Microclimate intelligent monitoring system based on edge computing

Abstract

The application discloses a microclimate intelligent monitoring system based on edge computing and particularly relates to the technical field of microclimate monitoring, comprising an edge monitoring center, a microclimate and human health index monitoring module, a microclimate multidimensional monitoring fusion module, a microclimate monitoring primary control module, a microclimate monitoring complexity evaluation module, a microclimate intelligent monitoring cloud optimization module and a microclimate intelligent monitoring updating and feedback module; the microclimate and human health index monitoring module synchronously collects multidimensional microclimate variable data of a target region and performs preprocessing to obtain a preprocessed microclimate variable data set; through the microclimate monitoring complexity evaluation module, the comprehensive complexity of edge computing is evaluated, a cloud optimization request is triggered according to high comprehensive complexity, efficient cooperation between the edge and the cloud is realized, and meanwhile, the edge end calls a local reinforcement learning model to optimize a control parameter vector, so that the system can continuously maintain high regulation and control precision in a complex environment.

Description

Technical Field

[0001] This invention relates to the field of microclimate monitoring technology, specifically to a microclimate intelligent monitoring system based on edge computing. Background Technology

[0002] With the rapid development of urbanization and infrastructure construction on the Qinghai-Tibet Plateau, the quality of the living environment in buildings on the plateau directly affects the health of typical groups such as local residents, migrant workers, and tourists. The Qinghai-Tibet Plateau has unique climatic characteristics such as high altitude, low air pressure, and low oxygen. Dramatic fluctuations in the microclimate (temperature, humidity, air pressure, oxygen concentration, etc.) of buildings can easily trigger health problems such as altitude sickness and respiratory diseases. Therefore, the research and development of intelligent microclimate monitoring and control based on monitoring results has important practical significance.

[0003] Existing intelligent monitoring systems for building microclimates on the Qinghai-Tibet Plateau fall into two categories. One type is a simple monitoring system, which cannot adapt to the unique climate of the Qinghai-Tibet Plateau. Its data collection lacks synchronization, and the large deviation in timestamps of multi-dimensional data leads to low fusion accuracy and no control functions. The other type is a monitoring and control system based on cloud processing. It uploads all collected microclimate data to the cloud, where data processing, anomaly identification, and control strategy generation are completed, and then control commands are sent to the terminal execution devices.

[0004] However, the aforementioned intelligent microclimate monitoring still has limitations: firstly, all collected data needs to be uploaded to the cloud, consuming a large amount of network bandwidth and cloud storage resources; secondly, the lack of synchronization in data collection leads to low fusion accuracy; and thirdly, it relies on cloud servers for real-time computation, which can cause excessive delays in control commands when the network is interrupted or the cloud load is too high. Although edge controllers are involved, they can only achieve simple data processing and fixed strategy control, lacking a dynamic evaluation mechanism for complexity and unable to flexibly trigger cloud optimization based on the environment and the difficulty of sensor perception, resulting in inefficient edge-cloud collaboration. Therefore, developing an intelligent microclimate monitoring and control system with real-time closed-loop control capabilities, relying on edge computing to reduce cloud dependence, and adapting to the special climate of the Qinghai-Tibet Plateau has become an urgent technical problem to be solved. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a microclimate intelligent monitoring system based on edge computing to address the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a microclimate intelligent monitoring system based on edge computing, comprising: Edge Monitoring Center: Receives and stores the results data from other modules, performs local edge real-time processing of microclimate data, issues control commands to the corresponding actuators based on the processing results, adjusts microclimate parameters, and builds a microclimate intelligent monitoring database; Microclimate and Human Health Indicator Monitoring Module: Through a sensing execution unit array deployed in the target area, the module synchronously collects microclimate data and human health indicators of the target area and performs preprocessing to obtain preprocessed microclimate variable datasets and human health indicator datasets, respectively. Microclimate Data and Human Health Analysis Module: The edge monitoring center first generates microclimate data vectors for the target area based on the microclimate variable dataset, and then obtains an environment-health correlation model based on the microclimate data vectors and the human health indicator dataset, which are stored in the edge monitoring center and the cloud platform respectively. Microclimate monitoring primary control module: The edge monitoring center obtains the primary control results of microclimate variables based on the deviation between the target area microclimate data vector and the target value vector, as well as the health correction weights of microclimate variables, and sends them to the implementing agency, which must meet the dual rule verification. Microclimate monitoring complexity assessment module: Based on the edge monitoring center's sensing execution array of the target area, the target area microclimate data vector, and the human health indicator dataset, it calculates the comprehensive complexity of intelligent microclimate monitoring and triggers cloud optimization requests based on the comprehensive complexity results; Microclimate Intelligent Monitoring Cloud Optimization Module: Based on cloud optimization requests, reinforcement learning algorithms are used to collaboratively optimize the control parameters and human health early warning deviation in the primary control module of microclimate monitoring, obtain the optimized control parameter vector, and send it to the edge monitoring center; Microclimate Intelligent Monitoring, Update and Feedback Module: The edge monitoring center updates the destandardized control parameter vectors of all microclimate variables to the module containing the parameters, and at the same time feeds back the update operation results to the management terminal for human-computer interaction.

[0007] The technical effects and advantages of this invention are as follows: 1. This invention, through a microclimate and human health indicator monitoring module, integrates multi-dimensional microclimate monitoring and constructs an environment-health correlation model suitable for the Qinghai-Tibet Plateau, realizing the linkage analysis of microclimate parameters and human health indicators, integrating health protection needs into microclimate regulation strategies, and improving the accuracy of data fusion. 2. This invention completes data fusion, human health data processing, and primary regulation locally at the edge. Only optimization requests for complex scenarios that meet the comprehensive complexity are uploaded to the cloud, avoiding the network latency of uploading all data to the cloud. This reduces the response time of microclimate parameter regulation from the second level in the cloud to the millisecond level at the edge, thereby reducing control command latency and improving control response efficiency. 3. This invention uses a microclimate monitoring complexity assessment module to calculate the overall complexity at the edge. Based on the high overall complexity, it triggers cloud optimization requests, reducing cloud dependency and achieving efficient collaboration between the edge and the cloud. At the same time, the edge calls a local reinforcement learning model to optimize the control parameter vector, forming a collaborative control process of local closed loop and optimization iteration, ensuring that the system maintains high control accuracy in complex environments. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the overall process of the present invention.

[0009] Figure 2 This is a flowchart illustrating the primary control module for microclimate monitoring in this invention. Detailed Implementation

[0010] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0011] Please see Figure 1 As shown, the present invention provides a microclimate intelligent monitoring system based on edge computing, including an edge monitoring center, a microclimate and human health indicator monitoring module, a microclimate multi-dimensional monitoring fusion module, a microclimate monitoring primary control module, a microclimate monitoring complexity assessment module, a microclimate intelligent monitoring cloud optimization module, and a microclimate intelligent monitoring update and feedback module. The edge monitoring center is connected to all other modules. The microclimate multi-dimensional monitoring fusion module is connected to the microclimate and human health indicator monitoring module and the microclimate monitoring primary control module, respectively. The microclimate monitoring complexity assessment module is connected to the microclimate intelligent monitoring cloud optimization module, and the microclimate intelligent monitoring update and feedback module is connected to the microclimate monitoring primary control module and the microclimate intelligent monitoring cloud optimization module, respectively.

