An indoor air quality real-time monitoring system based on multi-sensor fusion

Abstract

The application discloses a kind of indoor air quality real-time monitoring system based on multi-sensor fusion, specifically relates to indoor environment monitoring and health protection technical field, and sensitive population breathing zone data is collected by transient exposure perception module and generates transient exposure data, time alignment fusion module carries out time alignment to multiple sensor data and generates multidimensional pollution exposure vector, risk driving control module is mapped into health risk utility value and generates control instruction with minimization as target, and health feedback calibration module generates new individualized exposure response boundary parameter according to health outcome feedback and reversely adjusts the sampling strategy and determination criterion of perception module, and forms closed loop optimization;The application realizes accurate exposure evaluation by breathing zone transient exposure perception and time alignment fusion;Through health risk utility driven control, realize the collaborative optimization of multiple pollutants;Through health outcome reverse calibration, closed-loop adaptive mechanism is constructed, and the health protection effect for sensitive population is significantly improved.

Description

Technical Field

[0001] This invention relates to the field of indoor environmental monitoring and health protection technology, and more specifically, to a real-time indoor air quality monitoring system based on multi-sensor fusion. Background Technology

[0002] Indoor air quality has a significant impact on human health, especially for sensitive groups such as those with asthma, chronic obstructive pulmonary disease (COPD), and allergic rhinitis. Exposure to indoor pollutants is often a key factor in triggering acute health events. In recent years, with the development of the Internet of Things (IoT) and sensor technology, indoor air quality monitoring equipment has become increasingly widespread. Common monitoring systems use sensors deployed indoors to collect real-time data on the concentrations of pollutants such as particulate matter, formaldehyde, and carbon dioxide, and then provide environmental information to users via mobile devices.

[0003] Existing technologies have made some progress in the field of indoor air quality monitoring. For example, some systems use multi-sensor arrays to collect data on various pollutants and upload the data to a cloud platform for storage and display via wireless communication. Some systems further integrate equipment control functions, automatically activating air purifiers or fresh air systems when pollutant concentrations exceed preset thresholds, achieving basic automatic adjustment. These technological solutions provide effective means for indoor environmental monitoring and control.

[0004] However, existing technologies still have limitations in studying the health effects on sensitive populations. First, existing systems typically monitor the average concentration in the indoor environment, rather than the actual exposure dose in the respiratory zone of non-sensitive individuals, making it difficult to accurately reflect the true level of pollutant inhalation for each individual. Second, existing systems use periodic sampling and fixed threshold determination, resulting in a mismatch between their temporal resolution and the timescale of health risk, creating blind spots for monitoring transient high-exposure events occurring in sensitive populations while they are active. Third, the varying response times of different types of sensors lead to misalignment of data from multiple pollutants in combined pollution events over time, making it difficult to accurately characterize the health impact of combined exposures. Finally, existing systems' control strategies aim to achieve single pollutant concentration targets, lacking a unified decision-making mechanism guided by the health risks of sensitive populations, and failing to incorporate health outcome feedback into the system optimization loop, thus failing to achieve adaptive adjustments to individual health protection strategies. Therefore, this invention proposes a real-time indoor air quality monitoring system based on multi-sensor fusion to address the aforementioned problems. Summary of the Invention

[0005] To achieve the above objectives, the present invention provides the following technical solution: A real-time indoor air quality monitoring system based on multi-sensor fusion includes: The transient exposure sensing module is used to collect concentration data of multiple pollutants in the respiratory zone of sensitive populations at a preset sampling frequency, and simultaneously collect human activity status data to generate transient exposure data in the respiratory zone. The temporal alignment and fusion module is used to dynamically align the concentration data of multiple pollutants with different sensor response times in the transient exposure data of the respiratory zone, and generate a time-aligned multidimensional pollution exposure vector. The risk-driven control module is used to map time-aligned multidimensional pollution exposure vectors into health risk utility values ​​that uniformly characterize the health risks of sensitive populations, and to generate a sequence of control instructions with the goal of minimizing the health risk utility values. The health feedback calibration module is used to collect health outcome feedback data of sensitive populations after the control command sequence is executed, generate new individualized exposure response boundary parameters based on the health outcome feedback data, and input the new individualized exposure response boundary parameters into the transient exposure perception module to adjust the sampling strategy and exposure judgment criteria of the transient exposure perception module.

[0006] In a preferred embodiment, the transient exposure sensing module generates transient exposure data of the respiratory zone by including the following steps: The location information of the breathing zone of sensitive people is obtained, and multiple pollutant sensors and human activity status sensors are deployed in the breathing zone according to the location information. The human activity status sensors include accelerometers and heart rate sensors. Simultaneously collect concentration data of multiple pollutants in the respiratory zone and human activity status data at a preset sampling frequency; The activity intensity index is calculated based on the acceleration data in the human activity state data. The logarithmically transformed value of the activity intensity index is fused with the heart rate data. The real-time ventilation of the sensitive population is estimated through a preset ventilation volume mapping model. The ventilation volume mapping model is a linear regression model pre-established based on the sensitive population sample. The model uses the logarithmically transformed value of the activity intensity index and the heart rate value as input variables and ventilation volume as output variable. The concentration data of multiple pollutants are aligned and matched with the real-time ventilation volume in the time dimension to calculate the inhalation volume per unit time of each pollutant. The inhalation volume per unit time is obtained by multiplying the instantaneous concentration value of the pollutant by the real-time ventilation volume.

[0007] Transient exposure data in the respiratory zone is obtained by encapsulating data including instantaneous concentration values ​​of multiple pollutants, real-time ventilation values, and the inhalation volume of each pollutant per unit time.

[0008] In a preferred embodiment, the time-aligned fusion module generates a time-aligned multidimensional contamination exposure vector by the following sub-steps: Acquire transient exposure data in the respiratory zone, extract concentration data of multiple pollutants and their corresponding timestamps, and identify the sensor type and response time characteristics corresponding to each pollutant concentration data; The pollutant concentration data with the shortest response time is selected from the multi-type pollutant concentration data as the reference time axis. This reference time axis is used to mark the actual occurrence time of the pollution event. For the other types of pollutant concentration data, feature point matching is performed with the reference time axis to identify the starting point of the rising edge, the peak point, and the inflection point of the falling edge in the pollutant concentration curve. Calculate the time offset of the pollutant concentration data relative to the reference time axis based on the matched feature points; The pollutant concentration data is resampled on the time axis according to the time offset, and the time points of the pollutant concentration data are mapped to the time grid of the reference time axis through interpolation, so that the pollutant concentration data and the reference time axis are aligned point-to-point in the time dimension. The time-aligned multi-pollutant concentration data are merged with the instantaneous pollutant concentration values, real-time ventilation values, and inhalation volume per unit time from the transient exposure data of the respiratory zone to generate a time-aligned multi-dimensional pollution exposure vector.

[0009] In a preferred embodiment, the step of calculating the time offset of the pollutant concentration data relative to the reference time axis is implemented using a sliding window cross-correlation algorithm, specifically including: Set the sliding window length to divide the pollutant concentration sequence on the reference time axis and the pollutant concentration sequence to be aligned into multiple subsequences respectively; Calculate the cross-correlation coefficient between each pair of subsequences and determine the sliding window position corresponding to the maximum cross-correlation coefficient; The time offset of the pollutant concentration sequence to be aligned relative to the reference time axis is determined by the sliding window position corresponding to the maximum cross-correlation coefficient.

[0010] In a preferred embodiment, the risk-driven control module maps time-aligned multidimensional pollution exposure vectors to health risk utility values ​​in the following way: Obtain time-aligned multidimensional pollution exposure vectors, which include instantaneous concentration values ​​of multiple pollutants on a unified time axis and synchronized real-time ventilation values; Obtain individual characteristic parameters of the sensitive population, including age, disease type, and allergen spectrum. When the sensitive population includes multiple individuals, obtain the individual characteristic parameters of each individual separately. Based on the individual characteristic parameters of each individual, a risk weight function for each type of pollutant is constructed for that individual. The risk weight function is used to characterize the contribution of the pollutant to the individual's health risk per unit concentration of exposure. For each individual, the instantaneous concentration value of each type of pollutant corresponding to that individual is multiplied by the corresponding risk weight function to obtain the instantaneous health risk contribution value of that pollutant to that individual. The instantaneous health risk contribution values ​​of all pollutants are weighted and summed. The result of the weighted sum is then multiplied by the individual's real-time ventilation value to obtain the instantaneous health risk utility value of that individual at the current moment. For each individual, the instantaneous health risk utility value of that individual within a preset time window is accumulated to obtain the cumulative health risk utility value of that individual. With the goal of minimizing the sum of cumulative health risk utility values ​​of all individuals, and combining the current operating status of multiple devices with outdoor air quality data, the optimal combination of device control parameters is solved through optimization algorithms to generate a sequence of control commands.

[0011] In a preferred embodiment, the step of constructing a risk weight function for each type of pollutant for each individual based on their individual characteristic parameters specifically includes: For each individual, determine whether each type of pollutant is a triggering factor for that individual based on the individual's disease type. If it is a triggering factor, assign a non-zero weight; otherwise, assign a zero weight. For pollutants that are triggering factors, the baseline sensitivity coefficient is determined based on the individual's age. The baseline sensitivity coefficient shows a peak distribution that first increases and then decreases with age, and is used to characterize the inherent sensitivity of different age groups to pollutants. The specificity sensitivity coefficient is determined based on the individual's allergen spectrum. When the individual has a positive reaction to an allergen corresponding to a certain type of pollutant, the specificity sensitivity coefficient is set to a preset value greater than one; otherwise, it is set to one. The risk weight value of the pollutant in different concentration ranges is determined by a preset risk weight model. The risk weight model is an exponential nonlinear function. The function takes the pollutant concentration as input and the risk weight value as output. The risk weight value is obtained by multiplying the preset basic weight coefficient, basic sensitivity coefficient, specific sensitivity coefficient and the exponential function value of the pollutant concentration. The risk weight value increases exponentially with the increase of pollutant concentration. The risk weight value is the function value of the risk weight function at the corresponding pollutant concentration.

[0012] In a preferred embodiment, the step of generating the control command sequence specifically includes: Establish an indoor air quality dynamic model, which is used to describe the time-domain evolution of the concentration of various pollutants in the room under given equipment control parameters and outdoor air quality conditions. Set an optimization time window and divide the optimization time window into multiple discrete moments; Starting from the current moment, within the optimization time window, the equipment control parameter sequence is used as the decision variable, and the outdoor air quality data is used as the known disturbance input. The indoor air quality dynamic model is used to predict the multidimensional pollution exposure vector at each discrete moment within the optimization time window. Based on the predicted multidimensional pollution exposure vector, the sum of the cumulative health risk utility values ​​of all individuals within the optimization time window is calculated and used as the optimization objective function; The optimal sequence of equipment control parameters that minimizes the objective function is obtained by using a model predictive control algorithm. Select the equipment control parameters corresponding to the current moment from the optimal equipment control parameter sequence, generate the control command sequence for the current moment, and continuously update and optimize the solution process in the next moment.

[0013] In a preferred embodiment, the health feedback calibration module generates new individualized exposure response boundary parameters in the following way: Obtain environmental data after the execution of the control command sequence generated by the risk-driven control module. The environmental data includes a multi-dimensional pollution exposure vector during and after the execution of the control commands. Health outcome feedback data of sensitive populations were collected. The health outcome feedback data included the time and severity of respiratory symptoms, the time and dosage of medication use, and the physiological parameter changes collected synchronously by wearable devices. By correlating and matching health outcome feedback data with environmental data over time, pollutant exposure characteristics within a preset time window before a health event occurs are identified. These pollutant exposure characteristics include the average concentration, peak concentration, exposure duration, and concentration rise rate of various pollutants. Pollutant exposure features and corresponding health outcome feedback data are used to form exposure-response sample pairs, which are then input into a pre-defined health effect learning model. The health effect learning model is a one-dimensional convolutional neural network model. This model takes the pollutant exposure feature vector as input and the probability of health event occurrence as output. It extracts the time series patterns in the exposure features through convolutional layers, outputs the probability values ​​through fully connected layers, and uses an incremental learning method to update the model parameters after each acquisition of a new exposure-response sample pair. Based on the updated health effect learning model, the probability thresholds for health events under different pollutant exposure levels are recalculated, and new individualized exposure response boundary parameters are generated. These individualized exposure response boundary parameters include the safety thresholds for various pollutants under different exposure durations and the corresponding early warning triggering conditions.

