A method and system for multi-parameter environmental monitoring and intelligent identification of pollution sources

Abstract

This invention discloses a method and system for multi-parameter environmental monitoring and intelligent identification of pollution sources, comprising: collecting and preprocessing multi-source environmental monitoring data to generate an environmental monitoring data sequence; inputting the monitoring data sequence, identifying phase states in segments, and outputting a net disturbance sequence; constructing a monitoring grid, a pollution source grid, and a meteorological grid to establish a three-layer constraint grid structure; projecting the net disturbance sequence onto the monitoring grid, performing multi-level mapping and trimming to generate candidate pollution sources; constructing a pollution prototype matrix, establishing an inversion matrix, and calculating the pollution contribution; sorting and screening the main pollution sources according to the pollution contribution, and outputting the identification result score. This invention achieves pollution disturbance extraction, propagation path screening, and accurate identification of main pollution sources from multi-parameter environmental data by constructing a disturbance interpretation structure, a constraint grid structure, and an inversion solution structure.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring and data processing technology, and in particular to a multi-parameter environmental monitoring and intelligent identification method and system for pollution sources. Background Technology

[0002] Existing environmental monitoring technologies primarily rely on fixed monitoring stations, online monitoring instruments, and conventional sensor networks to collect multi-source environmental data, including air, water, soil, and meteorological data. Abnormal changes are identified through trend analysis, threshold comparison, or empirical models. However, in real-world monitoring scenarios, various environmental parameters are affected by seasonal and diurnal variations, topographic disturbances, and natural fluctuations, resulting in highly time-varying and complex environmental monitoring data. Traditional methods typically treat monitoring data directly as continuous curves, making it difficult to distinguish sudden disturbances caused by pollution events from mixed background fluctuations, leading to insufficient pollution identification accuracy. Existing technologies generally lack the ability to uniformly model the relationships between multiple parameters and media; spatial information and propagation logic between different monitoring points are often ignored, making it difficult to form an interpretable pollution propagation analysis framework.

[0003] In pollution source tracing, existing technologies rely heavily on empirical judgments such as wind direction and water flow direction, or path inference methods based on statistical regression and physical simulation. However, these methods typically require a large amount of validation data, complex models, or precise environmental parameter inputs, limiting their real-time performance and deployability. When there are many pollution sources or multiple sources overlap, traditional source tracing methods struggle to decouple the effects of different sources and cannot quantitatively distinguish the contribution of each source to the monitored anomalies. Existing technologies often rely on single propagation assumptions, making it difficult to accurately characterize pollution events with cross-media propagation characteristics. For example, the interconnected pollution transmission paths between air and water bodies cannot be accurately described using simple rules.

[0004] In summary, existing environmental monitoring and pollution source tracing technologies generally suffer from three key shortcomings: First, they lack the ability to extract structured pollution disturbances from complex monitoring data, making it difficult to accurately distinguish between background changes and pollution events; second, they lack interpretable pollution propagation path constraint mechanisms, making it difficult to effectively tailor and verify propagation links based on spatial, temporal, and media characteristics; and third, they lack quantitative inversion methods for cases with multiple pollution sources superimposed, making it impossible to accurately assess the contribution of each pollution source, resulting in unreliable pollution source identification results.

[0005] Therefore, how to provide a multi-parameter environmental monitoring and intelligent pollution source identification method and system is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose a multi-parameter environmental monitoring and intelligent pollution source identification method and system. This invention introduces techniques such as phase transition layered interpretation of pollution disturbances, three-layer constraint lattice path pruning, and pollution effect prototype inversion calculation to achieve disturbance removal, pollution propagation link screening, and quantitative solution of pollution source contribution in multi-source environmental monitoring data. During data processing, this invention constructs a net disturbance extraction structure, a cross-media propagation constraint structure, and a multi-source inversion structure, enabling accurate identification of pollution events in complex background changes, screening of feasible propagation paths based on spatiotemporal constraints, and precise decomposition of multiple pollution source superposition scenarios. This invention possesses advantages such as strong interpretability, high identification accuracy, wide applicability, and real-time operation.

[0007] A multi-parameter environmental monitoring and intelligent pollution source identification method according to an embodiment of the present invention includes:

[0008] Collect multi-source environmental monitoring data within the target area, preprocess the multi-source environmental monitoring data, and form a multi-parameter environmental monitoring data sequence;

[0009] The multi-parameter environmental monitoring data sequence is input into the pollution disturbance phase transition hierarchical interpreter for phase state segmentation processing, identification of stable phase, transition phase and diffusion phase, construction of disturbance operator chain, inverse reconstruction of phase state buffer, and output of net disturbance sequence.

[0010] A three-layer constraint grid structure consisting of monitoring grids, pollution source grids, and meteorological grids is constructed. Each monitoring point is mapped to a monitoring grid, the emission domain of pollution sources is mapped to a pollution source grid, and the wind direction grid, water flow grid, and rainfall impact grid are mapped to a meteorological grid.

[0011] The net disturbance sequence is projected onto the monitoring grid, and the pollution propagation path is trimmed through the multi-level grid mapping relationship between the monitoring grid, meteorological grid and pollution source grid to generate a set of candidate pollution sources;

[0012] Based on the set of candidate pollution sources, a pollution effect prototype matrix is ​​constructed, the pollution effect prototype vector of each candidate pollution source is obtained, a perturbation inversion matrix is ​​constructed and inversion solution is performed to obtain the pollution contribution.

[0013] Candidate pollution sources are ranked and main pollution sources are screened based on their pollution contribution, and the target pollution source identification results and corresponding contribution scores are output.

[0014] Optionally, the multi-source environmental monitoring data includes air parameters, water parameters, soil parameters, and meteorological parameters. The air parameters include PM2.5, PM10, sulfur dioxide, nitrogen oxides, and ozone concentrations. The water parameters include chemical oxygen demand, ammonia nitrogen, dissolved oxygen, and conductivity. The soil parameters include moisture content and ion concentration. The meteorological parameters include wind speed, wind direction, precipitation, and water flow direction.

[0015] Optionally, the preprocessing of multi-source environmental monitoring data includes performing time alignment, outlier removal, missing value completion, and normalization encoding on the multi-source environmental monitoring data.

[0016] Optionally, the output net perturbation sequence includes:

[0017] Based on the multi-parameter environmental monitoring data sequence, the rate of change, acceleration of change and fluctuation amplitude at each sampling time are calculated to generate the corresponding feature distribution vector. A baseline statistical template is established based on historical pollution-free periods to obtain the initial value set of phase determination threshold. The feature distribution vector, the baseline statistical template and the initial value set of phase determination threshold are written into the feature statistical cache area.

[0018] A contamination perturbation phase transition hierarchical interpreter is constructed, which consists of a self-calibrated phase state segmentation layer, a cross-parameter coupled perturbation coding layer, and a bidirectional reversible perturbation chain buffer layer.

[0019] Within the self-calibrated phase segmentation layer, the feature statistics buffer is updated by a fixed-length rolling window. Threshold self-calibration is performed based on the feature distribution vector and the initial value set of phase determination thresholds. For each sampling time, a phase label for a stable phase, transition phase, or diffusion phase is generated. Data segments that are time-continuous and have consistent labels are divided into phase buffers.

[0020] Within the cross-parameter coupling perturbation coding layer, the phase buffer and feature statistics buffer are read, the actual perturbation operator type is identified according to the sampling time order, the start time, duration, amplitude information and parameter coupling label of each perturbation operator are recorded, and the perturbation operator chain is written into the bidirectional reversible perturbation chain buffer layer in the recording order.

[0021] Within the bidirectional reversible perturbation chain buffer layer, the inverse operators of each perturbation operator are first called according to the reverse index table to perform reverse reconstruction of the phase buffer. Then, forward consistency verification is performed according to the registration order. After bidirectional consistency, the net perturbation sequence is output.