[0012] Edge Monitoring Center: Receives and stores the results data from other modules, performs local edge real-time processing of microclimate data, issues control commands to the corresponding actuators based on the processing results, adjusts microclimate parameters, and builds a microclimate intelligent monitoring database; This embodiment specifically describes the microclimate monitoring system, which is adapted to the unique climate of the Qinghai-Tibet Plateau, including temperature, humidity, oxygen concentration, wind speed, light intensity, and particulate matter concentration. Core indicators for human health monitoring include blood oxygen saturation (SpO2), heart rate, systolic / diastolic blood pressure, and respiratory rate, covering typical health risk-related indicators of the Qinghai-Tibet Plateau. The actuators include oxygenation equipment, ventilation equipment, humidification equipment, sunshade and UV protection equipment, dehumidification equipment, heating equipment, and backup execution modules, adapting to the environmental control needs of the low-oxygen, dry, and strong-radiation environment of the Qinghai-Tibet Plateau. Each device primarily controls a microclimate variable (referred to as its master control variable), for example: the master control variable for a heater is temperature, for a humidifier it is humidity, and for a ventilation fan it may be CO2 concentration. The edge monitoring center is implemented by an edge computing gateway deployed within the target area (within buildings on the Qinghai-Tibet Plateau). This gateway has multiple sensor interfaces, actuator control interfaces, network communication modules, and runs the various functional modules described in this invention.

[0013] Microclimate and Human Health Indicator Monitoring Module: Through a sensing execution unit array deployed in the target area, it synchronously collects microclimate data and human health indicators of the target area and performs preprocessing to obtain preprocessed microclimate variable datasets and human health indicator datasets, including: A1: Microclimate Variables Dataset A1.1: The sensing execution unit array deployed within the target area consists of n distributed sensing nodes. Each sensing node integrates a sensor corresponding to the microclimate variable to be monitored. It synchronously collects the raw data D of the j-th type of microclimate variable collected by the i-th sensing node at time t, according to a preset frequency (e.g., once every 30 seconds, which can be remotely adjusted via the edge monitoring center). i,j (t), i=1,2,...,n, j=1,2,...,m, where n is the number of sensor nodes and m is the number of microclimate variables; the data acquisition timestamp deviation is ≤ preset value (e.g., 500ms); D i,j (t)=S i,j (t)+ε i,j (t), S i,j (t) represents the true value of the j-th type of microclimate variable, ε i,j (t) represents the noise from the collection of the j-th type of microclimate variable, which follows a normal distribution, ε i,j (t)~N(0,σ j 2 ), σ j For the sensor accuracy threshold of the j-th type of variable (e.g., temperature σ) j =±0.3℃, humidity σ j =±2%RH, etc.); This embodiment specifically describes the deployment of sensor nodes in the target area (within buildings on the Qinghai-Tibet Plateau, including living rooms, bedrooms, and office areas) at preset intervals (e.g., 5-10m) to ensure no blind spots in the target area. The microclimate variables to be monitored include temperature, humidity, air pressure, wind speed, light intensity, PM2.5 / PM10 particulate matter concentration, and oxygen concentration (for the low-oxygen environment of the plateau). The corresponding sensors include temperature and humidity sensors, air pressure sensors, wind speed sensors, light intensity sensors, and particulate matter concentration sensors. For the special environment of the Qinghai-Tibet Plateau, the sensor selection meets the requirements of wide-temperature operation (-55℃ to +85℃), UV resistance, low-oxygen environment, and waterproofing.

