Cooperative control method and device for measurable, displayable, adjustable and controllable environment system

Abstract

The invention relates to the technical field of environment monitoring, and provides a collaborative control method and device for a measurable, displayable, adjustable and controllable environment system, and the method comprises the steps: collecting an original data set of a monitoring region, carrying out the preprocessing, and constructing a multi-dimensional environment data knowledge graph containing time-space correlation attributes; through a neural network learning environment parameter and health risk space-time correlation characteristic, outputting a healthy human settlement environment comprehensive evaluation result based on a fuzzy comprehensive evaluation matrix of adaptive threshold adjustment; inputting the comprehensive evaluation result of the healthy human settlement environment into a Kriging interpolation algorithm, generating an indoor four-dimensional space-time dynamic thermodynamic diagram in combination with environment parameter space-time distribution characteristics, generating a health risk prediction thermal layer based on risk weight superposition, and forming a visual composite atlas; outputting a regulation and control strategy by taking energy consumption, equipment cost and a health risk value as optimization targets; and controlling the household cooperative equipment. According to the invention, predictive control is carried out on the environmental equipment, and the accuracy and effectiveness of environmental regulation and control are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of environmental monitoring, and particularly to a collaborative control method and device for an environmental system that can be measured, displayed, adjusted, and controlled. Background Art

[0002] With the improvement of the requirements for the quality of the living environment, the construction of a healthy living environment has become an important research direction. The living environment system is a comprehensive environmental monitoring and regulation system for the family living space. Through the intelligent device collaborative control network, the system can accurately adjust various environmental parameters of the living space and provide a healthy living environment solution for family users.

[0003] However, traditional living environment monitoring methods mostly use linear models or simple statistical analyses, which cannot effectively capture the spatio-temporal variation characteristics and their complex correlations of environmental parameters. In particular, it is more difficult to evaluate the health impacts under the interaction of multiple parameters. In addition, the existing regulation strategies for the living environment system are mostly reactive controls, lacking predictability and foresight, unable to pre-identify potential health risks and take preventive measures, and at the same time, the collaborative control mechanism between various environmental adjustment devices is not perfect, making it difficult to achieve balance optimization under multi-objective constraints. Summary of the Invention

[0004] The present invention provides a collaborative control method and device for an environmental system that can be measured, displayed, adjusted, and controlled to solve the defects of the prior art.

[0005] The present invention provides a collaborative control method for an environmental system that can be measured, displayed, adjusted, and controlled, including: S1: Based on the octave space environmental parameters, collect the original data set of the monitoring area; S2: Preprocess the original data set and construct a multi-dimensional environmental data knowledge graph containing spatio-temporal correlation attributes; S3: According to the multi-dimensional environmental data knowledge graph, learn the spatio-temporal correlation characteristics between environmental parameters and health risks through a neural network, and based on the fuzzy comprehensive evaluation matrix adjusted by an adaptive threshold, output a comprehensive evaluation result of a healthy living environment including risk prediction weights; S4: Input the comprehensive evaluation result of the healthy living environment into the Kriging interpolation algorithm, generate an indoor four-dimensional spatio-temporal dynamic heat map in combination with the spatio-temporal distribution characteristics of environmental parameters, and generate a health risk prediction heat layer based on the superposition of risk weights to form a visual composite map integrating health risk prediction; S5: Use the health risk prediction heat layer parameters in the visual composite map as the constraint conditions of the multi-objective genetic algorithm, construct a Pareto front solution set with energy consumption, equipment cost, and health risk values as the optimization objectives, and solve the Pareto front solution set to output a regulation strategy; S6: Parse the regulation strategy into a device control instruction set, and perform indoor environmental parameter adjustment through the home collaborative device control network.

[0006] The present invention also provides a collaborative control device for a measurable, displayable, adjustable, and controllable environmental system, including: Collection module: used to collect the original data set of the monitored area based on the octave space environmental parameters; Multi-dimensional environmental data knowledge graph generation module: used to preprocess the original data set and construct a multi-dimensional environmental data knowledge graph including spatio-temporal correlation attributes; Evaluation module: used to learn the spatio-temporal correlation characteristics between environmental parameters and health risks through a neural network according to the multi-dimensional environmental data knowledge graph, and output a comprehensive evaluation result of a healthy living environment including risk prediction weights based on a fuzzy comprehensive evaluation matrix adjusted by an adaptive threshold; Visualization compliance graph generation module: used to input the comprehensive evaluation result of the healthy living environment into the Kriging interpolation algorithm, generate a four-dimensional spatio-temporal dynamic heat map of the indoor environment in combination with the spatio-temporal distribution characteristics of environmental parameters, and generate a health risk prediction heat layer based on the superposition of risk weights to form a visualization composite graph integrating health risk prediction; Optimization module: used to use the health risk prediction heat layer parameters in the visualization composite graph as health constraint conditions for a multi-objective genetic algorithm, construct a Pareto front solution set with energy consumption, equipment cost, and health risk values as optimization objectives, and solve the Pareto front solution set to output a regulation strategy; Control module: used to parse the regulation strategy into a device control instruction set, and perform indoor environmental parameter adjustment through the home collaborative device control network.

[0007] A collaborative control method and device for a measurable, displayable, adjustable, and controllable environmental system provided by the present invention collects environmental parameters in the octave space through a hierarchical and zonal deployment strategy, realizing all-round and multi-dimensional perception of the environmental information in the living space, overcoming the limitations of single indicators and scattered collection in traditional monitoring methods, and establishing a unified and systematic environmental parameter collection system; secondly, a heterogeneous data processing pipeline is used to preprocess the original data and construct a multi-dimensional environmental data knowledge graph containing spatio-temporal correlation attributes, which not only solves the problem of heterogeneous data fusion, but also establishes a complex correlation network among environmental parameters, laying a solid foundation for subsequent analysis; subsequently, the spatio-temporal correlation characteristics between environmental parameters and health risks are learned through a neural network. The system cleverly utilizes the advantages of deep learning algorithms in processing time series data and complex pattern recognition, and can identify non-linear relationships and long-term dependence characteristics that are difficult to capture by traditional linear models, so as to more accurately evaluate the health impacts under the interaction of multiple parameters. The combination of this neural network structure and a fuzzy comprehensive evaluation matrix with adaptive threshold adjustment realizes the dynamic optimization of evaluation criteria, enabling the system to perform personalized adjustments according to the health needs of different family members; the Kriging interpolation algorithm shows unique advantages in processing spatially distributed data. The four-dimensional spatio-temporal dynamic heat map generated by this algorithm intuitively presents the spatio-temporal distribution law of environmental parameters. Combining with the health risk prediction heat layer generated by risk weight superposition, a visual composite map integrating health risk prediction is formed, greatly enhancing the interpretability and intuitiveness of the system; in multi-objective optimization, by constructing a Pareto front solution set with energy consumption, equipment cost, and health risk value as optimization objectives, the best balance point can be sought among health, economy, and energy efficiency, and the generated regulation strategy is scientific and feasible; finally, the regulation strategy is parsed into a device control instruction set and executed through a home collaborative device control network, realizing the collaborative optimization control of multiple devices, effectively avoiding the parameter imbalance problem that may be caused by traditional single-device control, and greatly improving the accuracy and effectiveness of environmental regulation.