[0014] In a preferred embodiment, the steps of adjusting the sampling strategy and exposure determination criteria of the transient exposure sensing module specifically include: The safety thresholds for various pollutants under different exposure durations in the new individualized exposure response boundary parameters are transmitted to the transient exposure sensing module. These thresholds serve as the new benchmark thresholds for the module to make real-time judgments on pollutant concentration data. The new benchmark thresholds replace the original thresholds to determine whether the current exposure has reached a dangerous level. The warning triggering conditions in the new individualized exposure response boundary parameters are transmitted to the transient exposure sensing module as a new triggering rule for the module to generate warning signals. When the monitored pollutant exposure characteristics meet the warning triggering conditions, the module adds a warning label while generating transient exposure data in the respiratory zone. Based on the changes in the safety thresholds of various pollutants in the new individualized exposure response boundary parameters, the sampling frequency of the transient exposure sensing module for various pollutant sensors is adjusted. Specifically, the adjustment method is as follows: with the preset basic sampling frequency as a benchmark, when the safety threshold of a certain type of pollutant is updated, the safety threshold stored at the time of the last update is read as the safety threshold before the update. The safety threshold before the update is divided by the safety threshold after the update to obtain the safety threshold change ratio. The basic sampling frequency is multiplied by the safety threshold change ratio to obtain the new sampling frequency. An upper limit and a lower limit are set for the new sampling frequency so that the new sampling frequency is not higher than the preset maximum sampling frequency and not lower than the preset minimum sampling frequency. The adjusted sampling frequency is then applied to the corresponding pollutant sensor. The transient exposure sensing module collects multi-pollutant concentration data of the respiratory zone of sensitive populations at an adjusted sampling frequency based on the updated sampling strategy and exposure judgment criteria. It then uses the adjusted new benchmark thresholds and early warning triggering conditions to make real-time judgments on the collected data and generate updated transient exposure data for the respiratory zone.

[0015] The technical effects and advantages of this invention are as follows: This invention uses a transient exposure sensing module to collect concentration data of multiple pollutants in the respiratory zone of sensitive populations at a preset sampling frequency, while simultaneously collecting data on human activity status. This generates transient exposure data for the respiratory zone, shifting the monitoring focus from the average concentration in the indoor environment to the actual respiratory area of ​​sensitive populations. Simultaneously, a time-series alignment and fusion module dynamically aligns the concentration data of multiple pollutants with varying sensor response times within the transient exposure data for the respiratory zone, generating a time-aligned multidimensional pollution exposure vector. This solves the problems of missed reporting of instantaneous high exposure events and time misalignment of complex pollution caused by fixed sampling periods and differences in sensor response times in existing technologies. The system can accurately capture the transient exposure dose in the respiratory zone of sensitive populations during activity, significantly improving the temporal accuracy of exposure assessment and the reliability of time synchronization of multi-pollutant data. This provides a high-fidelity exposure data foundation for research on the relationship between indoor health hazards and the health effects on sensitive populations.

[0016] This invention maps time-aligned multidimensional pollution exposure vectors into a unified health risk utility value that characterizes the health risks of sensitive populations through a risk-driven control module. This health risk utility value integrates the instantaneous concentrations of multiple pollutants, individual characteristic parameters of sensitive populations, and real-time ventilation information, transforming complex and variable compound pollution exposures into a single quantitative risk indicator. It generates a sequence of control commands with the goal of minimizing the health risk utility value, overcoming the limitations of existing technologies that aim to achieve single pollutant concentration standards, suffer from multi-device control logic conflicts, and cannot balance the health hazards of different pollutants. This invention achieves multi-pollutant collaborative dynamic optimization control oriented towards the individual health risks of sensitive populations, enabling the system to autonomously decide on the optimal combination of equipment control parameters under dynamically changing indoor and outdoor environmental conditions, effectively reducing the cumulative health risk exposure level of sensitive populations.

[0017] This invention collects health outcome feedback data from sensitive populations after the execution of a control command sequence through a health feedback calibration module. This data includes the onset and severity of respiratory symptoms actively recorded by the sensitive population, the timing and dosage of medication use, and physiological parameter changes synchronously collected by wearable devices. Based on this health outcome feedback data, new individualized exposure response boundary parameters are generated and input into a transient exposure sensing module to adjust its sampling strategy and exposure determination criteria. This constructs a complete closed-loop optimization mechanism from environmental monitoring to health outcomes and back to monitoring strategies. The system can gradually learn individual exposure-response patterns as health events accumulate in the sensitive population, automatically calibrating sampling frequency, safety thresholds, and early warning triggering conditions, thus achieving adaptive evolution of health protection strategies for sensitive populations. Attached Figure Description

[0018] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a schematic diagram of a real-time indoor air quality monitoring system based on multi-sensor fusion, as described in this invention. Detailed Implementation

[0019] 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 of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0020] Reference Figure 1 The following examples were obtained: Example 1: A real-time indoor air quality monitoring system based on multi-sensor fusion, comprising: The transient exposure sensing module collects concentration data of various pollutants in the respiratory zone of sensitive populations at a preset sampling frequency, and simultaneously collects data on human activity status to generate transient exposure data in the respiratory zone.

[0021] Multiple pollutant sensors and human activity status sensors, including accelerometers and heart rate sensors, are deployed in the respiratory zone of sensitive populations. The module synchronously collects multi-pollutant concentration data and human activity status data in the respiratory zone at a preset sampling frequency. An activity intensity index is calculated based on the accelerometer data from the human activity status data. This index is then logarithmically transformed and fused with heart rate data. A preset ventilation volume mapping model is used to estimate the real-time ventilation volume of the sensitive population. This ventilation volume mapping model is a pre-established linear regression model based on the sensitive population sample, using the logarithmically transformed activity intensity index and heart rate as input variables, and ventilation volume as the output variable. The module aligns and matches the multi-pollutant concentration data with the real-time ventilation volume over time, calculating the inhalation volume per unit time for each pollutant. The inhalation volume per unit time is obtained by multiplying the instantaneous pollutant concentration value by the real-time ventilation volume. The module encapsulates the data containing the instantaneous concentration values ​​of multiple pollutants, the real-time ventilation volume values, and the inhalation volume per unit time for each pollutant to generate transient exposure data for the respiratory zone. The core significance of this module lies in shifting the monitoring focus from the average concentration in indoor environments to the respiratory zone of sensitive populations, and expanding exposure assessment from static concentration to instantaneous inhaled dose that includes individual ventilation, thereby providing accurate input data for subsequent health risk assessment.

[0022] The temporal alignment and fusion module dynamically aligns the concentration data of various pollutants with different sensor response times in the transient exposure data of the respiratory zone, generating a time-aligned multidimensional pollution exposure vector.

[0023] The module acquires transient exposure data in the respiratory zone, extracts multi-type pollutant concentration data and their corresponding timestamps, and identifies the sensor type and response time characteristics corresponding to each pollutant concentration data. The pollutant concentration data with the shortest response time is selected as a reference timeline, used to mark the actual occurrence time of the pollution event. For the remaining pollutant concentration data, the module performs feature point matching with the reference timeline, identifying the rising edge start point, peak point, and falling edge inflection point in the pollutant concentration curve. The time offset of the pollutant concentration data relative to the reference timeline is calculated based on the matched feature points. The pollutant concentration data is then resampled based on this time offset, and the time points of the pollutant concentration data are mapped to the time grid of the reference timeline through interpolation, achieving point-to-point alignment between the pollutant concentration data and the reference timeline in the time dimension. Finally, the module merges the time-aligned multi-type pollutant concentration data with the instantaneous pollutant concentration values, real-time ventilation values, and inhalation volume per unit time from the transient exposure data in the respiratory zone to generate a time-aligned multi-dimensional pollution exposure vector. The core significance of this module lies in solving the problem of time misalignment in complex pollution events caused by differences in the response time of different sensors, ensuring that data on multiple types of pollutants are accurately synchronized on a unified time axis, and providing a time-aligned data foundation for the analysis of the health effects of complex exposures.

[0024] The risk-driven control module maps time-aligned multidimensional pollution exposure vectors into health risk utility values ​​that uniformly characterize the health risks of sensitive populations, and generates a sequence of control instructions with the goal of minimizing the health risk utility values.

[0025] The module acquires a time-aligned multidimensional pollution exposure vector, which includes instantaneous concentrations of multiple pollutants on a unified time axis and synchronized real-time ventilation values. It also acquires individual characteristic parameters of the sensitive population, including age, disease type, and allergen spectrum. When the sensitive population includes multiple individuals, the module acquires the individual characteristic parameters for each individual separately. Based on each individual's characteristic parameters, a risk weight function is constructed for each pollutant class, representing the contribution of each pollutant to the individual's health risk per unit concentration exposure. For each individual, the module multiplies the instantaneous concentration value of each pollutant class corresponding to that individual with the corresponding risk weight function to obtain the instantaneous health risk contribution value of that pollutant to the individual. The module then performs a weighted sum of the instantaneous health risk contribution values ​​of all pollutants, and multiplies the result by the individual's real-time ventilation value to obtain the individual's instantaneous health risk utility value at the current moment. For each individual, the module accumulates the instantaneous health risk utility values ​​within a preset time window to obtain the individual's cumulative health risk utility value. Finally, with the objective of minimizing the sum of cumulative health risk utility values ​​for all individuals, and combining current operating status of multiple devices with outdoor air quality data, an optimization algorithm is used to solve for the optimal combination of device control parameters, generating a sequence of control commands. The core significance of this module lies in transforming multi-pollutant combined exposure into a unified quantitative indicator of health risk, and taking the minimization of population health risk as the control objective, thus realizing a paradigm shift from "environmental concentration compliance" to "health risk driven" decision-making.

[0026] After the control command sequence is executed, the health feedback calibration module collects health outcome feedback data from sensitive populations, generates new individualized exposure response boundary parameters based on the health outcome feedback data, and inputs the new individualized exposure response boundary parameters into the transient exposure perception module to adjust the sampling strategy and exposure judgment criteria of the transient exposure perception module.

[0027] The module acquires environmental data following the execution of control command sequences generated by the risk-driven control module. This environmental data includes multi-dimensional pollution exposure vectors during and after the execution of the control commands. It also collects health outcome feedback data from sensitive populations, including the time and severity of respiratory symptoms actively recorded by the sensitive populations, the time and dosage of medication use, and physiological parameter changes synchronously collected by wearable devices. The module correlates and matches the health outcome feedback data with the environmental data over time to identify pollutant exposure characteristics within a preset time window before the health event. These characteristics include the average concentration, peak concentration, exposure duration, and concentration rise rate of various pollutants. The module forms exposure-response sample pairs by combining the pollutant exposure characteristics with the corresponding health outcome feedback data and inputs them into a preset health effect learning model. This model is a one-dimensional convolutional neural network model that takes the pollutant exposure feature vector as input and the probability of health event occurrence as output. It extracts time-series patterns from the exposure features through convolutional layers, outputs probability values ​​through fully connected layers, and updates the model parameters using incremental learning after acquiring new exposure-response sample pairs. Based on the updated health effect learning model, the module recalculates the probability thresholds of health events at different pollutant exposure levels, generating new individualized exposure response boundary parameters. These parameters include safety thresholds for various pollutants at different exposure durations and corresponding warning triggering conditions. Finally, these parameters are input into the transient exposure perception module to update its sampling strategy and exposure determination criteria. The core significance of this module lies in using the real health outcomes of sensitive populations as feedback signals to reverse-calibrate the system's exposure determination standards, enabling the system to adaptively adjust protection strategies as individual health conditions change, thus forming a data-driven closed-loop optimization capability.

[0028] The transient exposure sensing module generates transient exposure data of the respiratory zone through the following steps: In the homes of sensitive populations, the primary areas of daily activity for these individuals are first identified, such as the area around the bed in a child with asthma's bedroom, the play area in the living room, or the area next to their study desk. These areas are marked as breathing zones. Based on the location of these breathing zones, various types of pollutant sensors are deployed within approximately 30 to 50 centimeters of the sensitive individual's mouth and nose. These include laser dust sensors for detecting particulate matter, electrochemical sensors for detecting formaldehyde, and infrared sensors for detecting carbon dioxide. Simultaneously, accelerometers and heart rate sensors are deployed close to the body. The accelerometer is worn on the wrist or waist, while the heart rate sensor is integrated into a smart bracelet or chest strap. By deploying sensors in actual breathing zones rather than at fixed indoor locations, the collected pollutant concentration data accurately reflects the actual air composition inhaled by the sensitive individual, avoiding monitoring blind spots caused by spatial differences between environmental sensors and the human body. For example, in a child with asthma's bedroom, if sensors are installed on the wall in a corner of the room, the pollutant concentration at the corner when the child is active in bed may differ by several times from the actual exposure concentration at the mouth and nose. The breathing zone deployment method effectively eliminates this difference.

[0029] The system synchronously collects data from all sensors deployed in the respiratory zone at a preset sampling frequency, which is set to once per second or higher depending on the monitoring requirements. Synchronous acquisition means that at each sampling point, the system simultaneously records particulate matter concentration, formaldehyde concentration, carbon dioxide concentration, acceleration data, and heart rate data, ensuring strict temporal correspondence between different types of data. This synchronous acquisition mechanism avoids distortion in exposure assessment caused by misaligned sampling times. For example, when a child is jumping and playing indoors, the particulate matter concentration may surge from 10 micrograms per cubic meter to 150 micrograms per cubic meter within two seconds, while the heart rate increases from 80 beats per minute to 120 beats per minute. If the particulate matter sensor and the heart rate sensor collect data at different times (by several seconds), the system cannot accurately capture the real-world exposure scenario where high concentration exposure and high ventilation occur simultaneously. Synchronous acquisition, however, ensures a precise match between pollutant concentration and physiological state.