[0014] A1.2: Raw data D for the j-th type of microclimate variable i,j (t) Perform preprocessing operations, including outlier detection and noise filtering, to obtain the preprocessed j-th type of microclimate variable data D. i,j y (t), traversing all microclimate variables of all sensing nodes and performing preprocessing operations to obtain the preprocessed microclimate variable dataset D(t), which is then fed back to the edge monitoring center; outlier detection and noise filtering operations are used to remove the collected noise ε. i,j (t); The outlier detection: Based on the multi-dimensional microclimate variable data collected by all sensor nodes at a set time period (e.g., the first hour) before time t according to the collection frequency (e.g., if t=10:00, the time range is 9:00 to 9:59:30, the collection frequency is 1 time / 30 seconds, and the data volume = number of nodes m × collection frequency × 1 hour (e.g., if n=12, m=5, collection frequency 1 time / 30 seconds, then the data volume in 1 hour = 12 × 5 × 120 = 7200 records)), outliers are removed using the 3σ criterion, and the same type of data from adjacent sensor nodes is used to fill in the gaps using linear interpolation; The noise filtering: The dataset after outlier detection is filtered by mean filtering with a sliding window size of W (the sliding window covers the data collected at time t + the data collected W-1 times before, e.g., if the window size W=6, it includes the 6 data points t-5, t-4, t-3, t-2, t-1, and t); A2: Human Health Indicator Dataset: Raw data on human health indicators, including core indicators such as blood oxygen saturation, heart rate, blood pressure, and respiratory rate, are collected through fixed monitoring terminals (pulse oximeters, heart rate meters, blood pressure monitors, etc.) deployed in buildings on the Qinghai-Tibet Plateau and wearable devices (smart bracelets, pulse oximeters, etc.) worn by typical populations. The raw data is preprocessed (using the same methods as for microclimate variable data preprocessing) to generate a preprocessed human health indicator dataset H(t) = [H1(t), H2(t), ..., H...]. k (t)],H k(t) represents the k-th type of human health indicator, where k is the number of health indicators, and is fed back to the edge monitoring center; Microclimate Data and Human Health Analysis Module: The edge monitoring center first generates a microclimate data vector for the target area based on the microclimate variable dataset. Then, based on the microclimate data vector and the human health indicator dataset, it obtains an environment-health correlation model, which is stored in both the edge monitoring center and the cloud platform, including: B1: Edge monitoring center calls fusion function E j (t), for each microclimate variable data D in the preprocessed microclimate variable dataset D(t). i,j y (t), and the corresponding reliability coefficient γ of the sensing node. i,j We perform weighted fusion of the discrete variables to generate a microclimate data vector E(t) = [E1(t), E2(t), ..., E...]. j (t),...,E m [(t)],E j (t) represents the data after fusing the j-th type of microclimate variables obtained through the fusion function, where m is the number of microclimate variables, and it is stored in the data cache unit of the edge monitoring center; B2: The above D i,j y (t) represents the preprocessed data of the j-th type of microclimate variable, and n1_j represents the confidence coefficient γ of the j-th type of microclimate variable. i,j The number of effective sensor nodes obtained; taking temperature variable j=1 as an example, the number of effective sensor nodes n1_j=3 (i=1,2,3) is the input parameter, and the preprocessed temperature data D 1,1 y (t) = 21.5℃, D 2,1 y (t) = 22.3℃, D 3,1 y (t) = 21.8℃, node reliability coefficient γ 1,1 =0.9 (full weight), γ 2,1 =0.45 (half weight), γ 3,1 =1.0 (full weight), numerator calculation (21.5×0.9)+(22.3×0.45)+(21.8×1.0)=51.185℃, denominator calculation 0.9+0.45+1.0=2.35, fused temperature data E1(t)=51.185℃ / 2.35≈21.78℃; B3: Before deploying the sensor corresponding to the j-th type of microclimate variable, place it in a standard environmental chamber (constant temperature and humidity standard source) to obtain the standard value S. j refThe measured value D of the variable at the i-th sensing node i,j Calculate the static calibration error ΔD i,j =|S j ref -D i,j |;Based on ΔD i,j The maximum permissible error ΔD for the j-th type of microclimate variable jmax and the range of confidence coefficients for microclimate variables [γ] min ,γ max ] (e.g., [γ] min =0,γ max =1.0], offline calibration parameters. Since the reliability coefficient is a relative reliability index and is independent of the physical dimensions of variables such as temperature, humidity, and air pressure, all microclimate variables share the same set of reliability intervals and calculation rules. There is no need to set upper and lower limits according to the variable type. The upper limit of 1.0 represents the ideal state of complete reliability, no error, and consistency with the standard true value. The lower limit means that sensors with errors exceeding the limit are directly eliminated, and the weight approaches 0. If γmin is too high (e.g., >0.6), nodes with low to medium accuracy but still effective will be excessively eliminated. Based on the linear negative correlation between sensor static calibration error and reliability, the initial reliability index γ of the j-th type of microclimate variable is calculated. i,j 0 γ i,j 0 =max(0,1-(ΔD i,j / ΔD jmax )), ΔD jmax =3×|σ j |,σ j The sensor accuracy threshold for the j-th type of variable; when the sensor is running online, the mean μ of the j-th type of microclimate variable is obtained from n sensing nodes. j (t), calculate the data consistency coefficient η of the j-th type of microclimate variable at the i-th sensing node. i,j (t)=max[0,1-|μ j (t)-D i,j (t)| / (k1×δ j )],δ j The standard deviation of the historical data of the variable for a set time period before time t is defined, and k1 is an adjustable sensitivity coefficient, usually taken as 2≤k1≤3, such as k=3; then the sensor status of the variable at the sensing node is collected in real time by the edge monitoring center, including the relative scaling factor f of the power supply voltage deviating from the rated value. v The communication packet loss factor f is collected at a preset frequency within a set time period before time t. p The factor f of the signal strength RSSI RSSI Calculate the sensor health status coefficient f of this variable. i,j (t), f i,j(t)=⌊10×[f v ×(1-f p )×f RSSI ]+0.5⌋ / 10, where ⌊⌋ is the floor function, f v =max(0,1-|V-V0| / V0), where V and V0 are the actual supply voltage and the rated operating voltage, respectively, and f p The ratio of the number of lost microclimate variables to the total number of microclimate variables that should have been received is used when the RSSI value is greater than or equal to the first preset threshold RSSI_th1 (e.g., −75 dBm). RSSI =1, when the second preset threshold RSSI_th2 (e.g., −95dBm) ≤ RSSI < RSSI_th1, f RSSI =(RSSI-RSSI_th2) / (RSSI_th2-RSSI_th1), when RSSI<RSSI_th2, f RSSI =0; based on η i,j (t) and f i,j (t) Obtain the online dynamic reliability correction coefficient ζ for the j-th type of variable at the i-th sensor node. i,j (t), ζ i,j (t)=a1×η i,j (t)+a2×f i,j (t), where a1 and a2 represent the corresponding weights, for example, a1=0.6 and a2=0.4. The weights a1 and a2 are determined based on the historical data consistency coefficient and sensor health status coefficient using the entropy weight method or the AHP method (based on scenario experience); based on γ... i,j 0 and ζ i,j (t) Obtain the final credibility index γ of the j-th type of variable of the i-th sensor node. i,j (t)=γ i,j 0 ×ζ i,j (t), if γ i,j (t) < unavailable threshold γ th (e.g., 0.4–0.5), then directly remove the microclimate variable data and corresponding sensor nodes at that node at that time. If γ th ≤γ i,j (t)<γ max Participate in fusion based on confidence level; if γ i,j (t)≥γ max Then the credibility coefficient is taken as γ. max Traverse all sensor nodes to obtain the number of effective sensor nodes n1_j for the j-th type of microclimate variable participating in the fusion; taking oxygen concentration sensor i=2 and variable j=2 as an example, the input parameter static calibration: standard value S2 ref =20.95%Vol, measured value D 2,2=20.3%Vol, σ²=±0.5%Vol, confidence range [γ] min =0,γ max =1.0], Online data: Node mean μ2 = 20.7%Vol, D 2,2 (t) = 20.4%Vol, δ2 = 0.3%Vol, k1 = 3, sensor state f v =0.95、f p =0.05 and f RSSI =1.0, weights a1=0.6 and a2=0.4, calculate the static calibration error ∆D 2,2 =|20.95-20.3|=0.65%Vol, maximum permissible error ΔD 2max =3×|±0.5|=1.5%Vol, initial confidence level γ 2,2 0 =max(0,1-(0.65 / 1.5))=0.567, data consistency coefficient η 2,2 (t)=max[0,1-|20.7-20.4| / (3×0.3)]≈0.667, sensor health state coefficient f 2,2 (t) = ⌊10 × [0.95 × (1 - 0.05) × 1.0] + 0.5⌋ / 10 = 0.9, dynamic correction coefficient ζ 2,2 (t) = 0.6 × 0.667 + 0.4 × 0.9 = 0.7602, the final credibility index γ 2,2 (t)=0.567×0.7602≈0.431 (≥γ th =0.4, with full weights participating in the fusion); B4: At T time points (e.g., once per minute, for 1 hour, T=60), for any microclimate variable E j (t) and H k (t) Perform Z-score standardization on each, and calculate the normalized E. j 1 (t) and H k 1 (t) Absolute value of Pearson correlation coefficient |r j,k |, filter out|r j,k |≥ Set the first significance threshold r th1 (e.g., |r) j,k The microclimate variable j corresponding to |≥0.7) yields the set of significant microclimate influencing variables L. k L k ={j||r j,k |≥r th}; For each normalized health indicator H k 1(t), with the filtered set of microclimate variables L k As input, a PLSR (Partial Least Squares Regression) model is constructed. PLSR addresses collinearity among variables by extracting latent variables (principal components). The formula is as follows: H k y (t) represents the health indicator value predicted by the model, β k,0 The intercept term is the regression coefficient β. k,j The solution is obtained using the nonlinear iterative partial least squares (NIPALS) algorithm or the kernel partial least squares (PLSR) method. This algorithm is well-known in the field. The core steps include: extracting latent variables (principal components, i.e., microclimate variables) from the independent and dependent variable matrices respectively, obtaining the latent variables that maximize the covariance, and updating the residual matrix until the preset accuracy requirement is met or the maximum number of principal components is reached; the number of principal components A is determined through ten-fold cross-validation, with minimizing the predicted residual sum of squares (PRESS) as the criterion, and β... k,j The coefficient of influence of microclimate variable j on health index k; after the PLSR model is constructed, the coefficient of determination R is used. 2 The model error is evaluated; if R... k 2 With a value ≥0.8, the PLSR model can be directly deployed and applied as an environment-health association model. If the preset R... 2 Threshold (e.g., R) 2 =0.6)≤R k 2 <0.8, the model is deployable, but real-time monitoring needs to be strengthened. If R k 2 If the value is less than 0.6, the model's cloud-based optimization process is triggered, which involves expanding the model to the second significance threshold r. th2 (e.g., |r) j,k |≥0.5)), supplement microclimate variables, if R after supplementing variables k 2 If the result is still <0.6, perform 10-fold cross-validation again, expand the search range for the number of principal components to [1, 2Amax], select the new number of principal components A that minimizes PRESS, and retrain the model until R < 0.6. k 2 ≥0.6; Please see Figure 2 As shown, the microclimate monitoring primary control module: Based on the deviation between the target area microclimate data vector and the target value vector, and the health correction weights of microclimate variables, the edge monitoring center obtains the primary control results of microclimate variables and sends them to the implementing agency. These results must meet dual-rule verification, including: C1: Data E obtained by fusing the j-th type of microclimate variables from the microclimate data vector of the target area. j(t) and the corresponding pre-set target value E j,tar By comparing (t), the deviation e of the j-th type of microclimate variable is obtained. j (t), e j (t)=E j,tar (t)-E j (t), if the absolute deviation of each microclimate variable |e j (t)|Enters the tolerance threshold range of each variable and exceeds the preset minimum steady-state holding time T. st (For example, temperature tolerance thresholds of ≤0.6℃ (2×0.3℃) and humidity of ≤4%RH (2×2%RH), etc., the tolerance thresholds should be greater than the accuracy threshold of the sensor used and the standard deviation of the natural fluctuation of the environmental parameters under steady-state conditions. Typically, the larger of the two values ​​is taken as 2 to 3 times to ensure that the controller responds to real environmental changes rather than measurement noise. The steady-state holding time is 3 to 5 minutes.) This indicates good intelligent monitoring; conversely, through PID control algorithms, the deviation e... j (t) performs independent variable operations, where each variable has an independent proportionality coefficient K. p,j Integral coefficient K I,j and differential coefficient K d,j (First, the ZN method is used to calculate the pre-control quantity U by considering the dynamic characteristics of the target area object (such as the time constant τ of temperature regulation and the lag time τ0), and then it is obtained through on-site experimental debugging and fine-tuning. It can be optimized and updated through the cloud during operation.) j (t) (represents the expected change required to bring the main control variable j closer to the target value, where the proportional coefficient is dimensionless, the integral coefficient is in units of 1 / time t, and the differential coefficient is in units of time t, such as seconds); Simultaneously, based on the system's built-in altitude-physiological index reference database (this database establishes reference intervals according to three dimensions: altitude, age group, and gender, integrating high-altitude medicine research results, including: reference intervals for blood routine tests (RBC, HGB, HCT, etc.) for healthy individuals at 1400-4100m in western Sichuan plateau, reference ranges for blood indicators for Tibetans at altitudes above 4500m in Ali region of Tibet, reference ranges for white blood cell parameters for healthy individuals at 2260m in Xining region, and altitude-stratified reference values ​​for heart rate and blood oxygen saturation (based on NIH review data), etc.), the system adopts a dual-track early warning mechanism: for short-term stayers (e.g., stay duration ≤ 30 days), the upper limit of the normal range of various human health indicators is preset based on the group reference values ​​of the same altitude group (3000~4500 meters), age group (31-45 years old), and gender [H k,low H k,highFor long-term stays (e.g., stays > 30 days), based on individual baseline values, the data collection period is as follows: starting from the cumulative number of days set at the plateau (e.g., day 20), health indicators are collected daily in a resting state upon waking for a set number of times (e.g., 3 times), with a set interval between each collection (e.g., 10 minutes). The average value is taken as the individual baseline value, and it is recalculated periodically (e.g., every 90 days). The upper limit of the individual warning threshold is [H]. k,low =max(H k,low Individual baseline × (1 - allowable fluctuation range)), H k,high =min(H k,high The permissible fluctuation range is calculated as follows: [Individual baseline × (1 + allowable fluctuation range)], where the allowable fluctuation range is calculated from the mean μ and standard deviation σ of any indicator from healthy population data. The allowable fluctuation range = 1.96 × σ (corresponding to the 95% confidence interval). If any measured health indicator is outside the normal range, the actual health indicator is compared with the warning threshold in the preset warning rules. Based on the comparison result, a corresponding warning level is issued. The warning levels range from low to high, including Level 1, Level 2, and Level 3 warnings. The response actions are respectively management terminal prompt, local audible and visual alarm, and emergency notification to medical staff. The warning information includes the abnormal health indicator and its corresponding value, the warning level, and the response action. The warning threshold in the preset warning rules is a different health indicator warning level preset based on altitude, age group, gender, and medical knowledge. For example, Level 1 warning is H... k If the value exceeds the threshold but the deviation is <0.2, a Level 2 warning is issued if the deviation is 0.2 ≤ deviation <0.5, and a Level 3 warning is issued if the deviation is ≥0.5 or H. k Reaching the absolute danger threshold (based on the absolute danger threshold of each health indicator obtained in the field of medical research); for altitude or age profile systems not covered by the altitude-physiological index reference database, piecewise linear interpolation is used to estimate: R(h,a,s)=R(h1,a1,s)+[(R(h2,a2,s)-R(h1,a1,s)) / (h2-h1,a2-a1)]×(h-h1,a-a1), where R(h,a,s) is the reference value corresponding to altitude h, age a, and sex s; This embodiment requires specific explanation of the target values ​​for the j-th type of microclimate variable. These are pre-configured locally at the edge, remotely written to the cloud, and loaded by the system according to altitude / time period, representing a "configurable dynamic threshold." At the system's factory gate, based on medical knowledge and altitude, default suitable ranges for each variable are built-in as initialization baseline values, such as: altitude 3000–4500 meters, temperature: 16–18℃ in winter and 20–22℃ in summer, humidity 30–50%RH, oxygen concentration 31.5%, and light intensity 300 lx–500 lx, etc. Then, the user… Target values ​​can be manually entered via cloud platform or local touchscreen, overriding default values. Secondly, the system has a built-in altitude scheduling module for the target area, automatically switching target values ​​according to altitude. For example, for crop monitoring: 06:00–18:00 crop growth period: T=25℃, H=65%; 18:00–06:00 dormancy period: T=18℃, H=55%, etc., with preset segmented thresholds. Finally, in each intelligent monitoring cycle, the system executes: reading the current time, matching the target value group corresponding to the altitude, and loading E... j,tar (t) is the target value for the current period.