[0022] Optionally, the construction of the three-layer constrained grid structure consisting of a monitoring grid, a pollution source grid, and a meteorological grid includes:

[0023] A monitoring grid is established in the target area. The net disturbance sequence is spatially discretized according to the honeycomb grid cells with fixed side lengths. A unique hierarchical code is generated for each monitoring grid cell, and the coordinates, elevation information and media type of the monitoring point are recorded.

[0024] Pollution source grids are established around the monitoring grids. Known emission outlets, overflow points and potential leakage sources are spatially mapped according to the same honeycomb geometric parameters. The emission direction, emission intensity level and radius of action are stored for each pollution source grid cell. A one-hop mapping table is established with neighboring monitoring grid cells through the hierarchical coding of the pollution source grid cells.

[0025] Meteorological grids are built on top of monitoring grids and pollution source grids. Wind field grids and hydrodynamic grids are superimposed into dynamic layers in time slices. Current wind speed and direction, rainfall intensity or water flow speed information is recorded for each meteorological grid. A three-dimensional raster that can be updated on a rolling basis is formed by timestamp and spatial index.

[0026] Based on the spatial adjacency list of monitoring grid cells and pollution source grid cells, as well as the wind direction, water flow and rainfall information recorded by meteorological grid cells, spatial accessibility constraints, directional consistency constraints and time window constraints are formulated to construct a three-layer grid structure cross-layer association rule set and generate a cross-layer path candidate directory.

[0027] According to the cross-layer association rule set, the monitoring grid cells, meteorological grid cells and pollution source grid cells are matched. When the conditions of spatial accessibility, directional consistency and time window are met, the corresponding pollution source grid cell is registered as a candidate pollution source of the monitoring grid cell.

[0028] Optionally, generating the candidate pollution source set includes:

[0029] Project the net disturbance sequence according to the monitoring grid, open the path record list in the three-level constraint grid data structure according to the hierarchical encoding of the monitoring grid cells, and establish path seeds for the monitoring grid cells where the disturbance start time occurs.

[0030] Based on the cross-layer association rule set, a forward expansion is performed between the monitoring grid, meteorological grid and pollution source grid starting from each path seed, and a forward feasible path is generated according to spatial accessibility constraints, directional consistency constraints and time window constraints;

[0031] For each pollution source grid cell, reverse verification is performed in reverse order from pollution source grid to monitoring grid. The forward feasible path is compared segment by segment using the time sequence and amplitude information of the perturbation operator chain. Path segments that do not meet the consistency of perturbation order or amplitude progression are deleted, and paths that simultaneously meet the requirements of forward expansion and reverse verification are retained, thus completing the bidirectional consistency pruning.

[0032] Perform phase alignment and medium connectivity checks on the bidirectional clipped path:

[0033] Only paths that are continuously connected within the transition or diffusion phase are retained;

[0034] For paths that cross air, water, or soil interfaces, cross-layer connectivity is checked based on the monitoring cell medium type and the wind field, water flow, and rainfall data recorded by meteorological cells, and paths that do not conform to the rules are eliminated.

[0035] The remaining paths are merged according to the pollution source grid cells to generate the minimum redundant path set. For each pollution source grid cell, the shortest evidence chain covering all disturbance monitoring grid cells is retained. The merged pollution source grid cells are registered as candidate pollution sources to form a candidate pollution source set.

[0036] Optionally, obtaining the pollution contribution includes:

[0037] Within the time slice set, according to the candidate pollution source set and the covered monitoring cell list, the observation data sequence with the same order is extracted from the net disturbance sequence, spliced ​​according to the rule of monitoring cell priority and time slice priority, and the phase label is segmented and weighted to form a phase-weighted observation data sequence.

[0038] Based on the cross-layer association relationship of the three-layer constraint lattice, a lattice constraint mask is constructed. Combinations of pollution sources-monitoring lattice cells-time slices that are unreachable or disconnected are marked as invalid, while combinations that simultaneously satisfy spatial accessibility, directional consistency and time window constraints are marked as valid.

[0039] For each candidate pollution source, a pollution effect prototype vector is generated that is consistent with the order of the observed data sequence. Amplitude normalization, time alignment and phase segmentation are performed. The phase weights are written into the data segments corresponding to the pollution effect prototype vector. The standardized pollution effect prototype matrix is ​​then assembled by column.

[0040] A lattice constraint mask structure is superimposed on the standardized pollution effect prototype matrix to obtain the mask constraint matrix. Column sparsity constraints and bounded condition number constraints are introduced. The rank determination and stability check are performed by a numerical decomposition step with column pivot selection to generate the perturbation inversion matrix.

[0041] The phase-weighted observation data sequence is input into the perturbation inversion matrix to obtain the pollution contribution sequence of candidate pollution sources. The contribution sequence is checked according to the non-negative pruning rule and the phase continuity check rule. The components that fail the check are set to zero or rolled back to the suboptimal solution. The pollution contribution and candidate pollution source number that pass the check are output.

[0042] Optionally, the output target pollution source identification results and corresponding contribution scores include:

[0043] Summarize the pollution contribution values ​​and compose a complete record for each candidate pollution source, its pollution contribution value, the corresponding evidence chain identifier, and the associated time slice;

[0044] Perform nonnegation and integrity checks on the summary records. Change records with pollution contribution values ​​less than zero to zero, fill in zeros for records with missing pollution contribution values, and keep the original order of the records unchanged.

[0045] Sort the pollution contribution values ​​from high to low, set a pollution contribution threshold, delete candidate pollution source records below the threshold, and retain candidate pollution source records above or equal to the threshold to generate a list of main pollution sources and corresponding pollution contribution scores.

[0046] The main pollution source list, pollution contribution score, evidence chain identifier, and associated time slices are integrated and output to form the final target pollution source identification result.

[0047] A multi-parameter environmental monitoring and intelligent pollution source identification system according to an embodiment of the present invention includes the following modules:

[0048] The data acquisition and preprocessing module is used to collect multi-source environmental monitoring data within the target area and perform preprocessing to form a multi-parameter environmental monitoring data sequence.

[0049] The perturbation layered interpretation module is used to receive multi-parameter environmental monitoring data sequences, construct a perturbation operator chain, perform inverse reconstruction on the phase buffer, and output a net perturbation sequence.

[0050] The constraint grid construction module is used to construct a three-layer constraint grid structure, mapping monitoring points to monitoring grids, mapping pollution source emission domains to pollution source grids, and mapping wind direction grids, water flow grids, and rainfall impact grids to meteorological grids.

[0051] The path trimming module is used to project the net disturbance sequence onto the monitoring grid, trim the pollution propagation path, and generate a set of candidate pollution sources;

[0052] The prototype and inversion module is used to construct a pollution effect prototype matrix based on the set of candidate pollution sources, obtain the pollution effect prototype vector of the candidate pollution sources, perform inversion solution, and obtain the pollution contribution.

[0053] The source tracing and screening output module is used to sort candidate pollution sources and screen main pollution sources according to their pollution contribution, and output the target pollution source identification results and corresponding contribution scores.

[0054] The beneficial effects of this invention are:

[0055] This invention constructs a pollution disturbance phase transition hierarchical interpreter to structurally analyze complex, multi-parameter environmental monitoring data according to phase characteristics. It identifies stable, abrupt, and diffusion changes in monitoring curves hierarchically and reversibly separates pollution disturbances using a perturbation operator chain, effectively solving the problem of traditional methods' difficulty in distinguishing between natural fluctuations and pollution disturbances. The output net disturbance sequence more realistically reflects the changing characteristics of pollution events, laying a reliable foundation for propagation path trimming and pollution source contribution calculation, and improving the accuracy and sensitivity of pollution event identification.

