CN-121601071-B - Multi-dimensional integrated air quality evaluation computing system based on artificial intelligence

CN121601071BCN 121601071 BCN121601071 BCN 121601071BCN-121601071-B

Abstract

The invention discloses a multidimensional integrated air quality evaluation and calculation system based on artificial intelligence, and relates to the technical field of artificial intelligence environment monitoring. The system comprises a multi-dimensional data acquisition terminal for acquiring multi-source data, a multi-scale space-time feature extraction module for generating dynamic pollution field characterization by constructing space-time collaborative tensor and utilizing a diffusion-graph convolution network and a path integration attention mechanism, a multi-dimensional integral calculation module for carrying out space-time volume integration by adopting an interpretable neural integrator to output dynamic air quality comprehensive indexes and space-time accumulation flux, an air quality contribution decomposition module for generating a multi-dimensional contribution map based on contribution tensor decomposition technology, and a visualization and interaction output module for rendering and publishing results. The invention realizes dynamic, collaborative and interpretable accurate assessment of regional air quality and pollution contribution tracing.

Inventors

Zou Muxi
CHEN YUXUAN
WANG LEIJUN
ZHAO YUTAO
ZHAI JINYE
DENG HUAN

Assignees

南京师范大学

Dates

Publication Date: 20260508
Application Date: 20260129

Claims (4)

1. The system is characterized by comprising a multi-dimensional data acquisition terminal, a multi-scale space-time feature extraction module, a multi-dimensional integral calculation module, an air quality contribution decomposition module and a visualization and interaction output module; the multi-dimensional data acquisition terminal acquires multi-source monitoring data in the target area; The multi-scale space-time feature extraction module is connected with the multi-dimensional data acquisition terminal, receives and processes multi-source monitoring data, constructs a space-time collaborative tensor, extracts multi-scale space-time features through a diffusion-graph convolution network and a path integration attention mechanism, and generates dynamic pollution field characterization; the multidimensional integral calculation module is connected with the multiscale space-time feature extraction module, takes dynamic pollution field characterization as input, adopts an interpretable neural integrator to perform space-time volume integral calculation on a target evaluation area within a specified time window, and outputs a dynamic air quality comprehensive index and space-time accumulated flux; The air quality contribution decomposition module is connected with the multidimensional integral calculation module and the multi-scale space-time feature extraction module, and is used for decomposing the dynamic air quality comprehensive index and the space-time cumulative flux into quantized results of different pollution source types and different transmission path contributions based on a contribution degree tensor decomposition technology to generate a multidimensional contribution degree decomposition map; The visual and interactive output module is connected with the multidimensional integral calculation module and the air quality contribution decomposition module, performs fusion visual rendering on the dynamic air quality comprehensive index, the space-time accumulated flux and the multidimensional contribution decomposition map, and externally issues the map through a standard data interface; The multi-scale space-time feature extraction module comprises a space-time collaborative tensor construction unit, a diffusion-graph convolution network unit and a path integration attention mechanism unit, and the process of generating dynamic pollution field characterization by the multi-scale space-time feature extraction module is as follows: The space-time collaborative tensor construction unit fuses and ascends dimensions of the preprocessed pollutant concentration time sequence data, the multidimensional meteorological field data and the static geographic information data according to a unified space-time grid to construct a space-time collaborative tensor; The diffusion-graph convolution network unit takes space-time collaborative tensor as input, utilizes graph convolution operators embedded with an unsteady diffusion equation to carry out message transmission and feature aggregation on a dynamic graph formed by monitoring stations and virtual grid nodes, and explicitly models the space transmission process of pollutants under the action of non-uniform wind fields and turbulent diffusion to output node feature graphs; the path integration attention mechanism unit receives the node characteristic diagram, takes the pollution state at the previous moment as initial distribution, takes the wind speed and wind direction sequence as a driving field, calculates pollution mass transportation contribution along a plurality of virtual tracks through the learnable attention weight, and integrates the pollution mass transportation contribution to the grid state at the current moment, so as to generate the dynamic pollution field characterization.
2. The system of claim 1, wherein the core calculation formula of the path integration attention mechanism unit is: ; Wherein, the Is at Dynamic pollution field characterization at moment; Is a fusion function; Is a node At the moment of time Node feature vectors updated via a diffusion-graph convolution network element; is a time index variable, traverses the historical moment, and is characterized by that To the point of ; The length of the time window for the path backtracking; Is at the time of history All possible wind farm transport influences the current target node Is a source node set of (1); is a source node belonging to a collection ; Is a learnable attention weight, which is calculated by a multi-head attention network; Is the source node Historical time of day Is defined by the node feature vector of (a); is a double summation for all historic times and all possible source nodes in each historic time Weighted summation is performed.
3. The system of claim 1, wherein the multi-dimensional integration calculation module comprises an interpretable neural integrator unit and a space-time cumulative flux calculation unit, and the process of outputting the dynamic air quality comprehensive index and the space-time cumulative flux by the multi-dimensional integration calculation module is as follows: the interpretable neural integrator unit discretizes the dynamic pollution field representation on a three-dimensional space domain and a specified time window of a target area, and adopts an integral kernel function parameterized by a neural network and provided with physical constraints to carry out weighted summation on each space-time unit so as to realize space-time volume integration under physical guidance and output a dynamic air quality comprehensive index; The space-time cumulative flux calculating unit synchronously records the pollutant flux components passing through all boundary surfaces of the evaluation area and accumulates along the time dimension in the integration process of the neural integrator unit, and calculates the pollutant net mass input into the area from different directions in the evaluation time as the space-time cumulative flux.
4. The system of claim 1, wherein the air quality contribution decomposition module comprises a contribution tensor construction unit, a tensor decomposition unit, a contribution quantization and map generation unit, and the process of generating the multi-dimensional contribution decomposition map by the air quality contribution decomposition module is as follows: The contribution degree tensor construction unit correlates the dynamic pollution field characterization, pollution source list data and the conveying path label simulated by the reverse track model to construct a high-order contribution degree tensor, wherein the dimensionality of the high-order contribution degree tensor corresponds to the space grid, the time step, the pollution source type and the conveying path type respectively; The tensor decomposition unit adopts a non-negative tensor decomposition algorithm to perform dimensionality reduction and decomposition on the high-order contribution tensor to obtain a group of low-rank space-time mode factors, source contribution factors and path contribution factors; The contribution degree quantization and map generation unit carries out reverse mapping and proportion distribution on various factors obtained by decomposition, dynamic air quality comprehensive indexes and space-time accumulated flux, quantitatively calculates the contribution percentages of various pollution sources and various conveying paths to a final integral result, and fuses the contribution percentages in a thermodynamic diagram, a flow diagram and a stacked diagram form to generate a multidimensional contribution degree decomposition map.

