A method and system for regional linkage monitoring, feedback and purification control of multiple units

Abstract

The present invention discloses a method and system for regional linkage monitoring, feedback and purification control of multiple units. The method comprises: utilizing an Internet of Things architecture to distributely collect operating status data of operating parameters of each node of the units in the region at the current moment; utilizing a tensor splicing method to analyze the operating status data and extract multiple features in the operating status data; splicing the multiple features in the operating status data to obtain a three-dimensional joint feature tensor; performing predictive analysis on the three-dimensional joint feature tensor to obtain an abnormal probability distribution; and dynamically adjusting the current operating parameters of the operating status data of each unit node and an air purification device based on the abnormal probability distribution to perform purification control on the current region where the multiple units are located.

Description

Technical Field

[0001] The present invention relates to the field of air purification, and in particular to a method and system for regional linkage monitoring and feedback purification control of multiple units. Background Art

[0002] With the acceleration of urbanization and the continuous advancement of industrialization, air pollution is becoming increasingly serious, especially in enclosed environments such as subways, where air quality has a particularly significant impact on passenger health and equipment operation. Traditional air purification systems often rely on a single device to treat pollutants, lacking targeted and intelligent coordinated control. Not only are these systems unable to cope with sudden pollution emergencies in complex environments, they also suffer from significant deficiencies in energy efficiency optimization and resource scheduling. Therefore, finding a more intelligent and efficient way to coordinate multiple units for real-time monitoring, feedback, and purification control has become a pressing issue.

[0003] Currently, many air purification control systems still rely on a single monitoring method to detect pollutants, such as air quality monitoring instruments to detect PM2.5, CO2 concentration, and other pollutant indicators. While these monitoring methods can provide real-time pollutant data, they often fail to fully consider the synergy between various units and the feedback loop on their environmental impact. More importantly, traditional control systems lack dynamic adjustment mechanisms based on real-time data analysis and prediction. They are unable to immediately dispatch and optimize the operating status of various units and purification equipment after pollution source detection, resulting in slow pollution control response and low operational efficiency.

[0004] Furthermore, existing air purification systems often fail to fully utilize advanced methods such as the Internet of Things (IoT) and deep learning to jointly analyze operating data from multiple units, effectively failing to extract multi-dimensional environmental and equipment operational characteristics. This is particularly true in complex environments (such as confined spaces like subways and tunnels), where the interrelationship between pollutant diffusion and equipment load is extremely complex. Traditional control methods are unable to accurately reflect the potential connections between various environmental factors in real time, nor can they predict potential pollution or equipment failure risks based on big data analysis.

[0005] Therefore, how to analyze air pollution and unit power consumption through the linkage of multiple units to adjust control parameters and achieve efficient and intelligent pollution purification control has become the focus of research and application. Summary of the Invention

[0006] The object of the present invention is to provide a method and system for regional linkage monitoring, feedback and purification control of multiple units, which solves the above-mentioned technical problems pointed out in the prior art.

[0007] The present invention provides a regional linkage monitoring feedback purification control method for multiple units, comprising the following steps: utilizing the Internet of Things architecture to collect distributed operating status data of operating parameters of each node of the units in the area at the current moment;

[0008] Analyzing the operating status data using a tensor splicing method to extract multiple features from the operating status data; splicing the multiple features in the operating status data to obtain a three-dimensional joint feature tensor; performing predictive analysis on the three-dimensional joint feature tensor to obtain an abnormality probability distribution;

[0009] Dynamically adjust the current operating parameters of the operating status data of each unit node and the air purification device according to the abnormal probability distribution, and perform purification control on the current area where the multiple units are located;

[0010] The operating status data includes: air quality parameters, unit operating temperature, fluid component concentration and power consumption data of each node; the multiple features in the operating status data include: standard material abundance matrix, radial heat flux density characteristics and power consumption data characteristics of each node.

[0011] Compared with the prior art, the embodiments of the present invention have at least the following technical advantages: Analysis of the above-mentioned regional linkage monitoring feedback purification control method and system for multiple units provided by the present invention shows that in specific applications, by performing peak normalization processing on the air quality parameters, fluid component concentration, unit operating temperature and power consumption data, the influence of different data scales is eliminated, and all data are mapped to a unified standard range, so that different types of data can be directly compared and fused; after the air quality parameters and fluid component concentrations are peak normalized, a standard material abundance matrix is ​​formed. This matrix uses tensor splicing technology to fuse the data of multiple monitoring points together to reflect the overall changes in air quality and pollutant concentrations in the area. In this way, collaborative analysis of multi-point data can be achieved, environmental changes can be monitored in real time, and a basis for the dynamic adjustment of the operating status of subsequent purification equipment can be provided; by calculating the local gradient of the operating temperature of the computer group, the temperature change trend and the direction of heat transfer are further described, and the calculated radial heat flux intensity provides a physical quantitative indicator for the heat exchange process inside the equipment;

[0012] This indicator provides data support for the operating status and energy efficiency optimization of the equipment, and can reveal the direction and intensity of heat flow, thereby providing a basis for the linkage adjustment of the units; the normalization of power consumption data allows the power consumption characteristics of each node to be compared on a unified scale. This processing helps to reduce data fluctuations caused by measurement errors or equipment fluctuations, making the power consumption data of different nodes smoother and more stable; the normalization of power consumption data not only improves the reliability of the data, but also makes energy efficiency analysis more accurate, which helps to optimize the overall energy consumption in the process of multi-unit linkage; the standard material abundance matrix, radial heat flux density characteristics and power consumption data are channel-expanded and merged into a three-dimensional tensor through the tensor splicing method; the three-dimensional joint feature tensor combines monitoring data of different dimensions together, which can fully reflect the status of each unit in the environmental purification process. Through the joint analysis of multi-dimensional data, the system can monitor the operating status of the equipment in real time, and adjust the equipment operation in time according to data changes to ensure the optimization of the overall purification effect and avoid instability caused by uncoordinated operation between equipment;

[0013] Furthermore, by initializing the number of channels in the three-dimensional joint feature tensor and using the attention mechanism to assign different weights to each channel, the system can precisely adjust the influence of each feature. This initialization method ensures that data from all units can be processed at the same scale, avoiding processing inconsistencies caused by data differences. Guided by the attention mechanism, the system can focus on the feature data that has the greatest impact on environmental quality (such as air quality and temperature), providing more accurate data input for subsequent analysis. Next, a bidirectional recurrent neural network (RNN) reads the data in a forward and reverse time series.