[0056] This invention constructs a three-layer constraint grid structure consisting of monitoring grids, pollution source grids, and meteorological grids. Through multi-dimensional constraint rules such as spatial accessibility, directional consistency, and time windows, it automatically prunes pollution propagation paths, achieving cross-media and cross-spatial pollution propagation link screening. Compared with existing technologies that rely on empirical inference, this structure can form propagation links with clear spatial logic and interpretability. By utilizing the correlation information of the three-layer grid elements, this invention can accurately eliminate unreasonable transmission paths in complex environments, improving the accuracy of propagation path judgment and the transparency of the source tracing process.

[0057] In the pollution source identification stage, this invention utilizes a pollution effect prototype matrix and a perturbation inversion structure to quantitatively invert candidate pollution sources. This decomposes the comprehensive perturbation in cases of multiple superimposed pollution sources into the independent contribution of each source, achieving multi-source pollution separation capabilities that traditional techniques cannot handle. By ranking and filtering the contributions, this invention can accurately locate the main pollution source and output a corresponding contribution score, providing clear quantitative evidence for the source tracing results. This invention not only improves the accuracy and interpretability of pollution identification but also enhances its practicality and adaptability in complex environments and multi-source pollution scenarios. Attached Figure Description

[0058] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0059] Figure 1 This is a flowchart of a multi-parameter environmental monitoring and intelligent pollution source identification method proposed in this invention;

[0060] Figure 2 This is a schematic diagram of the structure of a multi-parameter environmental monitoring and intelligent pollution source identification system proposed in this invention. Detailed Implementation

[0061] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0062] refer to Figure 1 A multi-parameter environmental monitoring and intelligent pollution source identification method, comprising:

[0063] Collect multi-source environmental monitoring data within the target area, preprocess the multi-source environmental monitoring data, and form a multi-parameter environmental monitoring data sequence;

[0064] The multi-parameter environmental monitoring data sequence is input into the pollution disturbance phase transition hierarchical interpreter for phase state segmentation processing, identification of stable phase, transition phase and diffusion phase, construction of disturbance operator chain, inverse reconstruction of phase state buffer, and output of net disturbance sequence.

[0065] A three-layer constraint grid structure consisting of monitoring grids, pollution source grids, and meteorological grids is constructed. Each monitoring point is mapped to a monitoring grid, the emission domain of pollution sources is mapped to a pollution source grid, and the wind direction grid, water flow grid, and rainfall impact grid are mapped to a meteorological grid.

[0066] The net disturbance sequence is projected onto the monitoring grid, and the pollution propagation path is trimmed through the multi-level grid mapping relationship between the monitoring grid, meteorological grid and pollution source grid to generate a set of candidate pollution sources;

[0067] Based on the set of candidate pollution sources, a pollution effect prototype matrix is ​​constructed, the pollution effect prototype vector of each candidate pollution source is obtained, a perturbation inversion matrix is ​​constructed and inversion solution is performed to obtain the pollution contribution.

[0068] Candidate pollution sources are ranked and main pollution sources are screened based on their pollution contribution, and the target pollution source identification results and corresponding contribution scores are output.

[0069] In this embodiment, the multi-source environmental monitoring data includes air parameters, water parameters, soil parameters, and meteorological parameters. The air parameters include PM2.5, PM10, sulfur dioxide, nitrogen oxides, and ozone concentrations. The water parameters include chemical oxygen demand, ammonia nitrogen, dissolved oxygen, and conductivity. The soil parameters include moisture content and ion concentration. The meteorological parameters include wind speed, wind direction, precipitation, and water flow direction.

[0070] In this embodiment, the preprocessing of multi-source environmental monitoring data includes performing time alignment, outlier removal, missing value completion, and normalization encoding on the multi-source environmental monitoring data.

[0071] In this embodiment, the output net perturbation sequence includes:

[0072] Based on the multi-parameter environmental monitoring data sequence, the rate of change, acceleration of change, and fluctuation amplitude at each sampling moment are calculated to generate corresponding feature distribution vectors. A baseline statistical template is established based on historical pollution-free periods to obtain an initial set of phase determination thresholds. The feature distribution vectors, baseline statistical templates, and initial set of phase determination thresholds are written into the feature statistical cache. The baseline statistical template is stored in matrix form, with row indices corresponding to each monitoring parameter and column indices corresponding to statistical indicators. Each cell records the mean, standard deviation, maximum, minimum, and upper limit of the 95% confidence interval for each parameter during pollution-free periods. The lower limit is given by subtracting twice the standard deviation from the mean. The baseline statistical template also stores the timestamp range. The calculation of the rate of change, acceleration of change, and fluctuation amplitude at each sampling moment is specifically as follows:

[0073] The rate of change is obtained by dividing the difference between the current monitored value and the monitored value at the previous sampling time by a fixed sampling period.

[0074] After obtaining two consecutive rates of change, the difference between the two is calculated and then divided by the sampling period to reflect the increasing or decreasing trend of the rate of change itself.

[0075] The fluctuation amplitude is extracted from the maximum and minimum values ​​of the monitored values ​​within a sliding time window centered on the current time. The difference between the two values ​​describes the magnitude of the monitoring curve during that period, which is used to determine abnormal fluctuations.

[0076] Stable phase thresholds: absolute value of rate of change ≤ 2 μg / m³·min, absolute value of acceleration ≤ 0.5 μg / m³·min², fluctuation amplitude ≤ 10 μg / m³;

[0077] Transition phase threshold: the rate of change is between 2 and 8 micrograms per cubic meter per minute, or the acceleration of change is between 0.5 and 2 micrograms per cubic meter per minute², and the fluctuation range is ≤25 micrograms per cubic meter;

[0078] Diffusion phase threshold: rate of change ≥ 8 μg / m³·min, or acceleration of change ≥ 2 μg / m³·min², or fluctuation amplitude ≥ 25 μg / m³;

[0079] A contamination perturbation phase transition hierarchical interpreter is constructed, which consists of a self-calibrating phase state segmentation layer, a cross-parameter coupled perturbation coding layer, and a bidirectional reversible perturbation chain buffer layer, wherein:

[0080] The self-calibrating phase segmentation layer takes the feature distribution vector and the initial set of phase determination thresholds as input, and outputs a phase label and a phase buffer. Specifically, the output phase label and phase buffer are as follows:

[0081] The feature distribution vector at the current sampling time is compared item by item with the initial value set of phase state determination thresholds. Based on the results, it is directly labeled as a stable phase, transitional phase, or diffusion phase, and the label is written into the phase state sequence index.

[0082] When the phase flags of two adjacent sampling times are consistent, the current buffer is continuously expanded, and the start time, the latest time, and the corresponding data pointer are recorded.

[0083] When a change in phase marker is detected or a preset duration threshold is reached, the existing buffer is immediately sealed, the start and end times, sample index, and phase type are written into the buffer list, a new buffer is created, and recording begins from the current time.

[0084] The cross-parameter coupling perturbation coding layer generates step perturbation operators, pulse perturbation operators, incremental diffusion perturbation operators, and fallback perturbation operators within the phase buffer and attaches parameter coupling tags. Specifically, the generation of the step perturbation operator, pulse perturbation operator, incremental diffusion perturbation operator, and fallback perturbation operator involves:

[0085] Within the phase buffer, the curves of each monitoring parameter are scanned with a fixed window. When the monitored value shows a significant jump between two consecutive samples and then maintains a new level, the segment of the curve is marked as a step perturbation operator.

[0086] Within the same buffer zone, short-term spikes or troughs are detected. If the monitored value increases or decreases rapidly in a very short time and then immediately returns to the original level, the instantaneous period is marked as a pulse disturbance operator.

[0087] Fit the slow segments in the buffer that show a continuous upward or downward trend. When the monitored value gradually climbs to the peak and then stabilizes or gradually falls back to the baseline level, generate the incremental diffusion perturbation operator and the fall perturbation operator respectively. Record the index of the same type of feature segment that appears synchronously with other parameters in each operator as the coupling label between parameters.

