An indoor air quality self-adaptive control method and system applying deep learning

Abstract

The application discloses an indoor air quality self-adaptive control method and system applying deep learning, and relates to the field of air quality optimization.The method comprises the following steps: constructing an air conditioning deployment grid according to air conditioning equipment deployment characteristics; constructing an indoor-outdoor linkage monitoring matrix, performing indoor air quality target mining, and obtaining an indoor air quality target matrix; performing multivariate abnormal cause-effect modeling on real-time indoor air quality data, and establishing an air quality abnormal cause-effect graph model; performing multilevel adjustment optimization on the air conditioning deployment grid to obtain an air conditioning first strategy; performing air quality trend smoothing optimization under deep learning to obtain an air conditioning second strategy; and performing multi-trend characteristic interference correction on the air conditioning deployment grid to obtain an air conditioning third strategy.The method solves the technical problems of insufficient dynamic adaptability and full-link collaborative optimization of existing self-adaptive control, and achieves the technical effect of improving the dynamic adaptability and full-link collaborative optimization level.

Description

Technical Field

[0001] This application relates to the field of air quality optimization, and in particular to an adaptive control method and system for indoor air quality using deep learning. Background Technology

[0002] Indoor air quality control is directly related to people's health and comfort, as well as the safety of their living and working environment. It is a core requirement in the fields of building environmental engineering and intelligent control, and is of great significance for improving the level of building intelligence and ensuring the quality of the living environment. Currently, indoor air quality control mainly relies on traditional sensor monitoring and fixed threshold control. Specifically, this involves deploying sensors for temperature, humidity, PM2.5, etc., to collect air quality data. When the data exceeds a preset fixed threshold, it triggers the start / stop or speed switching of equipment such as air conditioners and fresh air systems. However, existing methods do not fully consider the coupling characteristics of indoor and outdoor environments and the synergistic effects of equipment linkage, leading to insufficient control accuracy, inability to adapt to dynamic environmental changes, and a lack of prediction of equipment operating trends, which can easily cause excessive equipment wear and energy waste.

[0003] At present, the adaptive control of indoor air quality has technical problems such as insufficient dynamic adaptability and insufficient whole-chain collaborative optimization. Summary of the Invention

[0004] This application provides a deep learning-based adaptive indoor air quality control method and system. It employs several techniques: constructing an air quality control deployment grid based on the deployment characteristics of indoor air conditioning equipment; building an indoor-outdoor linkage monitoring matrix and mining indoor air quality targets to form an indoor air quality target matrix; combining the monitoring matrix and the target matrix to perform multivariate anomaly causal modeling on real-time indoor air quality data to establish an air quality anomaly causal graph model; based on the target matrix and the causal graph model, performing multi-level adjustment and optimization on the air quality control deployment grid to obtain a first air quality control strategy; using deep learning to smooth and optimize the air quality situation to generate a second air quality control strategy; and combining the multi-trend characteristics of the air quality control deployment grid to interfere and correct the second strategy, finally obtaining a third air quality control strategy. These techniques address the technical problems of insufficient dynamic adaptability and end-to-end collaborative optimization in existing adaptive indoor air quality control systems, achieving the technical effect of improving the dynamic adaptability and end-to-end collaborative optimization level of control.

[0005] This application provides an adaptive indoor air quality control method using deep learning, comprising: constructing an air control deployment grid based on the deployment characteristics of indoor air control equipment; constructing an indoor-outdoor linkage monitoring matrix, and mining indoor air quality targets based on the indoor-outdoor linkage monitoring matrix to obtain an indoor air quality target matrix; performing multivariate anomaly causal modeling on real-time indoor air quality data based on the indoor-outdoor linkage monitoring matrix and the indoor air quality target matrix to establish an air quality anomaly causal graph model; performing multi-level adjustment and optimization on the air control deployment grid based on the indoor air quality target matrix and the air quality anomaly causal graph model to obtain a first air control strategy; performing air quality situation smoothing optimization under deep learning based on the first air control strategy to obtain a second air control strategy; and performing multi-trend characteristic interference correction on the air control deployment grid based on the second air control strategy to obtain a third air control strategy.

[0006] In a possible implementation, indoor air quality target mining is performed based on the indoor-outdoor linkage monitoring matrix to obtain an indoor air quality target matrix. The following processing is then performed: indoor air quality qualified sample retrieval is performed based on the indoor-outdoor linkage monitoring matrix to obtain an indoor air quality qualified sample space; air quality index characteristics are classified in the indoor air quality qualified sample space to obtain each air quality qualified sample area; confidence evaluation and cleaning are performed based on each air quality qualified sample area to establish each air quality confidence region that satisfies the confidence constraints; central tendency analysis is performed based on each air quality confidence region to generate the indoor air quality target matrix.

[0007] In a possible implementation, multivariate anomaly causal modeling is performed on real-time indoor air quality data based on the indoor-outdoor linkage monitoring matrix and the indoor air quality target matrix to establish an air quality anomaly causal graph model. The following processes are performed: An indoor air quality matrix is ​​constructed based on the real-time indoor air quality data; anomalies are identified in the indoor air quality matrix based on the indoor air quality target matrix to generate multivariate air quality anomaly characteristics; multivariate correlation analysis is performed on the multivariate air quality anomaly characteristics based on the indoor-outdoor linkage monitoring matrix to obtain a multivariate anomaly correlation sequence; the multivariate anomaly correlation sequence guides the indoor-outdoor linkage monitoring matrix to perform causal tracing of the multivariate air quality anomaly characteristics to obtain the causal pointing relationships of each anomaly; the air quality anomaly causal graph model is constructed using the multivariate air quality anomaly characteristics as multiple air quality anomaly nodes and the causal pointing relationships of each anomaly as multiple anomaly causal edges.

[0008] In a possible implementation, the air control deployment grid is optimized through multi-level adjustment based on the indoor air quality target matrix and the air quality anomaly causal graph model to obtain a first air control strategy. The following processes are then performed: real-time operating parameters of each device within the air control deployment grid are collected to generate an air control operation grid; multi-level joint adjustment of the air control operation grid is performed based on the air quality anomaly causal graph model to obtain a first air control group; air quality adaptation evaluation is performed on the first air control group based on the indoor air quality target matrix to generate a second air control group; and energy consumption minimization is performed on the second air control group to generate the first air control strategy.

[0009] In a possible implementation, the first air control group is evaluated and optimized for air quality adaptation based on the indoor air quality target matrix to generate a second air control group. The following processes are then performed: A P-th air control scheme is extracted from the first air control group, where P is a positive integer; based on the indoor-outdoor linkage monitoring matrix and the real-time indoor air quality data, indoor air quality is predicted according to the P-th air control scheme to obtain a P-th indoor air quality sequence; a twin comparison evaluation is performed on the P-th indoor air quality sequence based on the indoor air quality target matrix to obtain a P-th air quality adaptation value; it is determined whether the P-th air quality adaptation value is greater than or equal to an air quality adaptation threshold; if the P-th air quality adaptation value is greater than or equal to the air quality adaptation threshold, the P-th air control scheme is added to the second air control group.

[0010] In a possible implementation, the air quality situation smoothing optimization under deep learning is performed based on the first air control strategy to obtain the second air control strategy, and the following processing is performed: Deep learning is performed on the indoor air quality situation record set to generate an air quality situation prediction model; the indoor-outdoor linkage monitoring matrix, the real-time indoor air quality data, and the first air control strategy are input into the air quality situation prediction model to obtain a multi-dimensional air quality situation curve; abrupt changes are identified based on the multi-dimensional air quality situation curve to obtain a multi-dimensional situation abrupt change segment set; attention is allocated to the impact of abrupt changes based on the multi-dimensional situation abrupt change segment set to determine the impact weight of each situation abrupt change; based on the impact weight of each situation abrupt change, the first air control strategy is optimized with differentiated smoothing constraints according to the multi-dimensional situation abrupt change segment set to generate the second air control strategy.