[0015] C2: Based on the number of microclimate variables m, an m×m dimensional coupling influence coefficient matrix C is predefined, where the diagonal elements of the matrix are c. jj =1 (for the variable's own gain), and the off-diagonal element c in the matrix. jJ Let J be the coupling influence coefficient of variable J on variable j (i.e., the intensity and direction of the interference caused by the action of the actuator of the J-th type of microclimate variable on the j-th type of variable), J≠j; then based on the pre-control quantity U of the subvariables. J (t) and c jJ Calculate the weighted sum ΔU of the coupling effects of all variables J on variable j. j (t), U J (t) represents the pre-control quantity of variable J, and the final control quantity UU after coupling of the j-th type of microclimate variable is obtained. j (t), UU j (t)=W j (t)×[c jj ×U j (t)+ΔU j (t)],W j (t) represents the health-corrected weights of microclimate variable j; W j (t) Through health indicators H k (t) When the warning is triggered, the coefficient β of the PLSR model is used. k,j and the set of variables with significant influence on microclimate L k Calculate the effect of microclimate variable j on H k Contribution α of (t) k,j (t), ej (t) represents the bias of the j-th type of microclimate variable. , λ k The importance weights of the preset health indicator k are determined (based on medical common sense, health indicators are directly scored according to their impact on human body monitoring, and then normalized to obtain the weights, such as heart rate = 0.3, SpO2 = 0.5, etc.). The sum of the weights of all health indicators is 1. When indicator k issues an alert, I... k (t)=1, otherwise=0, K is the number of warning indicators; traverse all microclimate variables to obtain the final control vector UU(t) of the microclimate variables, that is, the primary control result of the microclimate variables, UU(t)=[UU1(t),UU2(t),...,UU j (t),...,UU m [(t)], where m is the number of microclimate variables; taking oxygen concentration variable j=2 and health indicator blood oxygen saturation k=1 as an example, the input parameters are: blood oxygen warning state I1(t)=1, importance weight of health indicator 1 λ1=0.5, set of significant microclimate influence variables L1={2,4} (j=2: oxygen concentration, j=4: air pressure), oxygen concentration E2(t)=18.5%Vol, target value E 2，tar =21%Vol, average value at T time points = 21%Vol, standard deviation = 2%Vol, air pressure E4(t) = 55kPa, target value E 4，tar =60kPa, mean =60kPa, standard deviation =5kPa, PLSR regression coefficient β 1,2 =0.8, β 1,4 =0.2, calculate the standardized E2. 1 (t) = (18.5 - 21) / 2 = -1.25, E 2，tar 1 =0, e2 1 (t) = -1.25 - 0 = -1.25, and similarly, E4 1 (t)=-1, E 4，tar 1 =0, e4 1 (t) = -1 - 0 = -1, calculate the contribution numerator (j = 2): 0.8 × (-1.25) = -1.0, denominator |0.8 × (-1.25) + 0.2 × (-1)| = 1.2, contribution α 1,2 (t)=-1.0 / 1.2≈-0.833, W2(t)=1+(0.5×-0.833×sign(85%-90%)×1)≈1.417; In this embodiment, it should be specifically explained that the coupling effect matrix C is predetermined through physical mechanism analysis and standard operating condition experimental calibration. First, the physical mechanism determines the coupling direction, such as temperature rise leading to humidity decrease (negative coupling), enhanced light leading to temperature rise (positive coupling), ventilation leading to CO2 concentration decrease (negative coupling), etc. jJ >0, positive coupling, the control action of J will cause j to change in the same direction, c jJ <0, negative coupling, the control action of J will cause j to change in the opposite direction, c jJ =0, indicating no direct coupling between the two types of variables; then, standard operating condition experimental calibration is performed, including experimental design, data acquisition, and coupling coefficient calculation; Experimental design: In a controlled environment (such as an experimental roc), all other variables are fixed, and only a step change is applied to the J-th type variable to obtain the microclimate change Δu of the control input. J (e.g., temperature increases by 1℃, humidity increases by 1%RH, etc.); Data acquisition: Continuously monitor the changes of all m controlled variables (temperature, humidity, CO2, etc.), for each controlled variable D j , calculate its average rate of change D within the time window Δt. j,avg (t), D j,avg (t)=|D j (t)-D j (t-Δt)| / Δt; Set a steady-state threshold D for each variable. j,th This threshold is typically determined based on the measurement noise level and the material properties of the variables (e.g., temperature change rate ≤ 0.1℃ / min, humidity change rate ≤ 0.5% / min, etc.). When the change rates D of all controlled variables... j,avg (t) simultaneously satisfies D j,avg (t)≤D j,th When the timer continues to exceed the preset minimum steady-state hold time T, a steady-state timer is started. st If the time is 3-5 minutes, the system is determined to have entered a new steady state, and the values ​​D of each controlled variable at this time are recorded. j ss (t), calculate the steady-state change ΔD j =D j ss (t)-D j in (t), D j in (t) represents the initial steady-state value before the step disturbance is applied. Calculate the initial coupling coefficient c. jJ =ΔD j / Δu J J≠j, c jJ The unit is the ratio of the units of the two variables. Each experiment is repeated a specified number of times (e.g., 3 times). After removing outliers, the average value is taken as the final initial value c. JjWhen J=j, c Jj =1; For example, the effect of heating on humidity: when the temperature increases by 1℃ (Δu J This will cause a 1.5% decrease in RH (ΔD). j That is, the influence coefficient c of temperature J on humidity j. jJ =-1.5 (%RH / ℃), the effect of humidification on temperature: when humidity increases by 1%RH (Δu J This will cause a slight temperature increase of 0.05℃ (ΔD). j That is, the influence coefficient c of humidity J on temperature j. jJ =0.05(℃ / %RH), the current PID calculates: required temperature increase: U1(t) = 2.0℃, required humidification: U2(t) = 3.0%RH, and the weighted ΔU is used to calculate the coupled effect of temperature J on humidity j. j (t) = -1.5(%RH / ℃) × 2.0℃ = -3.0%RH, where ΔU is the weighted sum of the coupled effects of humidity J and temperature j. j (t)=0.05(℃ / %RH)×3.0%RH=0.15℃.