[0088] A bidirectional reversible perturbation chain buffer layer establishes a perturbation operator chain and an inverse index table, and registers the inverse operators of each perturbation operator. Specifically, establishing the perturbation operator chain and the inverse index table involves:

[0089] The step, pulse, incremental diffusion and fallback perturbation operators extracted in time sequence from the phase buffer are connected sequentially to form a forward perturbation operator chain, and the position index of the starting time of each operator in the chain is recorded.

[0090] The forward chain index order is reversed to generate a reverse index table, so that the chain tail operator is at the beginning of the table and the chain head operator is at the end of the table. For each operator, its termination time and the index pointer of the previous operator in the chain are recorded in the table.

[0091] For each perturbation operator in the chain, register the corresponding inverse operator reference in the buffer and store it in the inverse operator mapping table at the same level as the inverse index table;

[0092] Within the self-calibrating phase segmentation layer, the feature statistics buffer is updated using a fixed-length rolling window. Threshold self-calibration is performed based on the feature distribution vector and the initial set of phase determination thresholds. For each sampling moment, a phase label (stable phase, transition phase, or diffusing phase) is generated. Time-continuous and consistently labeled data segments are divided into phase buffers, where:

[0093] A stable phase refers to a situation where the rate of change, acceleration of change, and amplitude of fluctuation of each monitored parameter all fall within the normal fluctuation range of the baseline statistical template, without any obvious trend of increase or decrease, and the data curve remains flat. When the self-calibration threshold judgment result shows that all characteristic indicators are in the normal range, the current sampling time is marked as a stable phase.

[0094] The transition phase refers to the first time that the rate of change or acceleration of change of at least one monitoring parameter exceeds the stable range, but the overall fluctuation amplitude has not continued to expand. The curve shows a fluctuating turning point or a short-term upward trend. When the self-calibration threshold judgment result shows that a single indicator has entered the warning range while the other indicators are still in the normal or slightly abnormal range, the sampling time is marked as the transition phase.

[0095] The diffusion phase refers to the simultaneous and sustained significant deviation of two or more monitoring parameters, with the rate of change and the amplitude of fluctuation exceeding the upper limit of the warning interval, and maintaining the same direction of growth or diffusion characteristics in adjacent time periods. When the self-calibration threshold judgment result shows that multiple indicators are continuously in the abnormally high range and show a diffusion trend, the corresponding sampling time is marked as the diffusion phase. The time periods of the stable phase, transition phase or diffusion phase that are continuously marked with the same characteristics are automatically merged to form the corresponding phase buffer.

[0096] Within the cross-parameter coupling perturbation coding layer, the phase buffer and feature statistics buffer are read, the actual perturbation operator type is identified according to the sampling time order, the start time, duration, amplitude information and parameter coupling label of each perturbation operator are recorded, and the perturbation operator chain is written into the bidirectional reversible perturbation chain buffer layer in the recording order.

[0097] Within the bidirectional reversible perturbation chain buffer layer, the inverse operators of each perturbation operator are first called according to the reverse index table to perform reverse reconstruction of the phase buffer. Then, forward consistency verification is performed according to the registration order. After bidirectional consistency is achieved, the net perturbation sequence is output, where:

[0098] The phase buffer is reconstructed in reverse by calling the inverse operators of each perturbation operator according to the reverse index table. Specifically:

[0099] Based on the reverse index table, the perturbation operator is read item by item from the end of the chain. The data segment at the end time in the phase buffer is located. The numerical recovery operation is performed on the data segment according to the registered inverse operator type to gradually cancel out the perturbation effect from the latest time to an earlier time.

[0100] After each reverse recovery is completed, the index pointer is moved forward to continue searching for the previous perturbation operator and the corresponding data segment. The reverse processing is repeated until the head of the chain is reached, and the contents of the buffer are continuously updated during the process.

[0101] Once the reverse processing of all perturbation operators in the chain is completed, the buffer is traversed to confirm that all data segments have been restored to the baseline level, and the start and end times of this round of reverse reconstruction, the list of used reverse operators, and the recovery completion flag are recorded.

[0102] Then, a forward consistency check is performed according to the registration order, specifically as follows:

[0103] Following the original registration order, each perturbation operator is reapplied sequentially from the beginning to the end of the chain. The data after reverse reconstruction is superimposed and restored segment by segment. The overlapping area of ​​each step is compared with the original perturbation amplitude before reverse reconstruction in real time. If the error exceeds the set tolerance, the calculation is rolled back until it is consistent.

[0104] After completing the full-chain forward overlay, the scan buffer confirms that the error between the restored curve and the original monitoring baseline is within the tolerance range at all sampling points. If all passes, a verification success flag is recorded; otherwise, an anomaly flag is triggered and the upper-level logic readjusts the threshold or expands the buffer.

[0105] In this embodiment, the construction of a three-layer constrained grid structure consisting of a monitoring grid, a pollution source grid, and a meteorological grid includes:

[0106] A monitoring grid is established in the target area. The net disturbance sequence is spatially discretized according to the honeycomb grid cells with fixed side lengths. A unique hierarchical code is generated for each monitoring grid cell, and the coordinates, elevation information and media type of the monitoring point are recorded.

[0107] A pollution source grid is established around the monitoring grid. Known emission outlets, overflow points, and potential leakage sources are spatially mapped using the same honeycomb geometric parameters. The emission direction, emission intensity level, and radius of effect are stored for each pollution source grid cell. A one-hop mapping table is established between the pollution source grid cells and neighboring monitoring grid cells through hierarchical coding. Specifically, the spatial mapping of known emission outlets, overflow points, and potential leakage sources using the same honeycomb geometric parameters involves:

[0108] Outside the monitoring grid, the same honeycomb side length and hexagonal layout parameters as the monitoring grid are used to establish a honeycomb grid framework covering all discharge outlets, overflow points and potential leakage sources in the overall coordinate system of the park.

[0109] Obtain the latitude and longitude or planar coordinates of each discharge outlet, overflow point and potential leakage source, and project the coordinates onto the honeycomb grid frame. When the coordinate is located inside a single honeycomb cell, the honeycomb cell is directly marked as the pollution source cell. When the coordinate is located in the area where two honeycomb cells share the same edge, the honeycomb cell to which it belongs is selected according to the principle of being closest to the center point of the honeycomb and the boundary alignment mark is recorded.

[0110] Once the pollution source characteristics are determined, they are written into the corresponding cellular unit to generate pollution source grid cells. The emission direction, emission intensity level and radius of effect are saved, and hierarchical codes are automatically assigned according to the cellular numbering rules.

[0111] Meteorological grids are built on top of monitoring grids and pollution source grids. Wind field grids and hydrodynamic grids are superimposed into dynamic layers in time slices. Current wind speed and direction, rainfall intensity or water flow speed information is recorded for each meteorological grid. A three-dimensional raster that can be updated on a rolling basis is formed by timestamp and spatial index.

[0112] Based on the spatial adjacency list of monitoring grid cells and pollution source grid cells, as well as the wind direction, water flow, and rainfall information recorded by meteorological grid cells, spatial accessibility constraints, directional consistency constraints, and time window constraints are formulated to construct a three-layer grid structure cross-layer association rule set and generate a cross-layer path candidate directory, wherein:

[0113] Spatial accessibility constraints refer to the requirement that there must be a connecting path between the monitoring grid cell and the pollution source grid cell, consisting of continuous cellular units that does not cross terrain barriers, and that the total length of the path does not exceed the effective radius of the pollution source grid cell record.

[0114] The directional consistency constraint means that the main extension direction of the connecting path must be kept within the set angle tolerance of the instantaneous wind direction or water flow direction recorded by the meteorological grid. The consistency between the local direction of each segment of the path and the main direction is used as the judgment criterion. The angle tolerance is that the deviation between the main extension direction of the path and the instantaneous wind direction or water flow direction does not exceed 30°.

[0115] The time window constraint means that the pollutant propagation time calculated based on wind speed or water flow speed must fall within the preset allowable time difference between the time of emission event and the time of monitoring disturbance. Paths that exceed the range are considered to not meet the timing conditions.