[0008] Generally speaking, through the deep integration of artificial intelligence algorithms and environmental science, the present invention constructs an intelligent, personalized, and predictive method for monitoring and regulating the human settlement environment, providing a scientific and efficient health human settlement environment solution for household users. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0010] Figure 1Schematic flowchart of a collaborative control method for a measurable, displayable, adjustable, and controllable environmental system provided by an embodiment of the present invention; Figure 2 Schematic structural diagram of a collaborative control device for a measurable, displayable, adjustable, and controllable environmental system provided by an embodiment of the present invention. Detailed implementation manners

[0011] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention, and they should not be construed as limiting the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without creative efforts shall fall within the scope of protection of the present invention. In the description of the present invention, it should be understood that the terms used are only for the purpose of description and cannot be construed as indicating or implying relative importance.

[0012] The embodiments of the present invention will be described below with reference to the drawings.

[0013] As Figure 1 shown, the present invention provides a collaborative control method for a measurable, displayable, adjustable, and controllable environmental system, including: S1: Based on the octave space environmental parameters, collect the original data set of the monitoring area.

[0014] In step S1 of the present invention, first, the original data set is collected based on the octave space environmental parameters, which involves multi-dimensional and all-round environmental information collection. The octave space environmental parameters refer to dividing the human living environment into eight key dimensions, including temperature and humidity environment, light environment, air quality environment, volatile organic compound environment, sound environment, electromagnetic environment, biological environment, and air flow environment, which together constitute a complete indoor healthy human living environment evaluation system. Each dimension has a direct or indirect impact on human health.

[0015] Among them, the original data set in step S1 includes: Temperature and humidity data, light data, PM2.5 data, VOC data, sound pressure level data, electromagnetic radiation data, microbial concentration, air flow velocity parameter.

[0016] In a specific embodiment, taking a family residence as an example, the residence area is 120 square meters, including six functional areas: a living room, a master bedroom, a secondary bedroom, a study, a kitchen, and a bathroom. Based on the hierarchical and zonal deployment strategy, 32 sensor nodes are arranged in each functional area. During the data collection process, the raw data recorded by the temperature and humidity sensors in the master bedroom area shows that during the night sleep period (23:00 - 7:00), the temperature is stable at 24.5 ± 0.5 °C, and the relative humidity is 45 ± 3%. The system automatically adjusts the sampling interval to 10 minutes. After getting up in the early morning (7:00 - 8:00), the temperature rapidly rises to 26.8 °C, and the humidity increases to 58%. The system detects that the change rate of 1.2 °C / 30 min exceeds the preset threshold of 0.5 °C / 30 min and automatically shortens the sampling interval to 2 minutes, accurately capturing the rapid change process of the environmental parameters. The PM2.5 sensor in the living room area records that the PM2.5 concentration rapidly rises from 12 μg / m³ to 45 μg / m³ during the cooking period from 18:00 to 19:00. The system captures this mutation and triggers the collaborative collection of the VOC sensor, and finds that the TVOC concentration synchronously rises to 0.58 mg / m³, indicating the impact of kitchen activities on the air quality in the living room. The sound pressure level data in the study area shows that during the working period (9:00 - 12:00), the background noise L90 is 32 dB(A), and the occasional sound event LAmax reaches 68 dB(A). The spectral analysis shows that the main energy is concentrated in the frequency band of 500 - 2000 Hz, and the characteristics are consistent with the outdoor traffic noise. After subsequent preprocessing of these raw data, it provides a data basis for the health assessment and intelligent regulation of the living environment.

[0017] S2: Preprocess the original data set and construct a multi-dimensional environmental data knowledge graph containing spatio-temporal correlation attributes.

[0018] Among them, step S2 further includes: S21: Detect outliers in the original data set and use the cubic spline interpolation algorithm to fill and reconstruct the missing data segments to generate a cleaned environmental data set.

[0019] Furthermore, in step S21, outliers are detected and missing data is filled in the original data set. Outlier detection is to identify and process data points that deviate from the normal range in the data set. The method adopted in the present invention is a dual detection mechanism combining the statistical box plot method and the density-based local outlier factor (LOF) algorithm, which improves the accuracy and robustness of outlier detection; the cubic spline interpolation algorithm used for missing data filling is a high-precision interpolation method for time series data. Compared with linear interpolation and polynomial interpolation, cubic spline interpolation can better retain the curvature characteristics of the data while maintaining data continuity.

[0020] S22: Perform correlation analysis on the multi-dimensional parameters in the environmental dataset based on a graph neural network to construct a spatio-temporal correlation matrix between the parameters.

[0021] In step S22, a graph neural network (GNN) is used to perform correlation analysis on the cleaned environmental dataset to construct a spatio-temporal correlation matrix between the parameters. A graph neural network is a type of deep learning model specifically designed to process graph-structured data and is suitable for analyzing multi-dimensional data with complex correlation relationships. In this solution, the environmental parameters in the octave space are represented as nodes V in the graph G(V, E), and the correlation relationships between the parameters are represented as edges E. Specifically, a variant of the graph attention network (GAT) is adopted, which can adaptively learn the importance weights between different nodes and is applicable to the analysis of non-linear and dynamic correlation relationships between environmental parameters.

[0022] Specifically, in the correlation analysis process, first, each environmental parameter in the octave space is represented as a feature vector, which includes the temporal and spatial features of the parameter; then, an initial graph structure is constructed, with each parameter as a node, and the initial edges are set based on physical knowledge and prior correlations; next, the correlation strength between nodes is learned through a multi-layer graph attention mechanism; then, the node representations are updated by aggregating neighbor information; finally, the attention weights are extracted from the trained model to construct a spatio-temporal correlation matrix between the parameters, and the matrix elements in the matrix represent the correlation strength between two parameters in the spatio-temporal dimension.

[0023] S23: Perform dynamic grid partitioning on the octave space according to the spatio-temporal correlation matrix, and generate spatio-temporal feature vectors for multiple grid cells through a spatio-temporal encoder.

[0024] The dynamic grid partitioning in step S23 is a technique that adaptively adjusts the spatial grid density and time window size according to the spatio-temporal distribution characteristics of environmental parameters. The dynamic grid takes into account the parameter gradient changes and correlation strength, and uses a denser grid partitioning in areas where the parameters change violently or the correlation strength is high.

[0025] The specific process includes two parts: adaptive spatial grid partitioning and dynamic time window segmentation. Adaptive spatial grid partitioning is implemented based on a quadtree (two-dimensional) or octree (three-dimensional) structure. Starting from the root node of the entire monitoring space, it is determined whether to further divide according to the coefficient of variation of the environmental parameters within the grid and the correlation strength threshold; dynamic time window segmentation determines the optimal time window size according to the temporal change rate of the parameters. A smaller time window (such as 1 minute) is used during rapid parameter changes, and a larger time window (such as 10 minutes) is used during stable periods.