[0011] In a possible implementation, the air control deployment grid undergoes multi-trend characteristic interference correction according to the second air control strategy to obtain the third air control strategy, and the following processes are performed: simulation control is applied to the air control deployment grid according to the second air control strategy to obtain simulation datasets for each device's operation; fault trend characteristics are analyzed based on the simulation datasets for each device's operation to obtain fault trend characteristics for each device; aging trend characteristics are analyzed based on the simulation datasets for each device's operation to obtain aging trend characteristics for each device; adaptive interference influence correction is applied to the second air control strategy based on the fault trend characteristics and aging trend characteristics of each device to generate the third air control strategy.

[0012] In a possible implementation, an air conditioning deployment grid is constructed based on the deployment characteristics of indoor air conditioning equipment, and the following processes are performed: an air conditioning equipment installation grid is constructed based on the installation characteristics of the indoor air conditioning equipment; multi-head attention enhancement is performed on the air conditioning equipment installation grid based on the function characteristics and linkage characteristics of the indoor air conditioning equipment to generate the air conditioning deployment grid.

[0013] In a possible implementation, an indoor-outdoor linkage monitoring matrix is ​​constructed, and the following processing is performed: based on the sensor network, indoor space characteristic data, indoor personnel monitoring data, and outdoor environmental monitoring data are acquired in real time; the indoor space characteristic data, the indoor personnel monitoring data, and the outdoor environmental monitoring data are cleaned to generate the indoor-outdoor linkage monitoring matrix.

[0014] This application also provides an indoor air quality adaptive control system using deep learning, comprising: an air control deployment grid construction module for constructing an air control deployment grid based on the deployment characteristics of indoor air control equipment; an indoor air quality target mining module for constructing an indoor-outdoor linkage monitoring matrix and mining indoor air quality targets based on the indoor-outdoor linkage monitoring matrix to obtain an indoor air quality target matrix; a multivariate anomaly causal modeling module for performing multivariate anomaly causal modeling on real-time indoor air quality data based on the indoor-outdoor linkage monitoring matrix and the indoor air quality target matrix to establish an air quality anomaly causal graph model; a multi-level adjustment optimization module for performing multi-level adjustment optimization on the air control deployment grid based on the indoor air quality target matrix and the air quality anomaly causal graph model to obtain a first air control strategy; an air quality situation smoothing optimization module for performing deep learning-based air quality situation smoothing optimization based on the first air control strategy to obtain a second air control strategy; and a multi-trend characteristic interference correction module for performing multi-trend characteristic interference correction on the air control deployment grid based on the second air control strategy to obtain a third air control strategy.

[0015] This application proposes a deep learning-based adaptive indoor air quality control method and system. First, an air quality control deployment grid is constructed based on the deployment characteristics of indoor air conditioning equipment. Then, an indoor-outdoor linkage monitoring matrix is ​​built, and indoor air quality targets are mined based on this matrix to obtain an indoor air quality target matrix. Next, multivariate anomaly causal modeling is performed on real-time indoor air quality data based on the indoor-outdoor linkage monitoring matrix and the indoor air quality target matrix to establish an air quality anomaly causal graph model. Then, multi-level adjustment and optimization are performed on the air quality control deployment grid based on the indoor air quality target matrix and the air quality anomaly causal graph model to obtain a first air quality control strategy. Subsequently, air quality situation smoothing optimization under deep learning is performed based on the first air quality control strategy to obtain a second air quality control strategy. Finally, multi-trend characteristic interference correction of the air quality control deployment grid is executed based on the second air quality control strategy to obtain a third air quality control strategy. The proposed method and system achieve the technical effect of improving the dynamic adaptability of control and the level of end-to-end collaborative optimization. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0017] Figure 1 This is a flowchart illustrating an adaptive indoor air quality control method using deep learning, provided as an embodiment of this application.

[0018] Figure 2 This is a schematic diagram of an indoor air quality adaptive control system using deep learning, provided as an embodiment of this application.

[0019] Figure labeling: Air control deployment grid construction module 10, indoor air quality target mining module 20, multivariate anomaly causal modeling module 30, multi-level regulation optimization module 40, air quality situation smoothing optimization module 50, multi-trend characteristic interference correction module 60. Detailed Implementation

[0020] The above description is merely an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, specific embodiments of this application are given below.

[0021] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application will be provided in conjunction with the accompanying drawings. The described embodiments should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] In the following description, references to "some embodiments" describe a subset of all possible embodiments. However, it is understood that "some embodiments" can be the same or different subsets of all possible embodiments and can be combined with each other without conflict. The terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only.

[0023] This application provides an embodiment of an adaptive indoor air quality control method using deep learning, such as... Figure 1 As shown, the method includes:

[0024] Step S100: Construct an air control deployment grid based on the deployment characteristics of indoor air control equipment.

[0025] Specifically, installation characteristic data such as the installation location, equipment type, and installation height of indoor air conditioning equipment are collected. The indoor space is then divided into basic grids corresponding to the equipment installation locations—the air conditioning equipment installation grid. The functional and linkage characteristics of the equipment are extracted, such as the temperature control range of air conditioners, the purification radius of air purifiers, the linkage logic between air conditioners and humidifiers, and the collaborative relationship between fresh air systems and exhaust systems. A multi-head attention mechanism is used to strengthen the air conditioning equipment installation grid. The number of heads in the multi-head attention mechanism is set to the sum of the total number of types of functional and linkage characteristics of the equipment. Each attention head corresponds to one characteristic. By calculating the attention weights of different grid nodes and equipment characteristics, the functional range and linkage relationships of the equipment are embedded into the grid, ultimately generating an air conditioning deployment grid.

[0026] In one possible implementation, an air conditioning deployment grid is constructed based on the deployment characteristics of indoor air conditioning equipment. Step S100 further includes step S110, constructing a control equipment installation grid based on the installation characteristics of the indoor air conditioning equipment. Specifically, spatial information such as the size and shape of the indoor space is acquired using an indoor floor plan acquisition tool, and installation characteristic data such as the installation location, equipment type, installation height, and equipment number of the air conditioning equipment are acquired using equipment installation files. A spatial grid is generated, and the size of the grid cell is determined based on the indoor space size and equipment installation density. The side length of the grid cell is typically set to 1 / 3 of the minimum effective radius of the equipment to ensure accurate mapping of the equipment's effective range. The equipment installation characteristic data is associated with the corresponding grid nodes to generate a control equipment installation grid. Each node in the grid contains information such as whether equipment is installed at that node, the equipment type, and the equipment number.

[0027] Step S120: Based on the operational characteristics and linkage characteristics of the indoor air conditioning equipment, multi-head attention enhancement is applied to the air conditioning equipment installation grid to generate the air conditioning deployment grid. Specifically, operational characteristic parameters and linkage characteristic parameters of the air conditioning equipment are extracted. Operational characteristic parameters include the equipment's control range, control intensity, and control response time. Linkage characteristic parameters include the linkage triggering conditions, linkage control amplitude, and linkage delay time between equipment. A multi-head attention model is constructed. The model input is the node feature matrix of the air conditioning equipment installation grid. Node features include node coordinates, equipment installation status, and equipment type. The number of heads in the multi-head attention mechanism is set to the sum of the number of operational characteristic types and the number of linkage characteristic types. Each attention head corresponds to one characteristic. By calculating the similarity between the query vector, key vector, and value vector, the attention weight of each node for different characteristics is obtained. The attention weights and node features are weighted and fused to update the features of the grid nodes, generating an air conditioning deployment grid that includes the operational range and linkage relationships of the equipment.

