrance supply for highly fatigued users meets their sensory needs without exceeding the limit, which is different from the one-way matching of "people flow-supply" of general environmental monitoring equipment (without fatigue correlation).

[0012] Preferably, the multi-strategy collaborative dynamic release control unit includes: Based on fatigue level and environmental data, a refined, hierarchical dynamic strategy logic is constructed: Mild fatigue (grades 1-2, corresponding to 0.2-0.3 mg / m³ of residual exhaled gas) 3 When this occurs, only a short-duration pulse release is initiated, increasing the concentration by 15% compared to baseline, such as a baseline of 0.4 mg / m³. 3 →0.46mg / m 3Avoid overstimulating olfactory receptors; Moderate fatigue (grades 3-4, residual fatigue 0.3-0.4 mg / m²) 3 When performing concentration gradient adjustment, the concentration is reduced from the baseline of 100% to 40% in 5 steps, such as 100%→80%→60%→50%→40%, while simultaneously switching the fragrance type slightly. 70% of the original fragrance core ingredients are retained. For example, the citrus note retains sweet orange essential oil and adds 30% green tea essential oil to reduce the abruptness of the olfactory system. Severe fatigue (Level 5, residual >0.4 mg / m²) 3 When the concentration is triggered to zero for 15 minutes, the fresh air system is simultaneously activated at maximum speed to accelerate the removal of residues. Then, the fragrance is switched across different notes, such as woody notes to aquatic notes, with a component difference of >80%. During the recovery phase, a 10-second pulse is released every 2 minutes, and the concentration gradually rises from 20% to 80%. Simultaneously embed a health-first mechanism: If the VOCs concentration reaches 0.5 mg / m³ 3 The system will automatically reduce the fragrance concentration by 20% and continue monitoring for 5 minutes. If the concentration does not decrease, the fragrance release will be turned off. At the same time, a pop-up message will appear on the device display and mobile app indicating that the VOC concentration is too high and the release has been paused. If it reaches 0.6 mg / m 3 A strong early warning value directly cuts off the fragrance supply and activates fresh air purification; in addition, the fragrance library is iterated based on user feedback, and fragrances with a rating of ≥4 stars (out of 5 stars) are retained and given increased usage weight, while fragrances with <3 stars are directly eliminated. Every quarter, 2-3 new combinations are added based on seasonal characteristics, such as adding mint-ocean notes in summer and cedar-cinnamon notes in winter, providing a clear strategic optimization direction for personalized interaction and feedback incentive units; VOCs safe concentration correction formula:

[0013] Revised safe concentration thresholds for VOCs, dynamically adjusted upper limits for health, in mg / m³ 3 base value =0.6mg / m 3 ; This indicates the basic safe concentration threshold for VOCs, based on the limits specified in GB / T18883-2022 "Indoor Air Quality Standard", with units of mg / m³. 3 Fixed standard value mg / m 3 ; This represents the correction factor for the safety threshold of VOCs due to temperature. VOCs volatilization intensifies with increasing temperature, and the unit is °C. -1The scheme specifies the value. =0.01, meaning the threshold decreases by 1% for every 1°C increase in temperature; This represents the difference between the actual ambient temperature and the standard temperature, and is related to the formula for correcting the adaptive fragrance diffusion coefficient. Consistent; This represents the correction factor for relative humidity on the VOCs safety threshold (VOCs solubility increases with increasing humidity), in %RH. -1 The scheme specifies the value. =0.005, meaning that for every 1% increase in humidity (RH), the threshold increases by 0.5%. This represents the difference between the actual relative humidity and the standard humidity. The unit is %RH, and the data is collected by the SHT4x series sensor.

[0014] The stratification strategy of this unit is specifically designed for olfactory fatigue and is not universally applicable to strategies used in other devices: For mild fatigue (levels 1-2), the "short-term pulse release" (concentration increased by 15% from baseline, 10 seconds each time, 2-minute intervals) is based on the physiological characteristic that olfactory receptors can "recover sensitivity with short-term stimulation," rather than a general concentration adjustment; for moderate fatigue (levels 3-4), the "concentration gradient adjustment + fragrance micro-switching" (retaining 70% of the original fragrance core components) is to avoid the olfactory system's rejection reaction to "abrupt odor changes" (olfactory receptors have higher tolerance to familiar components); for severe fatigue (level 5), the "concentration to zero for 15 minutes + accelerated fresh air removal + cross-modality switching (component difference > 80%)" is designed to address the mechanism that olfactory receptors "need complete desensitization after prolonged high-concentration exposure"—all of these strategies are triggered by fatigue data from the multimodal unit and are linked to VOCs concentration and pedestrian flow data from the dynamic environment unit (e.g., VOCs reaching 0.5 mg / m³). 3 Even when experiencing mild fatigue, the concentration is preferentially reduced by 20%, forming a unique control logic with "olfactory fatigue as the core and environmental health as the constraint," which is different from the general release strategy of other devices (which is not related to olfactory physiological mechanisms).

[0015] Preferably, the personalized interaction and feedback incentive unit includes: Dual-platform multi-dimensional manual operation interface: Device panel and mobile APP: The device panel uses touch buttons and supports one-click switching of olfactory fatigue warning thresholds, with 1-5 levels selectable and corresponding to different colored indicator lights. The app offers a fragrance preference weight setting function, with four fragrance categories—floral, woody, aquatic, and citrus—as adjustable sliders. Users can precisely adjust these based on their mood for the day, such as setting the aquatic scent to 80 points on a rainy day. Three minutes after the strategy is executed, the system triggers feedback through an APP pop-up or device voice prompt. The feedback interface includes a 1-5 star rating, three options for effective fatigue relief, average, and ineffective, as well as a text note box. It supports inputting specific suggestions such as the fragrance being too strong or switching too frequently. After the feedback is submitted, a confirmation prompt of "Thank you for your feedback, it has been incorporated into the strategy optimization" is displayed immediately. The system pushes the olfactory health report for last week to the app, and the report is presented in charts: The line chart shows the daily fatigue level distribution, the bar chart compares the number of times each strategy was implemented and its effectiveness, and the number cards indicate the number of times VOCs exceeded the standard and the peak value. For example, if there were 2 instances of exceeding the standard this week, with the highest value being 0.52 mg / m³, the chart would be displayed. 3 Finally, based on user preferences and seasonal characteristics, we recommend fragrances for next week, such as a mint-marine scent for summer, noting its invigorating properties and VOC content as low as 0.3mg / m³. 3 ; Simultaneously, a tiered feedback incentive mechanism is designed: completing 3 valid feedback sessions unlocks 1 customized fragrance formula, supporting adjustments to the ratio of 3 essential oil components, such as citrus: green tea: peppermint = 7:2:1. After customization, it is automatically saved as a personal exclusive fragrance, effectively enhancing user engagement and the long-term adaptability of the system. The collected feedback data is synchronized in real time to the fragrance library of the multi-strategy collaborative dynamic release control unit.

