Air quality data screening method and system based on multi-factor bias analysis

Abstract

The application discloses a kind of based on multi-factor deviation analysis air quality data screening method and system, solve the problem of complex spatiotemporal data deviation that traditional method is difficult to handle.The method includes obtaining the air pollutant monitoring value of multiple cities, meteorological data, topographic data and monitoring station metadata in multiple time dimensions, data cleaning is carried out through three-level quality control, and spatiotemporal data cube is constructed;Dynamic establishment of regional reference value, calculate the combined deviation and identify the deviation mode by spatiotemporal DBSCAN method;Build city deviation feature vector and provincial deviation network, and use graph convolution network for sorting;Establish multi-layer causal diagram, and diagnose the root cause of deviation using PC algorithm and Shapley value decomposition technique;Define Markov decision process, dynamically optimize correction strategy.The application can accurately quantify composite deviation, distinguish regional pollution, equipment failure or weather disturbance, etc.Root cause, provide data support for regional collaborative governance.

Description

Technical Field

[0001] This invention relates to the field of air quality monitoring technology, specifically to an air quality data screening method and system based on multi-factor deviation analysis. Background Technology

[0002] With the acceleration of urbanization and the continuous expansion of industrial scale in my country, the accuracy and reliability of air quality monitoring data have become a core foundation for environmental governance and public health protection. However, air quality data is often affected by sensor errors, changes in the external environment, and other factors, resulting in data bias. First, monitoring data is susceptible to interference from sensor drift, extreme weather, and equipment failure. Traditional methods struggle to dynamically adapt to multi-source heterogeneous data, leading to insufficient outlier removal or incorrect corrections. Second, existing technologies mostly use single-pollutant thresholds for determination, ignoring pollutant interaction effects and spatiotemporal correlations, and failing to distinguish between regional pollution diffusion and equipment failure. Finally, bias detection relies heavily on traditional statistical methods, which are ill-suited for handling complex, multi-dimensional, and time-series data. Therefore, intelligent data screening technology based on deep learning has become an important direction for future data processing development. Summary of the Invention

[0003] The purpose of this invention is to provide an air quality data screening method and system based on multi-factor deviation analysis. Through multi-level data processing and multi-dimensional feature learning of air quality factors, it can effectively identify cities or provinces with large deviations and provide dynamic optimization and correction strategies.

[0004] To achieve the above objectives, the present invention provides the following technical solution:

[0005] In a first aspect, the present invention provides an air quality data screening method based on multi-factor deviation analysis, comprising:

[0006] We acquire air pollutant monitoring values, meteorological data, topographic data, and monitoring station metadata from multiple cities and time dimensions. We perform data cleaning through a three-level quality control system, interpolate and align the spatiotemporal data, and construct a spatiotemporal data cube.

[0007] A regional benchmark value is dynamically established by integrating spatially weighted average observations and historical averages over a time sliding window. The combined deviation is calculated based on the absolute value of individual deviations and pollutant interactions. The deviations are classified and deviation patterns are identified using the spatiotemporal DBSCAN method.

[0008] We construct urban deviation feature vectors to calculate the comprehensive urban deviation index, establish a provincial deviation network with cities as nodes and atmospheric influence coefficients as edges, and sort the provincial deviations through a graph convolutional network.

[0009] A multi-layered cause-effect graph containing equipment and environmental layers was constructed. The PC algorithm was used to identify pollution propagation paths, and the Shapley value decomposition technique was used to quantify the contribution of each factor to the total deviation.

[0010] Define a Markov decision process, select a correction strategy based on the current bias characteristics and environmental state, and evaluate the data quality improvement and optimize the strategy through a reward function.

[0011] Furthermore, the three-level quality control includes: first-level removal of abnormal values ​​due to equipment failure, second-level correction of abnormal values ​​due to meteorological conditions, and third-level standardization processing to convert the concentrations of each pollutant into comparable values ​​under standard conditions.

[0012] The process of interpolating and aligning spatiotemporal data and constructing a spatiotemporal data cube includes: uniformly interpolating data with different time resolutions into hourly data; using Kriging interpolation to generate regular grid data for spatially unevenly distributed sites; and establishing a spatiotemporal data cube with dimensions including [time × space × pollutants].