[0016] C3: The final control quantity UU after coupling the j-th type of microclimate variable in the final control quantity vector UU(t) of microclimate variables. j (t) is issued to the corresponding implementing agency, and must simultaneously satisfy the verification of the control quantity limit rule for the implementing agency corresponding to the j-th type of microclimate variable and the conflict rule between implementing agencies; the control quantity limit rule: UU j (t)=clip(UU j (t),UU j,min ,UU j,max ), UU j,min and UU j,max These are the minimum and maximum adjustment amplitude limits for the j-th actuator within a single control cycle (typically 30s, 60s, 1min, and 5min; 60s is commonly used for microclimate control), respectively, with clip() being the amplitude limiting function. The adjustment amplitude limits are based on the physical rate of change of the variable VV per unit time and the system control cycle T. c The product of these factors yields the theoretical adjustment range (e.g., Δ). 理论 =VV×T c Then, multiply the theoretical adjustment range by the safety reduction factor (usually 0.8 to 0.9) to obtain the maximum adjustment range limit UU. j,max Minimum adjustment range limit UU j,min The theoretical adjustment range is 0.1 to 0.2; if the adjustment range exceeds the limit, the process involves re-measuring the variable, recalculating the deviation, re-outputting the PID, re-limiting and verifying the range, and re-controlling. Ultimately, the absolute deviation of the actual microclimate variable, |e|, is monitored. j(t)| meets the judgment rule of being close to the corresponding target value, and the human health indicator returns to the normal range and maintains the minimum steady state for a duration of T. st Stop control; the proximity judgment rule: the absolute deviation of each microclimate variable |e j (t)|Enters the tolerance threshold range of each variable and exceeds the preset minimum steady-state holding time T. st The conflict rule between the actuators is: B ≤ b max B is the constraint matrix, which consists of the on / off signals of conflicting actuators corresponding to variables. These signals include an "on" value (1) and a "off" value (0). Conflicting actuators cannot be activated simultaneously; the higher-priority actuator is activated according to rules. Each row corresponds to a conflict rule for an actuator. max =1; the priority selection rule: the absolute deviation |e| after coupling of the j-th type of microclimate variables j The ratio of (t)| to the corresponding tolerance threshold multiplied by the health correction weight W of that variable. j (t), where a larger result value indicates a higher priority for the corresponding actuator. For example, it is prohibited to simultaneously require heating and cooling, to simultaneously require heat preservation and ventilation cooling, and to simultaneously require humidification and ventilation dehumidification. This embodiment requires specific explanation of the adjustment range limit, taking temperature as an example. In a standard test environment that is closed, windless, free from external interference, and without solar radiation, the heating actuator is run at full capacity for one control cycle, and the actual temperature increase is measured to obtain the measured upward adjustment limit. Conversely, the cooling / ventilation actuator is run at full capacity for one control cycle, and the actual temperature decrease is measured to obtain the measured downward adjustment limit. For example, if the temperature change rate V ≈ 1℃ / min, and the control cycle T... c =1min, theoretical maximum adjustment range limit =1×1=1℃; the control quantity limit rule takes temperature as an example, the system limits the temperature adjustment range within a single control cycle to no more than ±0.9℃, that is: if the PID calculates the output temperature control quantity UU j (t) = +2.6℃, the absolute value exceeds the upper limit of adjustment, and the output after limiting is +0.9℃; if the PID calculates the output UU j (t) = -2.6℃, the absolute value exceeds the lower limit, and the output after limiting is -0.9℃; if the PID calculates an output of +0.4℃, which is within a reasonable range, the original value remains unchanged after limiting; the conflict rules between actuators take the temperature control actuator as an example, such as a heater (u1), a refrigerator (u2), and a ventilation fan (u3) as three actuators; constraint matrix , The allowed state (1,0,0) turns on only the heater; the priority selection state (1,1,0) turns on the heater first and then the chiller.