[0016] Preferably, the edge computing collaborative control hub unit includes: It adopts a two-tier architecture of real-time processing at the edge and long-term iteration at the cloud: The edge device is equipped with an ARM Cortex-A53 quad-core processor and integrates a Linux real-time operating system. It can integrate high-frequency data from the first four units in real time and generate an instant release strategy with a latency of ≤1 second. At the same time, it stores the raw data of the last 24 hours locally to avoid latency issues caused by relying on cloud transmission. The cloud-based deployment is on Alibaba Cloud ECS servers. Model iterations begin every Sunday at 2 AM (during off-peak user periods). Iteration data is derived from anonymized user data nationwide. Specific optimization directions include: adjusting fatigue thresholds for different age groups (e.g., reducing the sensitivity threshold for 20-30 year olds from 0.3 mg / m³). 3 Reduced to 0.25 mg / m³ 3 For individuals aged 50-60, the tolerance threshold is 0.4 mg / m³. 3 Increased to 0.45 mg / m³ 3 The environmental coefficient of the diffusion model is corrected (e.g., when humidity > 60%, the diffusion coefficient is increased by an additional 8%), and the optimized model is automatically distributed to the edge after the iteration is completed; Establish a quantitative priority decision matrix, with a weighting system of 10 points: Health and safety (weight 6 points) take priority. When VOCs exceed the standard, the concentration is forcibly limited to 80% of the baseline. Even if the user manually adjusts it to 100%, it will be ineffective. A pop-up message will appear to indicate that the concentration has been limited in order to protect health. Fatigue relief (weight 3 points) is the second best option. In multi-person scenarios, spatial partitioning detection prioritizes reducing the concentration by 10% in areas where users with high fatigue levels are located (such as workstations near windows). User Preference (Weight 1 point) Finally, if the VOCs content of the user's preferred fragrance type (such as rich woody notes) is close to the limit, the system will automatically recommend similar low VOCs options, such as light woody notes, with the note "85% similarity to your preference, 30% lower VOCs"; In terms of privacy protection, all data uploaded from the edge to the cloud has the user ID and device MAC address removed. Physiological data (such as HRV) is averaged for 5 minutes before being uploaded. The cloud uses federated learning technology, where nodes in each region only exchange model parameters (without transmitting raw physiological data). This unit integrates all data to generate the final execution instructions.

[0017] This invention also provides an intelligent fragrance control method based on olfactory perception adaptation. Based on the system described in claim 6, the specific steps of this method are as follows: S1. Multimodal data acquisition and spatiotemporal fusion: fatigue monitoring data, environmental health data and user interaction data are collected at a fixed frequency at the edge. Outliers are removed using the 3σ principle. Spatiotemporal alignment is performed using UTC timestamps. HRV data is taken as the 5-minute moving average before being uploaded to the cloud. S2. Multi-target collaborative release strategy generation and distribution: The control center identifies the scenario, sorts the targets in the order of health and safety > fatigue relief > user preference, calls the personalized fatigue model to calculate the strategy, distributes the strategy and receives the status through the MQTT protocol, and adjusts the parameters when an exception occurs. S3. Closed-loop iterative optimization and anomaly handling: Use Python data analysis tools to regularly review data, optimize personalized fatigue models and strategy iterations, add exclusive strategies for multi-person sensitive scenarios, and initiate countermeasures when sensor failures or VOCs concentrations exceed the standard.

[0018] Preferably, the specific steps of the multimodal data acquisition and spatiotemporal fusion process in step S1 are as follows: S11. Three types of data are synchronously collected at a fixed frequency at the edge: Fatigue monitoring data is collected by MSP430 sensor for exhaled residuals, SHT3x sensor for nasal temperature and humidity, Bluetooth-connected wristband for GSR and HRV, and distributed probes for spatial concentration difference. Environmental health data: SHT4x collects temperature and humidity, FS3000 collects airflow, TGS2600 collects VOCs, and RF600 collects pedestrian flow. User interaction data, such as threshold settings on the device panel and feedback ratings from the app, is transmitted using the MQTT protocol. After data collection, a three-step fine preprocessing process is performed: The first step, outlier removal, employs the 3σ principle. The mean and standard deviation of the data are calculated, and outliers exceeding ±3 times the standard deviation are removed. For example, fragrance residue exceeding 0.2 mg / m³... 3 Sudden rise to 2 mg / m 3 Data deemed faulty by the sensor was discarded. The second step is spatiotemporal alignment, which uses UTC timestamps to synchronize data of different frequencies to a uniform granularity of 10 seconds / time. The spatial dimension is labeled with sensor location codes for each data point, such as living room-1 and bedroom-2, to ensure that the data corresponds accurately to the physical space. The third step is to desensitize physiological data. HRV data is taken as the 5-minute moving average before being uploaded to the cloud. The original HRV data is encrypted and stored locally for 7 days and then automatically deleted. The AES-256 encryption algorithm is used. S12. Finally, generate a three-dimensional matrix of time, data type, and space, stored in JSON format: The time dimension is divided into 10-second intervals, the data type dimension is divided into three categories: fatigue, environment, and interaction, and the spatial dimension is divided according to the sensor location. For example, living room-1 corresponds to one set of data. Each cell of the matrix contains a specific value and a status indicator. The data retrieval response time of this matrix is ​​≤500ms, and it fully covers the individual fatigue, environmental status, and user preference information required for decision-making.