[0013] Furthermore, the formula for calculating the absolute value of the individual deviation is:

[0014]

[0015] in, Let be the absolute value of the deviation of the i-th factor. For measured values, This is the standard value;

[0016] The formula for calculating the combined deviation is:

[0017]

[0018] in, The weight of the individual deviation, Let λ be the interaction coefficient between pollutants k and l, and λ be the weight of the interaction term. The interaction coefficient is determined by principal component analysis, which can identify the correlation between pollutants.

[0019] Furthermore, the deviation is classified and the deviation pattern is identified by the spatiotemporal DBSCAN method, specifically including spatial clustering deviation, time synchronization deviation, and random discrete deviation; spatial clustering deviation indicates possible regional pollution sources, time synchronization deviation indicates system errors, and random discrete deviation indicates local disturbances or random fluctuations.

[0020] Furthermore, the formula for calculating the urban comprehensive deviation index is:

[0021]

[0022] in, These are the feature weights of city i, dynamically calculated using the entropy weight method. It is the urban deviation feature vector of city i;

[0023]

[0024] in, Let i be the average deviation of city i at time t; The maximum deviation of city i at time t; The frequency of deviations greater than the threshold θ; This represents spatial autocorrelation.

[0025] Furthermore, the equipment layer focuses on the equipment correlation between monitoring stations. Each monitoring point is regarded as a node in the cause-effect graph. The nodes in the equipment layer represent each monitoring station, and the edges represent the data correlation between the stations. The environmental layer focuses on the impact of meteorological factors on the spread of pollutants.

[0026] Furthermore, the Markov decision process includes:

[0027] State space S t : Describes the current system status, including deviation type, meteorological conditions, and equipment status;

[0028] Action space a t : Represents the correction strategy selected at time t, including physical correction, statistical interpolation, and learning correction;

[0029] reward function r t : Measures the degree of improvement in the system state after performing a certain action. ,in, It is the improvement in data quality after the correction is performed. It is the cost of choosing a certain correction strategy;

[0030] Transition probability: The transition between states is determined based on the current state and the chosen action, and is estimated using historical data.

[0031] Secondly, the present invention provides an air quality data screening system based on multi-factor deviation analysis, comprising:

[0032] Data preprocessing module: used to perform multi-source data acquisition, three-level quality control and spatiotemporal data alignment, and to construct a spatiotemporal data cube;

[0033] Joint Deviation Detection Module: Used to dynamically generate regional benchmark values, calculate combined deviations, and identify deviation patterns;

[0034] Hierarchical sorting module: used to construct and sort the city-province deviation network;

[0035] Root cause diagnosis module: used to construct multi-layered causal graphs and quantify bias contributions;

[0036] Dynamic calibration module: Used to generate and optimize calibration strategies.

[0037] Furthermore, the joint deviation detection module includes:

[0038] The baseline generation unit integrates spatially weighted average observations with historical mean values ​​from a time sliding window, and then dynamically weights the average using a Bayesian model.

[0039] Interaction coefficient calculation unit: used to determine the interaction coefficients between pollutants.

[0040] Furthermore, the root cause diagnosis module includes:

[0041] Cause-effect graph construction unit: used to establish a two-layer cause-effect graph between the device layer and the environment layer;

[0042] Propagation Path Analysis Unit: Employs PC algorithms to identify key pollution propagation paths;

[0043] Contribution decomposition unit: The contribution of each factor to the total deviation is decomposed using Shapley values.

[0044] Based on the above technical solution, the embodiments of the present invention can produce at least the following technical effects:

[0045] This invention provides an air quality data screening method based on multi-factor deviation analysis. It combines dynamic benchmark values ​​with a multi-factor weighted deviation model to accurately quantify composite deviations. Through spatiotemporal DBSCAN clustering and causal graph analysis, it distinguishes root causes such as regional pollution, equipment malfunction, or meteorological interference. Utilizing the entropy weight method to dynamically weight city-level deviation characteristics and constructing a provincial GCN network based on an atmospheric diffusion model, it quantifies cross-city pollution propagation, providing data support for regional collaborative governance. This achieves closed-loop optimization of "detection-attribution-correction," providing core technical support for precise environmental supervision. Attached Figure Description

[0046] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0047] Figure 1 This is a flowchart of an embodiment of the present invention. Detailed Implementation

[0048] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In addition, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0049] The objective of this invention is achieved through the following technical solution:

[0050] Example 1: This application provides an air quality data screening method based on multi-factor deviation analysis, comprising the following steps:

[0051] Step 1: Data Preprocessing

[0052] We acquire air pollutant monitoring values, meteorological data, topographic data, and monitoring station metadata from multiple cities and time dimensions. We perform data cleaning through three-level quality control, interpolate and align the spatiotemporal data, and construct a spatiotemporal data cube.