[0026] The spatio-temporal encoder is a neural network structure that encodes the multi-dimensional features of grid cells into fixed-dimensional vectors, adopting a hybrid architecture composed of a temporal convolutional network (TCN) and a spatial encoding module. The temporal convolutional network processes temporal features, capturing patterns at different time scales through multi-layer dilated convolutions. The receptive field of the dilated convolution grows exponentially with the number of layers, effectively modeling long-term dependencies. The spatial encoding module combines positional encoding and spatial attention mechanisms to transform the three-dimensional coordinates (x, y, z) of grid cells into high-dimensional spatial feature vectors. Finally, the temporal feature vector and the spatial feature vector are merged into a comprehensive spatio-temporal feature vector through a fusion layer, which contains the complete spatio-temporal semantic information of the grid cells.

[0027] S24: Through the TransR algorithm, embed the spatio-temporal feature vector for entity relationship, generating a multi-dimensional environmental data knowledge graph.

[0028] The TransR algorithm in step S24 is a knowledge graph embedding algorithm. The core idea is to map entities and relationships to different semantic spaces, and realize the representation learning of relationships through a spatial transformation matrix. TransR can express complex relationships more accurately, such as one-to-many, many-to-one, and many-to-many relationships, and is suitable for modeling complex associations between environmental parameters.

[0029] In the process of entity relationship embedding, first, each grid cell and its spatio-temporal feature vector are regarded as entities in the knowledge graph. The entity types include spatial entities (grid cells), parameter entities (environmental parameters), and temporal entities (time windows); then define the relationship set, including "is located at" (spatial relationship), "measurement value is" (parameter relationship), "occurs at" (temporal relationship), "affects" (causal relationship); then construct an initial triple set, such as (grid G1, measurement value is, temperature 24.5°C), (temperature rise, affects, humidity drop), etc.; then apply the TransR algorithm for embedding learning. The algorithm learns entity vectors, relationship vectors, and relationship-specific mapping matrices by minimizing the energy function, where the mapping matrix maps entities from the entity space to the relationship space; finally, by adopting a negative sampling optimization strategy, negative triples are constructed for each positive triple for contrast learning. The optimization goal is to make the energy of positive triples low and the energy of negative triples high.

[0030] Taking the processing of temperature, humidity, and PM2.5 parameter data in the living room area of a home as an example, when performing dynamic grid division based on the association matrix, since the temperature and humidity parameters are relatively uniform in space, while PM2.5 changes drastically near the ventilation opening (coefficient of variation CV = 0.68 > threshold 0.5), the grid division is denser (0.5m × 0.5m) in the area near the ventilation opening, and larger grids (1m × 1m) are used in the area far from the ventilation opening; in terms of the time window, it is found that the temperature change rate increases after opening the window for ventilation at 15:00, and the time window automatically shrinks from 10 minutes to 2 minutes. For the time-series data of environmental parameters of each grid cell, features are extracted through a spatio-temporal encoder. For example, the spatio-temporal feature vector of grid cell G33 = [temperature gradient, humidity change rate, PM2.5 diffusion coefficient, spatial position encoding, time period feature].

[0031] Subsequently, a knowledge graph is constructed through the TransR algorithm, taking grid cells, environmental parameters, and time points as entities, and defining relationships such as "is located at", "measurement value is", "affects", etc. For example, the triple (G33, measurement value is, temperature 25.2°C), (temperature rise, affects, humidity drop), and / or (window-opening behavior, causes, PM2.5 concentration drop), etc. Through relational embedding learning, the rationality score of the triple is calculated. For example, the score of (window-opening behavior, causes, PM2.5 concentration rise) is low, indicating that the relationship is unreasonable, while the score of (window-opening behavior, causes, PM2.5 concentration drop) is high, which conforms to common sense. The finally generated multi-dimensional environmental data knowledge graph contains the spatio-temporal distribution characteristics of environmental parameters, the correlation relationships between parameters, and the variation laws, providing a structured knowledge basis for the subsequent steps of health risk assessment and control strategy formulation.

[0032] S3: According to the multi-dimensional environmental data knowledge graph, learn the spatio-temporal correlation characteristics between environmental parameters and health risks through a neural network, and based on the fuzzy comprehensive evaluation matrix adjusted by an adaptive threshold, output the comprehensive evaluation result of a healthy living environment including risk prediction weights.

[0033] Among them, step S3 further includes: S31: Extract the temporal characteristics of the multi-dimensional environmental data knowledge graph through a gated recurrent unit to generate a dynamic evolution sequence of environmental parameters.

[0034] Furthermore, starting from step S31, the gated recurrent unit (GRU) is used to extract the temporal characteristics of the multi-dimensional environmental data knowledge graph. The gated recurrent unit is a variant of the recurrent neural network designed specifically for processing sequence data. Compared with the traditional RNN, it solves the problem of gradient disappearance in long-sequence training and has a more concise and efficient structure.

[0035] In the present invention, the input of the GRU is the time series of environmental parameters extracted from the multi-dimensional environmental data knowledge graph, specifically including the values of each environmental parameter (temperature, humidity, PM2.5, etc.) at different time points and their correlation relationships. The GRU network is designed as a multi-layer structure. The first layer receives the original parameter sequence, and each subsequent layer receives the hidden state output of the previous layer. To process time series of different lengths, the present invention adopts variable-length sequence processing techniques, including sequence padding and masking mechanisms. The hyperparameter configuration of the GRU includes a hidden layer dimension of 128, 3 layers, a dropout rate of 0.2, and the sequence length is set according to the sampling frequency, which is set to the sampling points within 24 hours in the present invention. Through GRU processing, the static parameters in the knowledge graph are transformed into dynamic evolution sequences, capturing the trends, periodicities, and mutation characteristics of environmental parameters over time. The output environmental parameter dynamic evolution sequence is a vector sequence containing time series features, and each vector represents the environmental state characterization at a specific time point.

[0036] S32: Based on the Asami five-dimensional evaluation index, perform health risk prediction on the dynamic evolution sequence of the environmental parameters using a long short-term memory network, and generate risk probability distribution maps for multiple evaluation dimensions.

[0037] Furthermore, the Asami five-dimensional evaluation index used in step S32 of the present invention is a comprehensive framework for evaluating the quality of the living environment, including five dimensions: safety, health, convenience, comfort, and sustainability. The safety index evaluates physical and chemical hazards in the environment, such as fire risks, gas leaks, and structural safety; the health index focuses on the direct impact of environmental factors on human health, such as the impact of air quality, water quality, and noise on the respiratory system, immune system, etc.; the convenience index evaluates the functionality and convenience of the living environment, such as spatial layout and ease of use of equipment; the comfort index measures the impact of environmental parameters on sensory experience and psychological comfort, such as thermal comfort, light environment comfort, and sound environment comfort; the sustainability index focuses on the energy efficiency, resource consumption, and long-term environmental impact of environmental control.