[0017] Microclimate Monitoring Complexity Assessment Module: This module calculates the overall complexity of intelligent microclimate monitoring using the edge monitoring center's sensing execution array based on the target area, the target area's microclimate data vector, and the human health indicator dataset. Based on the overall complexity result, it triggers a cloud-based optimization request. It also assesses the number of effective sensor nodes n1 for the j-th type of variable in the target area's sensing execution array and the corresponding reliability coefficient γ. i,j The computational complexity of the sensing node is C. N (t); then, based on the fused data E of each type of microclimate variable in the target area microclimate data vector over the nearest N1 time period (N1 time period can be 3 days, 5 days, 10 days, etc.), j σ(t) j (t), the maximum permissible error ΔD of the j-th type of microclimate variable jmax The computational environment complexity C E (t); the incidence rate R of abnormalities based on human health indicators ab (t) and health risk warning level, calculate the health monitoring complexity C. H (t); based on C N (t), C E (t) and C H (t) Calculate the overall complexity C(t). If the number of consecutive control cycles is set (e.g., 3 times), C(t) ≥ the corresponding threshold C. th (e.g., 0.7) and the absolute deviation of at least one type of microclimate variable |e j (t)| If the tolerance threshold is exceeded and the steady-state maintenance time is not reached, or if human health indicators are abnormal and related to microclimate parameters, a cloud optimization request will be triggered if either of the two conditions is met; otherwise, it will not be triggered. , N tot N represents the number of valid sensor nodes for all variables. tot =∑n1_j, where m is the number of microclimate variables, R ab (t) represents the number of people N whose health indicators trigger an alert at time t. ab (t) is the ratio of the total number of people monitored in the building, C H (t)=R ab (t)×W al (t), W al (t) represents the warning level coefficient. ω l For the first l The weights for different levels of early warning are: Level 1 warning ω1=1, similarly ω2=2 and ω3=3, N l (t) is the th l The number of people under Level 1 warning, where δ is a small positive number (e.g., 10). -6 C(t) = b1 × C N(t)+b2×C E (t)+b3×clip[C H (t) / C H,max [(t),0,1],C H,max (t) represents the maximum value in the N1-th time period, where b1, b2, and b3 are the corresponding weights, with b1 < b3 < b2, for example, b1 = 0.2, b2 = 0.5, and b3 = 0.3. This is adapted to the characteristic that the environmental fluctuations of the Qinghai-Tibet Plateau are the core complexity factor, and the weights are obtained based on historical data using the entropy weight method. The microclimate intelligent monitoring cloud optimization module: Based on the cloud optimization request, a reinforcement learning algorithm is used to collaboratively optimize the control parameters and human health early warning thresholds in the microclimate monitoring primary control module, obtaining the optimized control parameter vector and early warning threshold vector, which are then sent to the edge monitoring center. Based on the cloud optimization request, a reinforcement learning algorithm is used to define the state space S(t), S(t) = [E...]. j (t),e j (t),UU j [(t-1), C(t), H(t)], E j (t), e j (t) and U j (t-1) represents the fused data of the j-th type of microclimate variables, the deviation of the j-th type of microclimate variables, the final control quantity of the j-th type of microclimate variables in the previous control period, the comprehensive complexity, and the abnormal health index dataset, respectively; then, the action space A is defined, which is the control parameter vector θ to be optimized, A=[K p,j ,K I,j ,K d,j ,c jJ [,b1,b2,b3,TH],K p,j K I,j K d,j c jJ b1, b2, b3, and TH represent the proportional coefficient, integral coefficient, and differential coefficient of each variable, the coupling influence coefficient of the off-diagonal elements in matrix C, the sensor node perception complexity weight, the environmental complexity weight, and the deviation dataset of abnormal health indicators, respectively. Then, the optimization aims to minimize the root mean square error of microclimate deviation and the false alarm rate R of human health early warning within an optimization period T1 (e.g., 24 hours). ri Let (t) be the optimization objective, and define the optimization cost function J. j (θ), c1 and c2 represent the corresponding weights, where c1 > c2. For example, c1 = 0.6 and c2 = 0.4. R ri(t) represents the total number of false alarms and missed alarms confirmed by subsequent verification (false alarms refer to the number of times the system issued an alert, but the actual situation did not reach the corresponding alert level threshold; missed alarms refer to the number of times the actual situation reached the alert threshold but no alert was issued) and the total number of alerts N in the past T1 period. al The ratio of N al =0 when R ri (t)=0, E j,tar b (t) represents the standardized target value of the j-th type of microclimate variable at time t, E j b (t,UU j (t,θ) represents the time t and UU j The standardized actual values ​​of the j-th type of microclimate variable jointly determined by (t, θ) are then defined; next, the strategy iterative optimization function θ is defined. new.j =θ old.j -η×∇J(θ old.j ), θ new.j and θ old.j These are the optimized and updated action parameter vectors and the current action parameter vectors, respectively, ∇J(θ). old Let θ be the gradient of the cost function with respect to the current parameters, and η be the learning rate (e.g., 0.005). If the current iteration number reaches the preset maximum iteration number (e.g., 100, 200 times) or if the following condition is met for n² consecutive iterations (n² = 5 to 10 times): cost function J(θ) ≤ comprehensive threshold J li Furthermore, the decrease in J(θ) is less than or equal to the corresponding convergence threshold ∆J. th (Usually 10) -3 ~10 -5 ), J li =c1×E j,th +c2×R th E j,th The mean squared deviation threshold is typically calculated as the ratio of the square of the tolerance threshold for variable j to the variance of that variable. For example, the temperature tolerance threshold is 0.6℃, and the temperature variance is 25℃. j,th =0.36 / 25=0.0144), R th If the false alarm rate threshold for human health warnings is set (typically expected to be below 5%–10%, such as 0.05), the iteration terminates, and the parameters of the current iteration step are used as the optimized control parameter vector θ. new.j For each element in the optimized deviation dataset, the health indicator value obtained through inverse calculation must be within the medical absolute danger threshold range (e.g., heart rate ≤140 beats / min and ≥40 beats / min). This is achieved by iterating through all microclimate variables to obtain the optimized and inversely standardized control parameter vector θ for all microclimate variables. newThe data is then sent to the edge monitoring center. Taking an optimization period of T1=24 hours and a temperature variable j=1 as an example, the input parameters are: weights c1=0.6 and c2=0.4, and the standardized target value E. 1,tar b (t)=0 (mean and target values ​​are both 22℃), standardized measured value (taking 3 key time points) E1 b (1) = 0.2, E1 b (12) = -0.1, E1 b (24) = 0.3 (the sum of squared deviations at the remaining 21 time points = 1.5), total number of warnings N al =10, False alarms + false negatives = 1, R ri (t) = 1 / 10 = 0.1, calculate the sum of squared deviations = (0.2 - 0) 2 +(-0.1-0) 2 +(0.3-0) 2 +1.5 = 1.64, mean squared deviation = 1.64 / 24 ≈ 0.068, cost function J j (θ)=0.6×0.068+0.4×0.1=0.081; This embodiment specifically explains the gradient of the cost function with respect to the current parameters, which is solved using automatic differentiation techniques (based on the element-wise differentiation mechanism of well-known reinforcement learning frameworks, which has been widely used in mainstream frameworks such as TensorFlow and PyTorch, and can be directly implemented by calling the framework API). In lightweight deployment scenarios at the edge, the central difference method (numerical differentiation) can be used for approximate calculation, and the dimension of the gradient vector is consistent with the action parameter vector. Based on the cloud optimization request, a historical data packet is uploaded, which includes at least the target value vector sequence {E} sampled according to the control period within a past period T_hist (e.g., the past 48 hours). j,tar (τ)}、fused data vector sequence {E j (τ)}, Final control vector sequence: {UU j (τ,θ)}, the comprehensive complexity sequence {C(t)}, and the vector of all control parameters used at the current edge side θ old After receiving the data packet, the cloud server first performs system identification and constructs a predictive model for simulation to approximate the response from the control command UU. j (τ,θ) to microclimate variable E j The dynamic relationship of (τ); a lightweight recurrent neural network (RNN) or long short-term memory network (LSTM) is used to [E j,tar (τ),E j (τ),UU j [(τθ)] is used as input to predict E. j(τ+1); Finally, on the prediction model built in the cloud, the state at the end of the historical data packet is used as the initial state E for simulation. j (0), simulate a complete optimization cycle T1 to obtain the optimized control parameter vector.