[0053] Specifically, it is necessary to acquire monitoring values ​​of various air pollutants from multiple cities at different time dimensions, simultaneously collect meteorological and topographic data, and obtain monitoring station metadata. Different time dimensions include daily, weekly, and monthly; multiple air pollutant monitoring values ​​include PM2.5, PM10, SO2, NO2, CO, and O3; meteorological data includes wind speed, wind direction, temperature, humidity, and air pressure; and monitoring station metadata includes equipment model, maintenance records, and surrounding environment. The acquired data undergoes a three-level quality control process: Level 1: removing outliers caused by equipment malfunctions (such as persistently constant values); Level 2: correcting for anomalies caused by meteorological conditions (such as PM10 during sandstorms); and Level 3: standardization processing, converting the concentrations of each pollutant into comparable values ​​under standard conditions. Finally, spatiotemporal alignment processing is performed, uniformly interpolating data from different time resolutions to hourly data. For spatially unevenly distributed stations, Kriging interpolation is used to generate regular grid data, establishing a spatiotemporal data cube with dimensions including [time × space × pollutant].

[0054] Step 2: Multi-factor joint bias detection

[0055] A regional benchmark value is dynamically established by integrating spatially weighted average observations and historical averages over a time sliding window. The combined deviation is calculated based on the absolute value of individual deviations and pollutant interactions. The deviations are classified and deviation patterns are identified using the spatiotemporal DBSCAN method.

[0056] Specifically, the regional background value for each time slice (t) is first dynamically calculated, and then two benchmark values ​​are fused using a Bayesian model averaging (BMA): the spatially weighted average observed value (after removing ±3σ outliers) and the historical mean of the time-sliding window (data from the same period over the past 30 days). For each monitoring point, the absolute value of the individual deviation for each air quality factor (such as SO2, CO, NO2, O3, PM2.5, PM10) is calculated.

[0057]

[0058] in, Let be the absolute value of the deviation of the i-th factor. For measured values, This is the standard value;

[0059] The combined bias is a weighted sum of multiple individual biases, taking into account the interactions between pollutants:

[0060]

[0061] in, The weight of the individual deviation, Let be the interaction coefficient between pollutants k and l, and λ be the weight of the interaction term. Let be the deviation value of the i-th factor and the k-th index at time t. Let be the deviation value of the i-th factor and the l-th index at time t, and the interaction coefficient. Principal component analysis (PCA) can be used to identify the correlations between pollutants.

[0062] Finally, the deviations were classified and deviation patterns were identified using the spatiotemporal DBSCAN (density clustering) method.

[0063] Spatial clustering bias: Indicates potential regional pollution sources. It can be measured using spatial autocorrelation (such as the Moran index); Temporal synchronization bias: Indicates systematic errors, which may be related to equipment failure or network transmission problems; Random dispersion bias: Indicates local disturbances or random fluctuations, which may be caused by equipment problems at individual monitoring points or short-term weather events.

[0064] Step 3: Spatiotemporal hierarchical sorting analysis

[0065] We construct urban deviation feature vectors to calculate the comprehensive urban deviation index, establish a provincial deviation network with cities as nodes and atmospheric influence coefficients as edges, and sort the provincial deviations through a graph convolutional network.

[0066] First, calculate the city deviation feature vector; the deviation feature vector for each city... Includes multiple indicators:

[0067]

[0068] in, Let i be the average deviation of city i at time t; The maximum deviation of city i at time t; The frequency of deviations greater than the threshold θ; Spatial autocorrelation is calculated using methods such as the Moran index.

[0069] Then, the city's comprehensive deviation index is calculated based on the weights of the eigenvectors:

[0070]

[0071] in, , , , The feature weights are dynamically calculated based on the entropy weight method.