[0028] Step S200: Construct an indoor-outdoor linkage monitoring matrix, and mine indoor air quality targets based on the indoor-outdoor linkage monitoring matrix to obtain an indoor air quality target matrix.

[0029] Specifically, a sensor network deployed indoors and outdoors collects real-time data on indoor space characteristics, indoor occupant monitoring, and outdoor environmental monitoring. Data cleaning is performed, including missing value imputation, outlier removal, and data standardization, integrating the collected multi-source data into a structured indoor-outdoor linked monitoring matrix. Based on this matrix, a qualified indoor air quality sample space is obtained through qualified sample retrieval. The sample space is then classified according to its indicator characteristics. The samples are cleaned using confidence level evaluation to establish confidence regions that meet the confidence level constraints. Through central tendency analysis, the mean, median, and other central tendency indicators of each confidence region are calculated to generate an indoor air quality target matrix.

[0030] In one possible implementation, an indoor-outdoor linked monitoring matrix is ​​constructed. Step S200 further includes step S210, which involves acquiring indoor space characteristic data, indoor occupant monitoring data, and outdoor environmental monitoring data in real time based on a sensor network. Specifically, a sensor network is deployed according to monitoring needs. Indoor sensors include temperature and humidity sensors, air quality sensors, occupant presence sensors, and space size sensors. Air quality sensors include PM2.5, formaldehyde, and CO2 sensors, while occupant presence sensors include infrared sensors and cameras. Outdoor sensors include meteorological sensors and air quality sensors, including temperature, humidity, wind speed, and wind direction sensors. The sampling frequency of the sensors is set based on the rate of change of the monitored data; for example, the sampling frequency of indoor sensors is once every 5 seconds, and the sampling frequency of outdoor sensors is once every minute. The sensor data acquisition module uses communication protocols such as Modbus and MQTT to receive the sensor data in real time and stores the data in a time-sensor type-data value format, thereby acquiring indoor space characteristic data, indoor occupant monitoring data, and outdoor environmental monitoring data.

[0031] Step S220 involves cleaning the indoor space characteristic data, the indoor personnel monitoring data, and the outdoor environmental monitoring data to generate the indoor-outdoor linkage monitoring matrix. Specifically, the three types of data are preprocessed. For missing values, linear interpolation based on time series or mean-filling based on data from similar sensors is used. For outliers, the 3σ principle or box plot method is used for identification and removal. To address the issue of inconsistent data formats, all data is converted to numerical data through format conversion. Data standardization is performed, such as min-max standardization or z-score standardization, to map the data to a unified numerical range and eliminate the influence of dimensions. Finally, the cleaned data is integrated into an indoor-outdoor linkage monitoring matrix according to a two-dimensional structure of monitoring dimension-time node. The rows of the matrix correspond to monitoring dimensions such as temperature, humidity, PM2.5 concentration, and number of personnel, while the columns correspond to different monitoring time nodes. The matrix elements are the standardized data values ​​of the corresponding dimension at the corresponding time.

[0032] In one possible implementation, indoor air quality target mining is performed based on the indoor-outdoor linkage monitoring matrix to obtain an indoor air quality target matrix. Step S200 further includes step S230, which involves retrieving qualified indoor air quality samples based on the indoor-outdoor linkage monitoring matrix to obtain a qualified indoor air quality sample space. Specifically, the qualified indoor air quality criteria are determined, including the limits for various indicators. Qualified samples are retrieved by retrieving the indoor air quality data corresponding to each time node in the indoor-outdoor linkage monitoring matrix row by row. If all indicators meet the qualified criteria, all monitoring data at that time node are considered as a qualified sample and added to the qualified sample space. All qualified samples are arranged in chronological order to form the qualified indoor air quality sample space.

[0033] Step S240: Classify the indoor air quality qualified sample space according to air quality index characteristics to obtain each air quality qualified sample area. Specifically, determine the classification dimensions of air quality indicators, including physical indicators, chemical indicators, and biological indicators. Physical indicators include temperature, humidity, and wind speed; chemical indicators include PM2.5, formaldehyde, CO2, and SO2; and biological indicators include total bacterial count and total fungal count. Use clustering algorithms, such as K-means clustering or hierarchical clustering, to cluster the samples in the qualified sample space. The cluster feature is the value of each air quality indicator, and the number of clusters equals the number of index classification dimensions. Each cluster's corresponding sample set is considered an air quality qualified sample area. The name of the sample area corresponds to the corresponding index classification dimension, such as a temperature qualified sample area, a PM2.5 qualified sample area, etc.

[0034] Step S250: Based on the air quality compliance sample areas, perform confidence evaluation and cleaning to establish confidence intervals for each air quality indicator that meet the confidence constraints. Specifically, determine the confidence constraints, for example, setting the confidence level to 95%, meaning that the samples in the sample area can represent the air quality characteristics of that area with a 95% probability. Use the confidence interval estimation method to calculate the confidence interval for each air quality indicator in each sample area. The formula for calculating the confidence interval is: Confidence interval = [sample mean] The critical value is calculated as [critical value × (sample standard deviation ÷ √sample size)], where the critical value is obtained from the t-distribution table or normal distribution table based on the confidence level and sample size. Samples with a difference between the upper and lower limits of the confidence interval exceeding a preset threshold are removed, and the cleaned sample set is used as the air quality confidence interval that satisfies the confidence level constraint.

[0035] Step S260: Perform central tendency analysis based on the air quality confidence zones to generate the indoor air quality target matrix. Specifically, determine the indicators for central tendency analysis, including mean, median, and mode. The mean describes the average level of the data, the median describes the intermediate level of the data, and the mode describes the most frequently occurring value of the data. Through statistical analysis, calculate the mean, median, and mode for each air quality indicator in each air quality confidence zone. Based on the grid division of the indoor space, assign each central tendency indicator value to the corresponding grid node to form the indoor air quality target matrix. The rows of the matrix correspond to the grid nodes, the columns correspond to the air quality indicators, and the matrix elements are the target values ​​of the corresponding indicators for the corresponding nodes.

[0036] Step S300: Based on the indoor-outdoor linkage monitoring matrix and the indoor air quality target matrix, perform multivariate anomaly causal modeling on the real-time indoor air quality data to establish an air quality anomaly causal graph model.

[0037] Specifically, an indoor air quality matrix is ​​constructed based on real-time indoor air quality data. Anomalies are detected within this matrix based on the target indoor air quality matrix, generating multivariate air quality anomaly characteristics. Correlation analysis is performed on these anomaly characteristics using an indoor-outdoor linked monitoring matrix to obtain multivariate anomaly correlation sequences. These correlation sequences guide the monitoring matrix in causal tracing of the anomaly characteristics, revealing the causal relationships between each anomaly. Finally, an air quality anomaly causal graph model is established, with anomaly characteristics as nodes and causal relationships as edges.

[0038] In one possible implementation, multivariate anomaly causal modeling is performed on real-time indoor air quality data based on the indoor-outdoor linkage monitoring matrix and the indoor air quality target matrix to establish an air quality anomaly causal graph model. Step S300 further includes step S310, constructing an indoor air quality matrix based on the real-time indoor air quality data. Specifically, the dimensions of the indoor air quality matrix are determined, with rows corresponding to grid nodes in the indoor space and columns corresponding to air quality indicators, such as temperature, humidity, PM2.5 concentration, and CO2 concentration. The update frequency of the matrix is ​​set to be consistent with the sampling frequency of the sensors. Air quality sensor data corresponding to each grid node in the indoor space is acquired in real time through the data acquisition module. The data is organized according to the format of grid node-air quality indicator-data value, and the organized data is filled into the corresponding positions in the matrix to generate the indoor air quality matrix. The matrix elements are the air quality indicator values ​​of the corresponding grid nodes at the corresponding times.