[0038] The long short-term memory network is a variant of the recurrent neural network specifically designed to handle long sequence dependency problems. It solves the gradient vanishing problem of traditional RNNs through gating mechanisms and memory units. The LSTM unit contains three gates: an input gate, a forget gate, and an output gate, as well as a memory unit. In the present invention, the LSTM network receives the dynamic evolution sequence of environmental parameters output by the GRU and constructs risk prediction models for the Asami five-dimensional indicators respectively. The LSTM network is configured as a bidirectional structure, which can capture sequence features from both the forward and backward directions simultaneously, enhancing the ability to model context dependencies. The network is designed as a deep stacked structure, including an input layer, multiple bidirectional LSTM layers, an attention layer, and a fully connected output layer.

[0039] To meet the risk prediction requirements of different dimensions, the LSTM network adopts a multi-task learning framework. The underlying feature extraction network is shared among five dimensions, but each has an independent top-level prediction head. Each prediction head outputs the risk probability distribution of the corresponding dimension, represented in a classification or regression manner. For example, the output of the health dimension is the probability distribution of different health risk levels (no risk, low risk, medium risk, high risk); the output of the comfort dimension is the continuous value of the PMV-PPD thermal comfort index and its uncertainty estimate. Combining the outputs of all dimensions, a multi-dimensional risk probability distribution map is generated. This distribution map is a multi-dimensional tensor, representing the risk distribution at different spatial positions and different time points on each evaluation dimension.

[0040] S33: Calculate the membership degrees of multiple evaluation dimensions through a fuzzy clustering algorithm, and generate a dynamic weight allocation matrix in combination with the risk probability distribution map.

[0041] Among them, step S33 further includes: S331: Calculate the membership degrees of multiple evaluation dimensions corresponding to the Asami five-dimensional evaluation index through a fuzzy clustering algorithm, and generate an initial fuzzy relation matrix.

[0042] In S331, the present invention calculates the membership degrees of the Asami five-dimensional evaluation index through a fuzzy clustering algorithm. Fuzzy clustering is a clustering method that allows data points to partially belong to multiple clusters. Different from traditional hard clustering (such as K-means) that strictly divides each point into a single category, fuzzy clustering assigns a membership degree to each point for each category. The membership degree value is in the interval [0, 1], indicating the degree to which the point belongs to the category.

[0043] In the present invention, the input data of the FCM algorithm is the standardized scoring vector of the Asami five-dimensional index. Each spatial grid cell corresponds to a five-dimensional vector. The number of clusters is set to the number of risk levels. The output is the membership degree matrix of the grid cell for each risk level. This matrix represents the fuzzy belonging degree of each spatial position on each evaluation dimension to different risk levels, forming an initial fuzzy relation matrix.

[0044] In step S331, for the initial fuzzy relation matrix after introducing spatio-temporal constraints by adopting the improved fuzzy C-means clustering (FCM), the specific expression is: Among them, is the initial fuzzy relation matrix, is the spatio-temporal feature tensor corresponding to the risk probability distribution of the th grid cell, is the dimensional clustering center, is the risk probability distribution of the th grid cell, is the corresponding ideal risk distribution for the dimensional clustering center, is the dimension index value of the Asami five-dimensional evaluation index, represents the KL divergence, is the fuzzy index, taking , is the spatio-temporal - risk coordination coefficient, which is dynamically adjusted by the subsequent LSTM-GRU network.

[0045] S332: Use the entropy weight method to correct the weights of the risk probability distribution map to generate a dynamic weight allocation matrix.

[0046] In step S332, the entropy weight method is used to correct the weights of the risk probability distribution map. The entropy weight method is an objective weighting method based on the information entropy theory. Its basic idea is that the smaller the information entropy of an index, the greater the degree of variation of the index, and the greater the weight of the index in the comprehensive evaluation. In the present invention, the input of the entropy weight method is the risk probability values of each evaluation dimension in the risk probability distribution map, and the output is the dynamic weights of each dimension.

[0047] Specifically, considering the time-varying nature of risk prediction, the present invention introduces a risk sensitivity factor to adjust the weights. The risk sensitivity factor is a parameter determined based on the health status, age characteristics, and distribution of sensitive populations of residents, and is used to enhance the sensitivity to specific risk factors. For example, for residents with respiratory diseases, the risk sensitivity factor of air quality-related indicators will be set to a higher value; for the elderly, the risk sensitivity factor of thermal environment comfort will be correspondingly increased.

[0048] S333: Fuse the initial fuzzy relation matrix and the dynamic weight allocation matrix through the matrix Hadamard product to generate a weighted fuzzy evaluation matrix.

[0049] Among them, the expression of the weighted fuzzy evaluation matrix in step S333 is: Among them, is the finally generated weighted fuzzy evaluation matrix, is the risk sensitivity factor, is the th grid cell risk probability distribution, is the dimensional ideal risk distribution, is the total number of evaluation objects; Among them, is the dynamic weight introducing membership degree, is the initial fuzzy relation matrix, is the Hadamard product operation, is the dynamic weight corrected by the entropy weight method, is the spatiotemporal feature tensor corresponding to the risk probability distribution of the th grid cell, is the time step index, is the total time step length, is the encoded output of the spatiotemporal feature dynamic evolution sequence,

[0050] Step S333 fuses the initial fuzzy relation matrix and the dynamic weight assignment matrix through the matrix Hadamard product. The Hadamard product is the element-wise multiplication of matrices. In the present invention, the Hadamard product is used to multiply the fuzzy relation matrix and the dynamic weight matrix to obtain a weighted fuzzy evaluation matrix. This fusion method preserves the structural characteristics of the original fuzzy relation and at the same time introduces the influence of the dynamic weight, realizing an adaptive evaluation based on risk sensitivity.

[0051] S34: Online train the threshold adjuster of the LSTM-GRU hybrid neural network through the backpropagation mechanism, and output a fuzzy comprehensive evaluation matrix with adaptive threshold adjustment.

[0052] Step S34 online trains the threshold adjuster of the LSTM-GRU hybrid neural network through the backpropagation mechanism. The LSTM-GRU hybrid neural network is a composite architecture that combines the advantages of two recurrent neural networks. Among them, GRU is responsible for short-time sequence feature extraction, and LSTM is responsible for long-time sequence dependence modeling. The threshold adjuster is a component in the network architecture specifically used to dynamically adjust the risk evaluation threshold, including a fully connected layer and an adaptive normalization layer. The goal of threshold adjustment is to dynamically optimize the decision boundary of risk evaluation according to historical data and real-time feedback, improving the accuracy and adaptability of the evaluation.

[0053] S35: Non-linearly fit the fuzzy comprehensive evaluation matrix, and output the comprehensive evaluation result of the healthy human settlement environment with risk prediction weights.

[0054] In step S35, a non - linear fitting is performed on the fuzzy comprehensive evaluation matrix to output the comprehensive evaluation result of the healthy human settlement environment with risk prediction weights. The non - linear fitting is a process of converting a high - dimensional fuzzy evaluation matrix into an interpretable evaluation result. The methods adopted in the present invention are multi - layer perceptron (MLP) or support vector regression (SVR). In the present invention, the input of the non - linear fitting is the weighted fuzzy evaluation matrix, and the output is the comprehensive evaluation result including the overall score and the detailed scores of each dimension. The overall score adopts a standard score system of 0 - 100, reflecting the overall health level of the living environment; the scores of each dimension maintain the Asami five - dimensional framework, providing detailed scores for safety, health, convenience, comfort, and sustainability. In particular, the evaluation result also includes risk prediction weights, which represent the predicted risk change trends of each evaluation dimension within a future time window (such as 24 hours, 72 hours), and are used to guide preventive control measures.