[0072] Finally, provincial ranking is performed to establish a provincial network, where the comprehensive deviation index of each city is a node, and the edges between cities are atmospheric influence coefficients calculated based on atmospheric diffusion models (such as the CALPUFF model). A graph convolutional network (GCN) is used to perform convolution operations on the provincial deviation network.

[0073]

[0074] Where A is the adjacency matrix, representing the atmospheric influence coefficient between cities, and GCN is the graph convolution operation, which propagates and calculates the comprehensive deviation of each city through the adjacency relationship between nodes.

[0075] Step 4: Root Cause Diagnosis of Deviation

[0076] A multi-layered causal graph containing equipment and environmental layers is constructed. The PC algorithm is used to identify pollution propagation paths, and the Shapley value decomposition technique is used to quantify the contribution of each factor to the total deviation.

[0077] It's important to note that the core objective of propagation path analysis is to identify the propagation pathways of pollutants and determine which factors or variables significantly influence pollutant distribution and bias. The equipment layer focuses on the interrelationships between monitoring stations. Each monitoring point can be considered a node in a causal graph. Monitoring equipment may affect data accuracy due to sensor malfunctions, calibration problems, data transmission errors, etc. Therefore, nodes in the equipment layer represent individual monitoring stations, and edges represent data relationships between these stations. Equipment layer analysis helps identify biases caused by equipment problems; for example, higher pollutant readings in a certain area may be due to inaccurate calibration of monitoring equipment. The environmental layer primarily focuses on the impact of meteorological factors (such as wind speed, temperature, humidity, and air pressure) on pollutant propagation. Airborne pollutants diffuse or settle with meteorological factors such as wind direction, wind speed, and temperature, affecting pollutant concentration and distribution. For example, strong winds may cause pollutants to spread over a wide area, while high temperatures may exacerbate photochemical reactions leading to ozone formation. Through this causal relationship, the contribution of environmental factors to pollution bias can be inferred. A multi-layered causal graph is constructed to analyze pollutant propagation pathways.

[0078]

[0079] The equipment layer considers the correlation between equipment at monitoring stations, while the environmental layer considers the impact of meteorological factors on the diffusion of pollutants.

[0080] Specifically, causal discovery algorithms (such as the PC algorithm) are used to identify key pollution propagation paths. The PC algorithm is a causal discovery algorithm based on conditional independence tests. Initially, it assumes a causal relationship between all variables and progressively removes irrelevant edges through conditional independence tests. It then progressively checks the conditional independence between each pair of variables, calculating the independence of each pair while controlling for other variables. Statistical tests (such as those based on Pearson correlation or Granger causality tests) are used to determine the existence of a causal relationship. The PC algorithm can construct a directed acyclic graph (DAG), which can help identify pollutant propagation paths. For example, this causal graph might reveal that anomalous changes in certain pollutants are caused by equipment malfunction or diffusion effects under specific meteorological conditions.

[0081] It's important to note that the goal of contribution decomposition is to allocate the contribution of the total deviation to each factor or variable. Shapley values ​​are a method derived from cooperative game theory used to measure the contribution of each participant in a cooperative game. In our scenario, each factor (e.g., weather factors, equipment errors, etc.) can be considered a participant, and Shapley values ​​can be used to decompose the contribution of these factors to the total deviation. Using Shapley value decomposition, we can interpret the total deviation and decompose the contribution of each factor to the total deviation:

[0082]

[0083] in, Let be the Shapley value of the k-th factor, representing its contribution to the total bias. v(S) represents the bias explained by the subset S, and N is the set of all influencing factors. By calculating the Shapley value, we can quantitatively assess the impact of each factor on the total bias. Assuming there are multiple factors (such as different pollutant concentrations, meteorological factors, monitoring equipment errors, etc.), the Shapley value helps us identify which factors play a dominant role in pollutant bias.

[0084] Step 5: Dynamic Adaptive Correction

[0085] Define a Markov decision process, select a correction strategy based on the current bias characteristics and environmental state, and evaluate the data quality improvement and optimize the strategy through a reward function.