[0039] Step S320: Based on the indoor air quality target matrix, perform anomaly identification on the indoor air quality matrix to generate multi-dimensional air quality anomaly characteristics. Specifically, determine the threshold rules for anomaly identification. For example, a value in the air quality matrix exceeding the upper or lower limit of the corresponding value in the target matrix by 10% is considered an anomaly. For indicators with national standards, such as PM2.5 concentration, the national standard is directly used as the anomaly judgment threshold. By comparing corresponding elements of the two matrices element by element, determine whether they exceed the threshold range. Organize the anomaly identification results according to the format of anomaly grid node-anomaly indicator-anomaly degree to generate multi-dimensional air quality anomaly characteristics. The anomaly degree is divided into mild anomaly, moderate anomaly, and severe anomaly, with the division standard being the proportion exceeding the threshold. For example, exceeding by 10%-20% is mild anomaly, 20%-50% is moderate anomaly, and exceeding 50% is severe anomaly.

[0040] Step S330: Perform multivariate correlation analysis on the multivariate air quality anomaly characteristics based on the indoor-outdoor linkage monitoring matrix to obtain a multivariate anomaly correlation sequence. Specifically, determine the algorithm for multivariate correlation analysis, including Pearson correlation coefficient analysis algorithm, Spearman rank correlation coefficient analysis algorithm, etc. Pearson correlation coefficient is used for linearly correlated variables, and Spearman rank correlation coefficient is used for non-linearly correlated variables. Convert the multivariate air quality anomaly characteristics into numerical sequences, convert each monitoring dimension in the indoor-outdoor linkage monitoring matrix into numerical sequences, calculate the correlation coefficient between the two sequences, and organize the correlation analysis results according to the format of anomaly characteristic-related monitoring dimension-correlation coefficient-correlation direction to generate a multivariate anomaly correlation sequence. The correlation direction is divided into positive correlation and negative correlation. The larger the absolute value of the correlation coefficient, the stronger the correlation.

[0041] Step S340: Based on the multivariate anomaly correlation sequence, guide the indoor and outdoor linkage monitoring matrix to perform causal tracing of the multivariate air quality anomaly characteristics, obtaining the causal relationship of each anomaly. Specifically, determine the priority rules for causal tracing, prioritizing relevant monitoring dimensions according to the absolute value of the correlation coefficient from largest to smallest, tracing dimensions with larger absolute values ​​of correlation coefficients first. According to the priority rules, analyze each relevant monitoring dimension in turn, determining whether the change in that dimension precedes the appearance of the anomaly characteristic, and whether the change in that dimension can reasonably explain the occurrence of the anomaly characteristic. Organize the causal tracing results in the format of cause monitoring dimension - change - anomaly characteristic - influencing mechanism to obtain the causal relationship of each anomaly.

[0042] Step S350: Using the multivariate air quality anomaly characteristics as multiple air quality anomaly nodes and the various causal relationships as multiple causal edges, construct the air quality anomaly causal graph model. Specifically, determine the node attributes and edge attributes of the graph model. Node attributes include the name of the anomaly node, the degree of anomaly, and the grid node it belongs to. Edge attributes include the name of the causal edge, the intensity of its influence, and the delay time of its influence. Each multivariate air quality anomaly characteristic is treated as an air quality anomaly node, and a unique identifier and attribute information are assigned to each node. Each causal relationship is treated as an anomaly causal edge, and a unique identifier and attribute information are assigned to each edge. The direction of the edge points from the cause monitoring dimension to the anomaly characteristic node. Using graph model building tools such as NetworkX and Graphviz, the nodes and edges are connected and visualized according to the attribute information to construct the air quality anomaly causal graph model.

[0043] Step S400: Based on the indoor air quality target matrix and the air quality anomaly causal graph model, perform multi-level adjustment and optimization on the air control deployment grid to obtain the first air control strategy.

[0044] Specifically, real-time operating parameters of each device within the air control deployment grid are collected to generate an air control operation grid. The operation grid is adjusted based on an air quality anomaly causal graph model to obtain the first air control group. The first group is then evaluated and optimized based on the indoor air quality target matrix to generate the second air control group. Energy consumption minimization is then used to optimize the energy consumption of the second group, generating the first air control strategy.

[0045] In one possible implementation, the air control deployment grid is optimized through multi-level adjustment based on the indoor air quality target matrix and the air quality anomaly causal graph model to obtain the first air control strategy. Step S400 further includes step S410, which involves collecting real-time operating parameters of each device within the air control deployment grid to generate an air control operation grid. Specifically, the dimensions of the air control operation grid are determined, with rows of the matrix corresponding to nodes in the air control deployment grid and columns corresponding to the operating parameters of the devices, such as operating speed, temperature setpoint, wind speed, and purification efficiency. The sampling frequency of the operating parameters is set to be consistent with the control cycle of the devices. Through the device communication module, industrial communication protocols such as Modbus and BACnet are used to collect the operating parameters of each device within the grid in real time. The parameters are organized according to the format of grid node-device type-operating parameter-parameter value, and the organized data is filled into the corresponding positions of the operation grid to generate the air control operation grid. The grid elements are the operating parameter values ​​of the corresponding devices within the corresponding grid nodes.

[0046] Step S420: Perform multi-level joint adjustment of the air control operation grid according to the air quality anomaly causal graph model to obtain the first air control group. Specifically, determine the levels of the multi-level joint adjustment, divided into equipment layer, regional layer, and system layer. The equipment layer adjusts individual devices, the regional layer coordinates the adjustment of multiple devices within the same region, and the system layer performs global adjustment of all devices in the entire room. Based on the causal relationship in the air quality anomaly causal graph model, determine the adjustment equipment and adjustment method corresponding to each anomaly. For example, for PM2.5 concentration anomalies, adjust the settings of the fresh air system and the air purifier. Perform joint adjustment in the order of equipment layer, regional layer, and system layer to generate multiple different control schemes. Arrange all control schemes in order of adjustment intensity from low to high to obtain the first air control group.

[0047] Step S430: Based on the indoor air quality target matrix, perform air quality adaptation evaluation and optimization on the first air control group to generate the second air control group. Specifically, construct an air quality prediction model. The model's input includes an indoor-outdoor linkage monitoring matrix, real-time indoor air quality data, and control schemes. The output is a predicted indoor air quality sequence. Determine the air quality adaptation evaluation indicators, including air quality compliance rate, average deviation, and maximum deviation. The air quality compliance rate is the proportion of samples in the predicted sequence that meet the target matrix requirements out of the total number of samples. Set air quality adaptation thresholds, for example, setting an air quality compliance rate of no less than 95%, an average deviation not exceeding 5% of the target value, and a maximum deviation not exceeding 10% of the target value as adaptation thresholds. Through scheme-by-scheme evaluation, predict and evaluate each scheme in the first air control group, and select schemes that meet the adaptation thresholds to generate the second air control group.

[0048] Step S440: Based on the second air control group, perform energy consumption minimization optimization to generate the first air control strategy. Specifically, construct an equipment energy consumption model. The model's inputs are the equipment's operating parameters and operating time, and its output is the equipment's energy consumption value. The equipment energy consumption model is constructed based on the equipment's rated power and operating efficiency. The energy consumption value is calculated by multiplying the rated power by the operating efficiency by the operating time. Determine the energy consumption calculation cycle, i.e., the execution cycle of the control scheme, such as 30 minutes. Using a scheme-by-scheme energy consumption calculation, for each scheme in the second air control group, calculate the total energy consumption of all involved equipment within the execution cycle. Arrange all schemes in ascending order of total energy consumption, and select the scheme with the minimum total energy consumption to generate the first air control strategy.