[0019] Preferably, the specific steps for generating and issuing the multi-target collaborative release strategy in step S2 are as follows: S21. The control center first performs intelligent scene recognition based on a three-dimensional matrix, using the LightGBM classification model. It inputs four features: time, pedestrian flow, HRV value, and spatial location. It automatically identifies the scenario as a two-person fatigue scenario in the afternoon at home, then prioritizes the target in the order of health and safety > fatigue relief > user preferences. Next, it initiates a three-step dynamic strategy calculation: The first step is to invoke the personalized fatigue model and input four-dimensional data (exhaled residual 0.35mg / m³). 3 GSR 2μS, HRV 110ms, spatial concentration difference 0.05mg / m 3 The output is level 3 fatigue, corresponding to the basic strategy concentration gradient adjustment + aroma micro-switching; The second step involves calling the adaptive diffusion model, inputting a temperature and humidity of 25℃, an airflow of 0.2m / s, and a flow of 2 people. The calculation shows that the diffusion coefficient needs to be increased by 15% and the release angle needs to be adjusted by 10°. The third step combines health goals (VOCs 0.4 mg / m³). 3 <0.6mg / m 3(Limits), determine the final strategy parameters: concentration from 100%→80%→60%→50%→40%, fragrance from citrus to citrus-green tea, release angle 40°; S22. The final strategy is sent to the multi-strategy collaborative dynamic release control unit via the MQTT protocol, and the instruction format adopts structured data. The control center receives execution status feedback once every 5 seconds. For example, in the 5th minute, the concentration has dropped to 80% and the fragrance switching is completed. If there is an execution deviation, such as the concentration is still 75% in the 8th minute, which has not reached the 60% target, the compensation mechanism is automatically triggered: the fan speed of the fragrance machine is increased by 10%, and the maintenance time of this concentration level is extended by 2 minutes. If the target is still not met after 3 retries, a fault warning is triggered, the red light of the device flashes and the APP push is sent. This process ensures that the strategy is accurately adapted to the needs of the scenario, and the generated execution data is stored in real time.

[0020] Preferably, the specific steps of closed-loop iterative optimization and anomaly handling in step S3 are as follows: S31. Every Sunday morning at 10:00, we will conduct full-cycle optimization. The first step is to review the data: using Python data analysis tools, we will count the effectiveness of each fatigue level strategy and the number of concentration imbalances in different scenarios last week, and focus on analyzing abnormal scenarios. The second step is to initiate model and strategy iteration: optimize the personalized fatigue model, reduce the fatigue trigger threshold for sensitive users from level 2 to level 1, and add a sensitive user identification field for multi-person scenarios; revise the diffusion model, increasing the diffusion coefficient from 15% to 20% for each additional person in a multi-person scenario; add a special strategy for multi-person sensitive scenarios, setting the concentration in the area where the sensitive user is located to 70% of that in other areas, and verify it through A / B testing after iteration. S32. The third step is to establish a multi-level anomaly response mechanism: When a sensor fails, such as when the GSR sensor has no data transmission, the edge device switches to the backup monitoring scheme within 1 second, using two parameters, exhaled gas residue and user behavior records, to determine fatigue. At the same time, a pop-up window will display "Physiological sensor failure, backup mode has been activated". VOCs concentration reached 0.6 mg / m³ 3 During a strong alert, immediately shut off the fragrance release channel, activate the fresh air system at maximum fan speed, and check the concentration every minute. When the concentration drops to 0.4 mg / m³, continue monitoring. 3 After that, resume release and set the concentration to the baseline of 60%. After monitoring for 30 minutes without any abnormalities, gradually increase the concentration. If power grid fluctuations cause the equipment to restart, the system will automatically load the strategy parameters before the restart, such as executing a level 3 fatigue strategy with a concentration of 40% before restarting, to avoid user-perceived interruption.

[0021] The beneficial effects of this invention are as follows: 1. This invention uses a multimodal olfactory fatigue monitoring unit to capture real-time data on exhaled breath fragrance residue, skin conductance, and other dimensions. Combined with an LSTM-attention model, it accurately determines olfactory fatigue levels from 1 to 5. A dynamic release control unit then matches a tiered strategy: for example, mild fatigue triggers short-term pulse stimulation, while severe fatigue triggers fragrance modulation switching and concentration reduction to zero. Simultaneously, distributed spatial probes prevent localized high-concentration "false fatigue." This design ensures that fragrance release always adapts to the user's real-time olfactory state, completely solving the problem of "inability to effectively perceive fragrance after continuous exposure," ensuring users can stably perceive fragrance at different times of use, and significantly improving the user experience.

[0022] 2. This invention constructs a complete logical chain of multimodal monitoring, dual-target perception, and collaborative control: a multimodal olfactory fatigue monitoring unit provides user physiological fatigue data, while a dynamic environment-health dual-target perception unit collects environmental data such as temperature, humidity, pedestrian flow, and VOCs. These two data are integrated by an edge computing control center and a strategy is generated based on a priority order of "health and safety > fatigue relief > user preference." Simultaneously, a personalized interaction unit supports user-defined thresholds and preferences, with feedback data influencing the iteration of the fragrance library. This design breaks the limitations of "single-driven environmental parameters" and establishes an adaptive control system that links "user physiological state + dynamic environment + individual preference," fundamentally solving the problem of olfactory numbness and adapting to different scenarios and user needs with varying sensitivities.

[0023] 3. This invention achieves dual protection through multiple designs: On the one hand, the dynamic release control unit dynamically adjusts the concentration based on fatigue level and environmental data. For example, when the fatigue level is moderate, the concentration is reduced to 40% of the baseline to avoid ineffective high-concentration release. The supply is accurately matched based on the average fragrance consumption demand per person, which reduces the consumption rate of fragrance materials and significantly reduces waste. On the other hand, the environment-health dual-target sensing unit monitors the VOCs concentration in real time. When the concentration exceeds the standard, it triggers a warning to reduce the concentration or shut down the machine. The control center strictly follows the health-first decision-making logic, which reduces the health risks to users and reduces unnecessary consumption of fragrance materials, taking into account both economy and safety. Attached Figure Description

[0024] Figure 1 This is a flowchart of the intelligent fragrance control system based on olfactory perception adaptation of the present invention; Figure 2 This is a flowchart of the intelligent fragrance control method based on olfactory perception adaptation according to the present invention; Figure 3 This is a flowchart of the iterative optimization and exception handling process of this invention. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0026] like Figures 1 to 3 As shown, this embodiment of the invention provides an intelligent fragrance control system based on olfactory perception adaptation, the system comprising: Multimodal olfactory fatigue monitoring unit: integrates dual-parameter olfactory sensor, wearable linkage component, and distributed probe. It uses LSTM-attention model to process four-dimensional data of "gas + physiology + space + behavior", outputs fatigue level 1-5 and recovery time, and builds "sensitive / tolerant" personalized model to provide accurate fatigue benchmark for subsequent units. Dynamic Environment-Health Dual-Target Sensing Unit: Added VOCs and millimeter-wave human flow sensors, adaptive diffusion model combined with prior fatigue level adjustment coefficient to generate spatial concentration report, binding environmental adaptation and health and safety, providing dual-dimensional decision-making basis for releasing control unit; Multi-strategy collaborative dynamic release control unit: stratified strategies are set according to fatigue level (different for mild / moderate / severe), embedded VOCs excess response mechanism, and the fragrance library is iterated based on user feedback to provide optimization direction for interactive units, realizing "fatigue-health-preference" collaborative control; Personalized interaction and feedback incentive unit: Supports custom fatigue thresholds and fragrance preferences, post-strategy pop-up to receive feedback, weekly health reports, unlocking customized fragrances upon completion of feedback, and supplementing the preceding fragrance library with data to provide personalized support for the central unit; Edge computing collaborative control hub unit: Adopting an architecture of "real-time edge processing + cloud iteration", it builds a health-first decision matrix, performs edge-de-identified data transmission, cloud-based federated learning, generates final instructions, supports subsequent processes, and ensures efficient system collaboration.