[0030] The system acquires triaxial acceleration data from an accelerometer, calculates the composite acceleration value, and filters it to remove the static gravity component, obtaining an activity intensity index reflecting the intensity of physical activity. Since the relationship between human activity intensity and ventilation volume is non-linear—ventilation volume increases slowly during low-intensity activity and rapidly during high-intensity activity—the activity intensity index is logarithmically transformed to approximate a linear relationship with ventilation volume. The system uses the logarithmically transformed activity intensity index and heart rate data as two input variables, substituting them into a pre-defined ventilation volume mapping model. This model is a pre-established linear regression model based on a large sample of sensitive individuals, developed in a laboratory environment through cardiopulmonary exercise testing. The model parameters reflect the ventilation volume patterns under different combinations of activity intensity and heart rate. The model outputs real-time ventilation volume in liters per minute. For example, when a six-year-old child with asthma runs around in the living room, the accelerometer collects an activity intensity index of 5 meters per second squared, which, after logarithmic transformation, becomes 1.6. Simultaneously, the heart rate is monitored at 140 beats per minute. Based on these two input values, the ventilation volume mapping model calculates the child's current real-time ventilation volume as 25 liters per minute. This estimate is significantly higher than the child's resting ventilation volume of 6 liters per minute, accurately reflecting the actual amount of air inhaled during exercise.

[0031] The system aligns and matches the instantaneous concentrations of multiple pollutants collected synchronously with the real-time ventilation values ​​estimated in the third step at the same time point, ensuring that the pollutant concentration and ventilation volume at the same moment correspond to the same respiratory behavior. For each pollutant, its instantaneous concentration value is multiplied by the real-time ventilation volume to obtain the inhalation volume of that pollutant per unit time at the current moment, expressed in micrograms per minute or milligrams per minute. Inhalation volume per unit time reflects the actual mass of pollutants inhaled per minute by a sensitive population during their current activity state, and is a more physiologically significant exposure dose indicator than a simple concentration value. For example, in the above scenario of children running, the instantaneous particulate matter concentration is 120 micrograms per cubic meter, and the real-time ventilation volume is 25 liters per minute. After unit conversion, the particulate matter inhalation volume per unit time is calculated as 3 micrograms per minute. If only the environmental concentration is monitored, the system would record the value of 120 micrograms per cubic meter, but it cannot reflect the actual increase in exposure dose caused by the child's ventilation volume increasing fourfold due to vigorous activity. Inhalation volume per unit time, however, fully presents this comprehensive exposure risk. The system encapsulates the instantaneous concentration values ​​of various pollutants, real-time ventilation values, and inhalation volume per unit time to generate transient exposure data of the respiratory zone, which serves as the input basis for subsequent modules to perform time alignment and health risk analysis.

[0032] Taking a six-year-old asthmatic child playing in the living room as an example: The system first deploys sensors in the living room area where the child frequently plays. An accelerometer is worn on the wrist, and a heart rate sensor is integrated into a smart bracelet. When the child starts running, the system simultaneously collects particulate matter concentration, formaldehyde concentration, acceleration data, and heart rate data once per second. The acceleration data shows an activity intensity index of 5 meters per second squared, which, after logarithmic transformation, is 1.6. The heart rate is 140 beats per minute, and the ventilation volume mapping model calculates a real-time ventilation volume of 25 liters per minute. Simultaneously, the particulate matter sensor detects an instantaneous concentration of 120 micrograms per cubic meter due to the re-suspension of dust from the floor caused by the child's running. The system multiplies the instantaneous particulate matter concentration by the real-time ventilation volume to calculate the particulate matter inhalation rate per unit time as 3 micrograms per minute. Finally, the system encapsulates the particulate matter concentration, formaldehyde concentration, real-time ventilation volume, and inhalation rates per unit time into transient exposure data for the respiratory zone, providing accurate exposure dose information for subsequent health risk assessment.

[0033] Transient exposure data in the respiratory zone is obtained by encapsulating data including instantaneous concentration values ​​of multiple pollutants, real-time ventilation values, and the inhalation volume of each pollutant per unit time.

[0034] The time-aligned fusion module generates a time-aligned multidimensional contamination exposure vector through the following sub-steps: The system acquires transient exposure data from the respiratory zone via the transient exposure sensing module. This data encapsulates the instantaneous concentration values, real-time ventilation values, and inhalation volume per unit time for multiple pollutants. The system then parses the concentration data for each pollutant and its accompanying precise timestamp, recording the acquisition time of each data point. Simultaneously, the system identifies the sensor type corresponding to each pollutant concentration data based on its data source, such as particulate matter sensors, electrochemical formaldehyde sensors, and infrared carbon dioxide sensors. It then queries a pre-defined sensor parameter library to obtain the response time characteristics of each sensor type. Response time characteristics refer to the time required for a sensor to detect a change in pollutant concentration and output a stable signal; different sensor types exhibit significant differences in response time. For example, laser particulate matter sensors typically have a response time of one to two seconds, allowing for rapid response after a pollution event; while electrochemical formaldehyde sensors typically have a response time of two to three minutes, showing a significant delay in responding to pollution events. By identifying sensor types and response time characteristics, the system establishes foundational information for subsequent time alignment operations.

[0035] Taking a family setting with a child suffering from asthma as an example, the transient exposure data of the respiratory zone acquired by the system includes particulate matter concentration data, formaldehyde concentration data, and carbon dioxide concentration data. The system identified the particulate matter data as coming from a laser dust sensor with a response time of one second; the formaldehyde data as coming from an electrochemical sensor with a response time of two minutes; and the carbon dioxide data as coming from an infrared sensor with a response time of thirty seconds. This identification result provides a basis for subsequently determining the reference time axis and calculating the time offset.

[0036] After identifying the response time characteristics of all pollutant concentration data, the system selects the pollutant concentration data with the shortest response time as the reference timeline. The sensor with the shortest response time can most promptly reflect the occurrence of the pollution event, and its timeline is considered the benchmark closest to the actual pollution event timeline. The system uses the timestamp sequence of this type of pollutant concentration data as the reference timeline, and other pollutant concentration data are time-aligned based on this axis. The selection of the reference timeline ensures that the reference for time alignment has the highest time accuracy, avoiding overall time misalignment caused by using sensors with long response times as the benchmark.

[0037] In the aforementioned home scenario involving a child with asthma, the particulate matter sensor has the shortest response time of one second, the shortest among the three types of sensors. The system selects the timestamp sequence of particulate matter concentration data as the reference time axis. Assuming that at the instant a child runs and causes ground dust to be resuspended, the particulate matter concentration rises from 10 micrograms per cubic meter to 120 micrograms per cubic meter within one second, this rising point is precisely recorded on the reference time axis. However, due to its two-minute response time, the formaldehyde sensor's data will lag behind the actual pollution event by approximately two minutes in terms of the starting point of the rising point. The system will correct for this lag in subsequent steps.

[0038] The system performs feature point matching on the concentration curves of each pollutant category (excluding those corresponding to the reference time axis) against the concentration curves on the reference time axis. Feature points include the starting point of the rising edge, the peak point, and the inflection point of the falling edge in the concentration curves. The starting point of the rising edge indicates the moment when the pollutant concentration begins to rise significantly; the peak point indicates the moment when the concentration reaches its highest value; and the inflection point of the falling edge indicates the turning point where the concentration begins to decline from the peak. The system automatically identifies these feature points by analyzing characteristics such as the rate of change and slope changes of the concentration curves. The purpose of feature point matching is to find the corresponding key event moments in the concentration curves of the two pollutant categories, providing reliable matching anchor points for subsequent calculations of time offsets.

[0039] In a home setting with a child suffering from asthma, the system performs feature point matching between the formaldehyde concentration curve and a particulate matter reference time axis. The particulate matter reference time axis shows the rising edge starting at 0:00 seconds and reaching its peak at 20 seconds. Due to response delay, the formaldehyde concentration curve's rising edge starts at 120 seconds and its peak occurs at 140 seconds. The system identifies that the rising edge starting points of the formaldehyde curve and the particulate matter curve correspond to the same pollution event, and that the peak points of the formaldehyde curve and the particulate matter curve also correspond to the same pollution event, thus establishing a matching relationship between the feature points.

[0040] Based on the feature points matched in the previous step, the system calculates the time offset of the pollutant concentration data to be aligned relative to the reference time axis. In the scenario of a family with asthmatic children, the system selects the rising edge start point for offset calculation. The rising edge start point of particulate matter on the reference time axis occurs at 0:00, while the rising edge start point of the formaldehyde curve occurs at 120:00, with a difference of 120 seconds. The system also selects peak points for verification; the peak point of particulate matter is at 20:00, and the peak point of formaldehyde is at 140:00, with a difference of 120 seconds as well. The system confirms that the time offset of the formaldehyde concentration data relative to the reference time axis is 120 seconds, meaning that the formaldehyde sensor has a two-minute response delay.

[0041] The system resamples the pollutant concentration data to be aligned along the time axis based on the calculated time offset. Resampling essentially shifts the original time points of the pollutant concentration data forward by the offset, aligning them with the reference time axis in the time dimension. Since the shifted time points may not perfectly coincide with the time grid on the reference time axis, the system uses interpolation to calculate the corresponding concentration values ​​at the shifted time points, generating a concentration data sequence perfectly aligned with the reference time axis time grid. Linear interpolation or spline interpolation can be used to ensure the continuity and smoothness of the concentration curve. After resampling, each time point of the pollutant concentration data forms a point-to-point alignment with the corresponding time point on the reference time axis, achieving temporal synchronization between the two types of data.

[0042] In a home setting with a child suffering from asthma, the system resamples formaldehyde concentration data along a timeline. The original formaldehyde concentration data's timestamp corresponds to a concentration of 0.05 mg / m³ at 120 seconds and 0.12 mg / m³ at 140 seconds. The system shifts forward by 120 seconds, mapping 120 seconds to 0.00 seconds and 140 seconds to 20 seconds. Since the reference timeline's time grid is one point per second, the shifted time points are 0 seconds, 20 seconds, etc. The system performs linear interpolation on time points between 0 and 20 seconds to generate the corresponding formaldehyde concentration value per second. Ultimately, at 0.00 seconds, the interpolated formaldehyde concentration value is 0.05 mg / m³, at 1 second it is 0.053 mg / m³, and so on. After resampling, the formaldehyde concentration data and particulate matter concentration data are synchronized at every second.

[0043] The system merges the time-aligned concentration data of various pollutants with the instantaneous concentrations, real-time ventilation rates, and inhalation rates per unit time originally included in the transient exposure data of the respiratory zone. Since the time-aligned pollutant concentration data has replaced the corresponding parts in the original data, the merging operation essentially integrates the corrected concentration data with the synchronously collected ventilation rate and inhalation rate data to form a new data structure. This data structure contains the aligned concentration values ​​of various pollutants, the corresponding real-time ventilation rates, and the inhalation rates per unit time on a unified time axis, and is referred to as the time-aligned multidimensional pollution exposure vector. This vector serves as the input to the subsequent risk-driven control module, providing time-synchronized multidimensional exposure information for health risk assessment.

[0044] In a home setting with a child suffering from asthma, the system merges time-aligned formaldehyde concentration data with particulate matter concentration data, real-time ventilation volume, and inhalation volume per unit time. At 0:00, the merged data includes a particulate matter concentration of 10 micrograms per cubic meter, a formaldehyde concentration of 0.05 milligrams per cubic meter, a real-time ventilation volume of 6 liters per minute, a particulate matter inhalation volume of 0.06 micrograms per minute, and a formaldehyde inhalation volume of 0.3 micrograms per minute. At 20:00, the merged data includes a particulate matter concentration of 120 micrograms per cubic meter, a formaldehyde concentration of 0.12 milligrams per cubic meter, a real-time ventilation volume of 25 liters per minute, a particulate matter inhalation volume of 3 micrograms per minute, and a formaldehyde inhalation volume of 3 micrograms per minute. The system encapsulates this series of synchronized data on a unified timeline into a time-aligned multidimensional pollution exposure vector, which is then passed to the risk-driven control module for health risk assessment and control decisions.

[0045] Taking the example of dust suspended on the ground caused by a child with asthma running, the transient exposure data of the respiratory zone output by the transient exposure sensing module showed a two-minute time misalignment between particulate matter concentration data and formaldehyde concentration data. The time-series alignment and fusion module first identified the particulate matter sensor with the shortest response time and used its time axis as the reference time axis. Then, it extracted the rising edge start point and peak point of the formaldehyde concentration curve and matched them with the corresponding feature points in the particulate matter curve to calculate the 120-second time offset of the formaldehyde data relative to the particulate matter data. Based on this offset, the formaldehyde data was resampled on the time axis, and a formaldehyde concentration sequence aligned with the particulate matter time grid was generated through linear interpolation. Finally, the aligned particulate matter concentration, formaldehyde concentration, and synchronized real-time ventilation volume and inhalation volume per unit time were merged to generate a time-aligned multidimensional pollution exposure vector. In this vector, the particulate matter peak and formaldehyde peak are precisely aligned on the time axis, truly reflecting the combined pollution scenario of children being exposed to high concentrations of particulate matter and high concentrations of formaldehyde simultaneously during running, providing accurate time-synchronized data for subsequent health risk assessment.