[0086] Specifically, a Markov decision process includes: state space: states This is a description of the current system, including deviation characteristics and environmental conditions. For example, the conditions may include the current pollutant concentration deviation, meteorological conditions (such as wind speed, temperature, etc.), and the status of monitoring equipment (such as equipment fault indicators).

[0087] Action space: Action This represents the correction strategy chosen at time t. For example, the action could be to choose a physical method (such as chemical mass balance correction), a statistical method (such as EM algorithm interpolation), a learning method (such as LSTM correction), etc.

[0088] Reward function: Reward This measures the degree of improvement in the system's state after performing a certain action. Rewards can be measured by improvements in data quality, such as:

[0089]

[0090] in, It is the improvement in data quality after the correction is performed. It is the cost of choosing a certain correction strategy (such as computational complexity, time consumption, etc.).

[0091] Transition probability: The transition between states is based on the current state. and the chosen action The decision is made, and the transition probability can be estimated using historical data.

[0092] Example 2: This application provides an air quality data screening system based on multi-factor deviation analysis, comprising:

[0093] Data preprocessing module: used to perform multi-source data acquisition, three-level quality control and spatiotemporal data alignment, and to construct a spatiotemporal data cube;

[0094] Joint Deviation Detection Module: Used to dynamically generate regional benchmark values, calculate combined deviations, and identify deviation patterns;

[0095] Hierarchical sorting module: used to construct and sort the city-province deviation network;

[0096] Root cause diagnosis module: used to construct multi-layered causal graphs and quantify bias contributions;

[0097] Dynamic calibration module: Used to generate and optimize calibration strategies.

[0098] In this embodiment, the joint deviation detection module includes:

[0099] The baseline generation unit integrates spatially weighted average observations with historical mean values ​​from a time sliding window, and then dynamically weights the average using a Bayesian model.

[0100] Interaction coefficient calculation unit: used to determine the interaction coefficients between pollutants.

[0101] In this embodiment, the root cause diagnosis module includes:

[0102] Cause-effect graph construction unit: used to establish a two-layer cause-effect graph between the device layer and the environment layer;

[0103] Propagation Path Analysis Unit: Employs PC algorithms to identify key pollution propagation paths;

[0104] Contribution decomposition unit: The contribution of each factor to the total deviation is decomposed using Shapley values.

[0105] During the implementation process, the deviation analysis of air quality monitoring data in urban agglomerations in North China (January 2024) was conducted as follows:

[0106] Step 1: Data Preprocessing

[0107] (1) Data collection:

[0108] a. Cities: Shijiazhuang, Taiyuan, Beijing, Tianjin, Jinan, Zhengzhou (6 cities in total);

[0109] b. Time range: January 1-31, 2024 (hourly data);

[0110] c. Pollutants: PM2.5, PM10, SO2, NO2, CO, O3 (6 items in total);

[0111] d. Meteorological data: wind speed, wind direction, temperature, humidity, air pressure (Source: China Meteorological Administration);

[0112] e. Topographic data: city elevation, distribution of surrounding mountains (Source: Geographic Information System);

[0113] f. Equipment metadata: Equipment models (Thermo Fisher, etc.) and maintenance records (calibration time, fault logs) for each city monitoring station.

[0114] (2) Data cleaning:

[0115] Level 1: Exclude the PM2.5 level of 120 μg / m³ recorded at a certain monitoring station in Taiyuan on January 5th due to a sensor malfunction. 3 );

[0116] Level 2: Corrected PM10 data for the sandstorm in Shijiazhuang on January 15 (original value 800 μg / m³). 3 → Corrected to 620 μg / m based on meteorological models 3 );

[0117] Level 3: Convert all pollutant concentrations to standard conditions (25°C, 1 atm).

[0118] (3) Spatiotemporal alignment:

[0119] Interpolate daily / weekly data to hourly data (using linear interpolation);

[0120] Kriging interpolation was used to generate 1km×1km grid data for Jinan City, where the stations are sparse.

[0121] Construct the spatiotemporal cube dimensions: [Time: 744 hours × Space: 6 cities × 50 grid points × Pollutants: 6 items].