[0027] The multimodal olfactory fatigue monitoring unit includes: The dual-parameter olfactory sensor for exhaled fragrance residue and nasal temperature and humidity uses the MSP430 series gas-sensitive film, with a response time of ≤1 second. It can capture temperature and humidity fluctuations of 0.1℃ in the olfactory receptor mucosa, helping to judge subtle changes in the early stage of fatigue. Wearable linkage components connect to mainstream smart bracelets via Bluetooth 5.0 protocol to collect skin conductance response (GSR, sampling rate 100Hz) and heart rate variability (HRV, time resolution 1ms) in real time. Olfactory fatigue can cause a decrease in the excitability of sympathetic nerve endings. Three to five distributed spatial fragrance detectors, employing SHT3x series temperature and humidity compensation technology, are deployed in key areas such as the living room and bedroom, achieving a detection accuracy of 0.01 mg / m³. 3 This can rule out false fatigue misjudgments caused by localized high concentrations; At the algorithm level, an LSTM-attention mechanism deep learning model is used. The training data covers 100,000 time-series data from 500 users of different ages (20-60 years old) and different olfactory sensitivities. After inputting four-dimensional data of "gas residue, physiological signal, spatial concentration difference and user behavior records (on-hook duration, adjustment frequency)", the system outputs individual olfactory fatigue levels of 1-5 levels and recovery time accurate to the minute. The interaction process between the LSTM-attention model and the hardware: Data input stage: The edge device (equipped with an ARM Cortex-A53 quad-core processor) receives exhaled gas residue data from the MSP430 sensor (sampling frequency 1Hz) via the SPI interface, nasal temperature and humidity data from the SHT3x sensor (sampling frequency 0.5Hz) via the I2C interface, GSR (100Hz) and HRV (1ms time resolution) data from the smart bracelet via the Bluetooth 5.0 protocol, and spatial concentration difference data from the distributed probe via the RS485 interface (sampling frequency 0.2Hz). Data preprocessing stage: The edge device synchronizes the received multi-source heterogeneous data (aligned to 10-second granularity based on UTC timestamps), and uses the 3σ principle to remove outliers (such as a sudden increase of 2mg / m³ in exhaled residual data). 3 If the sensor is identified as faulty, the backup data completion logic is triggered, and the HRV data is calculated as a 5-minute moving average (the original data is stored locally with AES-256 encryption for 7 days). Model computation stage: Preprocessed data is input into an LSTM layer (64 hidden units, ReLU activation function) to capture temporal features (such as the decreasing trend of GSR with olfactory fatigue); then, the core features are enhanced through an attention layer (the weight coefficients are dynamically adjusted based on the influencing factors of olfactory fatigue, such as the weight of exhaled residual weight w1=2.0 and the weight of GSR weight w2=0.3); finally, the fatigue level of 1-5 and the recovery time are output through a fully connected layer (the output results are transmitted to the control center in real time via the MQTT protocol). Model iteration phase: Every Sunday at 2 AM, the edge device uploads anonymized fatigue data (removing user ID and device MAC address, HRV is the 5-minute average) to the cloud (Alibaba Cloud ECS server). The cloud uses federated learning technology, exchanging only model parameters (without transmitting raw data), iteratively adjusting the fatigue threshold for different age groups (e.g., the sensitivity threshold for 20-30 years old starts from 0.3 mg / m³). 3 Reduced to 0.25 mg / m³ 3The iterated model is delivered to the edge via OTA protocol (data packets are encrypted with AES-128, and the old model is replaced after the MD5 verification is passed). Establish a dynamic and personalized model. For example, the fatigue threshold for "sensitive" users (who have a strong daily reaction to floral fragrances) is set to level 1, and for "tolerant" users (who are exposed to fragrance environments for a long time) it is set to level 3. The model will automatically fine-tune the threshold based on the user's weekly interaction data, providing a fatigue benchmark that is accurate to the individual for the subsequent dynamic environment-health dual-objective perception unit. LSTM-Attention Model Data Labeling Standards: Level 1 fatigue corresponds to GSR ≥ 5 μS and exhaled residual temperature ≤ 0.2 mg / m³. 3 HRV ≤ 80ms, grades 2-5, decrease by 5% according to GSR, increase exhaled residual by 0.1mg / m³. 3 HRV incremented by 20ms for labeling; Adam optimizer was explicitly used for training, with a learning rate of 0.001, 100 iterations, stopping when MSE on the validation set ≤ 0.05, and cross-validation to calibrate the weights for each age group; Distributed spatial fragrance detector deployment rules: living room 20-30m 2 Set up 2 units, spaced at least 3 meters apart and avoiding ventilation openings; 15-20 meters apart in bedrooms. 2 Set up one unit, located in the center of the room; calibrate on the 1st of each month using 0.3 mg / m². 3 Standard fragrance gas, readings corrected to ±0.005 mg / m³ 3 The calibration records are stored locally for 7 days.

[0028] The dynamic environment-health dual-objective perception unit includes: VOCs concentration sensor, TGS2600 series, detection range 0-10 mg / m³ 3 With a response time of ≤5 seconds, it monitors the volatile organic compounds emitted by the fragrance in real time, strictly adhering to the standard of 0.6 mg / m³ in GB / T18883-2022. 3 The indoor air quality limit is set at 0.5 mg / m³. 3 A yellow warning is triggered, and a local buzzer sounds; the concentration reaches 0.6 mg / m³. 3 A strong red alert is triggered, and a notification is sent to the mobile app. The RF600 series millimeter-wave spatial crowd sensor has a detection range of 0.5-5m and an angle of 120°. It can penetrate thin curtains, glass and other obstructions, accurately count 1-10 people, and dynamically adjust the supply matching degree based on the formula for calculating the average fragrance consumption demand per person. An adaptive modified fragrance diffusion model is adopted, which is optimized based on the Navier-Stokes equations of fluid dynamics. It can adjust the diffusion coefficient in real time according to the fatigue level. The coefficient increases by 2% for every 1°C increase in temperature, 5% for every 0.1m / s increase in airflow, and 15% for every additional person in the flow of people. It also generates a visual spatial concentration balance report. This unit deeply integrates environmental adaptation with health and safety, avoiding concentration imbalance of the fixed model in scenarios such as open ventilation and large gatherings, and providing a dual-dimensional decision-making basis of environmental parameters and health thresholds for the multi-strategy collaborative dynamic release control unit. The adaptive diffusion model Navier-Stokes equations were used to optimize parameters, where F represents the fatigue correction force: 0.02 for levels 1-2, 0 for levels 3-4, and -0.05 for level 5. The concentration visualization report was generated using Matplotlib with a coordinate accuracy of 0.5m × 0.5m, with red markers indicating concentrations exceeding 0.5 mg / m³. 3 Area, Blue Label ≤0.3mg / m³ 3 The region will be synchronized to the app within 10 seconds of being generated. VOCs warning operation intervals: For a yellow alert, monitor the concentration every 2 minutes; if the concentration does not decrease within 5 minutes, shut off the supply. For a red alert, after supply is cut off, monitor the concentration every 1 minute until it drops to 0.4 mg / m³. 3 When the air conditioner is in use, maintain the maximum airflow speed for 3 minutes before resuming the fragrance. During the initial resuming process, measure the concentration every 5 minutes.