[0014] Comprehensively capture historical and future information in environmental data; enable each unit to not only evaluate its own status, but also perceive changes in other units, thereby strengthening the synergy between multiple units; this two-way information capture helps the system monitor environmental changes or pollution spread across units in real time, identify potential problems in advance and optimize resource allocation; extract dynamic change features in GRU, and the unit can sensitively capture subtle changes in the environment in real-time monitoring; ensure that even if the sensor changes of a certain unit are not obvious, the abnormalities of other units can be identified and responded to in time; this interconnected feature provides strong data support for dynamic adjustments across units, reducing the risk of pollution spread; in addition, the application of adaptive learning rate strategy can adjust the learning speed at different stages to avoid overfitting or underfitting during training The adaptive learning rate ensures the stability of the collaborative learning process of multi-unit data, avoids the negative impact of over-training of individual units on unit stability, and improves the overall efficiency of the unit. Through weighted summation of attention, it can focus on the most critical features and further improve the response capability to environmental changes. Finally, a graph neural network is used to establish a pollution transmission topology between unit nodes to ensure that the diffusion process of pollutants is monitored in real time. By combining the pollution diffusion rate with the power consumption coupling indicator, the unit can realize the prediction and early warning of pollution and power consumption anomalies, providing a scientific basis for environmental governance. The linkage effect of multiple unit nodes is reflected in the ability to capture and process cross-unit data changes in real time, optimize information exchange and resource allocation between units, effectively reduce pollution spread, and improve the accuracy and response speed of environmental monitoring and equipment management. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a main flow chart of a regional linkage monitoring feedback purification control method for multiple units in Example 1;

[0016] Figure 2 This is a flow chart of analyzing operating status data of a regional linkage monitoring feedback purification control method for multiple units in Example 1;

[0017] Figure 3 This is a flow chart of an attention mechanism analysis of a regional linkage monitoring feedback purification control method for multiple units in Example 1;

[0018] Figure 4 This is a flow chart of a graph neural network analysis of a regional linkage monitoring feedback purification control method for multiple units in Example 1;

[0019] Figure 5 This is a flow chart of a method for regional linkage monitoring, feedback, and purification control of multiple units using hidden Markov analysis according to Example 1;

[0020] Figure 6This is a main flow chart of an expanded scheme of a method for regional linkage monitoring, feedback, and purification control of multiple units according to Example 1;

[0021] Figure 7 A schematic diagram of pollution and power consumption of a regional linkage monitoring and feedback purification control method for multiple units in Example 1;

[0022] Figure 8 This is a flow chart of a regional linkage monitoring, feedback and purification control system for multiple units according to Example 2;

[0023] Label: acquisition module 10; analysis module 20; control module 30. DETAILED DESCRIPTION

[0024] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0025] The present invention will be further described in detail below through specific embodiments in conjunction with the accompanying drawings.

[0026] Example 1

[0027] like Figure 1 As shown, the present application provides a regional linkage monitoring feedback purification control method for multiple units, including the following steps:

[0028] S1: Using the IoT architecture to collect the current operating status data of the operating parameters of each node in the area where multiple units are located;

[0029] The operating status data includes: air quality parameters, unit operating temperature, fluid component concentration and power consumption data of each node;

[0030] It should be noted that in the area of ​​the subway where the air needs to be tested, the operating status data of multiple units in the area are collected. The unit operating status data includes: air quality parameters (usually including PM2.5, CO2, VOC and other indicators, which are used to measure the concentration of suspended particulate matter, harmful gases and other pollutants in the environment, reflect the level of air pollution in the area, find the source of pollution in time, and ensure that the regional air quality meets the predetermined standards), unit operating temperature (that is, the temperature environment in which the unit equipment is located during operation, evaluate the status of the unit and the surrounding environment, prevent the performance of the equipment from being affected by overheating or abnormal humidity, and at the same time, temperature will also affect the fluidity of pollutants around the unit), fluid component concentration (that is, the temperature of the unit equipment during operation, and the surrounding environment). The data also includes information on the concentration distribution of various substances in the fluid (such as particulate matter and chemical components suspended in the air), which is used to detect whether the fluid is contaminated or has undergone changes in chemical composition, helping to determine whether there are any abnormalities in the system. Fluid composition data helps determine the type and concentration of pollutants, thereby optimizing purification strategies. It also includes information on power consumption at each node (i.e., recording the energy consumption of each unit or sensor node during operation, reflecting the operating load and energy efficiency of the equipment or node, and helping to determine whether there are excessive energy consumption or abnormal fluctuations. In linkage analysis, power consumption data is combined with other parameters (such as temperature) to detect equipment anomalies or potential failures in advance, while providing data support for energy-saving optimization and scheduling control).

[0031] At the same time, air quality sensors, temperature sensors, triaxial acceleration sensors, fluid composition sensors, and power consumption monitoring modules are installed at each node of multiple units in the area to collect energy consumption data of each node in real time. The operating status data is collected through the various devices and set modules.

[0032] S2: Analyze the operating status data using a tensor splicing method to extract multiple features from the operating status data; splice the multiple features in the operating status data to obtain a three-dimensional joint feature tensor; perform predictive analysis on the three-dimensional joint feature tensor to obtain an abnormality probability distribution;

[0033] The multiple features in the operating status data include: standard material abundance matrix, radial heat flux density features, and power consumption data features of each node;

[0034] It should be noted that the tensor splicing method combines various operating status data (such as air quality, temperature, fluid component concentration, and power consumption data) to form a three-dimensional joint feature tensor. By splicing multidimensional data, this method can effectively extract the relationship between multiple environmental factors. Based on the spliced ​​three-dimensional feature tensor, machine learning or deep learning algorithms are used for predictive analysis to obtain the probability distribution of pollution and energy consumption anomalies for each unit or region. The joint analysis of data features can discover potential correlations between different factors in the environment. For example, an increase in unit temperature may cause pollutants to spread faster, and power consumption anomalies may be related to equipment failure or environmental pollution. By combining different data features, the operation of the unit and the working status of the air purification system can be more comprehensively analyzed, and potential problems can be discovered in advance.

[0035] Accurately extracting and analyzing multi-dimensional data features provides strong support for predicting abnormal events (such as air pollution or equipment failure). Multi-dimensional data analysis using tensor splicing can more accurately detect abnormalities (such as excessive pollutant concentrations or abnormal energy consumption), enabling the system to provide timely warnings and adjustments.

[0036] Through predictive analysis of three-dimensional feature tensors, the system can generate anomaly probability distributions that reflect the inter-unit feedback loop across the entire region. This analysis accurately reflects the potential risks of different units or nodes, providing a basis for subsequent control strategies.

[0037] S3: Dynamically adjust the current operating parameters of the operating status data of each unit node and the air purification device according to the abnormal probability distribution, and perform purification control on the current area where the multiple units are located.

[0038] It should be noted that based on the abnormal probability distribution obtained in step S2, the operating parameters of the unit are dynamically adjusted, such as temperature, flow, power consumption, etc., to ensure that the unit is in the best working condition; at the same time, the operating parameters of the air purification device are adjusted to make it more effective in dealing with pollution problems; based on real-time monitoring data and predictive analysis results, the system automatically adjusts the operating mode of the unit and purification device to optimize air quality, reduce pollution sources, and achieve energy efficiency optimization;

[0039] By dynamically adjusting system parameters, it can respond in real time to current pollution levels and unit operating status, thus avoiding over-operation or equipment waste and ensuring efficient, energy-saving, and environmentally friendly operation of the system. Especially in high-pollution or high-load situations, it can respond quickly and adjust the system to achieve optimal control effects. The automated control mechanism makes pollution control and energy efficiency optimization more intelligent, ensuring that air quality and energy consumption in the area are maintained at appropriate levels.

[0040] By optimizing the joint control of multiple units, the dual goals of energy conservation and reduction of environmental pollutants can be achieved. For example, during certain periods of time, units can work together to reduce energy consumption, while at the same time, the working intensity of air purification devices can be adjusted to ensure that the concentration of air pollutants is reduced to the standard range.