[0046] The step of calculating the time offset of the pollutant concentration data relative to the reference time axis is implemented using a sliding window cross-correlation algorithm, specifically including: The system pre-sets a sliding window length based on the response time characteristics and sampling frequency of the pollutant concentration data to be aligned. The selection of the sliding window length needs to comprehensively consider the differences in sensor response times, the duration of the pollution event, and computational efficiency; it is typically set to a time interval that can cover the complete evolution of a typical pollution event, such as 60 seconds or 120 seconds. After setting the sliding window length, the system divides the pollutant concentration sequence on the reference time axis and the pollutant concentration sequence to be aligned into multiple subsequences according to this window length. The division method uses a sliding window approach, meaning that there is an overlapping area between adjacent subsequences, with the overlap length typically being half the window length, to ensure continuous coverage and smooth transition on the time axis. Each subsequence represents a pollutant concentration change curve within a time interval, serving as the unit for subsequent cross-correlation calculations.

[0047] Taking a family setting with a child suffering from asthma as an example, the system needs to align the formaldehyde concentration sequence with the particulate matter reference time axis. The particulate matter sensor has a response time of one second, while the formaldehyde sensor has a response time of two minutes. The system sets the sliding window length to sixty seconds. The particulate matter concentration sequence is continuously recorded at a density of one data point per second. The system divides this sequence into multiple subsequences of sixty seconds each. The first subsequence covers the time from 0.05 to 59 seconds, the second subsequence covers the time from 30 to 89 seconds, and so on, with adjacent subsequences overlapping by thirty seconds. The formaldehyde concentration sequence is also divided according to the sixty-second window length, forming a corresponding set of subsequences. Through this division, the system converts the continuous long sequence into multiple time segments that are easier to calculate.

[0048] The system pairs each subsequence on the reference time axis with each subsequence in the pollutant concentration sequence to be aligned, calculating the cross-correlation coefficient between each pair. The cross-correlation coefficient is calculated by sliding the two subsequences relative to each other on the time axis, calculating the similarity between the two sequences at each relative displacement, resulting in a set of cross-correlation function values. The system extracts the maximum value from this set, which represents the optimal match between the two subsequences at a certain relative displacement. The sliding window position corresponding to this maximum value reflects the relative time offset between the two subsequences. The system iterates through all subsequence pairs, recording the maximum cross-correlation value and the corresponding sliding window position for each pair.

[0049] In a family setting with a child suffering from asthma, the system performs cross-correlation calculations between the first subsequence on the particulate reference timeline and every subsequence in the formaldehyde sequence. When pairing the particulate subsequence with the formaldehyde subsequence covering times from 60 to 119 seconds, the cross-correlation calculation shows a maximum value of 0.95 at a relative sliding shift of zero, indicating that the two subsequences are highly similar under this pairing and can be matched without additional sliding. When pairing the particulate subsequence with the formaldehyde subsequence covering times from 0.1 seconds to 59 seconds, the maximum cross-correlation value appears at a relative sliding shift of -60 seconds, with a maximum value of only 0.3, indicating a large temporal shift and low similarity between the two. The system records the calculation results for all subsequence pairs and determines the sliding window position corresponding to the maximum cross-correlation value of 0.95 as the pairing of the formaldehyde subsequence covering times from 60 to 119 seconds with the first particulate subsequence.

[0050] The system calculates the time offset of the pollutant concentration sequence to be aligned relative to the reference time axis based on the sliding window position corresponding to the maximum cross-correlation coefficient determined in the previous step. Specifically, the system compares the start time of the pollutant subsequence to be aligned on the time axis with the start time of the subsequence on the reference time axis; the difference between the two is the time offset. Because overlapping areas are included in the sliding window division, the system can smooth or average the calculation results of multiple adjacent subsequences, improving the robustness and accuracy of the offset estimation. The calculated time offset represents the overall time lag of the pollutant concentration data to be aligned due to sensor response delay.

[0051] In the context of families with children suffering from asthma, the pairing with the highest cross-correlation coefficient corresponds to the first subsequence of particulate matter and the subsequence of formaldehyde covering time intervals from 60 to 119 seconds. The start time of the first subsequence of particulate matter is 0 seconds, and the start time of the paired subsequence of formaldehyde is 60 seconds, with a difference of 60 seconds. The system further verified the pairing results of other subsequences, such as the second subsequence of particulate matter and the subsequence of formaldehyde covering time intervals from 90 to 149 seconds, which also achieved a high cross-correlation coefficient, with the start time difference also being 60 seconds. The system combined multiple pairing results and confirmed that the time offset of the formaldehyde concentration sequence relative to the particulate matter reference time axis is 60 seconds. This offset reflects the first 60 seconds of the two-minute response delay of the formaldehyde sensor. The system uses this offset for subsequent time axis resampling operations to achieve accurate alignment between formaldehyde concentration data and particulate matter data.

[0052] Taking the alignment of particulate matter and formaldehyde concentration data in families with children with asthma as an example, the system sets a sliding window length of 60 seconds, dividing the particulate matter reference timeline sequence into multiple overlapping subsequences, and also dividing the formaldehyde concentration sequence into corresponding overlapping subsequences. The system calculates the cross-correlation coefficient between each pair of subsequences and finds that the cross-correlation coefficient between the first particulate matter subsequence and the formaldehyde subsequence covering times from 60 seconds to 119 seconds reaches 0.95, the maximum value among all pairs. Based on this pairing, the starting time of the particulate matter subsequence is 0 seconds, and the starting time of the formaldehyde paired subsequence is 60 seconds. The system determines that the time offset of the formaldehyde concentration sequence relative to the particulate matter reference timeline is 60 seconds. Through the sliding window cross-correlation algorithm, the system can automatically and accurately identify the optimal matching position between two sequences under conditions of noise and local waveform changes, avoiding the subjective errors that may be introduced by relying on manual identification of feature points, and providing an accurate offset basis for subsequent timeline resampling.

[0053] The risk-driven control module maps time-aligned multidimensional pollution exposure vectors to health risk utility values ​​in the following way: The system obtains a time-aligned multidimensional pollution exposure vector from the time-aligned fusion module. This vector is a data structure that contains instantaneous concentration values ​​of multiple pollutants and synchronized real-time ventilation values ​​on a unified time axis. The instantaneous concentration values ​​of each pollutant have been time-aligned to eliminate time misalignment caused by differences in the response times of different sensors, ensuring that all pollutant concentration values ​​recorded at the same time point correspond to the same actual pollution event. The real-time ventilation values ​​and pollutant concentration values ​​are strictly synchronized in time, providing a physiological parameter basis for subsequent calculation of instantaneous inhaled dose.

[0054] Taking a family setting with a child suffering from asthma as an example, the time-aligned multidimensional pollution exposure vector obtained by the system includes a particulate matter concentration of 10 micrograms per cubic meter, a formaldehyde concentration of 0.05 milligrams per cubic meter, and a real-time ventilation rate of 6 liters per minute at time 0:00. At time 20:00, it includes a particulate matter concentration of 120 micrograms per cubic meter, a formaldehyde concentration of 0.12 milligrams per cubic meter, and a real-time ventilation rate of 25 liters per minute. This vector provides time-synchronized multidimensional exposure information for subsequent health risk assessment.

[0055] The system retrieves individual characteristic parameters of sensitive individuals from the user information database. These parameters include age, disease type, and allergen spectrum. Age is recorded in years and is used to assess the baseline sensitivity to pollutants at different age levels. Disease types include asthma, chronic obstructive pulmonary disease, allergic rhinitis, and cardiovascular disease, used to identify specific triggers for each individual. The allergen spectrum records the types of allergens identified through medical testing, such as dust mites, pollen, mold, and pet dander, demonstrating the correlation between these allergens and indoor pollutants. When the sensitive population includes multiple individuals, the system retrieves the individual characteristic parameters for each individual separately, ensuring that subsequent health risk assessments are personalized for each person.

[0056] Taking a family setting with a child suffering from asthma as an example, the sensitive population includes a six-year-old child and a thirty-five-year-old adult. The child's individual characteristics are: age six, disease type asthma, allergy to house dust mites, and a positive allergen spectrum for house dust mites. The adult's individual characteristics are: age thirty-five, no history of asthma, and a negative allergen spectrum for house dust mites. The system obtains the characteristic parameters of these two individuals separately, providing a basis for subsequent calculations of health risks.

[0057] The system constructs a risk weighting function for each pollutant class for each individual based on their unique characteristics. The risk weighting function is constructed based on the individual's disease type, age, and allergen spectrum, quantifying the contribution of different pollutants to health risk as a function of pollutant concentration. A higher risk weighting function value indicates a greater contribution of that pollutant to the individual's health risk per unit concentration of exposure. For pollutants that are not triggering factors, the risk weighting function value is zero; the specific implementation of this step will be described separately later.

[0058] For each individual, the system inputs the instantaneous concentration value of each type of pollutant into the risk weight function of that pollutant to obtain the risk weight value of the pollutant at the current concentration. Then, the instantaneous concentration value is multiplied by the risk weight value to obtain the instantaneous health risk contribution of the pollutant to the individual. This product reflects the immediate contribution of the pollutant to the individual's health risk at the current exposure level.

[0059] The system performs a weighted summation of the instantaneous health risk contributions of all pollutants. The weights in this summation are set based on existing epidemiological evidence and toxicological studies. Different pollutants have varying degrees of harm to specific health effects, and the weights reflect this difference. For example, for asthma patients, both particulate matter and formaldehyde are known triggering factors, but epidemiological studies show that particulate matter typically has a higher contribution weight to acute asthma attacks than formaldehyde. Therefore, the weight coefficient for particulate matter is set to 0.7, and the weight coefficient for formaldehyde is set to 0.3. The weight settings differ for different disease types. For example, particulate matter may have a higher weight for patients with chronic obstructive pulmonary disease, while volatile organic compounds may have a higher weight for patients with allergic rhinitis. The weight coefficients can be pre-set using clinical literature data or expert knowledge.

[0060] Multiplying the weighted summation result by the individual's real-time ventilation rate yields the instantaneous health risk utility value for that individual at the current moment. This multiplication operation is designed based on physiological principles: the health impact of pollutants depends not only on their concentration in the environment but also on the total amount of pollutants actually inhaled by the individual. Real-time ventilation rate reflects an individual's respiratory rate during their current activity state; the higher the ventilation rate, the larger the volume of air inhaled per unit time, and the greater the total amount of pollutants inhaled. At the same concentration, the pollutant dose inhaled by an individual during exercise is much higher than that at rest. Therefore, multiplying the weighted summation of pollutant risk contribution by real-time ventilation rate can transform the static environmental concentration risk into a dynamic individual inhaled dose risk, more accurately reflecting the true exposure level of sensitive populations.

[0061] Taking a family setting with a child suffering from asthma as an example, for a six-year-old child with asthma, the system obtains an instantaneous particulate matter concentration of 120 micrograms per cubic meter. Inputting this concentration into the particulate matter risk weight function yields a risk weight of 0.08. Multiplying the particulate matter concentration by 0.08 gives an instantaneous health risk contribution of 9.6. The instantaneous formaldehyde concentration is 0.12 milligrams per cubic meter. Inputting this concentration into the formaldehyde risk weight function yields a risk weight of 0.03. Multiplying this concentration by 0.036 gives an instantaneous health risk contribution of 0.0036. For weighted summation, the particulate matter weight coefficient is set to 0.7, and the formaldehyde weight coefficient is set to 0.3, resulting in a weighted sum of 6.72. Multiplying this result by the child's real-time ventilation of 25 liters per minute gives an instantaneous health risk utility of 168 for the child at the current moment. For a 35-year-old adult in the same scenario, due to the lack of a history of asthma, the risk weight function values ​​for both particulate matter and formaldehyde are lower, and the real-time ventilation is also lower. The calculated instantaneous health risk utility is significantly lower than that of the child, reflecting the differences in personalized risk assessment.

[0062] The system sets a preset time window for each individual, the length of which is determined according to the needs of the health effect study, such as five minutes, fifteen minutes, or one hour. The system accumulates the instantaneous health risk utility values ​​at each sampling moment within this time window to obtain the cumulative health risk utility value for that individual within that time window. This accumulation reflects the cumulative effect of exposure over time, because health risk depends not only on the intensity of the instantaneous exposure but also on its duration. A higher cumulative health risk utility value indicates a greater total health burden experienced by the individual within that time window.

[0063] Taking a family setting with a child suffering from asthma as an example, the system sets a preset time window of fifteen minutes. Within fifteen minutes of the child running, the system accumulates the instantaneous health risk utility value calculated every second. The instantaneous health risk utility value for the child reaches 168 at the peak of the running, and is lower before and after the running. The accumulated health risk utility value within this fifteen-minute window is 1200. The cumulative health risk utility value for adults during the same period is only 80. This comparison shows that the cumulative health risk of the same pollution event varies significantly among different individuals, and the system implements differentiated control accordingly.

[0064] The system sums the cumulative health risk utility values ​​of all individuals in the sensitive population to obtain the total cumulative health risk utility value. The system uses minimizing this total cumulative health risk utility value as its optimization objective. Combining the current operating status of multiple devices with outdoor air quality data, the system employs an optimization algorithm to find the optimal combination of device control parameters. Under the premise of satisfying the physical constraints of the devices, the optimization algorithm searches for the combination of device control parameters that minimizes the total cumulative health risk utility value and generates a sequence of control commands. The sequence of control commands includes specific execution parameters such as the opening degree of the fresh air unit, the setting of the air purifier, and the on / off status of the exhaust fan.