[0018] Microclimate Intelligent Monitoring Update and Feedback Module: The edge monitoring center updates the optimized and destandardized control parameter vectors of all microclimate variables to the module containing the parameters, and simultaneously feeds back the update operation results to the management terminal for human-computer interaction; the update operation results include the parameter vector before update, the parameter vector after update, and the update time; the proportional coefficient, integral coefficient, and differential coefficient of each variable, the coupling influence coefficient and deviation dataset in matrix C are updated to the microclimate monitoring primary control module, and the sensor node perception complexity weight, environmental complexity weight, and health monitoring complexity weight are updated to the microclimate monitoring complexity assessment module; Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A microclimate intelligent monitoring system based on edge computing, characterized in that: include: Edge Monitoring Center: Receives and stores the results data from other modules, performs local edge real-time processing of microclimate data, issues control commands to the corresponding actuators based on the processing results, adjusts microclimate parameters, and builds a microclimate intelligent monitoring database; Microclimate and Human Health Indicator Monitoring Module: Through a sensing execution unit array deployed in the target area, the module synchronously collects microclimate data and human health indicators of the target area and performs preprocessing to obtain preprocessed microclimate variable datasets and human health indicator datasets, respectively. Microclimate Data and Human Health Analysis Module: The edge monitoring center first generates microclimate data vectors for the target area based on the microclimate variable dataset, and then obtains an environment-health correlation model based on the microclimate data vectors and the human health indicator dataset, which are stored in the edge monitoring center and the cloud platform respectively. Microclimate monitoring primary control module: The edge monitoring center obtains the primary control results of microclimate variables based on the deviation between the target area microclimate data vector and the target value vector, as well as the health correction weights of microclimate variables, and sends them to the implementing agency, which must meet the dual rule verification. Microclimate monitoring complexity assessment module: Based on the edge monitoring center's sensing execution array of the target area, the target area microclimate data vector, and the human health indicator dataset, it calculates the comprehensive complexity of intelligent microclimate monitoring and triggers cloud optimization requests based on the comprehensive complexity results; Microclimate Intelligent Monitoring Cloud Optimization Module: Based on cloud optimization requests, reinforcement learning algorithms are used to collaboratively optimize the control parameters and human health early warning deviation in the primary control module of microclimate monitoring, obtain the optimized control parameter vector, and send it to the edge monitoring center; Microclimate Intelligent Monitoring, Update and Feedback Module: The edge monitoring center updates the destandardized control parameter vectors of all microclimate variables to the module containing the parameters, and at the same time feeds back the update operation results to the management terminal for human-computer interaction.

2. The microclimate intelligent monitoring system based on edge computing according to claim 1, characterized in that: The target area microclimate data vector: The edge monitoring center calls the fusion function E j (t), for each microclimate variable data D in the preprocessed microclimate variable dataset D(t). i,j y (t), and the corresponding reliability coefficient γ of the sensing node. i,j We perform weighted fusion of the discrete variables to generate a microclimate data vector E(t) = [E1(t), E2(t), ..., E...]. j (t),...,E m [(t)],E j (t) represents the data obtained by fusing the j-th type of microclimate variables through the fusion function, where m is the number of microclimate variables, and the data is stored in the data cache unit of the edge monitoring center.