[0049] In one possible implementation, the first group of air control systems is evaluated and optimized for air quality adaptation based on the indoor air quality target matrix to generate a second group of air control systems. Step S430 further includes step S431, extracting the Pth air control scheme from the first group of air control systems, where P is a positive integer. Specifically, the order of the schemes in the first group of air control systems is determined, arranged in order of adjustment intensity from low to high or generation time from early to late. Based on the arrangement order, the schemes in the first group are numbered, starting from 1 and incrementing sequentially. Based on the input positive integer P, the control scheme numbered P is extracted. The scheme content includes the equipment involved, the adjustment method of the equipment, and the operating parameters after adjustment.

[0050] Step S432: Based on the indoor-outdoor linkage monitoring matrix and the real-time indoor air quality data, predict indoor air quality according to the Pth air control scheme to obtain the Pth indoor air quality sequence. Specifically, construct a deep learning-based air quality prediction model. The model adopts an LSTM network structure. The input layer dimension of the network is the sum of the feature dimensions of the indoor-outdoor linkage monitoring matrix and the real-time air quality data. The hidden layer is set to 2 layers, each containing 64 neurons. The output layer dimension is the number of air quality indicators. Train the model using historical indoor-outdoor linkage monitoring matrices, real-time air quality data, control schemes, and corresponding air quality data. The training loss function is mean squared error, the optimizer is Adam, and the learning rate is set to 0.001. Use the indoor-outdoor linkage monitoring matrix, real-time indoor air quality data, and the Pth air control scheme as the model input. Set the prediction period to the execution period of the control scheme. Through forward propagation of the model, output the indoor air quality indicator values ​​within the prediction period. Arrange these values ​​in chronological order to obtain the Pth indoor air quality sequence.

[0051] Step S433: Evaluate the P-th indoor air quality sequence using a Siamese comparison based on the indoor air quality target matrix to obtain the P-th air quality fitness value. Specifically, a Siamese comparison model is constructed. The model employs a Siamese neural network structure, containing two identical sub-networks, used to extract features from the P-th indoor air quality sequence and the indoor air quality target matrix, respectively. The dimensions for feature extraction are determined, including the mean, variance, peak, and trough values ​​of air quality indicators. A similarity calculation method is set, using cosine similarity or Euclidean distance. The P-th indoor air quality sequence and the indoor air quality target matrix are input into the Siamese comparison model to extract features and calculate similarity. The similarity value is used as the P-th air quality fitness value; a higher value indicates better adaptability of the solution.

[0052] Step S434: Determine whether the Pth air quality adaptation value is greater than or equal to the air quality adaptation threshold. If the Pth air quality adaptation value is greater than or equal to the air quality adaptation threshold, add the Pth air control scheme to the second air control group. Specifically, determine the value of the air quality adaptation threshold. The threshold is determined based on statistical analysis of historical data and actual control needs, for example, set to 0.85, meaning that schemes with an adaptation value greater than or equal to 0.85 are considered compliant. Compare the Pth air quality adaptation value with the adaptation threshold. If the adaptation value is greater than or equal to the threshold, the judgment result is compliant; if the adaptation value is less than the threshold, the judgment result is non-compliant. Based on the judgment result, add the compliant Pth air control scheme to the second air control group, recording the scheme number, adaptation value, control content, and other information during addition.

[0053] Step S500: Based on the first air control strategy, perform deep learning-based air quality situation smoothing optimization to obtain the second air control strategy.

[0054] Specifically, a deep learning model is trained using a historical set of indoor air quality records to generate an air quality situation prediction model. The indoor-outdoor linkage monitoring matrix, real-time indoor air quality data, and the first air control strategy are input into the prediction model to obtain a multi-dimensional air quality situation curve. The situation curve is analyzed through abrupt change identification to obtain a set of multi-dimensional situation change segments. Attention allocation is used to assign influence weights to these abrupt change segments, determining the influence weight of each situation change. Based on these influence weights, differentiated smoothing constraint optimization is used to optimize the first air control strategy, generating a second air control strategy.

[0055] In one possible implementation, air quality situation smoothing optimization under deep learning is performed based on the first air control strategy to obtain the second air control strategy. Step S500 further includes step S510, performing deep learning based on the indoor air quality situation record set to generate an air quality situation prediction model. Specifically, an indoor air quality situation record set is constructed, which includes historical indoor and outdoor linkage monitoring data, real-time air quality data, control strategies, and corresponding air quality situation data. The situation data includes curves showing the changes of each air quality index over time. The structure of the deep learning model is determined, adopting a Transformer model structure. The input layer of the model is an embedding layer used to convert input data into embedding vectors. The encoder part includes a multi-head attention layer and a feedforward neural network layer. The decoder part includes a multi-head attention layer, an encoder-decoder attention layer, and a feedforward neural network layer. The output layer is a fully connected layer, and the output dimension is the number of air quality indicators. The model is trained with a batch size of 32, a learning rate of 0.001, an optimizer of Adam, and a loss function of mean squared error. An early stopping strategy is adopted during training, stopping training when the validation set loss no longer decreases for 10 consecutive rounds. The trained model is saved as an air quality situation prediction model. The model can output the future air quality situation curve based on the input monitoring data, real-time air quality data and control strategies.

[0056] Step S520: Input the indoor-outdoor linkage monitoring matrix, the real-time indoor air quality data, and the first air control strategy into the air quality situation prediction model to obtain a multi-dimensional air quality situation curve. Specifically, the input data is preprocessed: the indoor-outdoor linkage monitoring matrix and real-time indoor air quality data are converted into tensor formats matching the model input dimensions; the first air control strategy is converted into numerical codes, with the coding rule being a combination of equipment type number and adjustment method number. The model's prediction parameters are set, including the prediction period and time step. The prediction period is the execution period of the control strategy, and the time step is set to 1 minute to ensure the precision of the situation curve. The preprocessed input data is input into the air quality situation prediction model. The model's encoder extracts the spatiotemporal features of the input data, and the decoder generates predicted air quality index values ​​for the future prediction period. The predicted values ​​of each air quality index are arranged in chronological order to generate curves corresponding to indicators such as temperature, humidity, PM2.5 concentration, and CO2 concentration, i.e., the multi-dimensional air quality situation curve.

[0057] Step S530: Abrupt change identification is performed based on the multidimensional air quality situation curve to obtain a set of multidimensional situational change segments. Specifically, a sliding window difference method is used for abrupt change identification. A fixed-size sliding window is set, and the mean difference between two consecutive sub-windows within the window is calculated. If the difference exceeds a preset threshold, it is determined to be a change segment. The sliding window size is set to 5 time steps, and the threshold is set to 10% of the corresponding air quality indicator target value. The calculation is performed window-by-window for each air quality situation curve, recording the start time, end time, and magnitude of the change segment. All abrupt change segments in all dimensions are integrated according to the format of indicator type-change segment information to generate a set of multidimensional situational change segments.

[0058] Step S540: Attention is allocated based on the multi-dimensional situational change segment set to determine the impact weight of each situational change. Specifically, a weighted normalized attention algorithm is used to determine three impact factors: change magnitude factor, duration factor, and indicator importance factor. The weight value of each factor is calculated, and the three factor values ​​are multiplied to obtain the original attention weight. Finally, all original weights are normalized to obtain the final impact weight. The change magnitude factor is the ratio of the change magnitude to the corresponding indicator threshold, the duration factor is the ratio of the change segment duration to the prediction period, and the indicator importance factor is set according to national standards or actual needs. For each segment in the multi-dimensional situational change segment set, the three factor values ​​and the original weights are calculated sequentially. Finally, all original weights are normalized so that the sum of the weights is 1, thus determining the impact weight of each situational change.