[0065] Taking a family scenario with a child suffering from asthma as an example, the system calculates the cumulative health risk utility value as 1200 for the child and 80 for the adult, for a total cumulative health risk utility value of 1280. The system obtains the current device operating status as air purifier off and fresh air system off, and outdoor air quality data shows an outdoor particulate matter concentration of 80 micrograms per cubic meter. The optimization algorithm considers two control strategies: turning on the air purifier reduces indoor particulate matter concentration but cannot solve the formaldehyde problem; turning on the fresh air system introduces outdoor air to dilute formaldehyde but introduces outdoor particulate matter. The algorithm predicts the evolution of pollutant concentrations under the two strategies using an indoor air quality dynamic model, calculates the corresponding cumulative health risk utility value for each strategy, and selects the strategy that minimizes the total cumulative health risk utility value. Finally, the system generates a control command sequence, prioritizing the activation of the air purifier at a medium setting, while delaying the activation of the fresh air system until the outdoor particulate matter concentration decreases, thus minimizing the child's health risk.

[0066] Taking families with children suffering from asthma as an example, the system first obtains a time-aligned multidimensional pollution exposure vector, which includes the instantaneous concentrations of particulate matter and formaldehyde, as well as the real-time ventilation volume of children and adults. Then, it obtains the individual characteristic parameters of children and adults, constructing risk weight functions for each. For children, the system multiplies the particulate matter and formaldehyde concentrations by their corresponding risk weight function values ​​to obtain instantaneous health risk contribution values. These are then weighted and summed, multiplied by the child's real-time ventilation volume to obtain the child's instantaneous health risk utility value. This is accumulated over a 15-minute window to obtain a cumulative health risk utility value of 1200. For adults, the same calculation yields a cumulative health risk utility value of 80. The system aims to minimize the sum of these two values ​​(1280), combining the current device status and outdoor air quality. Through an optimization algorithm, it solves for the optimal device control parameters and generates a control command sequence to turn the air purifier to a medium setting, thus minimizing the health risk for sensitive populations.

[0067] The steps of constructing a risk weight function for each type of pollutant for each individual based on their individual characteristic parameters specifically include: The system reads the disease type information of each individual in the sensitive population. Disease types include asthma, chronic obstructive pulmonary disease, allergic rhinitis, cardiovascular disease, etc., and different types of diseases have specific correlations with pollutant sensitivity. Based on a pre-set disease-pollutant association knowledge base, the system determines whether each pollutant is a known trigger for that individual's disease type. If it is a trigger, the system assigns a non-zero weight to the pollutant, indicating that the pollutant actually contributes to the individual's health risk; if it is not a trigger, the system assigns a zero weight, indicating that the pollutant does not contribute to the individual's health risk assessment. This screening mechanism ensures that the subsequent construction of the risk weight function focuses on pollutants that truly pose a threat to the individual, avoiding the introduction of noise from irrelevant pollutants into the health risk assessment.

[0068] Taking a family scenario with a child suffering from asthma as an example, the system identifies the six-year-old child's condition as asthma. Based on medical knowledge, particulate matter, formaldehyde, and nitrogen dioxide are known triggers for asthma; therefore, the system assigns non-zero weights to particulate matter and formaldehyde. While high concentrations of carbon dioxide may cause discomfort, it is not a direct trigger for asthma, so the system assigns it a zero weight. For a 35-year-old adult in the same scenario, whose condition is no history of asthma, the system, based on a pre-defined knowledge base, determines that none of the pollutants are triggers for this individual. Therefore, all pollutants are assigned a zero weight, indicating that this adult does not contribute to the health risks associated with pollutants in this family environment.

[0069] For each type of pollutant identified as a triggering factor in the first step, the system determines a baseline sensitivity coefficient based on the individual's age. The baseline sensitivity coefficient reflects the inherent sensitivity of different age groups to pollutants, exhibiting a peak distribution that first increases and then decreases with age. Children, whose immune and respiratory systems are not yet fully developed, are more sensitive to pollutants; young adults, whose physical functions are at their peak, are relatively less sensitive; and the elderly, whose physical functions decline and immunity decreases, are more sensitive to pollutants again. The system uses a preset age-sensitivity mapping function to input the individual's age and output the corresponding baseline sensitivity coefficient. This coefficient typically ranges from one to three, with higher values ​​for children and the elderly, and lower values ​​for young adults.

[0070] Taking a family setting with a child suffering from asthma as an example, the system targets a six-year-old child. At this age, the child's respiratory and immune systems are not yet fully developed, resulting in a higher inherent sensitivity to pollutants. The system calculates a baseline sensitivity coefficient of 2.5 for six years old using an age-sensitivity mapping function. For a 25-year-old, the baseline sensitivity coefficient is approximately 1.2; for a 65-year-old, it is approximately 2.2. For adults in the same scenario, since all pollutants have already been determined to have zero weight in the first step, there is no need to calculate the baseline sensitivity coefficient.

[0071] The system reads the individual's allergen spectrum data for each type of pollutant identified as a triggering factor in the first step. The allergen spectrum records the types of allergens the individual has been identified by medical testing, such as dust mites, pollen, mold, and pet dander. The system determines whether the individual has a positive reaction to the allergen corresponding to a particular type of pollutant. For example, particulate matter often carries allergens such as dust mites and pollen; while formaldehyde itself is not an allergen, it is associated with certain volatile organic compounds. If the individual has a positive reaction to the allergen corresponding to that type of pollutant, the system sets the specificity sensitivity coefficient to a preset value greater than one, typically between 1.5 and 2.0, indicating that the allergic state amplifies the contribution of that pollutant to the individual's health risk. If the individual has no positive reaction to the corresponding allergen, the specificity sensitivity coefficient is set to one, indicating that the risk is not amplified.

[0072] Taking a family setting with a child suffering from asthma as an example, the system reads the allergen spectrum of a six-year-old child, showing a positive reaction to house dust mites. Particulate matter, as the main carrier of house dust mites, can introduce allergen stimuli upon exposure. Therefore, the system sets a specificity sensitivity coefficient of 1.8 for particulate matter, indicating that the house dust mite allergy amplifies the health risk contribution of particulate matter to the child by a factor of 1.8. Formaldehyde has no corresponding association in the allergen spectrum, so its specificity sensitivity coefficient is set to 1. For adults in the same setting, since all pollutants were determined to have zero weight in the first step, there is no need to calculate specificity sensitivity coefficients.

[0073] The system calculates the risk weight value of each pollutant belonging to the triggering factors at the current concentration using a preset risk weight model. The risk weight model employs an exponential nonlinear function, mathematically expressed as: the risk weight value equals the preset base weight coefficient multiplied by the base sensitivity coefficient multiplied by the specific sensitivity coefficient, and then multiplied by the exponential function value of the pollutant concentration. The exponential function value of the pollutant concentration has the natural constant e as the base and the pollutant concentration multiplied by the preset concentration sensitivity coefficient as the exponent. This design ensures that the risk weight value increases exponentially with increasing pollutant concentration, conforming to the nonlinear dose-response relationship in toxicology where risk increases slowly at low concentrations and sharply at high concentrations. The calculated risk weight value is the function value of the risk weight function at the current pollutant concentration. Subsequent steps will multiply this value by the instantaneous pollutant concentration to obtain the instantaneous health risk contribution value.

[0074] Taking a family setting with a child suffering from asthma as an example, the system calculates risk weight values ​​for particulate matter. The preset base weight coefficient is set to 0.05 based on the inherent hazard of particulate matter. The base sensitivity coefficient is set to 2.5 based on the child's age. The specific sensitivity coefficient is set to 1.8 based on dust mite allergies. The concentration sensitivity coefficient is set to 0.02. When the instantaneous particulate matter concentration is 120 micrograms per cubic meter, the exponential function value of the pollutant concentration is e^2.4, which is approximately 11.02. The risk weight value is calculated as 0.05 multiplied by 2.5 multiplied by 1.8 multiplied by 11.02, resulting in approximately 2.48. If the particulate matter concentration is lower, at 10 micrograms per cubic meter, the exponential function value is e^0.22, and the risk weight value is approximately 0.27. As can be seen from the comparison, when the concentration increases from 10 micrograms per cubic meter to 120 micrograms per cubic meter, the risk weight value increases from 0.27 to 2.48, which is much higher than the increase multiple of the concentration itself, reflecting the sensitive amplification effect of the exponential nonlinear function on high-risk concentrations.

[0075] For formaldehyde, the system calculates a risk weight value, with a preset base weight coefficient set at 0.03. The base sensitivity coefficient is 2.5. The specificity sensitivity coefficient is 1. The concentration sensitivity coefficient is set at 0.05. When the instantaneous formaldehyde concentration is 0.12 mg / m³, the exponential function value e raised to the power of 0.006 is approximately equal to 1.006. The risk weight value is calculated as 0.03 multiplied by 2.5 multiplied by 1 multiplied by 1.006, resulting in approximately 0.075. The exponential effect of formaldehyde risk weight value increasing with concentration is relatively gradual due to the low concentration sensitivity coefficient.

[0076] Taking a family setting with a child suffering from asthma as an example, the system first determines that particulate matter and formaldehyde are triggering factors based on the child's asthma condition (asthma) and assigns them non-zero weights, while carbon dioxide is not a triggering factor and is assigned a zero weight. Then, based on the child's age of six, a baseline sensitivity coefficient of 2.5 is determined, reflecting the high sensitivity to pollutants during childhood. Next, based on the child's allergen spectrum showing a positive result for dust mites, a specific sensitivity coefficient of 1.8 is set for particulate matter, and 1 for formaldehyde. Finally, using an exponential risk weighting model, a risk weight value of 2.48 is calculated when the particulate matter concentration is 120 micrograms per cubic meter, and a risk weight value of 0.075 is calculated when the formaldehyde concentration is 0.12 milligrams per cubic meter. These two risk weight values ​​are the function values ​​of the risk weighting function at the current concentration. These values ​​will then be multiplied by the instantaneous concentrations of particulate matter and formaldehyde, respectively, to obtain their respective instantaneous health risk contribution values. After weighted summation and correction for real-time ventilation, the final instantaneous health risk utility value for the child is obtained. This construction process fully integrates an individual's disease characteristics, age characteristics, and allergy status into risk quantification, achieving truly personalized health risk assessment.

[0077] The steps for generating the control command sequence specifically include: Step 1: Establishing an indoor air quality dynamic model. This model describes the time-domain evolution of indoor pollutant concentrations under given equipment control parameters and outdoor air quality conditions. The established indoor air quality dynamic model is a set of mathematical equations describing the dynamic changes in indoor pollutant concentrations over time under given equipment control parameters and outdoor air quality conditions. The model is based on the principle of mass conservation, treating the indoor space as a control volume. The rate of change of pollutant concentration equals the pollutant entry rate minus the pollutant discharge rate. The entry rate includes outdoor air infiltration, indoor source release, and fresh air system introduction; the discharge rate includes ventilation exhaust, air purifier removal, and surface deposition. The model uses equipment control parameters as input variables, including fresh air unit airflow, air purifier speed, and exhaust fan on / off status; outdoor air quality data as disturbance input, including outdoor particulate matter concentration and outdoor formaldehyde concentration; and indoor pollutant concentrations as state variables. The model describes the evolution of state variables over time using differential equations or difference equations.

[0078] In the specific setup process, the system first acquires the room's geometric parameters, including room volume, air exchange rate, and door / window airtightness. Then, it establishes independent differential equations for each type of pollutant. Taking particulate matter as an example, its concentration change rate is expressed as: outdoor particulate matter permeability minus natural ventilation exhaust rate, plus the outdoor particulate matter velocity introduced by the fresh air system minus the indoor particulate matter velocity exhausted by the fresh air system, minus the air purifier removal rate, plus the indoor particulate matter release rate. The indoor particulate matter release rate is dynamically estimated based on data on the activity status of sensitive populations; for example, the release rate of resuspended dust from the ground when children are running is positively correlated with activity intensity. The impact of each device on pollutant concentration is described through device performance parameters, such as the clean air delivery rate of the air purifier and the ventilation efficiency of the fresh air system.

[0079] Taking a family with a child suffering from asthma as an example, the system establishes a dynamic model of indoor air quality in the family's living room. The living room has a volume of 50 cubic meters, with a natural air exchange rate of 0.5 times per hour. The clean air delivery rate (CADR) of the air purifier is 200 cubic meters per hour. The maximum airflow of the fresh air system is 100 cubic meters per hour. The current outdoor particulate matter concentration is 80 micrograms per cubic meter. When the child is running, the estimated indoor particulate matter release rate is 50 micrograms per cubic meter per minute. Based on these parameters, the model establishes a differential equation for particulate matter concentration to predict the evolution curve of particulate matter concentration within a future time window under different combinations of device control parameters.

[0080] Step 2: Set an optimization time window, dividing it into multiple discrete moments. The system sets an optimization time window, the length of which determines the time range for forward-looking optimization. The optimization time window setting needs to balance prediction accuracy and computational complexity, typically set to ten to thirty minutes. A window that is too long will accumulate prediction errors, while a window that is too short will not fully reflect the long-term effects of equipment control. The system divides the optimization time window into multiple discrete moments, with the time interval between these moments consistent with the system sampling frequency, typically one second or several seconds. These discrete moments serve as the time grid for model prediction and control optimization. The system will calculate pollutant concentration and health risk utility value at each discrete moment.

[0081] Taking a family setting with a child suffering from asthma as an example, the system sets an optimization time window of fifteen minutes, divided into nine hundred discrete moments, each one second apart. Within the next nine hundred seconds, the system will predict changes in pollutant concentrations and the evolution of health risks with an accuracy of one point per second.