[0116] According to the cross-layer association rule set, the monitoring grid cells, meteorological grid cells and pollution source grid cells are matched. When the conditions of spatial accessibility, directional consistency and time window are met, the corresponding pollution source grid cell is registered as a candidate pollution source of the monitoring grid cell.

[0117] In this embodiment, generating the candidate pollution source set includes:

[0118] Project the net disturbance sequence according to the monitoring grid, open the path record list in the three-level constraint grid data structure according to the hierarchical encoding of the monitoring grid cells, and establish path seeds for the monitoring grid cells where the disturbance start time occurs.

[0119] Based on the cross-layer association rule set, a forward expansion is performed between the monitoring grid, meteorological grid, and pollution source grid, starting from each path seed. A forward feasible path is generated according to spatial accessibility constraints, directional consistency constraints, and time window constraints. Specifically, the forward expansion involves:

[0120] Starting from the monitored cell path seed, prioritize moving along the adjacent cell that is closest to the real-time wind direction or water flow direction, and add the cells passed to the temporary path stack.

[0121] During the forward movement, the cumulative distance value of each segment is continuously checked to see if it exceeds the radius of action of the pollution source cell, and each direction must be kept within the tolerance of directional consistency. If the conditions are met, the path segment is locked and the expansion continues; if not, other adjacent cells are retrieved back.

[0122] When the expected propagation time of the extended path falls within the allowable time window between the emission time and the time when the monitoring disturbance occurs, and is connected to the target pollution source cell, the entire path is confirmed as a forward feasible path and written into the path set; otherwise, the extension is terminated and marked as invalid.

[0123] For each pollution source grid cell, a reverse check is performed in reverse order from the pollution source grid to the monitoring grid. The forward feasible path is compared segment by segment using the time sequence and amplitude information of the perturbation operator chain. Path segments that do not meet the consistency of perturbation order or amplitude progression are deleted, and paths that simultaneously satisfy both forward expansion and reverse check are retained, completing the bidirectional consistency pruning. Specifically, the reverse check is performed as follows:

[0124] Traverse the forward feasible path from the endpoint of the pollution source grid cell in reverse order, and compare the time order of arrival at the monitoring grid cell of each segment with the order of occurrence of the disturbance recorded by the disturbance operator chain. If the order does not match, mark the segment of the path as invalid.

[0125] If the order is correct, check whether the disturbance amplitude of each monitoring cell in the path segment shows a continuous increasing or decreasing trend, and whether it is consistent with the amplitude change direction of the same type of operator in the disturbance operator chain. If the trend is inconsistent, delete the segment.

[0126] Once all path segments have passed the sequence check and magnitude trend check, the path is marked as a valid path with bidirectional consistency. Paths that fail the check are removed as a whole and the reason for the removal is recorded.

[0127] Perform phase alignment and medium connectivity checks on the bidirectional clipped path:

[0128] Only paths that are continuously connected within the transition or diffusion phase are retained;

[0129] For paths that cross air, water, or soil interfaces, cross-layer connectivity is checked based on the monitoring cell medium type and the wind field, water flow, and rainfall data recorded by meteorological cells, and paths that do not conform to the rules are eliminated.

[0130] The remaining paths are merged according to the pollution source grid cells to generate the minimum redundant path set. For each pollution source grid cell, the shortest evidence chain covering all disturbance monitoring grid cells is retained. The merged pollution source grid cells are registered as candidate pollution sources to form a candidate pollution source set.

[0131] In this embodiment, obtaining the pollution contribution rate includes:

[0132] Within the time slice set, according to the candidate pollution source set and the covered monitoring cell list, the observation data sequence with the same order is extracted from the net disturbance sequence, spliced ​​according to the rule of monitoring cell priority and time slice priority, and the phase label is segmented and weighted to form a phase-weighted observation data sequence.

[0133] Based on the cross-layer association relationship of the three-layer constraint lattice, a lattice constraint mask is constructed. Combinations of pollution sources-monitoring lattice cells-time slices that are unreachable or disconnected are marked as invalid, while combinations that simultaneously satisfy spatial accessibility, directional consistency and time window constraints are marked as valid.

[0134] For each candidate pollution source, a pollution effect prototype vector is generated that matches the order of the observed data sequence. Amplitude normalization, time alignment, and phase segmentation are performed, and phase weights are written into the data segments corresponding to the pollution effect prototype vector. These are then assembled column-wise to form a standardized pollution effect prototype matrix. Specifically, generating a pollution effect prototype vector for each candidate pollution source that matches the order of the observed data sequence involves:

[0135] Based on the time index of the observation data sequence, the historical emission records and typical diffusion curves of candidate pollution sources are resampled to the same time step;

[0136] Fill the empty spaces with the resampled emission intensity in the order of the monitoring parameters in the observation data sequence, and fill in zero for time periods when there is no emission impact;

[0137] The filled vector is segmented according to phase state and inserted into pre-calculated phase state weight coefficients to form a pollution effect prototype vector that is completely consistent with the order of the observed data sequence.

[0138] A lattice-constrained mask structure is superimposed on the standardized pollution prototype matrix to obtain a mask constraint matrix. Column sparsity constraints and bounded condition number constraints are introduced. A numerical decomposition step with column pivot selection is used for rank determination and stability verification to generate a perturbation inversion matrix. The lattice-constrained mask structure is a binary mask constructed on the same dimension as the standardized pollution prototype matrix. The row index corresponds to the combination of each monitoring lattice cell and time slice in the observation data sequence, and the column index corresponds to the candidate pollution source. Row-column cells that do not meet the conditions of spatial reachability, directional consistency, or time window are assigned zero values, and cells that meet all constraints are assigned one value. If row-column cells span different media but are still connected, the weight bits corresponding to the media penalty coefficient are recorded in the mask, ultimately forming the constraint matrix.

[0139] Among them, column sparsity constraint sets a non-zero upper limit for each candidate pollution source column in the mask constraint matrix. By reducing or zeroing weakly correlated column entries through threshold reduction, the number of effective columns is kept within the preset sparsity range, thereby highlighting the influence of the main pollution sources and suppressing the interference of redundant columns.

[0140] The bounded condition number constraint is used to dynamically evaluate the condition number of the matrix after column sparsity processing, and the condition number is limited to not exceeding a preset stability threshold; when the condition number exceeds the limit, the column vectors are scaled, rearranged, or collinear columns are removed.

[0141] The numerical decomposition step with column pivoting selection is used for rank determination and stability verification, specifically as follows:

[0142] First, arrange the mask constraint matrix in descending order of column norm, then select the column with the largest value as the pivot and record the column swapping order;

[0143] During the decomposition process, the size of the principal component is detected in real time. If the principal component is lower than the set tolerance, the column and dependent column are marked as out-of-rank columns to determine the actual rank of the matrix and remove redundant columns.

[0144] After decomposition, the matrix condition number is estimated based on the principal component sequence of the records. If the condition number exceeds the stability threshold, the columns with the smallest contribution are compressed or removed in the order of column swapping until the condition number returns to within the threshold, thus completing the stability check. The threshold is the stability threshold of 10,000, meaning the matrix condition number must not exceed 10,000.

[0145] The phase-weighted observation data sequence is input into the perturbation inversion matrix to obtain the pollution contribution sequence of candidate pollution sources. The contribution sequence is then checked according to the non-negative pruning rule and the phase continuity check rule. Components that fail the check are either set to zero or rolled back to a suboptimal solution. The pollution contribution sequences that pass the check, along with the candidate pollution source numbers, are output.

[0146] The pollution contribution sequence of candidate pollution sources is obtained by inputting the phase-weighted observation data sequence into the perturbation inversion matrix, specifically:

[0147] The phase-weighted observation data are filled into the solution input vector according to the row and column indices of the perturbation inversion matrix;

[0148] The perturbation inversion matrix and the input vector are operated column by column using matrix-vector multiplication. The dot product of each column vector and the input vector is calculated and accumulated to obtain the original value of the impact of each candidate pollution source on the overall perturbation.