[0041] Specifically, if Figure 2 As shown, in step S2, the operating status data is analyzed using a tensor splicing method to extract multiple features from the operating status data; the multiple features in the operating status data are spliced ​​to obtain a three-dimensional joint feature tensor; the three-dimensional joint feature tensor is predictively analyzed to obtain an abnormal probability distribution. The specific operation steps are as follows:

[0042] S21: performing peak normalization processing on the air quality parameters and the fluid component concentrations using a tensor splicing method to obtain a standard substance abundance matrix;

[0043] It should be noted that in order to eliminate the influence of different data scales, each data series is normalized according to its peak value (i.e., the peak value is usually the maximum value of the air quality parameter and fluid component concentration, and the air quality parameter and fluid component concentration can both reflect the air quality in the area. Therefore, they can be combined together for peak normalization and combined analysis) to map the data to a unified and standardized range (such as 0 to 1); for example, if the maximum value of a certain air quality parameter is A, each node is divided by A to obtain the normalized value; similar processing is performed on the fluid component concentration data; after this processing, all data are on a similar scale, which is conducive to subsequent comparison and fusion; after normalization, each set of data becomes a vector or matrix; using the tensor splicing method (i.e., merging different data "along" a certain dimension), the air quality and fluid composition data are spliced ​​into a multi-channel joint tensor; the spliced ​​tensor can usually be regarded as a standard substance abundance matrix, in which each channel represents a standardized substance concentration information, reflecting the "abundance" of each component (i.e., a multi-dimensional matrix);

[0044] In environmental monitoring, the interconnectedness of the units can collaboratively reflect changes in regional environmental quality by normalizing and fusing multiple different types of data (such as air quality parameters and fluid component concentrations) into a unified standard substance abundance matrix. Fusion of data from multiple monitoring points using unified standards facilitates the coordinated operation of the real-time monitoring system, ensuring accurate detection of ambient air quality or pollutant concentrations, and enabling subsequent purification equipment to dynamically adjust its operating status based on changes in air quality or pollutant concentrations. This data fusion improves the overall system response speed, ensuring that equipment can be collaboratively optimized under different conditions, thereby achieving more accurate and efficient environmental purification control.

[0045] S22: Calculating a local temperature gradient of the unit operating temperature, calculating a gradient direction of the unit operating temperature based on the local temperature gradient, describing the gradient direction of the unit operating temperature in a radial direction, and calculating radial heat flux intensity; summarizing the radial heat flux intensities of each node to obtain a radial heat flux density characteristic; ;

[0046] Where, Indicates the The gradient direction of the unit operating temperature of the node Radial heat flux intensity in each sub-direction;

[0047] Indicates that in this The number of valid measurement segments sampled in the sub-direction of the gradient direction of the unit operating temperature of each node;

[0048] Expressed as The node In the direction of the child The radial projection factor of each sampling point;

[0049] Indicates the The node The first direction The effective temperature gradient of a sampling point (that is, the effective temperature gradient of the location where the temperature of one of the units is running at the multiple unit nodes);

[0050] The effective temperature gradient is obtained by correcting (for example, weighting or modifying) the temperature gradient and the local temperature gradient, and is defined as follows: ;

[0051] Where, Expressed as temperature gradient;

[0052] Expressed as The node In the direction of the child The humidity factor of the corresponding position of each sampling point;

[0053] Expressed as correction factor;

[0054] Explanation: In the above formula, Indicates that All in the direction of The average of the sampling segments is used to obtain the overall radial heat flow index; the weighted average of the inner layer is: It means that in each sampling segment, The local sampling points are averaged to reduce the influence of individual point errors; the radial projection factor Project the temperature gradient component of each sampling point to the ideal radial direction to ensure that the heat flux component along the radial direction is finally calculated. By introducing a humidity correction factor, the data can simultaneously reflect the impact of temperature on heat conduction. For example, humidity can enhance or weaken the heat conduction effect under certain conditions. Absolute value processing represents the absolute value of the outer layer, ensuring that the heat flux intensity (i.e., positive value) is obtained, regardless of the positive or negative direction, which facilitates subsequent optimization and statistics. Using the above formula, temperature data can be converted into radial heat flux intensity information, thus providing a physically reasonable quantitative factor for linked monitoring feedback control and multi-objective optimization.

[0055] At the same time, by calculating the unit operating temperature data, the gradient direction is calculated, which indicates the direction of the fastest change. This can be used to determine the direction of heat transfer or humidity diffusion, and the rate of change of the measured data in space or time. The gradient can reflect the trend of temperature or humidity changes (such as increase or decrease).

[0056] Calculating the local temperature gradient and the corresponding heat flux intensity can provide physical quantitative indicators for monitoring the unit's heat transfer process. By describing the radial direction of the heat flux, the direction and intensity of heat flow can be intuitively demonstrated. Temperature has a significant impact on heat conduction. Calculating its gradient and heat flux intensity can help reveal the heat exchange and humidity distribution within the unit, providing important data support for the unit's operating status and energy efficiency optimization. The description of the radial direction of the heat flux provides a physical basis for multi-objective optimization, helping to adjust the system to achieve optimal efficiency.

[0057] By performing gradient calculations on the unit's temperature data, the heat transfer and humidity diffusion within the equipment can be reflected. In purification equipment, this can optimize internal heat conduction and humidity distribution, ensuring that the equipment operates under appropriate temperature conditions. The linkage of the units allows equipment in different areas to adjust according to local temperature changes to ensure overall operating results. For example, when the temperature gradient of a unit is too large, the system can link adjustments with other units to maintain the stability and efficiency of the entire purification system, avoiding local overheating or uneven humidity that may lead to a decrease in purification effect.

[0058] S23: Normalizing the power consumption data of each node to obtain normalized power consumption data features of each node;

[0059] It should be noted that the raw values ​​of the power consumption data of each node may vary significantly by orders of magnitude. To ensure that this data is not adversely affected by scale issues when subsequently integrated with other features (such as heat flow and air quality), the power consumption data needs to be normalized. Normalization can also reduce data fluctuations caused by measurement errors or equipment fluctuations, thereby obtaining the power consumption data characteristics of each node, making the power consumption data of different nodes smoother and more stable overall.

[0060] Normalization of power consumption data is designed to eliminate the effects of magnitude differences in the raw power consumption data, making it comparable with other features (such as temperature and humidity) in subsequent analysis. Normalization brings power consumption data to the same scale, reducing the impact of equipment fluctuations or measurement errors. The stability and consistency of power consumption data helps ensure data quality when merging multi-channel features, leading to more accurate assessments of the unit's operating status and energy efficiency.

[0061] By normalizing the power consumption data of each unit, the impact of power consumption differences between devices can be eliminated, allowing the energy efficiency status of each unit to be uniformly compared. In environmental monitoring and purification, unit linkage is reflected in real-time monitoring of the energy efficiency and power consumption status of the equipment to ensure energy-saving operation of the equipment. When the power consumption of a certain device is abnormal, the system can adjust the operating status of other units in a coordinated manner to avoid excessive overall energy consumption and optimize the energy utilization efficiency of the purification system. At the same time, the synchronous adjustment of unit power consumption helps to ensure that the system avoids excessive energy consumption while meeting environmental requirements, thereby improving the cost-effectiveness of the entire environmental purification process.