[0082] Step 3: Starting from the current moment, within the optimization time window, the system uses the equipment control parameter sequence as the decision variable and outdoor air quality data as the known disturbance input. It then uses the indoor air quality dynamic model to predict the multidimensional pollution exposure vector at each discrete moment within the optimization time window. The system uses the current moment as the prediction starting point and, within the optimization time window, uses the equipment control parameter sequence as the decision variable. The equipment control parameter sequence is a combination of control parameters that change over time, such as setting the air purifier's speed or the fresh air system's operating level at each second. The system uses outdoor air quality data as the known disturbance input, which can be set as a known sequence based on weather forecast data or real-time monitoring data within the optimization time window. The system inputs the equipment control parameter sequence and outdoor air quality data into the indoor air quality dynamic model. The model solves the differential equation using numerical integration and outputs the indoor pollutant concentration values ​​at each discrete moment within the optimization time window. The system combines these concentration values ​​with the synchronized real-time ventilation rate value to generate the predicted multidimensional pollution exposure vector value for each discrete moment within the optimization time window.

[0083] Taking a family setting with a child suffering from asthma as an example, the system starts at second zero and sets the device control parameters to turn on the air purifier at medium speed for the first 300 seconds and turn it off for the next 600 seconds. Outdoor air quality data predicts that the particulate matter concentration will gradually decrease from 80 micrograms per cubic meter to 50 micrograms per cubic meter over the next 15 minutes. The system substitutes these inputs into the indoor air quality dynamic model to obtain the predicted indoor particulate matter concentration for each second within the next 15 minutes. Combined with the predicted real-time ventilation volume of the child, a multi-dimensional pollution exposure vector is generated for each second, including particulate matter concentration, formaldehyde concentration, and real-time ventilation volume.

[0084] Step 4: Based on the predicted multidimensional pollution exposure vector, calculate the sum of the cumulative health risk utility values ​​of all individuals within the optimization time window, and use this sum as the optimization objective function. The system calculates the instantaneous health risk utility value of each individual in the sensitive population at each discrete moment within the optimization time window, based on the multidimensional pollution exposure vector predicted in Step 3, following the aforementioned method for calculating health risk utility values. Then, for each individual, the instantaneous health risk utility values ​​at all discrete moments within the optimization time window are summed to obtain the cumulative health risk utility value for each individual. Finally, the cumulative health risk utility values ​​of all individuals are summed to obtain the total cumulative health risk utility value within the optimization time window. The system uses this total cumulative health risk utility value as the optimization objective function. The value of the optimization objective function depends on the selection of the equipment control parameter sequence; different combinations of control parameters will lead to different pollutant concentration evolution trajectories, thus producing different objective function values.

[0085] Taking a family setting with a child suffering from asthma as an example, the system calculates the instantaneous health risk utility value of the child for each second within the optimization time window based on the predicted multidimensional pollution exposure vector, and accumulates these values ​​to obtain the child's cumulative health risk utility value. Adults, lacking health risk weights, have a cumulative health risk utility value of zero. The system uses the child's cumulative health risk utility value as the optimization objective function. For one sequence of control parameters, the calculated objective function value is 1200; for another sequence, it is 800. The system will select the control parameter sequence that minimizes the objective function value.

[0086] Step 5: The system employs Model Predictive Control (MMCC) to solve for the optimal sequence of equipment control parameters that minimizes the objective function. MMCC is an advanced process control method. Its core idea is to predict the system behavior over a future period based on the system model at each control time. The optimal control sequence is obtained by solving a finite-time optimization problem online. Only the first control action is executed, and prediction and optimization are repeated at the next time step, achieving rolling optimization and feedback correction.

[0087] The relationship between Model Predictive Control (MMCC) and the optimization algorithms mentioned earlier is as follows: the phrase "solving through optimization algorithms" is a general description of the optimization process, while MMCC is the specific implementation of this optimization algorithm. MMCC incorporates the core mechanism of optimization, namely, finding the optimal control parameters by minimizing the objective function. It also possesses enhanced characteristics such as rolling time domain and feedback correction, making it more suitable for real-time control in dynamic environments.

[0088] Within the framework of model predictive control (MMC) algorithms, the system formulates the optimization problem as follows: under the condition of satisfying equipment physical constraints, find the sequence of equipment control parameters that minimizes the objective function. Equipment physical constraints include discrete values ​​of equipment speed settings, minimum start-up and shutdown durations, and maximum equipment power. The system employs numerical optimization methods to solve this problem, such as gradient descent, genetic algorithms, or dynamic programming. The resulting optimal equipment control parameter sequence covers the entire optimization time window.

[0089] Taking a family setting with a child suffering from asthma as an example, the system employs a model predictive control algorithm to solve for the optimal sequence of device control parameters. The optimization problem is: under the constraint that the air purifier can only be switched between four discrete values—off, low, medium, and high—find the sequence of settings that minimizes the cumulative health risk utility for the child over the next fifteen minutes. The algorithm calculates the objective function values ​​corresponding to different setting sequences through enumeration or heuristic search, ultimately determining the optimal sequence as: medium setting for the first 300 seconds, low setting for the middle 300 seconds, and off for the last 300 seconds.

[0090] Step 6: Select the equipment control parameters corresponding to the current moment from the optimal equipment control parameter sequence, generate the control command sequence for the current moment, and continuously update the optimization solution in the next moment. The system selects the equipment control parameters corresponding to the current moment from the optimal equipment control parameter sequence obtained in Step 5, generates the control command sequence for the current moment, and executes it immediately. The system does not execute all control sequences within the entire optimization time window, but only executes the control commands for the first moment. This is because model predictions have uncertainties, and the actual system state may deviate from the prediction. In the next sampling moment, the system obtains the latest actual state information and re-executes Steps 1 to 5 for a new round of prediction and optimization, i.e., the continuous update optimization solution process. This continuous optimization mechanism enables the system to continuously correct control decisions based on actual operating conditions, improving the robustness and adaptability of control.

[0091] Taking a family scenario with a child suffering from asthma as an example, in the optimal device control parameter sequence obtained by the system, the control parameter corresponding to the current moment is the medium setting of the air purifier. The system generates the control command sequence for the current moment, turning the air purifier to the medium setting. In the next second, the system re-acquires the current actual particulate matter concentration and the child's ventilation volume, uses the new state as a starting point, resets the optimization time window, re-predicts the concentration evolution over the next fifteen minutes, re-solves the optimal control parameter sequence, and then selects the control parameter for the current moment for execution. This process is repeated continuously, enabling the system to achieve continuous optimization control in dynamic environments.

[0092] Taking a family setting with a child suffering from asthma as an example, the system first establishes a dynamic model of indoor air quality in the living room, including parameters such as room volume, air exchange rate, purifier performance, and outdoor pollution. The optimization time window is set at fifteen minutes, divided into nine hundred discrete moments. Starting from the current moment (zero second), the system uses the sequence of device control parameters as decision variables, substituting them into the model to predict the particulate matter concentration for each second within the next fifteen minutes. Combined with the child's real-time ventilation prediction, a multi-dimensional pollution exposure vector prediction is generated. The cumulative health risk utility value for the child is calculated based on the prediction values ​​as the optimization objective function. A model predictive control algorithm is used to find the optimal sequence of device control parameters that minimizes the objective function, resulting in a sequence of settings: medium speed for the first 300 seconds, low speed for the middle 300 seconds, and off for the last 300 seconds. The system selects the medium speed parameter at the current moment and generates a control command to turn on the air purifier at medium speed. The next second, the system re-acquires the actual state, performs prediction, optimization, and execution again, achieving rolling optimization control. Model predictive control (MMC) algorithms, as a specific implementation of optimization algorithms, enable the system to continuously minimize health risks to sensitive populations in dynamically changing indoor and outdoor environments through rolling time-domain optimization and feedback correction.

[0093] The health feedback calibration module generates new individualized exposure response boundary parameters by acquiring environmental data generated after the execution of the control command sequence from the risk-driven control module. The execution period of the control command sequence refers to the time interval from the issuance of the control command to the equipment's response and completion of adjustment; the post-execution period refers to the continuous observation period after the control command is completed. The environmental data includes a multi-dimensional pollution exposure vector within this time interval, which contains the instantaneous concentration values ​​of multiple pollutants on a unified time axis and the synchronous real-time ventilation volume value. This data records the actual change trajectory of indoor pollutant concentrations under specific control strategies, as well as the real-time exposure of sensitive populations during the control period, providing fundamental data for subsequent analysis of the correlation between control effectiveness and health response.

[0094] Taking a family setting with a child suffering from asthma as an example, the risk-driven control module generates a sequence of control commands to turn on the air purifier at a medium setting. The system acquires environmental data within 30 minutes after the command is executed, including data on the process of particulate matter concentration gradually decreasing from 120 micrograms per cubic meter to 20 micrograms per cubic meter per second, data on the child's real-time ventilation gradually recovering from 25 liters per minute while running to 6 liters per minute, and data on the slow decrease of formaldehyde concentration under the action of the purifier. This data completely records the entire process of environmental evolution after the control command is executed.

[0095] Health outcome feedback data from sensitive populations is collected through multiple channels. Active recording is accomplished via mobile applications or smart terminals. When sensitive individuals or their caregivers observe respiratory symptoms, they record the specific time of symptom occurrence, symptom type, and severity (e.g., cough, wheezing, chest tightness) in the system, and classify them according to a pre-set rating scale. The time, name, and dosage of medications used are also recorded. Passive data collection is achieved through wearable devices, such as smart bracelets, smartwatches, or medical-grade physiological monitoring devices, simultaneously collecting continuous data on changes in physiological parameters such as heart rate, respiratory rate, and blood oxygen saturation. Health outcome feedback data reflects the actual changes in the health status of sensitive populations after environmental exposure and control interventions, and is a key basis for calibrating the system's exposure determination criteria.

[0096] Taking a family scenario with a child suffering from asthma as an example, the child developed a cough fifteen minutes after the running event. The mother recorded the onset of the symptoms at 2:30 PM using a mobile application, classifying it as a mild cough. She also recorded the use of asthma inhaler medication at 2:35 PM, with a dosage of two puffs. Simultaneously, the child's smart bracelet recorded that from 2:30 PM onwards, the heart rate gradually increased from 120 beats per minute to 130 beats per minute, and the respiratory rate increased from 20 breaths per minute to 28 breaths per minute. The system integrates these actively recorded and passively collected data into health outcome feedback data.

[0097] The system correlates and matches health outcome feedback data with environmental data over time. The core operation of this correlation matching is to use the moment of the health event as an anchor point, trace back a pre-defined time window, and extract pollutant exposure data within that window to establish a temporal correlation between exposure and health outcome. The length of the pre-defined time window is set based on common timescales for exposure-response relationships in health effect studies, typically fifteen minutes, thirty minutes, or one hour, covering the time delay from the start of exposure to the appearance of symptoms. The system extracts pollutant exposure characteristics from the multi-dimensional pollution exposure vector within this time window, including the average concentration, peak concentration, exposure duration, and concentration rise rate of various pollutants. The average concentration reflects the overall exposure level within the window, the peak concentration reflects the instantaneous high-intensity exposure, the exposure duration reflects the cumulative effect of exposure, and the concentration rise rate reflects the suddenness of the pollution event.

[0098] Taking a family setting with a child suffering from asthma as an example, the child's cough symptoms occurred at 2:30 PM. The system set a preset time window of 30 minutes before the onset of symptoms, i.e., from 2:00 PM to 2:30 PM. The system extracted particulate matter exposure data within this time window: the average concentration was 65 micrograms per cubic meter, the peak concentration reached 120 micrograms per cubic meter at 2:02 PM, the exposure duration was defined as the total duration of the concentration exceeding 50 micrograms per cubic meter, which was 18 minutes, and the rate of increase in concentration was defined as the two minutes it took for the initial concentration of 10 micrograms per cubic meter to rise to the peak concentration of 120 micrograms per cubic meter, with an increase rate of 55 micrograms per cubic meter per minute. Similarly, formaldehyde exposure characteristics were extracted: the average concentration was 0.08 milligrams per cubic meter, the peak concentration was 0.12 milligrams per cubic meter, the exposure duration was 12 minutes, and the rate of increase in concentration was 0.06 milligrams per cubic meter per minute. These exposure characteristics formed a correlation pair with health outcomes.

[0099] The system combines pollutant exposure characteristics with corresponding health outcome feedback data to form an exposure-response sample pair. Exposure characteristics constitute the input part of the sample, while health outcome feedback data constitutes the label part. Whether a health event occurred serves as the classification label, and symptom level or medication use can serve as the regression label. The system inputs this sample pair into a pre-defined health effect learning model. This model employs a one-dimensional convolutional neural network structure, specifically designed for processing time-series data. The first part of the model is a convolutional layer, which automatically extracts local time-series patterns from the exposure characteristics by sliding multiple convolutional kernels along the time dimension, such as the rate of concentration increase, the decline after the peak, and the stationary features of continuous exposure. The output of the convolutional layer is then dimensionality-reduced by a pooling layer and input into a fully connected layer. The fully connected layer uses a non-linear transformation to map the extracted features to the probability of health event occurrence, outputting a probability value between zero and one.