3. The microclimate intelligent monitoring system based on edge computing according to claim 2, characterized in that: The reliability coefficient of the sensing node: Before deployment, the sensor corresponding to the j-th type of microclimate variable is placed in a standard environmental chamber, and the standard value S is obtained. j ref The measured value D of the variable at the i-th sensing node i,j Calculate the static calibration error ΔD i,j =|S j ref -D i,j |;Based on ΔD i,j The maximum permissible error ΔD for the j-th type of microclimate variable jmax and the range of confidence coefficients for microclimate variables [γ] min ,γ max ] Calculate the initial confidence index γ of the j-th type of microclimate variable. i,j 0 When the sensor is running online, it acquires the mean μ of the j-th type of microclimate variable from n sensing nodes. j (t), calculate the data consistency coefficient η of the j-th type of microclimate variable at the i-th sensing node. i,j (t); then the sensor status of this variable at the sensing node is collected in real time through the edge monitoring center, including the relative scaling factor f of the power supply voltage deviating from the rated value. v The communication packet loss factor f is collected at a preset frequency within a set time period before time t. p The factor f of the signal strength RSSI RSSI Calculate the sensor health status coefficient f of this variable. i,j (t); based on η i,j (t) and f i,j (t) Obtain the online dynamic reliability correction coefficient ζ for the j-th type of variable at the i-th sensor node. i,j (t); based on γ i,j 0 and ζ i,j (t) Obtain the final credibility index γ of the j-th type of variable of the i-th sensor node. i,j (t).

4. The microclimate intelligent monitoring system based on edge computing according to claim 1, characterized in that: The primary control module for microclimate monitoring includes: fusing the j-th type of microclimate variables from the microclimate data vector of the target area into data E. j (t) and the corresponding pre-set target value E j,tar By comparison, the deviation e of the j-th type of microclimate variable is obtained. j (t), if the absolute deviation of each microclimate variable |e j (t)|Enters the tolerance threshold range of each variable and exceeds the preset minimum steady-state holding time T. st This indicates that the intelligent monitoring is working well; otherwise, the PID control algorithm can be used to address the deviation e. j (t) Perform independent variable operations to obtain the independent variable pre-control quantity U. j (t).

5. The microclimate intelligent monitoring system based on edge computing according to claim 4, characterized in that: The primary control module for microclimate monitoring also includes: pre-defining an m×m dimensional coupling influence coefficient matrix C based on the number of microclimate variables m, where the diagonal elements c in the matrix are... jj =1, the off-diagonal element c in the matrix jJ Let J be the coupling influence coefficient of variable J on variable j, where J≠j; then, based on the pre-control quantity U of the discrete variables... J (t) and c jJ Calculate the weighted sum ΔU of the coupling effects of all variables J on variable j. j (t), to obtain the final control quantity UU after coupling of the j-th type of microclimate variable. j (t), UU j (t)=W j 1 (t)×[c jj ×U j (t)+ΔU j (t)],W j 1 (t) represents the normalized health correction weight of microclimate variable j; by iterating through all microclimate variables, the final control vector UU(t) of the microclimate variables is obtained.

6. The microclimate intelligent monitoring system based on edge computing according to claim 5, characterized in that: The primary control module for microclimate monitoring further includes: coupling the final control quantity UU of the j-th type of microclimate variable in the final control quantity vector UU(t) of the microclimate variables. j (t) is issued to the corresponding implementing agency, and must simultaneously satisfy the verification of the control quantity limit rule for the implementing agency corresponding to the j-th type of microclimate variable and the conflict rule between implementing agencies; the control quantity limit rule: UU j (t)=clip(UU j (t),UU j,min ,UU j,max ), UU j,min and UU j,max These are the minimum and maximum adjustment amplitude limits for the j-th type of actuator within a single control cycle, respectively, with clip() being the amplitude limiting function. If the adjustment amplitude exceeds the limit, the output is directly output after limiting. In the next cycle, the variable is remeasured, the deviation is recalculated, the PID is re-output, the amplitude limiting verification is performed again, and the control steps are repeated. Finally, when the absolute deviation of the actual microclimate variable is monitored, |e j (t)| meets the judgment rule of being close to the corresponding target value, and the human health indicator returns to the normal range and maintains the minimum steady state for a duration of T. st Stop control; the conflict rule between the actuators: satisfying B≤b max B is the constraint matrix, which consists of the on / off signals of conflicting actuators corresponding to variables. These signals include an "on" value (1) and a "off" value (0). Conflicting actuators cannot be activated simultaneously; the higher-priority actuator is activated according to rules. Each row corresponds to a conflict rule for an actuator. max The value is 1.

7. The microclimate intelligent monitoring system based on edge computing according to claim 1, characterized in that: The microclimate monitoring complexity assessment module includes: the number n1 of effective sensor nodes for the j-th type of variable in the perception execution array of the target area and the corresponding reliability coefficient γ. i,j The computational complexity of the sensing node is C. N (t); then based on the fused data E of each type of microclimate variable in the target area microclimate data vector over the N1 time period. j σ(t) j (t), the maximum permissible error ΔD of the j-th type of microclimate variable jmax The computational environment complexity C E (t); the incidence rate R of abnormalities based on human health indicators ab (t) and health risk warning level, calculate the health monitoring complexity C. H (t); based on C N (t), C E (t) and C H (t) Calculate the overall complexity C(t). If the number of consecutive control cycles C(t) ≥ the corresponding threshold C th And the absolute deviation of at least one type of microclimate variable |e j (t)| If the tolerance threshold is exceeded and the steady-state maintenance time is not reached, or if the human health indicators are abnormal and related to microclimate parameters, a cloud optimization request will be triggered if either of the two conditions is met; otherwise, it will not be triggered.

8. The microclimate intelligent monitoring system based on edge computing according to claim 1, characterized in that: The microclimate intelligent monitoring cloud optimization module includes: based on cloud optimization requests, using a reinforcement learning algorithm, defining a state space S(t), S(t) = [E...]. j b (t),e j b (t),UU j b (t-1), C b (t),H b [(t)], where the elements in S(t) are the standardized data of the fused j-th type of microclimate variable, the deviation of the j-th type of microclimate variable, the final control quantity of the j-th type of microclimate variable in the previous control period, the comprehensive complexity, and the abnormal health index dataset, respectively; then, the action space A is defined, which is the control parameter vector θ to be optimized, A=[K p,j b ,K I,j b ,K d,j b ,c jJ b [b1,b2,b3,TH], where A contains the standardized proportional coefficient, integral coefficient, and differential coefficient for each variable, the coupling influence coefficient of matrix C, the sensor node perception complexity weight, the environmental complexity weight, the health monitoring complexity weight, and the deviation dataset for abnormal health indicator warning levels. Then, with the optimization objective of minimizing the root mean square error of microclimate deviation and the false alarm rate of human health warnings within the optimization period T1, the optimization cost function J is defined. j (θ); secondly, define the strategy iterative optimization function θ. new.j If the current iteration count reaches the preset maximum iteration count or the following condition is met in n² consecutive iterations: cost function J(θ) ≤ comprehensive threshold J li Furthermore, the decrease in J(θ) is less than or equal to the corresponding convergence threshold ∆J. th The iteration is terminated, and the parameters of the current iteration step are used as the optimized control parameters and the human health early warning threshold vector θ. new.j By iterating through all microclimate variables, the optimized and inversely standardized control parameter vector θ is obtained. new And it was sent to the edge monitoring center.

Images