[0149] All original impact values ​​are scaled uniformly according to the column normalization coefficient, converted into contribution values ​​under the same dimension, and written into the contribution sequence in sequence while maintaining a one-to-one correspondence with the candidate pollution source number, thus obtaining the pollution contribution sequence of the candidate pollution source.

[0150] The non-negative pruning rule is as follows: when a contribution value less than zero appears in the pollution contribution sequence, the negative value is directly set to zero; if the contribution value is close to zero and lower than the preset micro-threshold, it is considered to have no effective contribution and is also set to zero; after pruning, only the non-negative contribution components that are greater than the micro-threshold are retained.

[0151] The phase continuity check rule is as follows: For each candidate pollution source, the effective components in the contribution sequence are checked in time order. Adjacent effective components must correspond to continuous or adjacent phase segments. If the phase segment corresponding to the contribution component is not continuous with the phase segments on both sides, or if the contribution only appears in a single isolated time slice, it is considered that the continuity requirement is not met, and the component is set to zero or rolled back to the suboptimal solution.

[0152] In this embodiment, the output of the target pollution source identification result and the corresponding contribution score includes:

[0153] Summarize the pollution contribution values ​​and compose a complete record for each candidate pollution source, its pollution contribution value, the corresponding evidence chain identifier, and the associated time slice;

[0154] Perform nonnegation and integrity checks on the summary records. Change records with pollution contribution values ​​less than zero to zero, fill in zeros for records with missing pollution contribution values, and keep the original order of the records unchanged.

[0155] Sort the pollution contribution values ​​from high to low, set a pollution contribution threshold, delete candidate pollution source records below the threshold, and retain candidate pollution source records above or equal to the threshold to generate a list of main pollution sources and corresponding pollution contribution scores.

[0156] The main pollution source list, pollution contribution score, evidence chain identifier, and associated time slices are integrated and output to form the final target pollution source identification result.

[0157] refer to Figure 2 A multi-parameter environmental monitoring and intelligent pollution source identification system includes the following modules:

[0158] The data acquisition and preprocessing module is used to collect multi-source environmental monitoring data within the target area and perform preprocessing to form a multi-parameter environmental monitoring data sequence.

[0159] The perturbation layered interpretation module is used to receive multi-parameter environmental monitoring data sequences, construct a perturbation operator chain, perform inverse reconstruction on the phase buffer, and output a net perturbation sequence.

[0160] The constraint grid construction module is used to construct a three-layer constraint grid structure, mapping monitoring points to monitoring grids, mapping pollution source emission domains to pollution source grids, and mapping wind direction grids, water flow grids, and rainfall impact grids to meteorological grids.

[0161] The path trimming module is used to project the net disturbance sequence onto the monitoring grid, trim the pollution propagation path, and generate a set of candidate pollution sources;

[0162] The prototype and inversion module is used to construct a pollution effect prototype matrix based on the set of candidate pollution sources, obtain the pollution effect prototype vector of the candidate pollution sources, perform inversion solution, and obtain the pollution contribution.

[0163] The source tracing and screening output module is used to sort candidate pollution sources and screen main pollution sources according to their pollution contribution, and output the target pollution source identification results and corresponding contribution scores. Example

[0164] To verify the feasibility of this invention in practice, it was applied to the environmental monitoring and emission control of an industrial park. The industrial park covers approximately 8 square kilometers and contains multiple fixed exhaust gas emission devices, solvent storage tank exhaust facilities, and auxiliary treatment equipment. The park has set up eight air monitoring points in accordance with environmental protection requirements, and simultaneously collects meteorological data such as wind speed and direction for daily emission monitoring and operational assessment.

[0165] During actual operation, the park management found that although some emission equipment did not show obvious excessive emissions, aging equipment, fluctuating operating conditions, or short-term abnormal operation often manifested as small, short-term, and non-continuous abnormal disturbances in multiple parameters in environmental monitoring data. Traditional threshold-based alarm monitoring systems struggle to identify such situations, often only triggering alarms after severe equipment abnormalities or continuous exceedances, leading to delayed maintenance, decreased equipment operating efficiency, and even increased risk of unplanned downtime. Therefore, the park hopes to use environmental monitoring data to reverse-analyze the operational status of pollution sources, enabling early warning and support for maintenance decisions.

[0166] In this scenario, from 13:30 to 14:10 on September 18, 2025, the park's monitoring system detected slight abnormal fluctuations in air quality indicators at some monitoring points. According to the method of this invention, multi-source environmental monitoring data were first collected and preprocessed to form a multi-parameter environmental monitoring data sequence. This data sequence was then input into a pollution disturbance phase transition stratification interpreter. Through comprehensive analysis of the rate of change, acceleration of change, and fluctuation amplitude, the data was segmented into phase states. Between 13:40 and 13:55, multiple monitoring points were identified as transitional and locally diffused phases, rather than normal stable phases, indicating the presence of non-background disturbance characteristics during this period.

[0167] Through perturbation operator chain construction and reverse reconstruction, the system successfully removed the natural fluctuations caused by the slight increase in wind speed on that day, outputting a net perturbation sequence. Although this net perturbation sequence did not reach the traditional exceedance threshold, it exhibited obvious multi-parameter synchronous offset characteristics, including the synchronous increase of PM2.5, PM10, and nitrogen oxides, and the duration of these parameters highly overlapped with the equipment operating shifts.

[0168] The system constructs a three-layer constrained grid structure consisting of monitoring grids, pollution source grids, and meteorological grids, and combines real-time wind direction information to trim the pollution propagation path. The trimming results show that the disturbance propagation path is highly concentrated towards a solvent storage tank exhaust device on the northwest side of the park. Further inversion solving is performed by combining the pollution effect prototype matrix and the disturbance inversion matrix to obtain the pollution contribution of each candidate pollution source. The results show that the pollution contribution of this storage tank exhaust device is significantly higher than that of other emission devices during this period, but the overall value is still in the low to moderate range.

[0169] By analyzing the trend of pollution contribution over time, the system determined that the equipment did not experience a sudden emission accident, but rather intermittent abnormal emissions caused by decreased emission efficiency or deterioration of sealing performance. Based on this determination, the system automatically marked the pollution source as a risk source of abnormal operating status and generated an equipment operating status assessment result, indicating that there may be problems with aging seals or unstable pressure control, and suggested arranging maintenance and inspection in advance.

[0170] The park's maintenance personnel inspected the storage tank's exhaust system over the following week and discovered slight aging of a valve seal. While this did not cause significant leakage, it did generate short-term emission fluctuations during load changes. By replacing the seal in advance, the equipment returned to stable operation, and no similar disturbances were subsequently observed in environmental monitoring data. The results demonstrate that this invention not only enables intelligent identification of pollution sources but also effectively reflects the operating status of emission equipment based on changes in pollution contribution, providing a reliable basis for early warning and maintenance.

[0171] Table 1. Correlation data between air quality and equipment operation at typical monitoring points in the park.

[0172]

[0173] As shown in Table 1, at 13:35, the air quality indicators at monitoring point B2 were relatively stable, with PM2.5 at 28 micrograms per cubic meter, PM10 at 46 micrograms per cubic meter, and nitrogen oxides at 18 micrograms per cubic meter. Combined with the wind speed of approximately 2.5 meters per second at that time, this can be considered a typical background condition. Subsequently, between 13:40 and 13:50, the three pollutant indicators at monitoring point B3 continued to rise: PM2.5 increased from 31 to 39 micrograms per cubic meter, PM10 from 50 to 58 micrograms per cubic meter, and nitrogen oxides from 19 to 24 micrograms per cubic meter. These changes exhibited continuity and synchronicity, significantly deviating from the normal fluctuation characteristics of this period, indicating signs of non-random background disturbance.