[0122] Step 2: Multi-factor joint bias detection

[0123] (1) Establishment of benchmark values:

[0124] Spatial weighted average: Remove outliers ±3σ (e.g., Beijing PM2.5 > 250 μg / m³). 3 (outliers)

[0125] Time sliding window: Takes the average of the same hours over the past 30 days (data from December 2023);

[0126] BMA fusion weights: spatial baseline weight 0.6, time baseline weight 0.4.

[0127] (2) Deviation quantification:

[0128] Calculate individual deviations:

[0129] The PM2.5 level measured at a monitoring station in Shijiazhuang at 12:00 PM on January 10th was 98 μg / m³. 3 The baseline value is 65 μg / m³. 3 →|δ i |=33μg / m 3 ;

[0130] Combined deviation calculation:

[0131] Weighting: PM2.5 weight 0.3 (primary pollutant), interaction term λ=0.2;

[0132] PCA analysis showed a strong correlation between PM2.5 and NO2 (correlation coefficient 0.7), with an interaction coefficient β_{PM2.5,NO2}=0.5.

[0133] (4) Deviation pattern recognition:

[0134] Spatiotemporal DBSCAN clustering results:

[0135] Spatial clustering: SO2 levels were synchronously high at all stations in Tianjin Binhai New Area (Moran index I=0.85, p<0.01), pointing to emissions from industrial areas;

[0136] Time synchronization type: Zhengzhou City experienced a sudden drop in NO2 across all sites from 9:00 to 11:00 on January 20 (equipment network transmission failure).

[0137] Random discrete type: short-term peak value of O3 at individual stations in Beijing (interference from local barbecue stalls).

[0138] Step 3: Spatiotemporal hierarchical sorting analysis

[0139] (1) City-level ranking:

[0140]

[0141] (2) Provincial ranking (Hebei Province as an example): Shijiazhuang has become a key pollution retention node due to its topography (blockage by the Taihang Mountains) (provincial deviation contribution of 32%).

[0142] Step 4: Root Cause Diagnosis of Deviation

[0143] (1) Propagation path analysis:

[0144] Device layer cause-effect graph: Taiyuan site A failure → affects adjacent site B through data communication protocol (edge ​​weight = 0.9);

[0145] Environmental layer cause-effect diagram: Northwest wind (wind speed > 4 m / s) → pollution from Shijiazhuang is transported to Jinan (path coefficient = 0.75);

[0146] PC algorithm identifies critical path: [wind direction → PM2.5 deviation] (p<0.05).

[0147] (2) Decomposition of contribution:

[0148]

[0149] Step 5: Dynamic Adaptive Correction

[0150] (1) MDP framework:

[0151] Status: S t ={ Deviation type = Spatial concentration, Wind speed = 3.2 m / s, Equipment status = Normal}

[0152] Action: Select physical correction method (reconstruct data using chemical mass balance model)

[0153] Reward: r t =Data quality improvement ΔQ - Cost C, ΔQ = Deviation reduction 22μg / m 3 (Quality improvement score +15), C = Model calculation time cost (-2)

[0154] Net reward: r t =13

[0155] (2) Strategy optimization:

[0156] Updating action value through Q-learning:

[0157] (Learning rate α=0.1, discount factor γ=0.9).

[0158] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for screening air quality data based on multi-factor bias analysis, characterized in that, include: We acquire air pollutant monitoring values, meteorological data, topographic data, and monitoring station metadata from multiple cities and time dimensions. We perform data cleaning through a three-level quality control system, interpolate and align the spatiotemporal data, and construct a spatiotemporal data cube. A regional benchmark value is dynamically established by integrating spatially weighted average observations and historical averages over a time sliding window. The combined deviation is calculated based on the absolute value of individual deviations and pollutant interactions. The deviations are classified and deviation patterns are identified using the spatiotemporal DBSCAN method. We construct urban deviation feature vectors to calculate the comprehensive urban deviation index, establish a provincial deviation network with cities as nodes and atmospheric influence coefficients as edges, and sort the provincial deviations through a graph convolutional network. The formula for calculating the urban comprehensive deviation index is: ; in, These are the feature weights of city z, dynamically calculated using the entropy weight method. It is the urban deviation feature vector of city z; ; in, The average deviation of city z over time t; The maximum deviation of city z at time t; The frequency of deviations greater than the threshold θ; This represents spatial autocorrelation. A multi-layered cause-effect graph containing equipment and environmental layers was constructed. The PC algorithm was used to identify pollution propagation paths, and the Shapley value decomposition technique was used to quantify the contribution of each factor to the total deviation. Define a Markov decision process, select a correction strategy based on the current bias characteristics and environmental state, and evaluate the data quality improvement and optimize the strategy through a reward function; The Markov decision process includes: State space S t : Describes the current system status, including deviation type, meteorological conditions, and equipment status; Action space a t : Represents the correction strategy selected at time t, including physical correction, statistical interpolation, and learning correction; reward function r t : Measures the degree of improvement in the system state after performing a certain action. ,in, It is the improvement in data quality after the correction is performed. It is the cost of choosing a certain correction strategy; Transition probability: The transition between states is determined based on the current state and the chosen action, and is estimated using historical data.