[0059] Step S550: Based on the influence weights of each situational abrupt change, the first air control strategy is optimized using differentiated smoothing constraints according to the multi-dimensional situational abrupt change segment set to generate the second air control strategy. Specifically, the correlation between abrupt change segments and control devices is established, i.e., determining which devices' adjustment behaviors cause each abrupt change segment. For example, PM2.5 concentration abrupt changes are related to the adjustment of fresh air systems and air purifiers, while CO2 concentration abrupt changes are related to the adjustment of exhaust systems and air conditioners. The rules for differentiated smoothing constraints are determined: for abrupt change segments with larger influence weights, the corresponding device adjustment amplitude is smaller and the adjustment time is longer. The specific constraint formula is: adjustment step size = original adjustment amplitude ÷ (influence weight × 10), adjustment interval = original adjustment interval × (influence weight × 10). For each device adjustment command in the first air control strategy, the smoothed adjustment step size and adjustment interval are calculated based on the influence weights of its associated abrupt change segments. The smoothed adjustment commands are then integrated to generate the second air control strategy.

[0060] Step S600: Perform multi-trend characteristic interference correction of the air control deployment grid according to the second air control strategy to obtain the third air control strategy.

[0061] Specifically, an equipment simulation platform is used to simulate and control the second air control strategy, outputting a simulation dataset of each device's operation. The simulation dataset is then analyzed for trends to obtain the fault trend characteristics and aging trend characteristics of each device. Based on the severity of the fault and aging trends, the correction coefficients for device adjustment are determined. Adaptive interference correction is then applied to the second air control strategy based on these correction coefficients to adjust the device's operating parameters and adjustment methods, thereby generating the third air control strategy.

[0062] In one possible implementation, the multi-trend characteristic interference correction of the air control deployment grid is executed according to the second air control strategy to obtain the third air control strategy. Step S600 further includes step S610, which involves performing simulation control on the air control deployment grid according to the second air control strategy to obtain a simulation dataset of each device's operation. Specifically, a simulation platform such as MATLAB / Simulink or EnergyPlus is used as the device simulation tool. This platform contains a physical model of the air control equipment, and the model parameters are consistent with the rated power, operating efficiency, and response characteristics of the actual equipment. The spatial parameters of the air control deployment grid and the equipment installation location parameters are imported into the simulation platform, and the simulation time step is set to be consistent with the strategy execution cycle to construct a simulation scenario. The adjustment commands of the second air control strategy are converted into control signals that the simulation platform can recognize, input into the simulation model, and the simulation program is run. Parameters such as operating current, operating temperature, vibration frequency, and energy consumption of each device are collected within each time step and organized in the format of device number-time step-parameter type-parameter value to obtain a simulation dataset of each device's operation.

[0063] Step S620: Analyze the fault trend characteristics based on the simulation dataset of each device's operation to obtain the fault trend characteristics of each device. Specifically, determine the key monitoring parameters for device faults. Different devices have different fault parameters; for example, the fault parameters for a fresh air system are operating current and fan speed, for an air conditioner are compressor temperature and refrigerant pressure, and for an air purifier are filter pressure difference and motor current. Set fault thresholds based on the device's factory standards and historical fault data. Use a sliding window linear regression method for trend analysis, setting the sliding window size to 10 time steps. Perform linear regression on the parameter data within the window and calculate the regression slope. If the absolute value of the slope exceeds the preset fault threshold, a fault trend is determined to exist. Calculate the key fault parameters for each device window by window, recording the parameter type, slope, start time, severity, and other information of the fault trend to obtain the fault trend characteristics of each device.

[0064] Step S630: Analyze the aging trend characteristics based on the simulation dataset of each device's operation to obtain the aging trend characteristics of each device. Specifically, determine the key monitoring parameters for device aging, such as the fan vibration frequency and bearing temperature for a fresh air system, the heat exchanger efficiency and compressor operating noise for an air conditioner, and the motor vibration amplitude and filter lifespan for an air purifier. Set aging thresholds based on the device's design life and historical aging data, and use an exponential fitting trend analysis method to analyze the trend. Fit the parameter data with an exponential function, calculate the growth rate of the fitted curve, and if the growth rate exceeds the preset aging threshold, an aging trend is determined to exist. Perform exponential fitting on the key aging parameters of each device, and record information such as the parameter type, growth rate, start time, and severity of the aging trend to obtain the aging trend characteristics of each device.

[0065] Step S640: Based on the fault trend characteristics and aging trend characteristics of each device, adaptive interference effect correction is applied to the second air control strategy to generate the third air control strategy. Specifically, a correlation matrix between trend characteristics and strategy correction is established to determine the device adjustment correction direction corresponding to each fault trend and aging trend. For example, a fault trend in operating current corresponds to reducing the device's operating level, and a motor vibration aging trend corresponds to reducing the adjustment amplitude. The calculation rules for the correction coefficient are determined. The correction coefficient is negatively correlated with the severity of the trend; the higher the severity, the smaller the correction coefficient. The specific formula is: Correction coefficient = 1 - Severity percentage ÷ 100, where severity percentage = (Actual slope / growth rate - threshold) ÷ threshold × 100%. For each device adjustment command in the second air control strategy, the correction coefficient is calculated based on its corresponding trend characteristics, and the target level or adjustment step of the device is adjusted. The corrected adjustment commands are integrated to generate the third air control strategy.

[0066] This application's embodiments employ techniques such as constructing an air control deployment grid based on the deployment characteristics of indoor air control equipment, building an indoor-outdoor linkage monitoring matrix and mining indoor air quality targets within it to form an indoor air quality target matrix, combining the aforementioned monitoring matrix and target matrix, performing multivariate anomaly causal modeling on real-time indoor air quality data to establish an air quality anomaly causal graph model, and based on the target matrix and causal graph model, performing multi-level adjustment and optimization on the air control deployment grid to obtain the first air control strategy, using deep learning to smooth and optimize the air quality situation to generate the second air control strategy, and combining the multi-trend characteristics of the air control deployment grid to perform interference correction on the second strategy, finally obtaining the third air control strategy. These techniques solve the technical problems of insufficient dynamic adaptability and end-to-end collaborative optimization in existing indoor air quality adaptive control, achieving the technical effect of improving the dynamic adaptability and end-to-end collaborative optimization level of control.

[0067] In the above text, refer to Figure 1 A method for adaptive indoor air quality control using deep learning, according to an embodiment of the present invention, is described in detail. Next, reference will be made to... Figure 2 An indoor air quality adaptive control system applying deep learning is described according to an embodiment of the present invention.

[0068] An indoor air quality adaptive control system based on deep learning, according to an embodiment of the present invention, addresses the technical problems of insufficient dynamic adaptability and end-to-end collaborative optimization in existing indoor air quality adaptive regulation, thereby improving the dynamic adaptability and end-to-end collaborative optimization level of regulation. The indoor air quality adaptive control system based on deep learning includes: an air regulation deployment grid construction module 10, an indoor air quality target mining module 20, a multivariate anomaly causal modeling module 30, a multi-level regulation optimization module 40, an air quality situation smoothing optimization module 50, and a multi-trend characteristic interference correction module 60.

[0069] The air control deployment grid construction module 10 is used to construct an air control deployment grid based on the deployment characteristics of indoor air control equipment; the indoor air quality target mining module 20 is used to construct an indoor-outdoor linkage monitoring matrix and mine indoor air quality targets based on the indoor-outdoor linkage monitoring matrix to obtain an indoor air quality target matrix; the multivariate anomaly causal modeling module 30 is used to perform multivariate anomaly causal modeling on real-time indoor air quality data based on the indoor-outdoor linkage monitoring matrix and the indoor air quality target matrix to establish an air quality anomaly causal graph model; the multi-level adjustment optimization module 40 is used to perform multi-level adjustment optimization on the air control deployment grid based on the indoor air quality target matrix and the air quality anomaly causal graph model to obtain a first air control strategy; the air quality situation smoothing optimization module 50 is used to perform deep learning-based air quality situation smoothing optimization based on the first air control strategy to obtain a second air control strategy; and the multi-trend characteristic interference correction module 60 is used to execute multi-trend characteristic interference correction on the air control deployment grid based on the second air control strategy to obtain a third air control strategy.