[0062] S24: using a tensor splicing method to perform channel expansion on the standard material abundance matrix, radial heat flux density features, and power consumption data features of each node, so that the channel dimensions of the standard material abundance matrix, radial heat flux density features, and power consumption data are the same, and the same channel dimensions of each feature after expansion are spliced ​​in a second dimension (i.e., a channel / feature category dimension) to form a three-dimensional tensor as a three-dimensional joint feature tensor;

[0063] It should be noted that by splicing the standard material abundance matrix, radial heat flux density characteristics, and normalized power consumption data, different feature data can be combined into a three-dimensional tensor. This ensures that each feature has the same channel dimension in the same dimension and allows for comprehensive analysis of multiple information. Tensor splicing expands and merges multiple features, allowing different types of information to be jointly analyzed simultaneously. The spliced ​​three-dimensional tensor has a higher information density, which helps to evaluate the unit status from multiple angles.

[0064] By combining different environmental monitoring data (such as air quality, temperature, and power consumption) into a three-dimensional joint feature tensor, the role of each unit in the environmental purification process can be fully reflected. Unit linkage is reflected in the ability to provide real-time feedback on the status of the entire purification system through joint analysis of multi-dimensional data. Multi-channel features enable the system to coordinate and adjust according to changes in multi-source data. When the system detects abnormal performance of a unit, other units can be adjusted synchronously, improving the overall purification effect and reducing system instability.

[0065] S25: Analyze the three-dimensional joint feature tensor using an attention mechanism to obtain a hidden change feature, assign weights to the hidden change feature to obtain an attention fusion feature; and predict an abnormal probability distribution based on the attention fusion feature.

[0066] It should be noted that the attention mechanism can dynamically assign weights to different features, highlighting important environmental monitoring data or equipment operating characteristics. During the environmental purification process, the linkage of the units can predict equipment failures or abnormal environmental indicators in advance by paying attention to the status changes of key equipment in the system (for example, changes in air quality or early signs of equipment failure). By predicting the probability distribution of abnormalities, the system can provide fault warnings and timely coordinate and adjust equipment operations to avoid environmental quality degradation or equipment damage. This linkage mechanism not only improves the system's response speed, but also enhances the reliability and accuracy of the environmental monitoring system.

[0067] Specifically, if Figure 3 As shown, in step S25, the three-dimensional joint feature tensor is analyzed using the attention mechanism to obtain hidden change features, and weights are assigned to the hidden change features to obtain attention fusion features; the abnormal probability distribution of the attention fusion features is predicted. The specific operation steps are as follows:

[0068] S251: confirming the number of channels of the three-dimensional joint feature tensor, and initializing the three-dimensional joint feature tensor using the attention mechanism and the number of channels to obtain an initialized three-dimensional joint feature tensor;

[0069] It should be noted that when the channel dimension of each feature is fused by the tensor splicing method in step S24, the channel dimension has been unified, so the number of channels of the three-dimensional joint feature tensor can be easily determined; the attention mechanism is used to assign an initial weight to each channel of the determined number of channels, thereby achieving the purpose of initializing the three-dimensional joint feature tensor;

[0070] The unification of channel dimensions ensures data compatibility, allowing features from different sources to be processed at the same scale, thereby improving the efficiency of data fusion. By using the attention mechanism to assign different initial weights to each channel, the influence of each feature can be more flexibly adjusted, strengthening the correlation between features and improving the model's adaptability to different factors in environmental monitoring. To accurately reflect environmental changes and pollution, the system can prioritize feature data (such as air quality and temperature) that have the greatest impact on environmental quality through the initialization of channel weights, thereby enhancing monitoring accuracy.

[0071] The sensor data from each unit is processed uniformly, ensuring that data from all units can be compared at the same scale, avoiding processing inconsistencies caused by data differences. The environmental data from each unit is fused together, and through the attention mechanism, the system can identify which units or sensors are more important, ensuring that the characteristics of key units are focused.

[0072] S252: Using a bidirectional recurrent neural network, all channels of each node in the initialized three-dimensional joint feature tensor are used as input sequence data;

[0073] The bidirectional recurrent neural network reads the sequence data in a forward and reverse direction, captures the corresponding past tense time series and future tense time series in the forward and reverse sequence data, and obtains the hidden layer dimension;

[0074] It should be noted that in the initialized joint feature tensor, the data of each channel in each node is regarded as a sequence input. The bidirectional RNN enters from the head (forward) and tail (reverse) of the sequence respectively, thereby capturing the "past" (historical information) and "future" (subsequent information) at the same time. The bidirectional RNN can process the past (historical information) and future (subsequent information) of the data simultaneously. This feature is particularly suitable for processing time series data and can more comprehensively understand the global trends and local changes in the process of environmental changes.

[0075] The hidden layer output of a bidirectional RNN is typically a vector array, whose dimension (hidden layer dimension) determines the amount of information it can retain. Hidden layer dimension refers to the number of neurons in the recurrent unit. Higher hidden layer dimensions can capture richer details (such as local temperature changes) but may introduce noise. Lower dimensions focus on key trends and facilitate faster calculations. Therefore, in practical engineering, a balance must be struck between accuracy and computational efficiency. By using RNNs to comprehensively analyze device operating data (such as temperature and power consumption), potential device failures or pollution spread trends can be detected in advance, allowing for dynamic adjustments to purification strategies.

[0076] The simultaneous capture of historical and future data for each unit enables real-time information exchange between units, helping to detect environmental changes or pollution spread across units. The bidirectional RNN allows each unit to assess its own status while also paying attention to changes in other units, leading to global optimization decisions based on multi-unit data.

[0077] S253: Using a gated recurrent unit (GRU) to extract dynamic change features from the hidden layer dimension to obtain hidden change features;

[0078] It should be noted that the GRU structure can maintain high efficiency and accuracy when processing long time series data, and can determine whether the system has anomalies by effectively extracting dynamic change features. The GRU can automatically learn the implicit correlations between different data (such as the relationship between temperature, humidity and pollution), and thus accurately detect abnormal changes through interactive analysis of different features. By extracting dynamic change features, the system can sensitively capture changes in air pollution or other environmental indicators in real-time monitoring, provide early warnings, and take measures.

[0079] By extracting dynamic features using GRU, multiple units can detect changing trends in the overall environment. This ensures that even if a unit's sensor changes are subtle, abnormalities in other units can help identify potential problems in the system. If a unit experiences dynamic changes (such as abnormal pollution or power consumption), other units can make coordinated adjustments based on the extracted features to reduce the spread of environmental pollution.

[0080] S254: Calculating a learning rate decay function for the past time series and the future time series using an adaptive learning rate strategy;

[0081] Calculating the attention weight of the learning rate by the learning rate decay function;

[0082] It should be noted that adaptive learning rates allow the model to adjust weights at different learning speeds at different stages, avoiding overfitting or underfitting during training. Calculating the learning rate and applying it to the attention mechanism helps to more flexibly adjust the influence of features, thereby improving the model's sensitivity to complex time series data. Adaptive learning strategies enable the system to better handle fluctuations in different data, thereby more accurately capturing subtle changes in the environment (such as the impact of climate change on pollution).

[0083] During the collaborative learning process of each unit, a consistent training pace is maintained to avoid deviations caused by overtraining of some unit data, thereby maintaining the overall stability of the system. Learning rate decay ensures the speed of collaborative training of multi-unit data, reduces the waste of computing resources caused by overtraining of a single unit, and improves overall efficiency.

[0084] S255: performing weighted summation on each channel in each node of the hidden change feature using the attention weight to obtain an attention fusion feature;

[0085] It should be noted that the attention mechanism can focus on the most influential features and weightedly fuse these features, making it more sensitive to key changes in the air. Through weighted attention, it can prioritize features that have the greatest impact on air pollution or environmental changes, improving the accuracy of anomaly detection.