[0029] The multi-strategy collaborative dynamic release control unit includes: Based on fatigue level and environmental data, a refined, hierarchical dynamic strategy logic is constructed: Mild fatigue (grades 1-2, corresponding to 0.2-0.3 mg / m³ of residual exhaled gas) 3 When this occurs, only a short-duration pulse release is initiated, increasing the concentration by 15% compared to baseline, such as a baseline of 0.4 mg / m³. 3 →0.46mg / m 3 Avoid overstimulating olfactory receptors; Moderate fatigue (grades 3-4, residual fatigue 0.3-0.4 mg / m²) 3 When performing concentration gradient adjustment, the concentration is reduced from the baseline of 100% to 40% in 5 steps, such as 100%→80%→60%→50%→40%, while simultaneously switching the fragrance type slightly. 70% of the original fragrance core ingredients are retained. For example, the citrus note retains sweet orange essential oil and adds 30% green tea essential oil to reduce the abruptness of the olfactory system. Severe fatigue (Level 5, residual >0.4 mg / m²) 3When the concentration is triggered to zero for 15 minutes, the fresh air system is simultaneously activated at maximum speed to accelerate the removal of residues. Then, the fragrance is switched across different notes, such as woody notes to aquatic notes, with a component difference of >80%. During the recovery phase, a 10-second pulse is released every 2 minutes, and the concentration gradually rises from 20% to 80%. Simultaneously embed a health-first mechanism: If the VOCs concentration reaches 0.5 mg / m³ 3 The system will automatically reduce the fragrance concentration by 20% and continue monitoring for 5 minutes. If the concentration does not decrease, the fragrance release will be turned off. At the same time, a pop-up message will appear on the device display and mobile app indicating that the VOC concentration is too high and the release has been paused. If it reaches 0.6 mg / m 3 A strong early warning value directly cuts off the fragrance supply and activates fresh air purification; in addition, the fragrance library is iterated based on user feedback, and fragrances with a rating of ≥4 stars (out of 5 stars) are retained and given increased usage weight, while fragrances with <3 stars are directly eliminated. Every quarter, 2-3 new combinations are added based on seasonal characteristics, such as adding mint-ocean notes in summer and cedar-cinnamon notes in winter, providing a clear strategic optimization direction for personalized interaction and feedback incentive units; The time parameters for the stratified strategy are as follows: for mild fatigue, the pulse release is 10 seconds each time with a 2-minute interval, lasting for 15 minutes and then pausing for 5 minutes; for moderate fatigue, the gradient adjustment is maintained for 5 minutes at each level with a 10-second switching interval; for severe fatigue, after 15 minutes of zeroing, the pulse recovery is 10 seconds each time with a 2-minute interval, and the concentration is increased by 10% each time to 80% of the baseline. For fragrance switching, the composition must be determined as follows: For cross-fragment switching, the difference between the core essential oil components of the original and new fragrances must be >80% (difference = number of different components / total number of components × 100%); for micro-switching, 70% of the original core components must be retained, and the odor threshold difference between the newly added 30% and the original components must be ≥0.1 mg / m³. 3 Concentration stability was measured 30 seconds after switching.

[0030] The personalized interaction and feedback incentive unit includes: Dual-platform multi-dimensional manual operation interface: Device panel and mobile APP: The device panel uses touch buttons and supports one-click switching of olfactory fatigue warning thresholds, with 1-5 levels selectable and corresponding to different colored indicator lights. The app offers a fragrance preference weight setting function, with four fragrance categories—floral, woody, aquatic, and citrus—as adjustable sliders. Users can precisely adjust these based on their mood for the day, such as setting the aquatic scent to 80 points on a rainy day. Three minutes after the strategy is executed, the system triggers feedback through an APP pop-up or device voice prompt. The feedback interface includes a 1-5 star rating, three options for effective fatigue relief, average, and ineffective, as well as a text note box. It supports inputting specific suggestions such as the fragrance being too strong or switching too frequently. After the feedback is submitted, a confirmation prompt of "Thank you for your feedback, it has been incorporated into the strategy optimization" is displayed immediately. The system pushes the olfactory health report for last week to the app, and the report is presented in charts: The line chart shows the daily fatigue level distribution, the bar chart compares the number of times each strategy was implemented and its effectiveness, and the number cards indicate the number of times VOCs exceeded the standard and the peak value. For example, if there were 2 instances of exceeding the standard this week, with the highest value being 0.52 mg / m³, the chart would be displayed. 3 Finally, based on user preferences and seasonal characteristics, we recommend fragrances for next week, such as a mint-marine scent for summer, noting its invigorating properties and VOC content as low as 0.3mg / m³. 3 ; Simultaneously, a tiered feedback incentive mechanism is designed: completing 3 valid feedback sessions unlocks 1 customized fragrance formula, supporting adjustments to the ratio of 3 essential oil components, such as citrus: green tea: peppermint = 7:2:1. After customization, it is automatically saved as a personal exclusive fragrance, effectively enhancing user engagement and the long-term adaptability of the system. The collected feedback data is synchronized in real time to the fragrance library of the multi-strategy collaborative dynamic release control unit. To determine the validity of APP feedback, valid feedback requires a 1-5 star rating, selection of fatigue relief effect, and a text note of ≥5 characters marked as "high-quality feedback". The pop-up prompt lasts for 15 seconds. If no action is taken, the feedback should be resubmitted after 1 minute. Three instances of no feedback are recorded as "no response". Weekly reports will summarize the number and percentage of no responses. Customizable essential oil parameters: The unlocked customizable fragrance supports 3 essential oils, each with a concentration ranging from 0.1% to 5%, adjustable in 0.1% increments; automatic VOCs estimation during storage, exceeding 0.5mg / m³. 3 A pop-up window prompts "It is recommended to reduce the concentration of XX essential oil" and provides a recommended value. After adjustment, the concentration is recalculated to meet the standard.