[0070] The detailed description of the specific configuration of the indoor air quality target mining module 20 is explained as follows: As mentioned above, indoor air quality target mining is performed based on the indoor-outdoor linkage monitoring matrix to obtain an indoor air quality target matrix. The indoor air quality target mining module 20 may further include: an indoor air quality qualified sample retrieval unit for retrieving indoor air quality qualified samples based on the indoor-outdoor linkage monitoring matrix to obtain an indoor air quality qualified sample space; an air quality index characteristic classification unit for classifying the indoor air quality qualified sample space according to air quality index characteristics to obtain each air quality qualified sample area; a confidence evaluation and cleaning unit for performing confidence evaluation and cleaning based on each air quality qualified sample area to establish each air quality confidence area that meets the confidence constraints; and a central tendency analysis unit for performing central tendency analysis based on each air quality confidence area to generate the indoor air quality target matrix.

[0071] The detailed description of the specific configuration of the multivariate anomaly causal modeling module 30 is explained as follows: As mentioned above, multivariate anomaly causal modeling is performed on real-time indoor air quality data based on the indoor-outdoor linkage monitoring matrix and the indoor air quality target matrix to establish an air quality anomaly causal graph model. The multivariate anomaly causal modeling module 30 may further include: an indoor air quality matrix construction unit for constructing an indoor air quality matrix based on the real-time indoor air quality data; an anomaly identification unit for identifying anomalies in the indoor air quality matrix based on the indoor air quality target matrix to generate multivariate air quality anomaly characteristics; a multivariate correlation analysis unit for performing multivariate correlation analysis on the multivariate air quality anomaly characteristics based on the indoor-outdoor linkage monitoring matrix to obtain a multivariate anomaly correlation sequence; a causal tracing unit for guiding the indoor-outdoor linkage monitoring matrix to perform causal tracing on the multivariate air quality anomaly characteristics based on the multivariate anomaly correlation sequence to obtain the causal pointing relationship of each anomaly; and an air quality anomaly causal graph model construction unit for constructing the air quality anomaly causal graph model with the multivariate air quality anomaly characteristics as multiple air quality anomaly nodes and the causal pointing relationship of each anomaly as multiple anomaly causal edges.

[0072] The detailed description of the specific configuration of the multi-level adjustment optimization module 40 is explained as follows: As mentioned above, the air control deployment grid is subjected to multi-level adjustment optimization based on the indoor air quality target matrix and the air quality anomaly causal graph model to obtain the first air control strategy. The multi-level adjustment optimization module 40 may further include: a real-time operating parameter acquisition unit for acquiring the real-time operating parameters of each device in the air control deployment grid to generate an air control operation grid; a multi-level joint adjustment unit for performing multi-level joint adjustment on the air control operation grid based on the air quality anomaly causal graph model to obtain the first air control group; an air quality adaptation evaluation optimization unit for performing air quality adaptation evaluation optimization on the first air control group based on the indoor air quality target matrix to generate the second air control group; and an energy consumption minimization optimization unit for performing energy consumption minimization optimization on the second air control group to generate the first air control strategy.

[0073] Specifically, the air quality adaptation evaluation and optimization of the first group of air control systems is performed based on the indoor air quality target matrix to generate a second group of air control systems. The air quality adaptation evaluation and optimization unit may further include: a P-th air control scheme extraction subunit for extracting the P-th air control scheme based on the first group of air control systems, where P is a positive integer; an indoor air quality prediction subunit for predicting indoor air quality based on the indoor-outdoor linkage monitoring matrix and the real-time indoor air quality data, according to the P-th air control scheme, to obtain the P-th indoor air quality sequence; a twin comparison evaluation subunit for performing twin comparison evaluation on the P-th indoor air quality sequence based on the indoor air quality target matrix to obtain the P-th air quality adaptation value; and a judgment and processing subunit for judging whether the P-th air quality adaptation value is greater than or equal to the air quality adaptation threshold. If the P-th air quality adaptation value is greater than or equal to the air quality adaptation threshold, the P-th air control scheme is added to the second group of air control systems.

[0074] The detailed description of the specific configuration of the air quality situation smoothing optimization module 50 is explained as follows: As mentioned above, the air quality situation smoothing optimization is performed under deep learning based on the first air control strategy to obtain the second air control strategy. The air quality situation smoothing optimization module 50 may further include: a deep learning unit for performing deep learning based on the indoor air quality situation record set to generate an air quality situation prediction model; an air quality situation prediction unit for inputting the indoor-outdoor linkage monitoring matrix, the real-time indoor air quality data, and the first air control strategy into the air quality situation prediction model to obtain a multi-dimensional air quality situation curve; a mutation identification unit for performing mutation identification based on the multi-dimensional air quality situation curve to obtain a multi-dimensional situation mutation segment set; a mutation impact attention allocation unit for performing mutation impact attention allocation based on the multi-dimensional situation mutation segment set to determine the impact weight of each situation mutation; and a differentiated smoothing constraint optimization unit for performing differentiated smoothing constraint optimization on the first air control strategy based on the impact weight of each situation mutation and the multi-dimensional situation mutation segment set to generate the second air control strategy.

[0075] The detailed description of the specific configuration of the multi-trend characteristic interference correction module 60 is explained as follows: As described above, the multi-trend characteristic interference correction of the air control deployment grid is executed according to the second air control strategy to obtain the third air control strategy. The multi-trend characteristic interference correction module 60 may further include: a simulation control unit for performing simulation control on the air control deployment grid according to the second air control strategy to obtain a simulation dataset of each device's operation; a fault trend characteristic analysis unit for performing fault trend characteristic analysis according to the simulation dataset of each device's operation to obtain the fault trend characteristics of each device; an aging trend characteristic analysis unit for performing aging trend characteristic analysis according to the simulation dataset of each device's operation to obtain the aging trend characteristics of each device; and an adaptive interference influence correction unit for performing adaptive interference influence correction on the second air control strategy according to the fault trend characteristics and aging trend characteristics of each device to generate the third air control strategy.

[0076] The detailed description of the specific configuration of the air control deployment grid construction module 10 is explained as follows: As mentioned above, an air control deployment grid is constructed based on the deployment characteristics of indoor air control equipment. The air control deployment grid construction module 10 may further include: a control equipment installation grid construction unit for constructing a control equipment installation grid based on the installation characteristics of the indoor air control equipment; and a multi-head attention enhancement unit for performing multi-head attention enhancement on the control equipment installation grid based on the function characteristics and linkage characteristics of the indoor air control equipment to generate the air control deployment grid.

[0077] The indoor air quality target mining module 20, which constructs an indoor-outdoor linkage monitoring matrix, may further include: a sensor data acquisition unit for acquiring indoor space characteristic data, indoor personnel monitoring data, and outdoor environmental monitoring data in real time based on a sensor network; and a data cleaning unit for cleaning the indoor space characteristic data, the indoor personnel monitoring data, and the outdoor environmental monitoring data to generate the indoor-outdoor linkage monitoring matrix.