[0086] By focusing on the environmental impact of each unit, higher weights can be assigned to important units, ensuring focus on critical units and improving the responsiveness of the entire system. Units can adjust their input weights based on current environmental conditions, optimize linkage behavior, and ensure optimal data integration.

[0087] S256: Using a graph neural network to determine the coordinates of each node of the unit, a pollution transmission topology structure between the unit nodes is constructed using the coordinates of the unit nodes; based on the pollution transmission topology structure between the unit nodes and the attention fusion feature, a concentration gradient diffusion rate between adjacent nodes is calculated; using the concentration gradient diffusion rate between adjacent nodes to introduce a radial heat flux density feature for calculation, a power consumption coupling index of each unit node is obtained; the concentration gradient diffusion rate between adjacent nodes and the power consumption coupling index of each unit node are predicted to obtain a distribution probability of pollution and power consumption anomalies (i.e., anomaly probability distribution, wherein the anomaly probability distribution is subjected to a step-by-step analysis of the concentration gradient diffusion rate and the power consumption coupling index);

[0088] It should be noted that graph neural networks (GNNs) can process graph-structured data and use the connection information between nodes for learning. They can determine the coordinates of each unit node and construct the pollution transmission topology between unit nodes based on these coordinate information. The node coordinates of each unit node determine its relationship in space and the pollution transmission path. The pollution transmission topology constructs the pollution transmission relationship between them based on the coordinates between the nodes. For example, pollution between adjacent nodes may be transmitted to each other through airflow or other physical phenomena. Based on the pollution transmission topology, the concentration difference of pollutants between adjacent nodes is calculated, and their diffusion rate is estimated. This is calculated by combining the topological information with air quality parameters and fluid component concentrations (such as pollution concentration) through the attention mechanism in the graph neural network. It can monitor the diffusion dynamics of pollutants between different unit nodes in real time, provide strong data support for environmental governance, and help predict the development trend of pollution. By introducing the radial heat flux density feature and combining the pollution diffusion rate, the power consumption coupling index of each unit node is calculated. Heat flux reflects the energy transfer rate, which can reveal the energy consumption of the unit under different environmental conditions. By coupling the concentration gradient diffusion rate between adjacent nodes with the power consumption indicator of each unit node, predictions can be made to provide early warnings of pollution and power consumption anomalies, enabling timely measures to purify the environment and adjust power consumption, ensuring equipment safety and reducing operating costs.

[0089] Each unit is linked based on pollution transmission paths to predict pollution diffusion trends in advance and optimize resource allocation to ensure that pollution sources are dealt with promptly. The Markov chain model can predict the probability of pollution spread for each unit, ensuring that pollution can be quickly identified and suppressed in other units, enhancing the system's linkage response capabilities, thereby reducing the impact of pollution on the environment and improving the overall purification effect.

[0090] During the analysis process, it is necessary to fully consider the pollution diffusion and power consumption effects of the unit nodes in order to more accurately monitor the spread of pollutants and the operating status of the equipment. The air pollutant concentration diffusion rate and power consumption of each unit node are two interrelated factors that jointly affect environmental quality and equipment performance. As an effective method for processing time series data, the Hidden Markov Model (HMM) can be used to dynamically predict the effective concentration diffusion rate and power consumption coupling indicators of the node. This can provide a prediction of the future state of the system, identify potential risks of excessive pollution or abnormal equipment operation in advance, and avoid accidents. Through the smooth prediction of these data, the distribution probability of pollutants and power consumption anomalies can be obtained, which helps the monitoring system to intervene and regulate impending abnormal situations at an early stage. Therefore, a detailed analysis of the distribution probability of pollution and power consumption anomalies is required:

[0091] Specifically, if Figure 4As shown, in step S256, the coordinates of the unit nodes of each node of the unit are determined by using a graph neural network, and a pollution transmission topology structure between the unit nodes is constructed based on the coordinates of the unit nodes; the concentration gradient diffusion rate between adjacent nodes is calculated based on the pollution transmission topology structure between the unit nodes and the attention fusion feature; the radial heat flux density feature is introduced by using the concentration gradient diffusion rate between adjacent nodes to calculate and obtain the power consumption coupling index of each unit node; the concentration gradient diffusion rate between adjacent nodes and the power consumption coupling index of each unit node are predicted to obtain the distribution probability of pollution and power consumption anomalies. The specific operation steps are as follows:

[0092] In steps S2561 and S2562, a specific solution is to introduce a graph neural network into each node of the unit through the Internet of Things architecture to determine the coordinates of the unit nodes, and to construct a pollution transmission topology structure between the unit nodes based on the coordinates of the unit nodes; and to analyze the standard material abundance matrix using the attention fusion feature to obtain the concentration of air pollutants collected by each unit node.

[0093] The concentration gradient diffusion rate between adjacent nodes of the air pollutant concentration is analyzed by the pollution transmission topology structure between the unit nodes. The specific operation steps of this scheme are as follows:

[0094] S2561: Introducing a graph neural network into each node of the unit according to the IoT architecture to determine the coordinates of the unit nodes;

[0095] The length of the pipes connecting each unit node is calculated using the coordinates of each unit node (i.e., the pipe length is the medium for connection, transmission, and linkage between unit nodes. At the same time, the pipe length can also be used as the distance between each unit node, that is, the spatial distance. The pipe length can also be used to understand the pipe diameter, bending angle, etc.);

[0096] The pollution transmission topology between the unit nodes is constructed by using the air pollutant diffusion principle and fluid dynamics through the coordinates of the unit nodes and the length of the pipes connecting the respective unit nodes;

[0097] It should be noted that the IoT architecture not only collects and transmits data between unit nodes, but also understands the coordinate location of unit nodes and, in combination with graph neural networks (GNNs), determines the coordinates of each unit node. These coordinates help define the spatial relationships between nodes and facilitate the calculation of pipeline lengths. By utilizing GNNs to process graph-structured data, the system can accurately map and manage the complex relationships between unit nodes, ensuring that spatial layout is effectively captured. Understanding node coordinates and pipeline connections, combined with principles of air pollutant diffusion and fluid dynamics, helps to more detailed and realistically simulate how pollutants propagate through the system, effectively implementing pollutant monitoring and control.

[0098] S2562: Analyze the standard substance abundance matrix using the attention fusion feature to obtain the air pollutant concentration collected by each unit node;

[0099] The length of the pipeline connecting each unit node of the pollution transmission topology structure between the unit nodes is used as the spatial distance between adjacent unit nodes;

[0100] The air pollution concentration of each unit node is used to calculate the difference in air pollution concentration between adjacent unit nodes;

[0101] The concentration gradient diffusion rate between adjacent nodes is calculated by the spatial distance between the adjacent unit nodes and the difference in air pollution concentration between the adjacent unit nodes. The calculation formula is: ;

[0102] Where, It is expressed as the concentration gradient factor of the air pollutant concentration between the unit nodes;

[0103] and are respectively represented as the air pollutant concentrations at the j-th unit node and the i-th unit node (i.e. It represents the difference in air pollution concentration between adjacent unit nodes);

[0104] It is expressed as the spatial distance between the j-th unit node and the i-th unit node (i.e., the pipeline length);

[0105] Expressed as a directional correction factor (i.e. expressed as directional differences);