[0031] The edge computing collaborative control hub unit includes: It adopts a two-tier architecture of real-time processing at the edge and long-term iteration at the cloud: The edge device is equipped with an ARM Cortex-A53 quad-core processor and integrates a Linux real-time operating system. It can integrate high-frequency data from the first four units in real time and generate an instant release strategy with a latency of ≤1 second. At the same time, it stores the raw data of the last 24 hours locally to avoid latency issues caused by relying on cloud transmission. The cloud-based deployment is on Alibaba Cloud ECS servers. Model iterations begin every Sunday at 2 AM (during off-peak user periods). Iteration data is derived from anonymized user data nationwide. Specific optimization directions include: adjusting fatigue thresholds for different age groups (e.g., reducing the sensitivity threshold for 20-30 year olds from 0.3 mg / m³). 3Reduced to 0.25 mg / m³ 3 For individuals aged 50-60, the tolerance threshold is 0.4 mg / m³. 3 Increased to 0.45 mg / m³ 3 The environmental coefficient of the diffusion model is corrected (e.g., when humidity > 60%, the diffusion coefficient is increased by an additional 8%), and the optimized model is automatically distributed to the edge after the iteration is completed; Establish a quantitative priority decision matrix, with a weighting system of 10 points: Health and safety (weight 6 points) take priority. When VOCs exceed the standard, the concentration is forcibly limited to 80% of the baseline. Even if the user manually adjusts it to 100%, it will be ineffective. A pop-up message will appear to indicate that the concentration has been limited in order to protect health. Fatigue relief (weight 3 points) is the second best option. In multi-person scenarios, spatial partitioning detection prioritizes reducing the concentration by 10% in areas where users with high fatigue levels are located (such as workstations near windows). User Preference (Weight 1 point) Finally, if the VOCs content of the user's preferred fragrance type (such as rich woody notes) is close to the limit, the system will automatically recommend similar low VOCs options, such as light woody notes, with the note "85% similarity to your preference, 30% lower VOCs"; In terms of privacy protection, user IDs and device MAC addresses are removed from all data uploaded to the cloud from the edge. Physiological data (such as HRV) is averaged over a period of 5 minutes before being uploaded. The cloud uses federated learning technology, where nodes in different regions only exchange model parameters (without transmitting raw physiological data). This unit integrates all data to generate the final execution instructions. The edge hardware system parameters are as follows: RK3399 chip (ARM Cortex-A53 architecture), 2GB LPDDR4 memory, 16GB eMMC storage, and Linux 5.4 kernel (including RT_PREEMPT real-time patch); local data is divided into "raw area" (24-hour cyclic overwrite) and "critical log area" (stored for 14 days), and the partition encryption uses AES-256; The cloud-based iterative data rules retrieve anonymized data from each edge device every Sunday at 2 AM, grouping it by age group (e.g., 20-30 years old), with each group containing ≥1000 data entries to adjust the threshold. After iteration, the model is distributed via OTA protocol, with data packets encrypted using AES-128. The edge device receives the data and verifies the MD5 hash; if it passes, the model is replaced; otherwise, the old model is rolled back.

[0032] This invention also provides an intelligent fragrance control method based on olfactory perception adaptation. Based on the system of claim 6, the specific steps of this method are as follows: S1. Multimodal data acquisition and spatiotemporal fusion: fatigue monitoring data, environmental health data and user interaction data are collected at a fixed frequency at the edge. Outliers are removed using the 3σ principle. Spatiotemporal alignment is performed using UTC timestamps. HRV data is taken as the 5-minute moving average before being uploaded to the cloud. S2. Multi-target collaborative release strategy generation and distribution: The control center identifies the scenario, sorts the targets in the order of health and safety > fatigue relief > user preference, calls the personalized fatigue model to calculate the strategy, distributes the strategy and receives the status through the MQTT protocol, and adjusts the parameters when an exception occurs. S3. Closed-loop iterative optimization and anomaly handling: Use Python data analysis tools to regularly review data, optimize personalized fatigue models and strategy iterations, add exclusive strategies for multi-person sensitive scenarios, and initiate countermeasures when sensor failures or VOCs concentrations exceed the standard.

[0033] The specific steps of the multimodal data acquisition and spatiotemporal fusion process in step S1 are as follows: S11. Three types of data are synchronously collected at a fixed frequency at the edge: Fatigue monitoring data is collected by MSP430 sensor for exhaled residuals, SHT3x sensor for nasal temperature and humidity, Bluetooth-connected wristband for GSR and HRV, and distributed probes for spatial concentration difference. Environmental health data: SHT4x collects temperature and humidity, FS3000 collects airflow, TGS2600 collects VOCs, and RF600 collects pedestrian flow. User interaction data, such as threshold settings on the device panel and feedback ratings from the app, is transmitted using the MQTT protocol. After data collection, a three-step fine preprocessing process is performed: The first step, outlier removal, employs the 3σ principle. The mean and standard deviation of the data are calculated, and outliers exceeding ±3 times the standard deviation are removed. For example, fragrance residue exceeding 0.2 mg / m³... 3 Sudden rise to 2 mg / m 3 Data deemed faulty by the sensor was discarded. The second step is spatiotemporal alignment, which uses UTC timestamps to synchronize data of different frequencies to a uniform granularity of 10 seconds / time. The spatial dimension is labeled with sensor location codes for each data point, such as living room-1 and bedroom-2, to ensure that the data corresponds accurately to the physical space. The third step is to desensitize physiological data. HRV data is taken as the 5-minute moving average before being uploaded to the cloud. The original HRV data is encrypted and stored locally for 7 days and then automatically deleted. The AES-256 encryption algorithm is used. S12. Finally, generate a three-dimensional matrix of time, data type, and space, stored in JSON format: The time dimension is divided into 10-second intervals, the data type dimension is divided into three categories: fatigue, environment, and interaction, and the spatial dimension is divided according to the sensor location. For example, living room-1 corresponds to one set of data. Each cell of the matrix contains a specific value and a status indicator. The data retrieval response time of this matrix is ​​≤500ms, and it fully covers the individual fatigue, environmental status, and user preference information required for decision-making.