[0078] The indoor air quality adaptive control system using deep learning provided in this embodiment of the invention can execute the indoor air quality adaptive control method using deep learning provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0079] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0080] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application. In some cases, the actions or steps described in this application can be performed in a different order than that shown in the embodiments and still achieve the desired results. Furthermore, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

Claims

1. An adaptive control method for indoor air quality using deep learning, characterized in that, The method includes: Based on the deployment characteristics of indoor air control equipment, an air control deployment grid is constructed; Construct an indoor-outdoor linkage monitoring matrix, and mine indoor air quality targets based on the indoor-outdoor linkage monitoring matrix to obtain an indoor air quality target matrix; Based on the indoor-outdoor linkage monitoring matrix and the indoor air quality target matrix, multivariate anomaly causal modeling is performed on the real-time indoor air quality data to establish an air quality anomaly causal graph model. Based on the indoor air quality target matrix and the air quality anomaly cause-effect graph model, the air control deployment grid is optimized through multi-level adjustment to obtain the first air control strategy. Based on the first air control strategy, a second air control strategy is obtained by performing deep learning-based air quality situation smoothing optimization. According to the second air control strategy, the multi-trend characteristic interference correction of the air control deployment grid is executed to obtain the third air control strategy.

2. The indoor air quality adaptive control method using deep learning as described in claim 1, characterized in that, Based on the aforementioned indoor-outdoor linkage monitoring matrix, indoor air quality target mining is performed to obtain an indoor air quality target matrix, including: Based on the indoor and outdoor linkage monitoring matrix, indoor air quality qualified samples are retrieved to obtain the indoor air quality qualified sample space. The indoor air quality qualified sample spaces are classified according to their air quality index characteristics to obtain each air quality qualified sample area; Based on the air quality qualified sample areas, confidence evaluation and cleaning are performed to establish confidence zones for each air quality that meet the confidence constraints. Based on the central tendency analysis of each air quality confidence zone, the indoor air quality target matrix is ​​generated.

3. The indoor air quality adaptive control method using deep learning as described in claim 1, characterized in that, Based on the indoor-outdoor linkage monitoring matrix and the indoor air quality target matrix, a multivariate anomaly causal model is performed on the real-time indoor air quality data to establish an air quality anomaly causal graph model, including: Based on the real-time indoor air quality data, construct an indoor air quality matrix; Based on the indoor air quality target matrix, anomalies are identified in the indoor air quality matrix to generate multi-dimensional air quality anomaly characteristics. Based on the indoor-outdoor linkage monitoring matrix, a multivariate correlation analysis is performed on the multivariate air quality anomaly characteristics to obtain the multivariate anomaly correlation sequence. Based on the multivariate anomaly correlation sequence, the indoor and outdoor linkage monitoring matrix is ​​guided to trace the causal relationship of the multivariate air quality anomaly characteristics and obtain the causal relationship of each anomaly. The air quality anomaly causal graph model is constructed by using the aforementioned multivariate air quality anomaly characteristics as multiple air quality anomaly nodes and the aforementioned anomaly causal relationships as multiple anomaly causal edges.

4. The indoor air quality adaptive control method using deep learning as described in claim 1, characterized in that, Based on the indoor air quality target matrix and the air quality anomaly causal graph model, the air control deployment grid is optimized through multi-level adjustment to obtain the first air control strategy, including: Collect real-time operating parameters of each device within the air control deployment grid to generate an air control operation grid; Based on the air quality anomaly cause-effect graph model, the air control operation grid is jointly adjusted at multiple levels to obtain the first air control group; Based on the indoor air quality target matrix, the first group of air control systems is evaluated for air quality adaptability and optimization to generate the second group of air control systems. Based on the second air control group, energy consumption minimization optimization is performed to generate the first air control strategy.

5. The indoor air quality adaptive control method using deep learning as described in claim 4, characterized in that, Based on the indoor air quality target matrix, an air quality adaptation evaluation and optimization is performed on the first group of air control systems to generate a second group of air control systems, including: Based on the first group of air control schemes, extract the Pth air control scheme, where P is a positive integer; Based on the indoor-outdoor linkage monitoring matrix and the real-time indoor air quality data, indoor air quality is predicted according to the Pth air control scheme to obtain the Pth indoor air quality sequence. Based on the indoor air quality target matrix, the Pth indoor air quality sequence is evaluated by twin comparison to obtain the Pth air quality fitness value; Determine whether the Pth air quality adaptation value is greater than or equal to the air quality adaptation threshold; If the Pth air quality adaptation value is greater than or equal to the air quality adaptation threshold, the Pth air control scheme is added to the second air control group.

6. The indoor air quality adaptive control method using deep learning as described in claim 1, characterized in that, Based on the first air control strategy, a second air control strategy is obtained by performing deep learning-based air quality situation smoothing optimization, including: Deep learning is used to generate an air quality situation prediction model based on the indoor air quality status record set. The indoor-outdoor linkage monitoring matrix, the real-time indoor air quality data, and the first air control strategy are input into the air quality situation prediction model to obtain a multi-dimensional air quality situation curve. Based on the multidimensional air quality situation curve, abrupt changes are identified to obtain a set of multidimensional situation abrupt change segments. Based on the multidimensional set of abrupt change segments, attention is allocated to the impact of abrupt changes, and the weight of each abrupt change impact is determined. Based on the influence weights of each situational change, the first air control strategy is optimized by differential smoothing constraints according to the set of multidimensional situational change segments to generate the second air control strategy.

7. The indoor air quality adaptive control method using deep learning as described in claim 1, characterized in that, Based on the second air control strategy, the multi-trend characteristic interference correction of the air control deployment grid is executed to obtain the third air control strategy, including: The air control deployment grid is simulated and controlled according to the second air control strategy to obtain a simulation dataset of each device's operation. Based on the simulation dataset of each device's operation, fault trend characteristics are analyzed to obtain the fault trend characteristics of each device. Based on the simulation dataset of each device's operation, the aging trend characteristics are analyzed to obtain the aging trend characteristics of each device. Based on the fault trend characteristics and aging trend characteristics of each device, the second air control strategy is adaptively modified to correct for interference effects, thereby generating the third air control strategy.

8. The indoor air quality adaptive control method using deep learning as described in claim 1, characterized in that, Based on the deployment characteristics of indoor air conditioning equipment, an air conditioning deployment grid is constructed, including: Based on the installation characteristics of the indoor air conditioning equipment, an installation grid for the conditioning equipment is constructed; Based on the operational characteristics and linkage characteristics of the indoor air conditioning equipment, the installation grid of the air conditioning equipment is enhanced with multi-head attention to generate the air conditioning deployment grid.

9. The indoor air quality adaptive control method using deep learning as described in claim 1, characterized in that, Construct an indoor-outdoor coordinated monitoring matrix, including: Based on the sensor network, real-time data on indoor space characteristics, indoor occupant monitoring data, and outdoor environmental monitoring data are acquired. The indoor space characteristic data, the indoor personnel monitoring data, and the outdoor environmental monitoring data are cleaned to generate the indoor-outdoor linkage monitoring matrix.

10. An indoor air quality adaptive control system applying deep learning, characterized in that, The system is used to implement the indoor air quality adaptive control method using deep learning as described in any one of claims 1-9, the system comprising: The air conditioning deployment grid construction module is used to construct an air conditioning deployment grid based on the deployment characteristics of indoor air conditioning equipment. The indoor air quality target mining module is used to construct an indoor-outdoor linkage monitoring matrix and mine indoor air quality targets based on the indoor-outdoor linkage monitoring matrix to obtain an indoor air quality target matrix. The multivariate anomaly causal modeling module is used to perform multivariate anomaly causal modeling on real-time indoor air quality data based on the indoor-outdoor linkage monitoring matrix and the indoor air quality target matrix, and to establish an air quality anomaly causal graph model. The multi-level adjustment and optimization module is used to perform multi-level adjustment and optimization on the air control deployment grid based on the indoor air quality target matrix and the air quality anomaly cause-effect graph model to obtain the first air control strategy. An air quality situation smoothing optimization module is used to perform deep learning-based air quality situation smoothing optimization based on the first air control strategy to obtain a second air control strategy. The multi-trend characteristic interference correction module is used to perform multi-trend characteristic interference correction of the air control deployment grid according to the second air control strategy, and obtain the third air control strategy.

Images