[0055] S4: Input the comprehensive evaluation result of the healthy human settlement environment into the Kriging interpolation algorithm, generate an indoor four - dimensional spatio - temporal dynamic heat map in combination with the spatio - temporal distribution characteristics of environmental parameters, and generate a health risk prediction heat layer based on the superposition of risk weights to form a visual composite map integrating health risk prediction.

[0056] Among them, step S4 further includes: S41: Perform spatial autocorrelation modeling on the improved Kriging interpolation algorithm based on variogram analysis to generate a covariance matrix of the octave space.

[0057] First, in step S41, spatial autocorrelation modeling is performed on the improved Kriging interpolation algorithm based on variogram analysis. The Kriging algorithm takes into account the spatial autocorrelation and anisotropy of data and can provide the best linear unbiased estimate (BLUE) and estimate error evaluation. The improved Kriging interpolation algorithm mentioned in the present invention is optimized for the distribution characteristics of indoor environmental parameters, and a multi - level variogram structure and an anisotropy adjustment mechanism are introduced.

[0058] The variogram is the core component of Kriging interpolation, which describes the relationship between spatial distance and data similarity. In the present invention, variogram analysis is performed separately for each environmental parameter in the octave space, considering the spatial distribution characteristics and physical meanings of the parameters. The specific steps include: first, calculating the experimental variogram, that is, directly calculating the variance at different distances from the measured data; then, performing theoretical model fitting, and common models include spherical model, exponential model, Gaussian model, etc.; then, determining the key parameters according to the fitting results, including nugget value (representing micro - scale variation), sill value (representing overall variation), and range (representing the influence range of spatial correlation); finally, verifying the effectiveness of the model, and evaluating the fitting effect through methods such as cross - validation.

[0059] The analysis results of the variogram of each environmental parameter form a covariance matrix, which describes the correlation structure between different positions in space. The covariance matrix C is an n×n matrix. The characteristic of the covariance matrix is symmetric positive definite, indicating the spatial law that the closer the distance, the stronger the correlation.

[0060] S42: Perform weighted correction on the covariance matrix according to the comprehensive evaluation result of the healthy living environment to obtain a risk-enhanced interpolation kernel function.

[0061] Step S42 performs weighted correction on the covariance matrix according to the comprehensive evaluation result of the healthy living environment to obtain a risk-enhanced interpolation kernel function. Specifically, it integrates risk assessment information into the spatial interpolation process, so that the interpolation result not only reflects the physical distribution of parameters, but also reflects the health risk correlation.

[0062] The risk weight calculation is based on the comprehensive evaluation result of the healthy living environment, especially the part with risk prediction weight. For each environmental parameter, calculate its risk weight: where, is the risk weight corresponding to the environmental parameter is a non-linear mapping, is the environmental parameter corresponding risk score, is the risk sensitivity, and the risk weights of each parameter form a weight vector.

[0063] The final kernel function construction is a process of fusing risk weights with the original covariance matrix. The risk-enhanced interpolation kernel function K is defined as the matrix Hadamard product of the covariance matrix and the risk adjustment matrix constructed based on the weight vector. The constructed kernel function, especially for the risk prediction part, can be adjusted by weights, and the weights decay as the prediction time increases. The risk-enhanced interpolation kernel function inherits the spatial autocorrelation characteristics of the covariance matrix and integrates health risk assessment information at the same time, realizing the transformation from the pure physical space description to the "risk perception" space description. As the kernel function of the subsequent Gaussian process regression, it directly affects the smoothness and accuracy of the interpolation result.

[0064] S43: Perform four-dimensional interpolation calculation on the spatio-temporal distribution characteristics of environmental parameters through Gaussian process regression to generate an initial spatio-temporal heat map.

[0065] ​Specifically, in step S43, Gaussian process regression is extended to a four-dimensional spatio-temporal domain. The input includes three-dimensional spatial coordinates and time coordinates, and the output is the environmental parameter value. The kernel function uses a risk-enhanced interpolation kernel function to achieve risk-aware interpolation estimation. To process multiple environmental parameters, the multi-output Gaussian process (MOGP) technology is used to consider the correlation between parameters and jointly model multiple output variables.

[0066] The specific steps of four-dimensional interpolation calculation include: First, determine the interpolation grid, set uniform or adaptive grid points in the spatial dimension, and set a fixed time interval in the time dimension; then for each grid point, apply GPR to calculate the predicted mean and variance of the environmental parameters; then construct a four-dimensional array according to the prediction results, where each element represents the parameter value at a specific spatio-temporal point; finally, use a four-dimensional interpolation function to fill the blank areas between the grids to generate a continuous spatio-temporal parameter distribution.

[0067] The initial spatio-temporal heat map is a visual representation of the four-dimensional interpolation result, including a sequence of three-dimensional spatial heat maps evolving over time. The heat map uses color coding to represent the parameter intensity. In this embodiment, a gradient color band from blue (low value) to red (high value) is used. A three-dimensional spatial heat map is generated at each time step, and a dynamic evolution sequence is formed by connecting them along the time axis. The heat map also includes isosurfaces and slice views to facilitate observing the parameter distribution from different angles.

[0068] S44: Superimpose the health risk prediction heat layer on the initial spatio-temporal heat map to generate a visual composite map.

[0069] The health risk prediction heat layer in step S44 is a visual representation of the risk distribution constructed based on the risk prediction weights generated in step S3. The risk heat layer is also a four-dimensional structure, representing the health risk levels at different spatio-temporal points. However, different from the environmental parameter heat map, the risk heat layer pays more attention to the cumulative effect and threshold effect of health hazards.

[0070] The process of constructing the risk heat layer includes: First, for each environmental parameter, calculate the risk index according to its health threshold and actual value; then consider the synergistic effect of multiple parameters to calculate the comprehensive risk index; finally, map the comprehensive risk index to a visual representation. In this embodiment, a color band from green (safe) to yellow (warning) to red (dangerous) is used.

[0071] The final overlay process is a technique that fuses the environmental parameter heat map and the risk heat layer into a single visual representation. The present invention uses transparency overlay, that is, visual fusion is achieved by adjusting the transparency of the two layers. The visual composite map that fuses the health risk prediction is finally presented in a four-dimensional interactive visualization form. Users can explore the environmental parameter distribution and the health risk prediction results from multiple angles and scales through operations such as time-axis control, spatial navigation, and parameter selection. The map also supports hotspot identification and trajectory tracking functions, automatically marking high-risk areas and tracking their spatio-temporal evolution trajectories.

[0072] S5: Use the health risk prediction heat layer parameters in the visual composite map as the constraint conditions of the multi-objective genetic algorithm, construct a Pareto front solution set with energy consumption, equipment cost, and health risk value as the optimization objectives, and solve the Pareto front solution set to output a regulation strategy.