[0034] The specific steps for generating and issuing the multi-target collaborative release strategy in step S2 are as follows: S21. The control center first performs intelligent scene recognition based on a three-dimensional matrix, using the LightGBM classification model. It inputs four features: time, pedestrian flow, HRV value, and spatial location. It automatically identifies the scenario as a two-person fatigue scenario in the afternoon at home, then prioritizes the target in the order of health and safety > fatigue relief > user preferences. Next, it initiates a three-step dynamic strategy calculation: The first step is to invoke the personalized fatigue model and input four-dimensional data (exhaled residual 0.35mg / m³). 3 GSR 2μS, HRV 110ms, spatial concentration difference 0.05mg / m 3 The output is level 3 fatigue, corresponding to the basic strategy concentration gradient adjustment + aroma micro-switching; The second step involves calling the adaptive diffusion model, inputting a temperature and humidity of 25℃, an airflow of 0.2m / s, and a flow of 2 people. The calculation shows that the diffusion coefficient needs to be increased by 15% and the release angle needs to be adjusted by 10°. The third step combines health goals (VOCs 0.4 mg / m³). 3 <0.6mg / m 3 (Limits), determine the final strategy parameters: concentration from 100%→80%→60%→50%→40%, fragrance from citrus to citrus-green tea, release angle 40°; S22. The final strategy is sent to the multi-strategy collaborative dynamic release control unit via the MQTT protocol, and the instruction format adopts structured data. The control center receives execution status feedback once every 5 seconds. For example, in the 5th minute, the concentration has dropped to 80% and the fragrance switching is completed. If there is an execution deviation, such as the concentration is still 75% in the 8th minute, which has not reached the 60% target, the compensation mechanism is automatically triggered: the fan speed of the fragrance machine is increased by 10%, and the maintenance time of this concentration level is extended by 2 minutes. If the target is still not met after 3 retries, a fault warning is triggered, the red light of the device flashes and the APP push is sent. This process ensures that the strategy is accurately adapted to the needs of the scenario, and the generated execution data is stored in real time.

[0035] The specific steps of closed-loop iterative optimization and anomaly handling in step S3 are as follows: S31. Every Sunday morning at 10:00, we will conduct full-cycle optimization. The first step is to review the data: using Python data analysis tools, we will count the effectiveness of each fatigue level strategy and the number of concentration imbalances in different scenarios last week, and focus on analyzing abnormal scenarios. The second step is to initiate model and strategy iteration: optimize the personalized fatigue model, reduce the fatigue trigger threshold for sensitive users from level 2 to level 1, and add a sensitive user identification field for multi-person scenarios; revise the diffusion model, increasing the diffusion coefficient from 15% to 20% for each additional person in a multi-person scenario; add a special strategy for multi-person sensitive scenarios, setting the concentration in the area where the sensitive user is located to 70% of that in other areas, and verify it through A / B testing after iteration. S32. The third step is to establish a multi-level anomaly response mechanism: When a sensor fails, such as when the GSR sensor has no data transmission, the edge device switches to the backup monitoring scheme within 1 second, using two parameters, exhaled gas residue and user behavior records, to determine fatigue. At the same time, a pop-up window prompts "Physiological sensor failure, backup mode has been activated". VOCs concentration reached 0.6 mg / m³ 3 During a strong alert, immediately shut off the fragrance release channel, activate the fresh air system at maximum fan speed, and check the concentration every minute. When the concentration drops to 0.4 mg / m³, continue monitoring. 3 After that, resume release and set the concentration to the baseline of 60%. After monitoring for 30 minutes without any abnormalities, gradually increase the concentration. If power grid fluctuations cause the equipment to restart, the system will automatically load the strategy parameters before the restart, such as executing a level 3 fatigue strategy with a concentration of 40% before restarting, to avoid user-perceived interruption.

[0036] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0037] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart fragrance control system based on olfactory perception adaptation, characterized in that, The system includes: Multimodal olfactory fatigue monitoring unit: integrates a dual-parameter olfactory sensor, wearable linkage components, and distributed spatial fragrance probes. It uses an LSTM-attention deep learning model to process four-dimensional data including gas residue, physiological signals, spatial concentration differences, and user behavior records, outputting fatigue level and recovery time, and establishing a dynamic personalized model. Dynamic Environment-Health Dual-Target Sensing Unit: Added VOCs sensor and millimeter-wave spatial flow sensor, adopts adaptive correction fragrance diffusion model combined with fatigue level to adjust diffusion coefficient in real time, generates spatial concentration balance report, and binds environmental adaptability and health and safety; Multi-strategy collaborative dynamic release control unit: constructs hierarchical strategy logic based on fatigue level and environmental data, embeds VOCs concentration exceeding the standard response mechanism, and iterates the fragrance library based on user feedback; Personalized interaction and feedback incentive unit: Supports customizable olfactory fatigue warning threshold and fragrance preference weight setting function. Feedback is received through APP pop-up after the strategy is executed. Health reports are pushed through APP regularly. Customized fragrance is unlocked after completing feedback. Edge computing collaborative control hub unit: It adopts a real-time processing at the edge + iterative architecture in the cloud, establishes a health-first decision matrix, uploads data to the cloud after de-identification at the edge, and uses federated learning technology in the cloud to generate the final execution instructions.

2. The intelligent fragrance control system based on olfactory perception adaptation according to claim 1, characterized in that, The multimodal olfactory fatigue monitoring unit includes: The exhaled breath fragrance residue and nasal cavity temperature and humidity dual-parameter olfactory sensor uses the MSP430 series gas-sensitive film. Wearable linkage components connect to mainstream smart bracelets via Bluetooth 5.0 protocol to collect skin conductance and heart rate variability in real time; Distributed spatial fragrance probes, employing SHT3x series temperature and humidity compensation technology; At the algorithm level, an LSTM-attention mechanism deep learning model is used. The training data consists of time-series data of users of different ages and different olfactory sensitivities. After inputting four-dimensional data of gas residue, physiological signal, spatial concentration difference and user behavior record, the model outputs the individual olfactory fatigue level and recovery time accurate to the minute. At the same time, a dynamic personalized model is established, which automatically fine-tunes the thresholds based on users' weekly interaction data.

3. The intelligent fragrance control system based on olfactory perception adaptation according to claim 2, characterized in that, The dynamic environment-health dual-objective perception unit includes: VOCs concentration sensor to monitor volatile organic compounds emitted by fragrances in real time; Millimeter-wave spatial crowd flow sensor dynamically adjusts supply matching degree through the formula: per capita fragrance consumption demand = total fragrance release / number of people; An adaptive modified fragrance diffusion model is adopted, which is optimized based on the Navier-Stokes equations of hydrodynamics. The diffusion coefficient is adjusted in real time in combination with fatigue level, and a visual spatial concentration balance report is generated, linking environmental adaptation with health and safety.