[0106] Expressed as a neighborhood concentration difference factor (i.e., using the concentration mean or concentration dispersion difference of each node (e.g., in its set of neighboring nodes) to describe local concentration fluctuations, that is, the local area is formed by all the unit nodes in the neighborhood around a unit node. Depending on the unit layout and installation, there may be one or more unit nodes in the neighborhood around a unit node);

[0107] It is expressed as the air pollutant concentration of the vth unit node in the neighborhood of the i-th unit node;

[0108] It is expressed as the factor of the radial heat flux density characteristics within the neighborhood of the i-th unit node (that is, N unit nodes, N(i) represents the set of unit nodes adjacent to the unit node i) (that is, the introduced radial heat flux intensity);

[0109] Expressed as attention weight (i.e., after the data are fused and deeply processed by the attention mechanism in step S264, an attention weight is calculated by the attention mechanism to adjust the comprehensive contribution of the channel (i.e., between nodes) and enhance the response of the key area);

[0110] It should be noted that in the above formula, Representation node and the basic concentration gradient between (concentration difference divided by distance); Represents the direction correction factor, when the inter-node connection is aligned with the preferred diffusion direction ( Smaller) its value is larger, thus amplifying the gradient effect; pass and The absolute value of is calculated, Representation node To Node The direction of the connection between them (that is, the unit node in one of the neighboring directions among the unit nodes in the numerous neighboring areas); Indicates the local main airflow or preferred diffusion direction (that is, the direction in which the airflow of the pollutant concentration may diffuse); Description Node The degree of unevenness of concentration within its neighborhood reflects local concentration fluctuations; is the weight calculated by the attention mechanism (i.e., used to adjust the comprehensive contribution of the channel (i.e., between nodes) and enhance the response of the key area);

[0111] By analyzing the standard substance abundance matrix (including air pollution concentration data collected by each unit node), an attention mechanism is used to help optimize the analysis and focus on more critical areas. The attention mechanism allows the system to prioritize nodes or connections that are more sensitive to pollution or significantly affected by surrounding environmental factors, thereby determining air pollutant concentrations. Since the spread of pollution concentration is also related to factors such as temperature and humidity, a radial heat flux density factor is introduced to calculate the pollution concentration propagation rate, thereby improving the accuracy of abnormal pollution.

[0112] By calculating the concentration gradient diffusion rate using spatial distance and pollution concentration differences, we can accurately measure and analyze the interconnected propagation of pollutants within the unit system. At the same time, with the help of the attention mechanism, we can enhance the monitoring capability of key pollution areas.

[0113] In steps S2563 and S2564, the specific solution is to introduce the factor that modifies the spatial attenuation factor and the radial heat flux density characteristic through the directional correction factor; the concentration gradient diffusion rate between adjacent nodes is updated through the modified spatial attenuation factor to obtain the effective concentration diffusion rate of each unit node;

[0114] The power consumption coupling index of each unit node is calculated by the factors of the effective concentration diffusion rate and radial heat flux density characteristics of each unit node and the power consumption data of each node. The specific operation steps of this solution are as follows:

[0115] S2563: introducing a modified spatial attenuation factor into the calculated concentration gradient diffusion rate between the adjacent nodes using the direction correction factor;

[0116] Using the modified spatial attenuation factor to correct the spatial restriction generated by the directional correction factor of the concentration gradient diffusion rate between adjacent nodes, updating the calculation formula of the concentration gradient diffusion rate between adjacent nodes, and obtaining the effective concentration diffusion rate of each unit node;

[0117] It should be noted that in actual environments, pollution transmission is not only affected by concentration gradients, but also by pipeline resistance, node layout, and spatial obstacles. Therefore, a spatial attenuation factor is introduced to correct the basic diffusion rate, making the assessment more consistent with the actual transmission process. ;in: is the attenuation constant, which can be determined based on parameters such as pipeline friction, pipe diameter, bending angle and node density; It is still the spatial distance between the nodes of the unit, reflecting the attenuation effect in spatial transmission;

[0118] A spatial attenuation factor is introduced to adjust the pollutant diffusion rate, taking into account factors such as pipeline friction, pipe diameter, and curvature to ensure that the diffusion of pollutants truly reflects physical limitations. If factors such as friction and spatial barriers are not considered, the prediction may be overly simple and inaccurate. This adjustment makes the pollution diffusion results closer to reality. In the real environment, there are many limiting factors (such as pipeline friction, curvature, node layout, etc.), which can significantly affect pollutant diffusion. Introducing these variables ensures that the calculation results can truly reflect the spread of pollutants.

[0119] S2564: Normalize the power consumption data of each node to obtain the normalized power consumption of each unit node (i.e., real-time power consumption).

[0120] Calculate the average effective diffusion rate between adjacent unit nodes of each unit node;

[0121] The power consumption coupling index of each unit node is obtained by introducing the factor of radial heat flux density characteristics according to the average effective diffusion rate between adjacent unit nodes and the normalized power consumption of each unit node. The calculation formula is: ;

[0122] in: It is represented as the set of unit nodes adjacent to unit node i;

[0123] Representation node The average effective diffusion rate between its adjacent nodes; Expressed as the normalized power consumption of each unit node (i.e., the impact of power consumption is incorporated so that the node reflects both the risk of air pollutant transmission and the equipment load or operating status in the overall coupling index);

[0124] It is expressed as the factor of the radial heat flux density characteristics within the range of all unit nodes in the surrounding area within the neighborhood of the i-th unit node;

[0125] It should be noted that high power consumption is often related to the operating status of the equipment and the local environmental thermal effect (that is, the radial heat flux density characteristics, that is, the radial heat flux intensity), which may affect the surrounding air flow and thus affect the transmission of pollutants. After standardizing the power consumption data of each node, the effective pollution diffusion rate of each node is combined with the normalized power consumption data and adjusted using the radial heat flux density factor. Power consumption affects the operating status of the machine and its impact on the environment. By combining energy consumption with pollution diffusion, the overall interactive and cooperative relationship of the unit can be reflected. High power consumption is usually related to factors such as equipment load and thermal effect. At the same time, thermal effect can also affect the surrounding air flow, thereby affecting the diffusion of pollutants and causing greater power consumption for the unit equipment. Combining these factors can more accurately reflect the impact of the equipment operating status on the environment. This not only helps monitor air quality, but also provides real-time understanding of the machine's operating status, optimize resource utilization, and improve environmental impact management.

[0126] S2565: Dynamically evolving the effective concentration diffusion rate of each unit node and the power consumption coupling index of each unit node using a hidden Markov algorithm to obtain a state transition matrix; performing smooth prediction using the state transition matrix to obtain distribution probabilities of pollution and power consumption anomalies;

[0127] It should be noted that the effective pollution diffusion rate and power consumption coupling index of each node are predicted using a hidden Markov model (HMM), thereby obtaining the distribution probability of pollution and power consumption anomalies. Hidden Markov models excel at processing time series data and can predict the future state of the system, which is crucial for anomaly detection and intervention. Because environmental data (such as pollution concentration and power consumption) are often time-series, using hidden Markov models can capture potential patterns in the data and predict when anomalies are likely to occur. By predicting pollution and power consumption anomalies, the system can intervene before problems occur, preventing pollution violations or equipment failures, thereby improving overall environmental health and safety.