[0073] Among them, step S5 further includes: S51: Discretize the health risk prediction heat layer parameters to generate a set of dynamic health constraint boundary conditions.

[0074] First, in step S51, discretize the health risk prediction heat layer parameters to generate a set of dynamic health constraint boundary conditions. The health risk prediction heat layer parameters refer to the numerical distribution representing the degree of health risk in the visual composite map generated in step S4, including quantitative indicators such as risk index, exceedance probability, and harm degree. These parameters exist in continuous value form and need to be converted into discrete constraint conditions that can be processed by the algorithm.

[0075] Discretization is the process of dividing the continuous parameter space into a finite number of intervals or categories, transforming a complex continuous optimization problem into a computable discrete optimization problem. In the present invention, discretization adopts a combination of the hierarchical threshold method and the grid partitioning method. The hierarchical threshold method sets multiple threshold points according to the risk level, mapping the continuous risk index to discrete risk levels. For example, the risk index R is divided into four levels: safe (R < 0.2), low risk (0.2 ≤ R < 0.5), medium risk (0.5 ≤ R < 0.8), and high risk (R ≥ 0.8). The grid partitioning method divides the three-dimensional space into discrete grids at a certain granularity, and each grid cell is assigned corresponding risk levels and constraint conditions.

[0076] The dynamic health constraint boundary condition set is a set of constraint functions that change over time, defining the allowable value ranges of environmental parameters at different times and different spatial positions. Each component corresponds to a constraint condition for an environmental parameter. The dynamic characteristics of the constraint condition set are reflected in: temporal dynamics, that is, the constraint conditions are updated over time, reflecting the temporal evolution of risk prediction; spatial dynamics, that is, the constraint conditions are differentially set at different spatial positions, considering the non-uniformity of spatial risk distribution. For example, when it is predicted that the formaldehyde concentration in a certain area will increase within the next 4 hours, the formaldehyde constraint conditions in this area will be tightened accordingly; when it is detected that the PM2.5 concentration in the bedroom area is relatively high, the PM2.5 constraint in this area will be more stringent than that in the living room area.

[0077] S52: Use the NSGA-III algorithm to perform multi-objective optimization on energy consumption, equipment cost, and health risk values to generate a non-dominated solution set.

[0078] A multi-objective optimization problem is a problem of simultaneously optimizing multiple potentially conflicting objective functions. Different from single-objective optimization, multi-objective optimization usually does not have a unique optimal solution, but there is a set of non-dominated solutions, that is, Pareto optimal solutions. The characteristic of non-dominated solutions is that any improvement in one objective function will necessarily lead to the deterioration of at least one other objective function.

[0079] In the present invention, the implementation details of the NSGA-III algorithm include a population size of 200, a maximum number of iterations of 100, a crossover probability of 0.9, a mutation probability of 0.1, using simulated binary crossover (SBX) and polynomial mutation operators. The reference points are generated using binomial distribution with a distribution parameter of 12. The final output of the algorithm is a set of non-dominated solutions, and each solution corresponds to a control strategy and its performance scores on three objectives.

[0080] Specifically, the expression of the objective function established in step S5 of the present invention is: Among them, is the energy consumption objective function, is the total number of controlled devices, is the index value of the controlled device, is the device at time instantaneous power, is the time step, is the device at time switch state, is the device cost objective function, is the cost of the device , is the device life attenuation factor, is the cumulative operating time of the device, The objective function dedicated to health and are the risk weight coefficients respectively, is the health risk proximity degree of the -th grid cell. The closer it is to 1, the higher the risk level. is the predicted value of the health risk of the -th grid cell, is the safety risk threshold.

[0081] Specifically, the expression of the constraint condition set established in step S5 of the present invention is: Among them, is the health risk constraint condition, is the indicator function, which takes 1 when the condition is met and 0 otherwise, is the dynamic risk threshold, is the global risk tolerance. For example, means allowing 10% of the grid cells to exceed the threshold, is the equipment collaboration constraint condition, is the maximum interval of equipment state switching, is the spatio-temporal validity constraint condition, is the digital twin simulation result. For example, the simulated value of the living room temperature is 26.3 °C, is the real environment parameter, is the simulation error tolerance.

[0082] S53: Based on the digital twin engine, conduct spatio-temporal trajectory simulation on the non-dominated solution set, and verify the stability of the strategy through the Monte Carlo method.

[0083] The digital twin in step S53 is a virtual replica of a physical entity or system in a digital environment. Through real-time data synchronization and model simulation, it provides predictions and optimizations for the behavior of the entity. The digital twin engine is the core computing module of digital twin technology, realizing the mapping and interaction between the physical system and the digital model.

[0084] Spatio-temporal trajectory simulation is to simulate the evolution process of environmental parameters over time and space under a specific regulation strategy in the digital twin environment. For each regulation strategy in the non-dominated solution set, the simulation engine calculates the predicted values of each environmental parameter at different spatial positions within a future time window (such as 24 hours), forming a four-dimensional spatio-temporal trajectory.

[0085] The Monte Carlo method is a numerical calculation technique based on random sampling, used to evaluate the performance stability of a model under uncertain conditions. In the present invention, the Monte Carlo method observes the sensitivity of the regulation strategy to these perturbations by randomly perturbing the input parameters (such as external meteorological conditions, occupant behavior patterns, equipment performance fluctuations, etc.) multiple times. The specific steps include: defining the set of uncertainty parameters and their probability distributions; performing multiple random samplings, generating a set of parameter values each time; conducting spatio-temporal trajectory simulations for each set of parameter values; statistically analyzing the simulation results, calculating the statistical characteristics of key performance indicators, such as mean, variance, quantiles, etc. The final stability results are used to screen and rank non-dominated solutions, eliminating the solutions that are unstable in the actual environment.

[0086] S54: Use the Pareto filter to perform crowding degree ranking on the verified solution set, and output the home environment context-aware regulation strategy.

[0087] The finally output regulation strategy is in the form of a time series control instruction set. Each instruction includes the execution time, device ID, operation parameters, and expected effect. The instruction set is sorted in chronological order to form a complete control plan. In addition, the regulation strategy also includes context adaptation rules for dynamic adjustment when the actual environment changes.

[0088] S6: Parse the regulation strategy into a device control instruction set, and execute the indoor environment parameter adjustment through the home collaborative device control network.

[0089] Among them, step S6 further includes: S61: Perform device instruction mapping on the home environment context-aware regulation strategy to generate an initial control instruction set.

[0090] In step S61, device instruction mapping is the process of converting the high-level strategy description into specific device-executable instructions, similar to the compilation from a high-level programming language to machine code. The initial control instruction set is the output result of the mapping process, containing all the device control instructions that need to be executed. Each instruction clearly defines the execution object, action, parameters, time, and conditions. The instruction set is stored in a structured data format and represented in JSON format for easy subsequent processing and transmission.

[0091] S62: Perform execution conflict detection on the initial control instruction set to generate a control instruction sequence.