4. The intelligent fragrance control system based on olfactory perception adaptation according to claim 3, characterized in that: The multi-strategy collaborative dynamic release control unit includes: Based on fatigue level and environmental data, a refined, layered dynamic strategy logic is constructed. When fatigue is mild, only a short-term pulse release is initiated; when fatigue is moderate, concentration gradient adjustment is performed; when fatigue is severe, the concentration is triggered to zero for 15 minutes, and the maximum wind speed of the fresh air system is simultaneously activated to accelerate the removal of residues, followed by fragrance cross-adjustment. Simultaneously, a health-first mechanism is embedded, and the fragrance library is iterated based on user feedback. Fragrances with a rating of ≥4 stars are retained and given increased usage weight, while fragrances with a rating of <3 stars are directly eliminated. New combinations are added every quarter based on seasonal characteristics, providing strategic optimization directions for personalized interaction and feedback incentive units.

5. The intelligent fragrance control system based on olfactory perception adaptation according to claim 4, characterized in that, The personalized interaction and feedback incentive unit includes: The device panel features touch-sensitive buttons that support one-click switching of olfactory fatigue warning thresholds; On the mobile app, a fragrance preference weight setting function is provided, and different types of fragrances can be set as sliding adjustment bars. Three minutes after the strategy is executed, the system will trigger feedback through an app pop-up or device voice prompt. The feedback interface includes a 1-5 star rating and three options: effective, average, and ineffective fatigue relief. The system pushes the olfactory health report of the previous week to the mobile APP. The report is presented in charts, with a line chart showing the distribution of daily fatigue levels, a bar chart comparing the number of times each strategy was executed and its effectiveness, and digital cards indicating the number of times VOCs exceeded the standard and the peak value. Finally, based on user preferences and seasonal characteristics, we recommend the next week's fragrance. We also designed a tiered feedback incentive mechanism: completing 3 valid feedback sessions unlocks 1 customized fragrance formula, and completing 10 feedback sessions unlocks the family shared formula function.

6. The intelligent fragrance control system based on olfactory perception adaptation according to claim 5, characterized in that, The edge computing collaborative control hub unit includes: It adopts a two-layer architecture of real-time processing at the edge and long-term iteration at the cloud. The edge integrates the high-frequency data of the first four units in real time to generate an immediate release strategy, while storing the raw data of the last 24 hours locally. The cloud starts model iteration, adjusts the fatigue threshold of different age groups, corrects the environmental coefficient of the diffusion model, and automatically sends the optimized model to the edge after the iteration is completed. Establish a quantitative priority decision matrix, with priorities from highest to lowest as follows: health and safety, fatigue relief, and user preferences last. In terms of privacy protection, user IDs and device MAC addresses are removed from all data uploaded to the cloud from the edge devices, and physiological data is averaged before being uploaded. The cloud uses federated learning technology, where nodes in different regions only exchange model parameters and integrate all data to generate the final execution instructions.

7. A smart fragrance control method based on adaptive olfactory perception, based on the system described in claim 6, characterized in that: The specific steps of this method are as follows: Multimodal data acquisition and spatiotemporal fusion: fatigue monitoring data, environmental health data and user interaction data are collected at a fixed frequency at the edge. Outliers are removed using the 3σ principle. Spatiotemporal alignment is performed using UTC timestamps. HRV data is taken as the 5-minute moving average before being uploaded to the cloud. Multi-target collaborative release strategy generation and distribution: The control center identifies the scenario, sorts the targets in the order of health and safety > fatigue relief > user preference, calls the personalized fatigue model to calculate the strategy, distributes the strategy and receives the status through the MQTT protocol, and adjusts the parameters when an exception occurs; Closed-loop iterative optimization and anomaly handling: Use Python data analysis tools to regularly review data, optimize personalized fatigue models and strategy iterations, add exclusive strategies for multi-person sensitive scenarios, and initiate countermeasures when sensor failures or VOCs concentrations exceed the standard.

8. The intelligent fragrance control method based on olfactory perception adaptation according to claim 7, characterized in that: The specific steps of the multimodal data acquisition and spatiotemporal fusion process are as follows: Fatigue monitoring data, environmental health data, and user interaction data are collected synchronously at the edge. After collection, three-step fine preprocessing is performed. The first step uses the 3σ principle to remove outliers. The second step is spatiotemporal alignment, which synchronizes data of different frequencies using UTC timestamps, and spatially labels each data point with sensor location codes. The third step is to desensitize the physiological data. The HRV data is taken as the 5-minute moving average and then uploaded to the cloud. Generate a three-dimensional matrix of time, data type, and space, stored in JSON format: the time dimension is divided into 10-second intervals, the data type dimension is divided into three categories: fatigue, environment, and interaction, and the spatial dimension is divided according to sensor location. Each cell of the matrix contains a specific value and a status identifier. The data retrieval response time of this matrix is ​​≤500ms, and it can be called by the strategy calculation module of the control center.

9. The intelligent fragrance control method based on adaptive olfactory perception according to claim 8, characterized in that: The specific steps for generating and issuing the multi-target collaborative release strategy are as follows: The control center performs intelligent scene recognition based on a three-dimensional matrix and adopts the LightGBM classification model. It takes four types of features as input: time, people flow, HRV value, and spatial location, and automatically judges it as a home afternoon fatigue scene for two people. Then, it sorts the targets in the order of health and safety > fatigue relief > user preference. Initiating dynamic strategy calculation: The first step is to call the personalized fatigue model, input four-dimensional data, and output level 3 fatigue; The second step is to invoke the adaptive diffusion model; The third step is to combine health goals to determine the final strategy parameters; The strategy is sent to the dynamic release control unit via the MQTT protocol, and the command format uses structured data; The control center receives execution status feedback. If an execution deviation occurs, the compensation mechanism is automatically triggered to increase the fan speed of the fragrance machine and extend the maintenance time of the concentration level. If the standard is still not met after 3 retries, a fault warning is triggered.

10. The intelligent fragrance control method based on olfactory perception adaptation according to claim 9, characterized in that: The specific steps of the closed-loop iterative optimization and anomaly handling are as follows: The first step is to review the data, using Python data analysis tools to statistically analyze the effectiveness of strategies for each fatigue level last week and the number of concentration imbalances under different scenarios; the second step is to start model and strategy iteration and optimize the personalized fatigue model. The diffusion model was revised, and a special strategy for sensitive scenarios involving multiple users was added. After iteration, the strategy was verified through A / B testing. The third step is to establish a multi-level anomaly response mechanism. When a sensor fails, the edge device switches to the backup monitoring scheme and a pop-up window indicates that the physiological sensor has failed and the backup mode has been activated. When the VOCs concentration reaches a strong warning level, the fragrance release channel is immediately cut off and the maximum airflow of the fresh air system is activated.

Images