2. The air quality data screening method based on multi-factor deviation analysis according to claim 1, characterized in that, The three-level quality control includes: Level 1: removing abnormal values ​​due to equipment failure; Level 2: correcting abnormal values ​​due to meteorological conditions; and Level 3: standardization processing, which converts the concentrations of each pollutant into comparable values ​​under standard conditions. The process of interpolating and aligning spatiotemporal data and constructing a spatiotemporal data cube includes: uniformly interpolating data with different time resolutions into hourly data; using Kriging interpolation to generate regular grid data for spatially unevenly distributed sites; and establishing a spatiotemporal data cube with dimensions including [time × space × pollutants].

3. The air quality data screening method based on multi-factor deviation analysis according to claim 1, characterized in that, The formula for calculating the absolute value of the individual deviation is: ; in, Let be the absolute value of the deviation of the i-th factor. For measured values, This is the standard value; The formula for calculating the combined deviation is: ; in, Let j be the combinational deviation of the j-th factor at time t. As the weight of the individual deviation, Let be the interaction coefficient between pollutants k and l, and λ be the weight of the interaction term. Let be the deviation value of the j-th factor at time t and the k-th pollutant. Let be the deviation value of the j-th factor at time t and the l-th pollutant.

4. The air quality data screening method based on multi-factor deviation analysis according to claim 1, characterized in that, The deviation is classified and the deviation pattern is identified by the spatiotemporal DBSCAN method, specifically including spatial clustering deviation, time synchronization deviation, and random discrete deviation. Spatial clustering deviation indicates possible regional pollution sources, time synchronization deviation indicates system errors, and random discrete deviation indicates local disturbances or random fluctuations.

5. The air quality data screening method based on multi-factor deviation analysis according to claim 1, characterized in that, The equipment layer focuses on the equipment correlation between monitoring stations. Each monitoring point is regarded as a node in the cause-effect graph. The nodes in the equipment layer represent each monitoring station, and the edges represent the data correlation between the stations. The environmental layer focuses on the impact of meteorological factors on the spread of pollutants.

6. An air quality data screening system based on multi-factor deviation analysis, used to implement the method described in any one of claims 1-5, characterized in that, include: Data preprocessing module: used to perform multi-source data acquisition, three-level quality control and spatiotemporal data alignment, and to construct a spatiotemporal data cube; Joint Deviation Detection Module: Used to dynamically generate regional benchmark values, calculate combined deviations, and identify deviation patterns; Hierarchical sorting module: used to construct and sort the city-province deviation network; Root cause diagnosis module: used to construct multi-layered causal graphs and quantify bias contributions; Dynamic calibration module: Used to generate and optimize calibration strategies.

7. The air quality data screening system based on multi-factor deviation analysis according to claim 6, characterized in that, The joint deviation detection module includes: The baseline generation unit integrates spatially weighted average observations with historical mean values ​​from a time sliding window, and then dynamically weights the average using a Bayesian model. Interaction coefficient calculation unit: used to determine the interaction coefficients between pollutants.

8. The air quality data screening system based on multi-factor deviation analysis according to claim 6, characterized in that, The root cause diagnosis module includes: Cause-effect graph construction unit: used to establish a two-layer cause-effect graph between the device layer and the environment layer; Propagation Path Analysis Unit: Employs PC algorithms to identify key pollution propagation paths; Contribution decomposition unit: The contribution of each factor to the total deviation is decomposed using Shapley values.

Images