[0092] Furthermore, an execution conflict refers to a situation where multiple instructions target the same device or affect the same environmental parameter. If not handled, it may lead to problems such as frequent state switching of the device, parameter oscillation, or mutual cancellation. Conflict detection is based on three dimensions: device conflict, function conflict, and effect conflict. Device conflict detection identifies multiple operation instructions for the same device within an overlapping time period; function conflict detection identifies cases where the functions of different devices overlap, such as the simultaneous operation of an air conditioner for dehumidification and a dehumidifier; effect conflict detection identifies whether the impacts of different instructions on environmental parameters cancel each other out, such as simultaneous heating and window ventilation.

[0093] S63: Perform dynamic priority scheduling on the control instruction sequence to generate a device collaborative execution queue.

[0094] Specifically, in the present invention, dynamic priority scheduling is also performed in step S63 to solve the sequence decision problem. Specifically, it is used to optimize the instruction execution order and timing to achieve better environmental regulation effects and device cooperation. The finally generated device collaborative execution queue, compared with simple static scheduling, reinforcement learning scheduling can adapt to environmental changes, optimize the device collaborative effect, and improve the regulation efficiency.

[0095] S64: Perform distributed deployment of the device collaborative execution queue through the message middleware in the home collaborative device control network to complete the adjustment of environmental parameters.

[0096] Finally, in step S64, the device collaborative execution queue is distributedly deployed through the message middleware in the home collaborative device control network to complete the adjustment of environmental parameters. The home collaborative device control network is a communication network connecting various environmental regulation devices, supporting device discovery, data exchange, and collaborative control. The message middleware is the core component in the network, responsible for message routing, protocol conversion, and device coordination.

[0097] In a specific embodiment, taking the summer indoor environmental regulation of a residential house as an example, the house is equipped with devices such as a central air conditioner (3 zones), a fresh air system, intelligent curtains, an air purifier, and a humidifier. The regulation strategy output by step S5 includes an environmental optimization plan for hot and humid weather.

[0098] First, perform device instruction mapping to convert the strategic goal of "controlling the temperature in the living room area at 24 - 26°C, the relative humidity at 45 - 55%, and the PM2.5 below 15 μg / m³" into a set of device instructions: air conditioner instruction {device, action: set temperature, parameter: 24°C, time: 12:00, duration: 8 hours}, dehumidification mode instruction {device, action: set mode, parameter: dehumidification, time: 13:30, duration: 3 hours}, fresh air instruction {device, action: set air volume, parameter: 200 m³ / h, time: 12:00, duration: 8 hours}, curtain instruction {device, action: close, time: 10:00, duration: 6 hours}.

[0099] Conflict detection finds that there is a functional overlap between the air conditioner cooling and the dehumidification mode during the period from 13:30 to 16:30. At the same time, it is detected that there is an effect conflict between the high - volume fresh air and dehumidification during the period from 13:30 to 16:30 (the fresh air introducing outdoor air may increase humidity). The conflict resolution strategy combines the air conditioner dehumidification and cooling into a "dry - cooling mode", and adds a "pre - dehumidification" parameter to the fresh air system. The optimized instruction sequence eliminates the functional conflict.

[0100] Subsequently, optimize the instruction sequence. Based on the trained DQN model in this embodiment, input the current state (temperature 28°C, humidity 68%, PM2.5 18 μg / m³). The model predicts the optimal execution order: first start to close the curtains to enclose the heat source; start the air conditioner cooling at 11:30 in advance to create a temperature gradient; start the low - volume fresh air mode at 12:30 to introduce fresh air; after the temperature drops to 26°C, then increase the fresh air volume to 200 m³ / h. This dynamic scheduling reduces the energy consumption by 12% and shortens the response time by 15 minutes compared with the static plan.

[0101] Finally, deploy and execute through the message middleware. The control system converts the optimized instruction sequence into MQTT messages, which are distributed to each device through the central message broker. For example, the air conditioner cooling instruction is sent to the topic "home / ac / livingroom / command", and the message content is in JSON format {command: "setTemp", value: 24, mode: "cool - dry", time: "11:30", duration: 480}. After each device receives the instruction, it performs the operation and sends the execution result and status feedback back to the system.

[0102] The system verifies the execution effect by monitoring the changes in environmental parameters. If it is found that the temperature in a certain area drops slowly, it automatically adjusts the corresponding air conditioner air supply angle and wind speed. The entire execution process forms a closed - loop control, continuously optimizing according to the real - time feedback, achieving precise and efficient indoor environment regulation, and effectively improving the living comfort and health.

[0103] As Figure 2 shown, the present invention also provides a collaborative control device for a measurable, displayable, adjustable and controllable environmental system, including: Acquisition module 100: configured to collect an original data set of a monitored area based on octave space environmental parameters; Multi-dimensional environmental data knowledge graph generation module 200: configured to preprocess the original data set and construct a multi-dimensional environmental data knowledge graph including spatio-temporal association attributes; Evaluation module 300: configured to, according to the multi-dimensional environmental data knowledge graph, learn spatio-temporal association features between environmental parameters and health risks through a neural network, and output a comprehensive evaluation result of a healthy human settlement environment including risk prediction weights based on a fuzzy comprehensive evaluation matrix adjusted by an adaptive threshold; Visualization compliance graph generation module 400: configured to input the comprehensive evaluation result of the healthy human settlement environment into a Kriging interpolation algorithm, generate an indoor four-dimensional spatio-temporal dynamic heat map in combination with spatio-temporal distribution characteristics of environmental parameters, and generate a health risk prediction heat layer based on risk weight superposition to form a visualization composite graph integrating health risk prediction; Optimization module 500: configured to use the health risk prediction heat layer parameters in the visualization composite graph as health constraint conditions of a multi-objective genetic algorithm, construct a Pareto front solution set with energy consumption, equipment cost and health risk values as optimization objectives, and solve the Pareto front solution set to output a regulation strategy; Control module 600: configured to parse the regulation strategy into a device control instruction set and execute indoor environmental parameter adjustment through a home collaborative device control network.

[0104] The device embodiments described above are merely illustrative, where the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative labor.

[0105] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A collaborative control method for a measurable, displayable, adjustable, and controllable environmental system, characterized in that, Including: S1: Based on the octave space environmental parameters, collect the original data set of the monitoring area; S2: Preprocess the original data set and construct a multi-dimensional environmental data knowledge graph containing spatio-temporal correlation attributes; S3: According to the multi-dimensional environmental data knowledge graph, learn the spatio-temporal correlation features between environmental parameters and health risks through a neural network, and based on a fuzzy comprehensive evaluation matrix with adaptive threshold adjustment, output a comprehensive evaluation result of a healthy living environment including risk prediction weights; S4: Input the comprehensive evaluation result of the healthy living environment into the Kriging interpolation algorithm, generate an indoor four-dimensional spatio-temporal dynamic heat map in combination with the spatio-temporal distribution characteristics of environmental parameters, and generate a health risk prediction heat layer based on the superposition of risk weights to form a visual composite map integrating health risk prediction; S5: Use the health risk prediction heat layer parameters in the visual composite map as the constraint conditions of the multi-objective genetic algorithm, construct a Pareto front solution set with energy consumption, equipment cost, and health risk value as the optimization objectives, and solve the Pareto front solution set to output a regulation strategy; S6: Parse the regulation strategy into a device control instruction set and execute the indoor environmental parameter adjustment through the home collaborative device control network.