[0174] Regarding spatial propagation characteristics, approximately 5 to 10 minutes after the disturbance occurred at monitoring point B3, a synchronous increase in pollutant concentrations was also detected at downstream monitoring point B4 at 13:55 and 14:00. Both PM2.5 and PM10 showed increases compared to previous times, while nitrogen oxides also showed a slight increase. Combined with the stable wind direction maintained between 229° and 231° and wind speed remaining around 2.6 meters per second, it can be seen that the pollution disturbance exhibited a gradual diffusion characteristic along the wind direction. This temporal relationship, where the disturbance first appeared at B3 and then manifested at B4, is consistent with the physical laws governing pollutant propagation with airflow, indicating that the source of the disturbance is more likely located in the upwind region of B3.

[0175] At 14:05, monitoring point B2 showed a slight fluctuation again, but the amplitude was significantly lower than that of B3 and B4, corresponding to PM2.5 of 30 micrograms per cubic meter and PM10 of 48 micrograms per cubic meter. This indicates that B2 was on the edge of the diffusion influence or affected by the tail of dilution. Overall, the data in the table reflects a pollution disturbance process with low intensity but short duration and a clear propagation path. It did not reach the traditional exceedance alarm threshold and had obvious spatiotemporal correlation, providing real and reasonable data support for identifying abnormal operating status of pollution sources based on multi-parameter disturbance analysis.

[0176] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for multi-parameter environmental monitoring and intelligent identification of pollution sources, characterized in that, include: Multi-source environmental monitoring data are collected within the target area. The multi-source environmental monitoring data is preprocessed to form a multi-parameter environmental monitoring data sequence. The multi-source environmental monitoring data includes air parameters, water parameters, soil parameters, and meteorological parameters. The air parameters include PM2.5, PM10, sulfur dioxide, nitrogen oxides, and ozone concentrations. The water parameters include chemical oxygen demand, ammonia nitrogen, dissolved oxygen, and conductivity. The soil parameters include moisture content and ion concentration. The meteorological parameters include wind speed, wind direction, precipitation, and water flow direction. The multi-parameter environmental monitoring data sequence is input into the pollution disturbance phase transition hierarchical interpreter for phase state segmentation processing, identification of stable phase, transition phase and diffusion phase, construction of disturbance operator chain, inverse reconstruction of phase state buffer, and output of net disturbance sequence. A three-layer constraint grid structure consisting of monitoring grids, pollution source grids, and meteorological grids is constructed. Each monitoring point is mapped to a monitoring grid, the emission domain of pollution sources is mapped to a pollution source grid, and the wind direction grid, water flow grid, and rainfall impact grid are mapped to a meteorological grid. The net disturbance sequence is projected onto the monitoring grid, and the pollution propagation path is trimmed through the multi-level grid mapping relationship between the monitoring grid, meteorological grid and pollution source grid to generate a set of candidate pollution sources; Based on the set of candidate pollution sources, a pollution effect prototype matrix is ​​constructed, the pollution effect prototype vector of each candidate pollution source is obtained, a perturbation inversion matrix is ​​constructed and inversion solution is performed to obtain the pollution contribution. Candidate pollution sources are ranked and main pollution sources are screened based on their pollution contribution, and the target pollution source identification results and corresponding contribution scores are output. The output net perturbation sequence includes: Based on the multi-parameter environmental monitoring data sequence, the rate of change, acceleration of change and fluctuation amplitude at each sampling moment are calculated to generate the corresponding feature distribution vector. A baseline statistical template is established based on historical pollution-free periods to obtain the initial value set of phase determination threshold. The feature distribution vector, the baseline statistical template and the initial value set of phase determination threshold are written into the feature statistical cache. A contamination perturbation phase transition hierarchical interpreter is constructed, which consists of a self-calibrating phase state segmentation layer, a cross-parameter coupled perturbation coding layer, and a bidirectional reversible perturbation chain buffer layer, wherein: The self-calibrating phase segmentation layer takes the feature distribution vector and the initial set of phase determination thresholds as input and outputs phase labels and phase buffers. The cross-parameter coupling perturbation coding layer generates step perturbation operators, pulse perturbation operators, incremental diffusion perturbation operators, and fallback perturbation operators in the phase buffer and adds inter-parameter coupling tags. A bidirectional reversible perturbation chain buffer layer establishes a perturbation operator chain and an inverse index table, and registers the inverse operators of each perturbation operator; Within the self-calibrated phase segmentation layer, the feature statistics buffer is updated by a fixed-length rolling window. Threshold self-calibration is performed based on the feature distribution vector and the initial value set of phase determination thresholds. For each sampling time, a phase label for a stable phase, transition phase, or diffusion phase is generated. Data segments that are time-continuous and have consistent labels are divided into phase buffers. Within the cross-parameter coupling perturbation coding layer, the phase buffer and feature statistics buffer are read, the actual perturbation operator type is identified according to the sampling time order, the start time, duration, amplitude information and parameter coupling label of each perturbation operator are recorded, and the perturbation operator chain is written into the bidirectional reversible perturbation chain buffer layer in the recording order. Within the bidirectional reversible perturbation chain buffer layer, the inverse operators of each perturbation operator are first called according to the reverse index table to perform reverse reconstruction of the phase buffer. Then, forward consistency verification is performed according to the registration order. After bidirectional consistency is achieved, the net perturbation sequence is output, where: The phase buffer is reconstructed in reverse by calling the inverse operators of each perturbation operator according to the reverse index table. Specifically: Based on the reverse index table, the perturbation operator is read item by item from the end of the chain. The data segment at the end time in the phase buffer is located. The numerical recovery operation is performed on the data segment according to the registered inverse operator type to gradually cancel out the perturbation effect from the latest time to an earlier time. After each reverse recovery is completed, the index pointer is moved forward to continue searching for the previous perturbation operator and the corresponding data segment. The reverse processing is repeated until the head of the chain is reached, and the contents of the buffer are continuously updated during the process. Once the reverse processing of all perturbation operators in the chain is completed, the buffer is traversed to confirm that all data segments have been restored to the baseline level, and the start and end times of this round of reverse reconstruction, the list of used reverse operators, and the recovery completion flag are recorded. Then, a forward consistency check is performed according to the registration order, specifically as follows: Following the original registration order, each perturbation operator is reapplied sequentially from the beginning to the end of the chain. The data after reverse reconstruction is superimposed and restored segment by segment. The overlapping area of ​​each step is compared with the original perturbation amplitude before reverse reconstruction in real time. If the error exceeds the set tolerance, the calculation is rolled back until it is consistent. After completing the full-chain forward overlay, the scan buffer confirms that the error between the restored curve and the original monitoring baseline is within the tolerance range at all sampling points. If all passes, the verification success flag is recorded; otherwise, an anomaly flag is triggered and the upper-level logic readjusts the threshold or expands the buffer. The construction of the three-layer constrained grid structure, consisting of monitoring grids, pollution source grids, and meteorological grids, includes: A monitoring grid is established in the target area. The net disturbance sequence is spatially discretized according to the honeycomb grid cells with fixed side lengths. A unique hierarchical code is generated for each monitoring grid cell, and the coordinates, elevation information and media type of the monitoring point are recorded. Pollution source grids are established around the monitoring grids. Known emission outlets, overflow points and potential leakage sources are spatially mapped according to the same honeycomb geometric parameters. The emission direction, emission intensity level and radius of action are stored for each pollution source grid cell. A one-hop mapping table is established with neighboring monitoring grid cells through the hierarchical coding of the pollution source grid cells. Meteorological grids are built on top of monitoring grids and pollution source grids. Wind field grids and hydrodynamic grids are superimposed into dynamic layers in time slices. Current wind speed and direction, rainfall intensity or water flow speed information is recorded for each meteorological grid. A three-dimensional raster that can be updated on a rolling basis is formed by timestamp and spatial index. Based on the spatial adjacency list of monitoring grid cells and pollution source grid cells, as well as the wind direction, water flow, and rainfall information recorded by meteorological grid cells, spatial accessibility constraints, directional consistency constraints, and time window constraints are formulated to construct a three-layer grid structure cross-layer association rule set and generate a cross-layer path candidate directory, wherein: Spatial accessibility constraints refer to the requirement that there must be a connecting path between the monitoring grid cell and the pollution source grid cell, consisting of continuous cellular units that does not cross terrain barriers, and that the total length of the path does not exceed the effective radius of the pollution source grid cell record. The directional consistency constraint means that the main extension direction of the connecting path must be kept within the set angular tolerance of the instantaneous wind direction or water flow direction recorded by the meteorological grid, and the consistency between the local direction of each segment of the path and the main direction is used as the judgment criterion. The time window constraint means that the pollutant propagation time calculated based on wind speed or water flow speed must fall within the preset allowable time difference between the time of emission event and the time of monitoring disturbance. Paths that exceed the range are considered to not meet the timing conditions. According to the cross-layer association rule set, the monitoring grid cells, meteorological grid cells and pollution source grid cells are matched. When the conditions of spatial accessibility, directional consistency and time window are met, the corresponding pollution source grid cell is registered as a candidate pollution source of the monitoring grid cell. The generated candidate pollution source set includes: Project the net disturbance sequence according to the monitoring grid, open the path record list in the three-level constraint grid data structure according to the hierarchical encoding of the monitoring grid cells, and establish path seeds for the monitoring grid cells where the disturbance start time occurs. Based on the cross-layer association rule set, a forward expansion is performed between the monitoring grid, meteorological grid and pollution source grid starting from each path seed, and a forward feasible path is generated according to spatial accessibility constraints, directional consistency constraints and time window constraints; For each pollution source grid cell, reverse verification is performed in reverse order from pollution source grid to monitoring grid. The forward feasible path is compared segment by segment using the time sequence and amplitude information of the perturbation operator chain. Path segments that do not meet the consistency of perturbation order or amplitude progression are deleted, and paths that simultaneously meet the requirements of forward expansion and reverse verification are retained, thus completing the bidirectional consistency pruning. Perform phase alignment and medium connectivity checks on the bidirectional clipped path: Only paths that are continuously connected within the transition or diffusion phase are retained; For paths that cross air, water, or soil interfaces, cross-layer connectivity is checked based on the monitoring cell medium type and the wind field, water flow, and rainfall data recorded by meteorological cells, and paths that do not conform to the rules are eliminated. The remaining paths are merged according to the pollution source grid cells to generate the minimum redundant path set. For each pollution source grid cell, the shortest evidence chain covering all disturbance monitoring grid cells is retained. The merged pollution source grid cells are registered as candidate pollution sources to form a candidate pollution source set. The obtained pollution contribution includes: Within the time slice set, according to the candidate pollution source set and the covered monitoring cell list, the observation data sequence with the same order is extracted from the net disturbance sequence, spliced ​​according to the rule of monitoring cell priority and time slice priority, and the phase label is segmented and weighted to form a phase-weighted observation data sequence. Based on the cross-layer association relationship of the three-layer constraint lattice, a lattice constraint mask is constructed. Combinations of pollution sources-monitoring lattice cells-time slices that are unreachable or disconnected are marked as invalid, while combinations that simultaneously satisfy spatial accessibility, directional consistency and time window constraints are marked as valid. For each candidate pollution source, a pollution effect prototype vector is generated that is consistent with the order of the observation data sequence. Amplitude normalization, time alignment and phase segmentation are performed. The phase weights are written into the data segments corresponding to the pollution effect prototype vector. The standardized pollution effect prototype matrix is ​​then assembled by column. A lattice-constrained mask structure is superimposed on the standardized pollution prototype matrix to obtain a mask constraint matrix. Column sparsity constraints and bounded condition number constraints are introduced. A numerical decomposition step with column pivot selection is used for rank determination and stability verification to generate a perturbation inversion matrix. The lattice-constrained mask structure is a binary mask constructed on the same dimension as the standardized pollution prototype matrix. The row index corresponds to the combination of each monitoring lattice cell and time slice in the observation data sequence, and the column index corresponds to the candidate pollution source. Row-column cells that do not meet the conditions of spatial reachability, directional consistency, or time window are assigned zero values, and cells that meet all constraints are assigned one value. If row-column cells span different media but are still connected, the weight bits corresponding to the media penalty coefficient are recorded in the mask, ultimately forming the constraint matrix. Among them, column sparsity constraint sets a non-zero upper limit for each candidate pollution source column in the mask constraint matrix. By reducing or zeroing weakly correlated column entries through threshold reduction, the number of effective columns is kept within the preset sparsity range, thereby highlighting the influence of the main pollution sources and suppressing the interference of redundant columns. The bounded condition number constraint is used to dynamically evaluate the condition number of the matrix after column sparsity processing, and the condition number is limited to not exceeding a preset stability threshold; when the condition number exceeds the limit, the column vectors are scaled, rearranged, or collinear columns are removed. The phase-weighted observation data sequence is input into the perturbation inversion matrix to obtain the pollution contribution sequence of candidate pollution sources. The contribution sequence is then checked according to the non-negative pruning rule and the phase continuity check rule. Components that fail the check are either set to zero or rolled back to a suboptimal solution. The pollution contribution sequences that pass the check, along with the candidate pollution source numbers, are output. The non-negative pruning rule is as follows: when a contribution value less than zero appears in the pollution contribution sequence, the negative value is directly set to zero; if the contribution value is close to zero and lower than the preset micro-threshold, it is considered to have no effective contribution and is also set to zero; after pruning, only the non-negative contribution components that are greater than the micro-threshold are retained.