[0100] The model employs incremental learning for updates. Incremental learning means that the model does not need to be trained from scratch every time new samples are acquired. Instead, it updates the parameters in small batches using only new samples, based on the current model parameters. Incremental learning achieves continuous model evolution by retaining the general knowledge already learned by the model while incorporating individual-specific information from new samples. This design allows the model to gradually converge to an exposure-response mapping that is highly adapted to the individual as health events in the sensitive population accumulate.

[0101] Taking a family setting with a child suffering from asthma as an example, the system combines extracted particulate matter exposure features, formaldehyde exposure features, and the health outcome of a child's cough symptoms into an exposure-response sample pair. This sample pair is input into a one-dimensional convolutional neural network model. The convolutional layer extracts a steep pattern where particulate matter concentration rapidly rises to a peak within two minutes and a high-concentration exposure pattern lasting eighteen minutes. The fully connected layer outputs a 78% probability of the health event based on these patterns. The system uses this sample pair for incremental learning, fine-tuning the model parameters based on this sample pair so that the output probability value more closely approximates the actual observed health outcome when encountering similar exposure features in the future.

[0102] Based on the updated health effect learning model through incremental learning, the probability thresholds for health events at different pollutant exposure levels are recalculated. Specifically, the system sets an acceptable probability of a health event, such as 1% or 5%, as the upper limit of risk tolerance. Then, using this probability threshold as a benchmark, the system uses the model to inversely calculate the exposure conditions required to reach this probability threshold at different pollutant exposure levels. The system iterates through combinations of various pollutants at different exposure durations, calculating the pollutant concentration value at which the probability of a health event is exactly equal to the set threshold; this concentration value is the safety threshold. Simultaneously, based on the model's analysis of the sensitivity of exposure characteristics, the system determines the warning trigger conditions; that is, when the monitored pollutant exposure characteristics reach or exceed the safety threshold, the system should issue a warning signal. The generated new individualized exposure response boundary parameters will replace the original parameters and serve as the basis for subsequent operation of the transient exposure sensing module.

[0103] Taking a family setting with a child suffering from asthma as an example, the system sets the acceptable probability of a health event at 5%. Using an updated health effect learning model, the system calculates the particulate matter concentration that would cause a 5% probability of coughing in a child under different exposure durations. The results show that the safe threshold is 80 micrograms per cubic meter when the exposure duration is 50 micrograms per cubic meter; 50 micrograms per cubic meter when the exposure duration is 15 minutes; and 35 micrograms per cubic meter when the exposure duration is 30 minutes. Simultaneously, the model's analysis of the concentration rise rate sensitivity shows that when the concentration rise rate exceeds 30 micrograms per cubic meter per minute, the probability of a health event increases significantly. Therefore, the warning trigger condition is set as follows: particulate matter concentration exceeding the safe threshold, or concentration rise rate exceeding 30 micrograms per cubic meter per minute. The system encapsulates these safe thresholds and warning trigger conditions into new individualized exposure response boundary parameters, inputting them into the transient exposure perception module. This allows the module to sample and determine based on a benchmark more suitable for the child's health condition during subsequent operation.

[0104] Taking a family setting with a child suffering from asthma as an example, the health feedback calibration module first acquires environmental data within 30 minutes after the control command is executed, including data on the process of particulate matter decreasing from 120 micrograms per cubic meter to 20 micrograms per cubic meter and data on changes in the child's ventilation. Simultaneously, it collects the time and severity of the child's cough symptoms, medication usage records, and heart rate and respiratory rate data from wearable devices. The system uses the time of the health outcome as an anchor point, tracing back 30 minutes to extract pollutant exposure characteristics within that time window, including an average particulate matter concentration of 65 micrograms per cubic meter, a peak concentration of 120 micrograms per cubic meter, an exposure duration of 18 minutes, and a concentration rise rate of 55 micrograms per cubic meter per minute. These exposure characteristics are then paired with cough symptoms to form exposure-response sample pairs, which are input into a one-dimensional convolutional neural network model. The model outputs the probability of the health event, and the model parameters are updated through incremental learning. The updated model recalculates the particulate matter safety thresholds that give a 5% probability of coughing, resulting in 80 micrograms per cubic meter for 5-minute exposure, 50 micrograms per cubic meter for 15-minute exposure, and 35 micrograms per cubic meter for 30-minute exposure. A warning trigger condition is set for a concentration rise rate exceeding 30 micrograms per cubic meter per minute. These new individualized exposure response boundary parameters are input into the transient exposure sensing module, enabling the monitoring system in the child's home to protect the child's respiratory health with stricter standards in subsequent operation. As more health events accumulate, the system will continuously learn and optimize the individualized exposure response boundary parameters, achieving adaptive evolution of the protection strategy.

[0105] The steps for adjusting the sampling strategy and exposure determination criteria of the transient exposure sensing module include: The health feedback calibration module transmits the safety threshold portion of the newly generated individualized exposure response boundary parameters to the transient exposure sensing module. These safety thresholds are organized according to pollutant type and exposure duration, such as safety thresholds for particulate matter exposure after five minutes, fifteen minutes, and thirty minutes. Upon receiving these safety thresholds, the transient exposure sensing module stores them in its local configuration and replaces the old thresholds. In subsequent real-time operation, after collecting pollutant concentration data, the module continuously monitors the current exposure duration. When the concentration of a pollutant exceeds the safety threshold for the corresponding exposure duration, the module determines that the current exposure has reached a dangerous level and triggers the corresponding protection mechanism. This dynamically updated safety threshold allows the system to continuously adjust the stringency of the danger assessment based on the latest health feedback data from sensitive populations.

[0106] Taking a family setting with a child suffering from asthma as an example, the health feedback calibration module learns from the child's coughing events and generates new individualized exposure response boundary parameters. The safety thresholds for particulate matter are updated as follows: 80 micrograms per cubic meter for five minutes of exposure, 50 micrograms per cubic meter for fifteen minutes, and 35 micrograms per cubic meter for thirty minutes. These safety thresholds are transmitted to the transient exposure sensing module, replacing the old thresholds. The old thresholds were 100 micrograms per cubic meter for five minutes, 70 micrograms per cubic meter for fifteen minutes, and 50 micrograms per cubic meter for thirty minutes. The new thresholds are significantly stricter, reflecting that the child's actual sensitivity to particulate matter is higher than the original default settings. If the child subsequently runs again, causing the particulate matter concentration to rise to 70 micrograms per cubic meter and persist for eight minutes, the module determines that the exposure has reached a dangerous level based on the new thresholds, whereas under the old thresholds, the exposure would not have been considered dangerous.

[0107] The health feedback calibration module transmits the early warning trigger conditions from the newly generated individualized exposure response boundary parameters to the transient exposure sensing module. The early warning trigger conditions are a set of rules used to determine whether an early warning is needed. These rules are based not only on pollutant concentration thresholds but may also include dynamic characteristics such as the rate of concentration increase and the trend of exposure duration. Upon receiving the early warning trigger conditions, the transient exposure sensing module stores them in its local configuration. During real-time monitoring, the module continuously calculates the current pollutant exposure characteristics, including real-time concentration, concentration change rate, and cumulative exposure time. When the monitored exposure characteristics meet the early warning trigger conditions, the module generates transient exposure data for the respiratory zone and attaches an early warning flag to this data. The early warning flag can be a flag bit in the data field; downstream modules can execute corresponding early warning pushes or control actions based on the early warning flag when processing the data.

[0108] Taking a family setting with a child suffering from asthma as an example, the health feedback calibration module generates a new warning trigger condition based on model analysis: a warning is triggered when the rate of increase in particulate matter concentration exceeds 30 micrograms per cubic meter per minute. This warning condition is transmitted to the transient exposure sensing module. One day, the child starts running, and the module detects that the particulate matter concentration rises from 10 micrograms per cubic meter to 15 micrograms per cubic meter within one second, calculating a rate of increase of 300 micrograms per cubic meter per minute, far exceeding the 30 micrograms per cubic meter per minute threshold. The module immediately determines that the warning trigger condition is met and adds a warning flag when generating the transient exposure data for the respiratory zone at the current moment. Upon receiving the data with the warning flag, the downstream risk-driven control module can activate the air purifier in advance, before the pollutant concentration reaches the safe threshold, achieving preventative control.

[0109] The transient exposure sensing module dynamically adjusts the sampling frequency of various pollutant sensors based on the relative changes between the new and old safety thresholds. The basic logic is as follows: a stricter safety threshold means that sensitive populations are more sensitive to the pollutant, requiring more intensive data collection to capture subtle exposure fluctuations; a more lenient safety threshold means that sensitive populations have increased tolerance to the pollutant, allowing for a reduction in sampling frequency to save energy and computing resources. Specifically, the module starts with a preset baseline sampling frequency, which is the system default, for example, once per second. When the safety threshold for a certain type of pollutant is updated, the module reads the previously stored safety threshold as the pre-update threshold, divides the pre-update threshold by the updated threshold, and obtains the percentage change. If the updated threshold is stricter than the pre-update threshold, the percentage change is less than one; if it is more lenient, the percentage change is greater than one. The module then multiplies the baseline sampling frequency by this percentage change to obtain the new sampling frequency. To prevent system overload due to excessively high sampling frequencies or monitoring failure due to excessively low sampling frequencies, the module sets upper and lower limits for the new sampling frequency, ensuring it does not exceed the preset maximum sampling frequency or fall below the preset minimum sampling frequency. For example, the maximum sampling frequency is five times per second, and the minimum sampling frequency is once every ten seconds. The adjusted sampling frequency is applied to the corresponding pollutant sensor, and subsequent data acquisition is performed at the new frequency.

[0110] Taking a family setting with a child suffering from asthma as an example, the preset basic sampling frequency for particulate matter in the transient exposure sensing module is once per second. The original safety threshold for particulate matter was set at 100 micrograms per cubic meter for five minutes of exposure. The updated safety threshold is 80 micrograms per cubic meter for five minutes of exposure. Dividing the original safety threshold by the updated one, 100 divided by 80, yields 1.25, a ratio greater than 1. However, this needs careful interpretation: when the safety threshold decreases from 100 to 80, it means the judgment standard becomes stricter, requiring more frequent sampling. The calculation method for the change in the safety threshold is the original value divided by the updated value; 100 divided by 80 equals 1.25. Multiplying the basic sampling frequency of once per second by 1.25 yields a new sampling frequency of 1.25 times per second, or once every 0.8 seconds. The increased sampling frequency aligns with the logic of requiring more frequent monitoring after the safety threshold becomes stricter. The module sets an upper and lower limit for this new frequency, assuming an upper limit of five times per second and a lower limit of once every ten seconds. 1.25 times per second falls within this range and is adopted. For formaldehyde, if the safety threshold is updated from 0.1 mg / m³ to 0.15 mg / m³, becoming more lenient, then 0.1 mg / m³ before the update divided by 0.15 mg / m³ after the update equals 0.66 mg / m³. The new sampling frequency is 0.66 times per second, that is, once every 1.5 seconds. The reduced sampling frequency is logical.

[0111] After adjusting the sampling frequency, replacing the new baseline threshold, and updating the warning trigger conditions, the transient exposure sensing module begins operating according to the new configuration. The module collects multi-pollutant concentration data from the respiratory zone of sensitive individuals at the adjusted sampling frequency; for example, the particulate matter sensor collects data at a frequency of 1.25 times per second, and the formaldehyde sensor at a frequency of once every 1.5 seconds. The module compares the collected real-time data with the new baseline threshold to determine whether the current exposure has reached a dangerous level. Simultaneously, the module continuously calculates exposure characteristics and matches them with the new warning trigger conditions to determine whether additional warning indicators are needed. The module encapsulates the collected pollutant concentration data, synchronously collected human activity data, real-time ventilation volume estimates, inhalation volume per unit time calculation results, and possible warning indicators to generate updated transient exposure data for the respiratory zone. This data serves as the starting point for the entire system's closed loop and is then input into the timing alignment and fusion module for further processing, thus achieving a complete closed loop from health feedback calibration to sensing strategy optimization.

[0112] Taking a family setting with a child suffering from asthma as an example, after the transient exposure sensing module completes its configuration update, the particulate matter sensor collects concentration data at a frequency of 1.25 times per second. One day, the child runs around again, and the module records changes in particulate matter concentration at the same frequency. As the concentration rises, the module continuously compares the concentration value for the current exposure duration with the new safety threshold. When the particulate matter concentration reaches 75 micrograms per cubic meter and persists for four minutes, the module determines that the exposure has reached a dangerous level. Simultaneously, the module detects that the concentration rise rate exceeds 30 micrograms per cubic meter per minute, meeting the warning trigger condition, and adds a warning label to the generated transient exposure data for the respiratory zone. The module encapsulates the data, including particulate matter concentration, formaldehyde concentration, real-time ventilation, inhalation volume per unit time, and the warning label, to generate updated transient exposure data for the respiratory zone, which is then passed to the time-series alignment and fusion module. This process achieves closed-loop optimization of the sensing module by the health feedback calibration module, enabling the system to collect data and determine exposure with parameters more suitable for the health status of this sensitive population in subsequent operation.