2. The collaborative control method of a measurable, displayable, adjustable, and controllable environmental system according to claim 1, wherein The original data set in step S1 includes: Temperature and humidity data, light data, PM2.5 data, VOC data, sound pressure level data, electromagnetic radiation data, microbial concentration, air flow velocity parameters.

3. The collaborative control method of a measurable, displayable, adjustable, and controllable environmental system according to claim 1, characterized in that Step S2 further includes: S21: Detect outliers in the original data set and use the cubic spline interpolation algorithm to fill and reconstruct the missing data segments to generate a cleaned environmental data set; S22: Perform correlation analysis on the multi-dimensional parameters in the environmental data set based on a graph neural network to construct a spatio-temporal correlation matrix between parameters; S23: Dynamically divide the octave space according to the spatio-temporal correlation matrix, and generate spatio-temporal feature vectors of multiple grid cells through a spatio-temporal encoder; S24: Through the TransR algorithm, perform entity relationship embedding on the spatio-temporal feature vectors to generate a multi-dimensional environmental data knowledge graph.

4. The collaborative control method of a measurable, displayable, adjustable and controllable environmental system according to claim 1, characterized in that, Step S3 further includes: S31: Extract temporal features from the multi-dimensional environmental data knowledge graph through a gated recurrent unit to generate a dynamic evolution sequence of environmental parameters; S32: Based on the Asami five-dimensional evaluation index, perform health risk prediction on the dynamic evolution sequence of environmental parameters based on a long short-term memory network to generate risk probability distribution maps for multiple evaluation dimensions; S33: Calculate the membership degrees of multiple evaluation dimensions through a fuzzy clustering algorithm, and generate a dynamic weight assignment matrix in combination with the risk probability distribution maps; S34: Online train the threshold adjuster of the LSTM-GRU hybrid neural network through a backpropagation mechanism, and output a fuzzy comprehensive evaluation matrix with adaptive threshold adjustment; S35: Perform non-linear fitting on the fuzzy comprehensive evaluation matrix and output a comprehensive evaluation result of a healthy living environment with risk prediction weights.

5. The collaborative control method of a measurable, displayable, adjustable, and controllable environmental system according to claim 4, characterized in that, Step S33 further includes: S331: Calculate the membership degrees of multiple evaluation dimensions corresponding to the Asami five-dimensional evaluation index through a fuzzy clustering algorithm to generate an initial fuzzy relation matrix; S332: The entropy weight method is used to correct the weights of the risk probability distribution map to generate a dynamic weight allocation matrix; S333: The initial fuzzy relation matrix and the dynamic weight allocation matrix are fused through the matrix Hadamard product to generate a weighted fuzzy evaluation matrix.

6. The collaborative control method of a measurable, displayable, adjustable and controllable environmental system according to claim 5, characterized in that, The expression of the weighted fuzzy evaluation matrix in step S333 is: Among them, is the finally generated weighted fuzzy evaluation matrix, is the dynamic weight introducing the membership degree, is the risk sensitivity factor, is the risk probability distribution of the th grid cell, is the dimensional ideal risk distribution, is the total number of evaluation objects; Among them, is the initial fuzzy relation matrix, is the Hadamard product operation, is the dynamic weight corrected by the entropy weight method, is the th spatio-temporal feature tensor corresponding to the risk probability distribution of grid cells, is the time step index, is the total time step length, is the encoded output of the spatio-temporal feature dynamic evolution sequence, is the non-linear transformation of the risk time series prediction vector.

7. The collaborative control method of a measurable, displayable, adjustable and controllable environmental system according to claim 1, characterized in that, Step S4 further includes: S41: Based on variogram analysis, spatial autocorrelation modeling is performed on the improved Kriging interpolation algorithm to generate a covariance matrix in the octave space; S42: The covariance matrix is weighted and corrected according to the comprehensive evaluation result of the healthy living environment to obtain a risk-enhanced interpolation kernel function; S43: Four-dimensional interpolation calculation is performed on the spatio-temporal distribution characteristics of environmental parameters through Gaussian process regression to generate an initial spatio-temporal heat map; S44: The healthy risk prediction heat layer is superimposed on the initial spatio-temporal heat map to generate a visual composite map.

8. The collaborative control method of a measurable, displayable, adjustable and controllable environmental system according to claim 1, characterized in that, Step S5 further includes: S51: Discretize the parameters of the healthy risk prediction heat layer to generate a set of dynamic healthy constraint boundary conditions; S52: Use the NSGA-III algorithm to perform multi-objective optimization on energy consumption, equipment cost, and healthy risk values to generate a non-dominated solution set; S53: Based on the digital twin engine, spatio-temporal trajectory simulation is performed on the non-dominated solution set, and the stability of the strategy is verified through the Monte Carlo method; S54: Use a Pareto filter to sort the verified solution set by crowding degree and output the family environment context-aware regulation strategy.

9. The collaborative control method of a measurable, displayable, adjustable and controllable environmental system according to claim 1, characterized in that, Step S6 further includes: S61: Map the family environment context-aware regulation strategy to device instructions to generate an initial control instruction set; S62: Detect execution conflicts in the initial control instruction set to generate a control instruction sequence; S63: Perform dynamic priority scheduling on the control instruction sequence to generate a device collaborative execution queue; S64: Through the message middleware in the home collaborative device control network, the device collaborative execution queue is distributedly deployed to complete the adjustment of environmental parameters.

10. A collaborative control device for an environment system that can be measured, displayed, adjusted, and controlled, characterized in that, It includes: Collection module: Used to collect the original data set of the monitoring area based on the octave space environmental parameters; Multi-dimensional environmental data knowledge graph generation module: Used to preprocess the original data set and construct a multi-dimensional environmental data knowledge graph containing spatio-temporal association attributes; Evaluation module: Used to learn the spatio-temporal association characteristics between environmental parameters and health risks through a neural network according to the multi-dimensional environmental data knowledge graph, and output the comprehensive evaluation result of the healthy living environment including risk prediction weights based on the fuzzy comprehensive evaluation matrix with adaptive threshold adjustment; Visualization composite map generation module: Used to input the comprehensive evaluation result of the healthy living environment into the Kriging interpolation algorithm, generate an indoor four-dimensional spatio-temporal dynamic heat map in combination with the spatio-temporal distribution characteristics of environmental parameters, and generate a healthy risk prediction heat layer based on risk weights to form a visual composite map integrating healthy risk prediction; Optimization module: It is used to take the health risk prediction hot layer parameters in the visualized composite map as the health constraint conditions of the multi-objective genetic algorithm, construct a Pareto front solution set with energy consumption, equipment cost and health risk value as the optimization objectives, solve the Pareto front solution set, and output a regulation strategy; Control module: It is used to parse the regulation strategy into a device control instruction set and execute indoor environment parameter adjustment through the home collaborative device control network.

Images