2. The multi-parameter environmental monitoring and intelligent pollution source identification method according to claim 1, characterized in that, The preprocessing of multi-source environmental monitoring data includes time alignment, outlier removal, missing value completion, and normalization encoding.

3. The multi-parameter environmental monitoring and intelligent pollution source identification method according to claim 1, characterized in that, The output target pollution source identification results and corresponding contribution scores include: Summarize the pollution contribution values ​​and compose a complete record for each candidate pollution source, its pollution contribution value, the corresponding evidence chain identifier, and the associated time slice; Perform nonnegation and integrity checks on the summary records. Change records with pollution contribution values ​​less than zero to zero, fill in zeros for records with missing pollution contribution values, and keep the original order of the records unchanged. Sort the pollution contribution values ​​from high to low, set a pollution contribution threshold, delete candidate pollution source records below the threshold, and retain candidate pollution source records above or equal to the threshold to generate a list of main pollution sources and corresponding pollution contribution scores. The main pollution source list, pollution contribution score, evidence chain identifier, and associated time slices are integrated and output to form the final target pollution source identification result.

4. A multi-parameter environmental monitoring and intelligent pollution source identification system, comprising executing the multi-parameter environmental monitoring and intelligent pollution source identification method according to any one of claims 1 to 3, characterized in that, Includes the following modules: The data acquisition and preprocessing module is used to collect multi-source environmental monitoring data within the target area and perform preprocessing to form a multi-parameter environmental monitoring data sequence. The perturbation layered interpretation module is used to receive multi-parameter environmental monitoring data sequences, construct a perturbation operator chain, perform inverse reconstruction on the phase buffer, and output a net perturbation sequence. The constraint grid construction module is used to construct a three-layer constraint grid structure, mapping monitoring points to monitoring grids, mapping pollution source emission domains to pollution source grids, and mapping wind direction grids, water flow grids, and rainfall impact grids to meteorological grids. The path trimming module is used to project the net disturbance sequence onto the monitoring grid, trim the pollution propagation path, and generate a set of candidate pollution sources; The prototype and inversion module is used to construct a pollution effect prototype matrix based on the candidate pollution source set, obtain the pollution effect prototype vector of the candidate pollution sources, perform inversion solution, and obtain the pollution contribution. The source tracing and screening output module is used to sort candidate pollution sources and screen main pollution sources according to their pollution contribution, and output the target pollution source identification results and corresponding contribution scores.

Images