[0128] Specifically, if Figure 5 As shown, Figure 6 As shown, in step S2565, the effective concentration diffusion rate of each unit node and the power consumption coupling index of each unit node are dynamically evolved by using a hidden Markov algorithm to obtain a state transition matrix; a smooth prediction is performed by using the state transition matrix to obtain the distribution probability of pollution and power consumption anomalies. The specific operation steps are as follows:

[0129] S25651: Use the hidden Markov algorithm to set the hidden state set of each unit node;

[0130] The hidden state set includes: normal hidden state probability, pollution abnormal hidden state probability, and power consumption abnormal hidden state probability;

[0131] It should be noted that this step first sets a hidden state set for each unit node. The hidden state set includes the following three states: normal hidden state, pollution abnormality hidden state and power consumption abnormality hidden state; normal hidden state probability: indicates the probability that the unit node is in normal operation; pollution abnormality hidden state probability: indicates the probability that the unit node has pollution abnormality (such as excessively high pollutant concentration); power consumption abnormality hidden state probability: indicates the probability that the unit node has power consumption abnormality (such as abnormally high energy consumption); Hidden Markov algorithm (HMM) can effectively process time series data. It can describe the "hidden state" of a system at each moment and the transition between these states; by setting these hidden state probabilities, the possible states of the system at different moments can be accurately described, and a basis for subsequent prediction and anomaly detection can be provided; it provides a clear state distinction for the system, facilitates monitoring of pollution and power consumption abnormalities, helps to discover potential risks in the environment in real time, and provides data support for decision-making;

[0132] S25652: Collect the effective concentration diffusion rate and power consumption coupling index of each unit node at time t;

[0133] The effective concentration diffusion rate and the power consumption coupling index at time t constitute an observation vector;

[0134] It should be noted that the effective concentration diffusion rate and power consumption coupling index at time t are collected from each unit node to form an observation vector. The effective concentration diffusion rate measures the diffusion speed and range of pollutants in the air. The power consumption coupling index measures the power consumption of each node and its correlation with pollution diffusion. In order to provide accurate observation information for the hidden Markov algorithm, the effective concentration diffusion rate and power consumption coupling index are used to construct an observation vector to better capture the relationship between pollution diffusion and power consumption, laying the foundation for subsequent prediction of pollution and power consumption anomalies. It provides real-time pollutant and power consumption data, supports pollution diffusion trends and power consumption optimization, and helps to accurately monitor and dynamically adjust environmental purification strategies.

[0135] S25653: Setting a priori probability vector at an initial moment for each hidden state probability in the hidden state set;

[0136] Using the attention weight, dynamically evolve the prior probability vector at the initial moment between hidden states to obtain a state transfer matrix;

[0137] It should be noted that a priori probability vector is set for each hidden state (normal, abnormal pollution, abnormal power consumption) at the initial moment, and then the attention weights are used to dynamically evolve between hidden states to obtain a state transition matrix. The initial priori probability vector represents the state distribution of the system at the initial moment, indicating the initial probability of each hidden state. Using the attention mechanism to dynamically evolve the transition between hidden states helps to further improve the algorithm's sensitivity and accuracy to different state transitions. By considering prior probabilities and dynamic evolution, the changing process of hidden states can be more accurately described. This method increases sensitivity to pollution and power consumption anomalies, helps to identify potential problems in advance, and implement effective control and regulation.

[0138] The specific scheme of steps S25654 to S25656 is as follows: according to the observation vectors at each moment before and after the moment t and the prior probability vector at the initial moment of each hidden state probability in the state transfer matrix, smoothing calculations are performed using the forward algorithm and the backward algorithm respectively to obtain the smoothed posterior probability at each moment;

[0139] The smoothed posterior probability at each moment is compared with the hidden state set to obtain the distribution probability of pollution and power consumption anomalies. The specific operation steps are as follows:

[0140] S25654: Calculate the prior probability of each hidden state at each moment using the forward algorithm based on the observation vector at each moment before moment t (i.e., from moment 1 to moment t) and the initial prior probability vector of each hidden state probability in the state transfer matrix.

[0141] It should be noted that the forward algorithm combines the observation vector and the state transition matrix to calculate the forward probability of each hidden state from time 1 to time t. The forward algorithm is an important tool for processing hidden Markov models. It can recursively calculate the hidden state probability at each moment based on historical observation data, providing strong support for subsequent predictions and inferences. Calculating forward probabilities using the forward algorithm can capture the evolution and potential changes in system states, helping to understand past behavioral patterns and providing a data foundation for further analysis. Providing accurate estimates of historical states helps understand long-term trends in pollution and power consumption changes, and provides a basis for adjusting response measures.

[0142] S25655: Based on the observation vector at each moment after time t (i.e., time t+1 to time n) and the prior probability vector of each hidden state at the initial moment in the state transfer matrix, a backward algorithm is used to calculate the posterior probability of each hidden state at each moment;

[0143] It should be noted that the backward algorithm combines the observation vector and the state transition matrix to calculate the backward probability of each hidden state from time t+1 to time n. The backward algorithm is used to infer state information at future moments. It can provide speculation about future states, thereby providing additional information for prediction at the current moment and enhancing prediction accuracy. By combining the backward algorithm with future observation information, the changing trends of states can be more comprehensively assessed, improving overall prediction accuracy. It also improves the ability to predict future pollution and power consumption anomalies, enabling the early identification of potential problems and the timely implementation of measures to reduce the impact of pollution and power consumption anomalies on the environment.

[0144] S25656: Smoothing the posterior probability of each hidden state at each moment using the prior probability of each hidden state at each moment to obtain a smoothed posterior probability at each moment;

[0145] The distribution probability of pollution and power consumption anomalies is obtained by comparing the smoothed posterior probability at each moment with the probability of each hidden state in the hidden state set;

[0146] It should be noted that the smoothed posterior probability at each moment is obtained by combining the prior probability and the posterior probability at each moment through smoothing calculation. The combination of the prior and posterior probabilities can improve the accuracy of state prediction because the prior probability uses historical information, while the posterior probability incorporates future information. Smoothing calculation can remove noise and improve the stability of the overall estimate. Through smoothing, the system's accurate estimation of the state at each moment is further improved, thereby enhancing the detection capability of pollution and power consumption anomalies. It provides more accurate moment estimates for the detection of pollutants and power consumption anomalies, helping to formulate timely response measures to reduce environmental pollution and energy consumption.

[0147] The smoothed posterior probability obtained at each moment is compared with the normal hidden state probability, the pollution abnormal hidden state probability, and the power consumption abnormal hidden state probability in the hidden state set. When the smoothed posterior probability at each moment is close to the normal hidden state probability, it means that the concentration of heavy air pollution is normal and the power consumption of the unit is also normal; when the smoothed posterior probability at each moment is close to the pollution abnormal hidden state probability, it means that the pollution concentration exceeds the standard and the power consumption of the unit is normal; when the smoothed posterior probability at each moment is close to the power consumption abnormal hidden state probability, it means that the unit consumes too much power when analyzing the transmission pollution concentration, which shows an abnormality; Figure 7 As shown in the figure (the orange dotted line marks pollution anomalies; the cyan dotted line marks power consumption anomalies; the red solid line represents pollution concentration (μg / m³), showing the dynamic change trend of pollutant concentration over time; the blue solid line represents power consumption (W), showing the change trend of power consumption of each unit node over time). In the subsequent comparison results, the distribution probability of pollution and power consumption anomalies is obtained, and further adjustments are made to the system equipment.