Description

Multi-dimensional integrated air quality evaluation computing system based on artificial intelligence Technical Field The invention relates to the technical field of artificial intelligence environment monitoring, in particular to an artificial intelligence-based multidimensional integrated air quality evaluation computing system. Background The existing air quality assessment system mainly relies on single index concentration measurement of discrete monitoring stations, and carries out linear weighting calculation through a fixed formula, the method regards the pollutants as independent static variables, and ignores the space-time synergistic effect, nonlinear superposition and dynamic migration processes of various pollutants under complex weather and geographic conditions. Although the evaluation model based on deep learning can capture partial nonlinear relations, most of the evaluation models are black box models, and the evaluation model lacks explicit physical mechanism embedding and contribution degree interpretability, so that dynamic integration and source contribution traceability of a space-time accumulation process of regional pollution are difficult. The traditional system can not carry out continuous integral evaluation on the pollution field in the three-dimensional space and the time dimension, and also can not be used for quantifying the specific contributions of different pollution sources and different transmission paths to the air quality of a target area. Therefore, it is necessary to design a multidimensional integrated air quality assessment computing system integrating an atmospheric physical mechanism and an artificial intelligence technology, so as to realize dynamic, collaborative and interpretable accurate assessment and contribution decomposition of regional air quality. Disclosure of Invention The invention provides a multi-dimensional integral air quality evaluation computing system based on artificial intelligence, which aims at the problem that the traditional air quality evaluation method ignores the space-time synergy and dynamic process of pollutants, and constructs a multi-dimensional integral computing framework integrating an atmospheric physical mechanism and interpretable artificial intelligence. The invention provides a multidimensional integrated air quality evaluation computing system based on artificial intelligence, which is suitable for air quality dynamic evaluation of urban, industrial park and regional scale, and comprises the following steps: The multi-dimensional data acquisition terminal acquires multi-source monitoring data in the target area; The multi-scale space-time feature extraction module is connected with the multi-dimensional data acquisition terminal, receives and processes multi-source monitoring data, constructs a space-time collaborative tensor, extracts multi-scale space-time features through a diffusion-graph convolution network and a path integration attention mechanism, and generates dynamic pollution field characterization; the multidimensional integral calculation module is connected with the multiscale space-time feature extraction module, takes dynamic pollution field characterization as input, adopts an interpretable neural integrator to perform space-time volume integral calculation on a target evaluation area within a specified time window, and outputs a dynamic air quality comprehensive index and space-time accumulated flux; The air quality contribution decomposition module is connected with the multidimensional integral calculation module and the multi-scale space-time feature extraction module, and is used for decomposing the dynamic air quality comprehensive index and the space-time cumulative flux into quantized results of different pollution source types, different transmission paths and different precursor contributions based on a contribution degree tensor decomposition technology to generate a multidimensional contribution degree decomposition map; the visualization and interaction output module is connected with the multidimensional integral calculation module and the air quality contribution decomposition module, performs fusion visualization rendering on the dynamic air quality comprehensive index, the space-time accumulated flux and the multidimensional contribution decomposition map, and externally issues the map through the standard data interface. Further, the multi-dimensional data acquisition terminal acquires multi-source monitoring data in a target area, and specifically comprises the following steps: S1, collecting pollutant concentration time sequence data by an air quality monitoring sensor deployed at a fixed site and a mobile platform of a target area; S2, accessing an area weather monitoring network to acquire multi-dimensional weather field data; Step S3, accessing a geographic information system to obtain static geographic information data, and forming multi-source monitoring data by the pollutant conc