[0113] Taking a family setting with a child suffering from asthma as an example, the health feedback calibration module generates new individualized exposure response boundary parameters and transmits them to the transient exposure perception module. The module first replaces the old thresholds with new safety thresholds for particulate matter: 80 micrograms per cubic meter for 5 minutes, 50 micrograms per cubic meter for 15 minutes, and 35 micrograms per cubic meter for 30 minutes. Simultaneously, it stores the new warning trigger condition—a concentration rise rate exceeding 30 micrograms per cubic meter per minute—as a warning rule. Then, it adjusts the sampling frequency based on changes in the safety thresholds. For example, when the particulate matter safety threshold decreases from 100 micrograms per cubic meter to 80 micrograms per cubic meter (a change of 1.25), the sampling frequency is adjusted from once per second to 1.25 times per second; when the formaldehyde safety threshold increases from 0.1 milligrams per cubic meter to 0.15 milligrams per cubic meter (a change of 0.66), the sampling frequency is adjusted from once per second to once every 1.5 seconds. The module collects data according to the adjusted sampling frequency, determines the exposure risk level using the new thresholds, and judges the warning conditions using the new rules. When the child runs again, the module detects a rapid increase in particulate matter concentration more frequently, attaching a warning sign before the concentration reaches a dangerous level and generating updated transient exposure data for the respiratory zone. This data, carrying richer health protection information, is fed into subsequent modules, achieving a complete closed loop from health feedback to perception optimization. This allows the system to automatically optimize its monitoring strategy after each health event.

[0114] The above-mentioned models or function formulas are all dimensionless and numerical calculations. The models or function formulas are obtained by software simulation based on a large amount of collected data to obtain the most recent real situation. The preset parameters in the models or function formulas are set by those skilled in the art according to the actual situation.

[0115] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0116] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0117] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0118] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations 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. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A real-time indoor air quality monitoring system based on multi-sensor fusion, characterized in that, include: The transient exposure sensing module is used to collect concentration data of multiple pollutants in the respiratory zone of sensitive populations at a preset sampling frequency, and simultaneously collect human activity status data to generate transient exposure data in the respiratory zone. The temporal alignment and fusion module is used to dynamically align the concentration data of multiple pollutants with different sensor response times in the transient exposure data of the respiratory zone, and generate a time-aligned multidimensional pollution exposure vector. The risk-driven control module is used to map time-aligned multidimensional pollution exposure vectors into health risk utility values ​​that uniformly characterize the health risks of sensitive populations, and to generate a sequence of control instructions with the goal of minimizing the health risk utility values. The health feedback calibration module is used to collect health outcome feedback data of sensitive populations after the control command sequence is executed, generate new individualized exposure response boundary parameters based on the health outcome feedback data, and input the new individualized exposure response boundary parameters into the transient exposure perception module to adjust the sampling strategy and exposure judgment criteria of the transient exposure perception module.

2. The real-time indoor air quality monitoring system based on multi-sensor fusion according to claim 1, characterized in that, The transient exposure sensing module generates transient exposure data of the respiratory zone through the following steps: The location information of the breathing zone of sensitive people is obtained, and multiple pollutant sensors and human activity status sensors are deployed in the breathing zone according to the location information. The human activity status sensors include accelerometers and heart rate sensors. Simultaneously collect concentration data of multiple pollutants in the respiratory zone and human activity status data at a preset sampling frequency; The activity intensity index is calculated based on the acceleration data in the human activity state data. The logarithmically transformed value of the activity intensity index is fused with the heart rate data. The real-time ventilation of the sensitive population is estimated through a preset ventilation volume mapping model. The ventilation volume mapping model is a linear regression model pre-established based on the sensitive population sample. The model uses the logarithmically transformed value of the activity intensity index and the heart rate value as input variables and ventilation volume as output variable. The concentration data of multiple pollutants are aligned and matched with the real-time ventilation volume in the time dimension to calculate the inhalation volume per unit time of each pollutant. The inhalation volume per unit time is obtained by multiplying the instantaneous concentration value of the pollutant by the real-time ventilation volume.

3. The real-time indoor air quality monitoring system based on multi-sensor fusion according to claim 2, characterized in that, Transient exposure data in the respiratory zone is obtained by encapsulating data including instantaneous concentration values ​​of multiple pollutants, real-time ventilation values, and the inhalation volume of each pollutant per unit time.

4. The real-time indoor air quality monitoring system based on multi-sensor fusion according to claim 3, characterized in that, The time-aligned fusion module generates a time-aligned multidimensional contamination exposure vector through the following sub-steps: Acquire transient exposure data in the respiratory zone, extract concentration data of multiple pollutants and their corresponding timestamps, and identify the sensor type and response time characteristics corresponding to each pollutant concentration data; The pollutant concentration data with the shortest response time is selected from the multi-type pollutant concentration data as the reference time axis. This reference time axis is used to mark the actual occurrence time of the pollution event. For the other types of pollutant concentration data, feature point matching is performed with the reference time axis to identify the starting point of the rising edge, the peak point, and the inflection point of the falling edge in the pollutant concentration curve. Calculate the time offset of the pollutant concentration data relative to the reference time axis based on the matched feature points; The pollutant concentration data is resampled on the time axis according to the time offset, and the time points of the pollutant concentration data are mapped to the time grid of the reference time axis through interpolation, so that the pollutant concentration data and the reference time axis are aligned point-to-point in the time dimension. The time-aligned multi-pollutant concentration data are merged with the instantaneous pollutant concentration values, real-time ventilation values, and inhalation volume per unit time from the transient exposure data of the respiratory zone to generate a time-aligned multi-dimensional pollution exposure vector.

5. The real-time indoor air quality monitoring system based on multi-sensor fusion according to claim 4, characterized in that, The step of calculating the time offset of the pollutant concentration data relative to the reference time axis is implemented using a sliding window cross-correlation algorithm, specifically including: Set the sliding window length to divide the pollutant concentration sequence on the reference time axis and the pollutant concentration sequence to be aligned into multiple subsequences respectively; Calculate the cross-correlation coefficient between each pair of subsequences and determine the sliding window position corresponding to the maximum cross-correlation coefficient; The time offset of the pollutant concentration sequence to be aligned relative to the reference time axis is determined by the sliding window position corresponding to the maximum cross-correlation coefficient.

6. The real-time indoor air quality monitoring system based on multi-sensor fusion according to claim 5, characterized in that, The risk-driven control module maps time-aligned multidimensional pollution exposure vectors to health risk utility values ​​in the following way: Obtain time-aligned multidimensional pollution exposure vectors, which include instantaneous concentration values ​​of multiple pollutants on a unified time axis and synchronized real-time ventilation values; Obtain individual characteristic parameters of the sensitive population, including age, disease type, and allergen spectrum. When the sensitive population includes multiple individuals, obtain the individual characteristic parameters of each individual separately. Based on the individual characteristic parameters of each individual, a risk weight function for each type of pollutant is constructed for that individual. The risk weight function is used to characterize the contribution of the pollutant to the individual's health risk per unit concentration of exposure. For each individual, the instantaneous concentration value of each type of pollutant corresponding to that individual is multiplied by the corresponding risk weight function to obtain the instantaneous health risk contribution value of that pollutant to that individual. The instantaneous health risk contribution values ​​of all pollutants are weighted and summed. The result of the weighted sum is then multiplied by the individual's real-time ventilation value to obtain the instantaneous health risk utility value of that individual at the current moment. For each individual, the instantaneous health risk utility value of that individual within a preset time window is accumulated to obtain the cumulative health risk utility value of that individual. With the goal of minimizing the sum of cumulative health risk utility values ​​of all individuals, and combining the current operating status of multiple devices with outdoor air quality data, the optimal combination of device control parameters is solved through optimization algorithms to generate a sequence of control commands.

7. The real-time indoor air quality monitoring system based on multi-sensor fusion according to claim 6, characterized in that, The steps of constructing a risk weight function for each type of pollutant for each individual based on their individual characteristic parameters specifically include: For each individual, determine whether each type of pollutant is a triggering factor for that individual based on the individual's disease type. If it is a triggering factor, assign a non-zero weight; otherwise, assign a zero weight. For pollutants that are triggering factors, the baseline sensitivity coefficient is determined based on the individual's age. The baseline sensitivity coefficient shows a peak distribution that first increases and then decreases with age, and is used to characterize the inherent sensitivity of different age groups to pollutants. The specificity sensitivity coefficient is determined based on the individual's allergen spectrum. When the individual has a positive reaction to an allergen corresponding to a certain type of pollutant, the specificity sensitivity coefficient is set to a preset value greater than one; otherwise, it is set to one. The risk weight value of the pollutant in different concentration ranges is determined by a preset risk weight model. The risk weight model is an exponential nonlinear function. The function takes the pollutant concentration as input and the risk weight value as output. The risk weight value is obtained by multiplying the preset basic weight coefficient, basic sensitivity coefficient, specific sensitivity coefficient and the exponential function value of the pollutant concentration. The risk weight value increases exponentially with the increase of pollutant concentration. The risk weight value is the function value of the risk weight function at the corresponding pollutant concentration.

8. The real-time indoor air quality monitoring system based on multi-sensor fusion according to claim 7, characterized in that, The steps for generating a control command sequence specifically include: Establish an indoor air quality dynamic model, which is used to describe the time-domain evolution of the concentration of various pollutants in the room under given equipment control parameters and outdoor air quality conditions. Set an optimization time window and divide the optimization time window into multiple discrete moments; Starting from the current moment, within the optimization time window, the equipment control parameter sequence is used as the decision variable, and the outdoor air quality data is used as the known disturbance input. The indoor air quality dynamic model is used to predict the multidimensional pollution exposure vector at each discrete moment within the optimization time window. Based on the predicted multidimensional pollution exposure vector, the sum of the cumulative health risk utility values ​​of all individuals within the optimization time window is calculated and used as the optimization objective function; The optimal sequence of equipment control parameters that minimizes the objective function is obtained by using a model predictive control algorithm. Select the equipment control parameters corresponding to the current moment from the optimal equipment control parameter sequence, generate the control command sequence for the current moment, and continuously update and optimize the solution process in the next moment.

9. A real-time indoor air quality monitoring system based on multi-sensor fusion according to claim 8, characterized in that, The health feedback calibration module generates new individualized exposure response boundary parameters in the following way: Obtain environmental data after the execution of the control command sequence generated by the risk-driven control module. The environmental data includes a multi-dimensional pollution exposure vector during and after the execution of the control commands. Health outcome feedback data of sensitive populations were collected. The health outcome feedback data included the time and severity of respiratory symptoms, the time and dosage of medication use, and the physiological parameter changes collected synchronously by wearable devices. By correlating and matching health outcome feedback data with environmental data over time, pollutant exposure characteristics within a preset time window before a health event occurs are identified. These pollutant exposure characteristics include the average concentration, peak concentration, exposure duration, and concentration rise rate of various pollutants. Pollutant exposure features and corresponding health outcome feedback data are used to form exposure-response sample pairs, which are then input into a pre-defined health effect learning model. The health effect learning model is a one-dimensional convolutional neural network model. This model takes the pollutant exposure feature vector as input and the probability of health event occurrence as output. It extracts the time series patterns in the exposure features through convolutional layers, outputs the probability values ​​through fully connected layers, and uses an incremental learning method to update the model parameters after each acquisition of a new exposure-response sample pair. Based on the updated health effect learning model, the probability thresholds for health events under different pollutant exposure levels are recalculated, and new individualized exposure response boundary parameters are generated. These individualized exposure response boundary parameters include the safety thresholds for various pollutants under different exposure durations and the corresponding early warning triggering conditions.

10. A real-time indoor air quality monitoring system based on multi-sensor fusion according to claim 9, characterized in that, The steps for adjusting the sampling strategy and exposure determination criteria of the transient exposure sensing module include: The safety thresholds for various pollutants under different exposure durations in the new individualized exposure response boundary parameters are transmitted to the transient exposure sensing module. These thresholds serve as the new benchmark thresholds for the module to make real-time judgments on pollutant concentration data. The new benchmark thresholds replace the original thresholds to determine whether the current exposure has reached a dangerous level. The warning triggering conditions in the new individualized exposure response boundary parameters are transmitted to the transient exposure sensing module as a new triggering rule for the module to generate warning signals. When the monitored pollutant exposure characteristics meet the warning triggering conditions, the module adds a warning label while generating transient exposure data in the respiratory zone. Based on the changes in the safety thresholds of various pollutants in the new individualized exposure response boundary parameters, the sampling frequency of the transient exposure sensing module for various pollutant sensors is adjusted. Specifically, the adjustment method is as follows: with the preset basic sampling frequency as a benchmark, when the safety threshold of a certain type of pollutant is updated, the safety threshold stored at the time of the last update is read as the safety threshold before the update. The safety threshold before the update is divided by the safety threshold after the update to obtain the safety threshold change ratio. The basic sampling frequency is multiplied by the safety threshold change ratio to obtain the new sampling frequency. An upper limit and a lower limit are set for the new sampling frequency so that the new sampling frequency is not higher than the preset maximum sampling frequency and not lower than the preset minimum sampling frequency. The adjusted sampling frequency is then applied to the corresponding pollutant sensor. The transient exposure sensing module collects multi-pollutant concentration data of the respiratory zone of sensitive populations at an adjusted sampling frequency based on the updated sampling strategy and exposure judgment criteria. It then uses the adjusted new benchmark thresholds and early warning triggering conditions to make real-time judgments on the collected data and generate updated transient exposure data for the respiratory zone.

Images