[0148] Example 2

[0149] like Figure 8 As shown, the present invention also provides a regional linkage monitoring feedback purification control system for multiple units, including: an acquisition module 10; an analysis module 20; a control module 30;

[0150] The collection module 10 is used to collect the operating status data of the operating parameters of each node of the unit in the area where multiple units are located at the current moment in a distributed manner using the Internet of Things architecture;

[0151] The analysis module 20 is configured to analyze the operating status data using a tensor splicing method to extract multiple features from the operating status data; splice the multiple features in the operating status data to obtain a three-dimensional joint feature tensor; and perform predictive analysis on the three-dimensional joint feature tensor to obtain an abnormality probability distribution.

[0152] The control module 30 is used to dynamically adjust the current operating parameters of the operating status data of each unit node and the air purification device according to the abnormal probability distribution, and perform purification control on the current area where the multiple units are located.

[0153] To sum up, the regional linkage monitoring, feedback and purification control method and system for multiple units proposed in the example of the present invention can be seen. By combining the pollution diffusion rate and power consumption coupling index, the unit can realize the prediction and early warning of pollution and power consumption anomalies, providing a scientific basis for environmental governance; the linkage effect of multiple unit nodes is reflected in the ability to capture and process cross-unit data changes in real time, optimize information exchange and resource allocation between units, effectively reduce pollution diffusion, and improve the accuracy and response speed of environmental monitoring and equipment management.

[0154] 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. A person skilled in the art may modify the technical solutions described in the above embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A regional linkage monitoring feedback purification control method for multiple units, characterized in that: The following steps are involved: The IoT architecture is used to collect the current operating status data of the operating parameters of each node of the units in the area where multiple units are located in a distributed manner; Analyze the running status data using the tensor splicing method and extract multiple features from the running status data; Multiple features in the running status data are spliced ​​together to obtain a three-dimensional joint feature tensor; Determine the number of channels of the three-dimensional joint feature tensor, initialize the three-dimensional joint feature tensor using the attention mechanism and the number of channels, and obtain an initialized three-dimensional joint feature tensor; A bidirectional recurrent neural network is used to take all channels of each node in the initialized three-dimensional joint feature tensor as input sequence data. The bidirectional recurrent neural network reads the sequence data in both forward and reverse directions, captures the corresponding past and future time series in the forward and reverse sequence data, and obtains the hidden layer dimension. The gated recurrent unit is used to extract the dynamic change features of the hidden layer dimension to obtain the hidden change features; The learning rate attenuation function is calculated for the past and future time series using an adaptive learning rate strategy; the attention weight of the learning rate is calculated using the learning rate attenuation function; The attention fusion feature is obtained by weighting and summing each channel in each node of the hidden change feature through the attention weight; Through the Internet of Things architecture, a graph neural network is introduced to each node of the unit to determine the coordinates of the unit nodes. Based on the coordinates of the unit nodes, the pollution transmission topology structure between the unit nodes is constructed; the attention fusion feature is used to analyze the standard material abundance matrix to obtain the concentration of air pollutants collected by each unit node; The concentration gradient diffusion rate between adjacent nodes of air pollutant concentration is analyzed through the pollution transmission topology structure between unit nodes; The radial heat flux density characteristics are introduced by using the concentration gradient diffusion rate between adjacent nodes to calculate and obtain the power consumption coupling index of each unit node. The concentration gradient diffusion rate between adjacent nodes and the power consumption coupling index of each unit node are predicted to obtain the distribution probability of pollution and power consumption anomalies. Dynamically adjust the current operating parameters of each unit node and the operating status data of the air purification device based on the abnormal probability distribution, and perform purification control on the current area where multiple units are located; Among them, the operating status data includes: air quality parameters, unit operating temperature, fluid component concentration and power consumption data of each node; multiple features in the operating status data include: standard material abundance matrix, radial heat flux density characteristics and power consumption data characteristics of each node.

2. The method for regional linkage monitoring, feedback and purification control of multiple units according to claim 1, characterized in that: The operating status data is analyzed using a tensor splicing method to extract multiple features from the operating status data; multiple features in the operating status data are spliced ​​to obtain a three-dimensional joint feature tensor. The specific operation steps are as follows: Performing peak normalization processing on the air quality parameters and the fluid component concentrations using a tensor splicing method to obtain a standard substance abundance matrix; Calculating the local temperature gradient of the unit operating temperature, calculating the gradient direction of the unit operating temperature based on the local temperature gradient, describing the gradient direction of the unit operating temperature in a radial direction, and calculating the radial heat flux intensity; summarizing the radial heat flux intensities of each node to obtain a radial heat flux density characteristic; Normalizing the power consumption data of each node to obtain normalized power consumption data characteristics of each node; The tensor splicing method is used to expand the channel dimensions of the standard material abundance matrix, radial heat flux density characteristics and power consumption data so that the channel dimensions of the standard material abundance matrix, radial heat flux density characteristics and power consumption data are the same. The same channel dimensions of each expanded feature are spliced ​​in the second dimension to form a three-dimensional tensor as a three-dimensional joint feature tensor.

3. The method for regional linkage monitoring, feedback and purification control of multiple units according to claim 2, characterized in that: The radial heat flux density characteristic is introduced by using the concentration gradient diffusion rate between adjacent nodes to calculate and obtain the power consumption coupling index of each unit node. The concentration gradient diffusion rate between adjacent nodes and the power consumption coupling index of each unit node are predicted to obtain the distribution probability of pollution and power consumption anomaly. The specific operation steps are as follows: The directional correction factor is used to introduce a factor to correct the spatial attenuation factor and the radial heat flux density characteristic; the concentration gradient diffusion rate between adjacent nodes is updated by the corrected spatial attenuation factor to obtain the effective concentration diffusion rate of each unit node; Calculating the power consumption coupling index of each unit node by using the factors of the effective concentration diffusion rate and radial heat flux density characteristics of each unit node and the power consumption data of each node; The effective concentration diffusion rate of each unit node and the power consumption coupling index of each unit node are dynamically evolved by using a hidden Markov algorithm to obtain a state transfer matrix; smooth prediction is performed through the state transfer matrix to obtain the distribution probability of pollution and power consumption anomalies.

4. The method for regional linkage monitoring, feedback and purification control of multiple units according to claim 3, characterized in that: The effective concentration diffusion rate of each unit node and the power consumption coupling index of each unit node are dynamically evolved by using a hidden Markov algorithm to obtain a state transfer matrix. The specific operation steps are as follows: The hidden state set of each unit node is set using a hidden Markov algorithm; the hidden state set includes: normal hidden state probability, pollution abnormal hidden state probability, and power consumption abnormal hidden state probability; Collecting the effective concentration diffusion rate and power consumption coupling index of each unit node at time t; forming an observation vector with the effective concentration diffusion rate and power consumption coupling index at time t; A priori probability vector at an initial moment is set for each hidden state probability in the hidden state set; and the priori probability vector at the initial moment is dynamically evolved between hidden states using the attention weight to obtain a state transfer matrix.

5. The method for regional linkage monitoring, feedback and purification control of multiple units according to claim 4, characterized in that: The state transition matrix is ​​used to perform smooth prediction to obtain the distribution probability of pollution and power consumption anomalies. The specific steps are as follows: According to the observation vectors at each moment before and after the time t and the prior probability vector at the initial moment of each hidden state probability in the state transfer matrix, the forward algorithm and the backward algorithm are used to smooth the calculation respectively to obtain the smoothed posterior probability at each moment; The smoothed posterior probability at each moment is compared with the hidden state set to obtain the distribution probability of pollution and power consumption anomalies.

Images