Precise transportation control method and system for remediation agent for in-situ remediation of underground water

Abstract

The invention provides an accurate transportation control method and system for a remediation agent for in-situ remediation of underground water, and relates to the technical field of underground water pollution remediation, and the method comprises the steps: collecting an image of a drilling rock core of a polluted site and sensor data, extracting features through a visual attention network and a time sequence attention network, and carrying out the feature extraction; the features are input into a conditional probability diffusion model and fused to generate hydrogeologic feature vectors, a high-precision hydrogeologic feature model is constructed in combination with space attention mapping and a generative diffusion model, the high-precision hydrogeologic feature model is input into a multi-region collaborative simulation environment for optimization calculation, and monitoring data are fused through edge calculation nodes. And training the lightweight control model to execute repair medicament transport control.

Description

Technical Field

[0001] The present invention relates to the technical field of groundwater pollution remediation, and particularly to a method and system for precise transport control of remediation agents for in-situ groundwater remediation. Background Art

[0002] As an important means of treating groundwater pollution, the core of in-situ groundwater remediation technology lies in the precise transport control of remediation agents. With the development of technology, the transport control methods for in-situ groundwater remediation have evolved from empirical control to intelligent control. In the early stage, the transport control mainly relied on manual experience, and the remediation agents were injected using single-point injection or simple grid layouts, often designing fixed injection parameter schemes based on static geological conditions;

[0003] With the progress of geological exploration and characterization technologies, means such as borehole geophysical prospecting and geophysical surveys have been gradually developed, and image recognition technology has been introduced to analyze the core structure characteristics. At the same time, multi-source sensors are used to obtain hydrogeological parameters, and numerical simulation methods are used to construct geological models. However, the existing technologies still have problems such as limited feature extraction accuracy of single data sources, difficulty in comprehensively depicting complex geological conditions, lack of multi-region cooperation mechanisms in the decision-making process, and poor adaptability of optimization strategies, making it difficult to cope with complex and variable geological conditions;

[0004] Therefore, there is an urgent need for a solution to solve the problems existing in the prior art. Summary of the Invention

[0005] Embodiments of the present invention provide a method and system for precise transport control of remediation agents for in-situ groundwater remediation, which can at least solve some of the problems existing in the prior art.

[0006] In the first aspect of the embodiments of the present invention, a method for precise transport control of remediation agents for in-situ groundwater remediation is provided, including:

[0007] Collect image data and sensor measurement data corresponding to the borehole cores of the polluted site, input the image data into a visual attention neural network for feature extraction to obtain soil structure feature data, input the sensor measurement data into a temporal attention neural network for analysis to obtain groundwater dynamic feature data, add the soil structure feature data and groundwater dynamic feature data to a conditional probability diffusion model for feature fusion, generate a hydrogeological feature vector and map it to a three-dimensional spatial grid through a spatial attention calculation module to obtain an initial hydrogeological feature model, and complete the missing areas in combination with a generative probability diffusion model to obtain a high-precision hydrogeological feature model;

[0008] The high-precision hydrogeological characteristic model is input into a multi-region collaborative simulation environment. Local optimization calculations are performed on different spatial regions through multiple decision-making units. The decision-making units share the optimization results through a communication network. The central evaluation network performs a global evaluation based on the shared results and collaboratively adjusts the optimization parameters of each decision-making unit. The adjusted transport strategy is verified by combining the Monte Carlo probability search method to output the optimal transport parameter solution.

[0009] Edge computing control nodes are deployed at the contaminated site, and the monitoring data of each area are fused and processed through a distributed model parameter aggregation algorithm to generate a global transport control model. This model is trained with the optimal transport parameter scheme to obtain a lightweight field control model. The lightweight field control model calculates the optimal transport control scheme based on the real-time monitoring data and sends it to the injection device for execution.

[0010] In an optional embodiment,

[0011] Image data and sensor measurement data corresponding to the borehole core of the contaminated site are collected. The image data is input into the visual attention neural network for feature extraction to obtain soil structure feature data. The sensor measurement data is input into the temporal attention neural network for analysis to obtain groundwater dynamic feature data. The soil structure feature data and groundwater dynamic feature data are added to the conditional probability diffusion model for feature fusion. The hydrogeological feature vector is generated and mapped to the three-dimensional spatial grid through the spatial attention calculation module to obtain the initial hydrogeological feature model. The missing areas are supplemented by the generative probability diffusion model to obtain a high-precision hydrogeological feature model, including:

[0012] Deploy a borehole sampling system at the contaminated site to collect core image data and groundwater sensor measurement data;

[0013] Inputting the core image data into an adaptive histogram equalization processing module to eliminate illumination interference to obtain a first processed image, performing local contrast sharpening on the first processed image to enhance structural details to obtain a second processed image, inputting the second processed image into a visual attention neural network, performing convolution operation on an encoder of the visual attention neural network to extract features, and reconstructing the soil structural feature data through a decoder;

[0014] Inputting the groundwater sensor measurement data into an outlier detection module to remove outliers to obtain first processed data, performing denoising on the first processed data through an adaptive sliding window to obtain second processed data, inputting the second processed data into a temporal attention neural network, and analyzing temporal associations through a bidirectional long short-term memory unit to obtain groundwater dynamic characteristic data;

[0015] Input the soil structure feature data and the groundwater dynamic feature data into the conditional probability diffusion model in a spliced manner, obtain the hydrogeological feature vector through dynamic noise reduction iteration, input the hydrogeological feature vector into the spatial attention calculation module, calculate the spatial weight according to the feature similarity and perform grid encryption, and obtain the initial hydrogeological feature model through feature interpolation;

[0016] Input the initial hydrogeological feature model into the generative probability diffusion model, perform feature generation on the missing area based on the known regional feature distribution, and obtain the high-precision hydrogeological feature model through evaluation and optimization by the multi-scale discriminator.

[0017] In an alternative embodiment,

[0018] Inputting the soil structure feature data and the groundwater dynamic feature data into the conditional probability diffusion model in a spliced manner and obtaining the hydrogeological feature vector through dynamic noise reduction iteration includes:

[0019] Perform feature splicing on the soil structure feature data and the groundwater dynamic feature data to obtain a mixed feature vector, input the mixed feature vector into the double-branch structure of the conditional probability diffusion model, obtain potential features through the main branch, obtain constraint conditions through the auxiliary branch, perform differential noise injection and noise reduction processing on the features, generate a fused feature by combining the constraint conditions and the attention mechanism, and reconstruct the hydrogeological feature vector after optimizing the parameters based on the feature quality evaluation, specifically including:

[0020] Perform feature splicing on the soil structure feature data and the groundwater dynamic feature data through a hierarchical cascading structure, splice the features of the same scale to obtain a same-scale feature vector, combine the same-scale feature vectors to form a mixed feature vector and input it into the conditional probability diffusion model with a double-branch parallel structure. The main branch of the conditional probability diffusion model maps the mixed feature vector to the latent space through a feature encoder to obtain potential features, and the auxiliary branch extracts physical law constraint information, spatial continuity constraint information, and temporal consistency constraint information through a constraint encoder to obtain a constraint condition vector;

[0021] Calculate the feature importance distribution based on the mixed feature vector, perform differential noise injection on different feature regions according to the feature importance distribution, inject strong noise into the feature regions with an importance lower than 0.3, inject weak noise into the feature regions with an importance higher than 0.75, and realize feature degradation through multi-level noise superposition to obtain a noise feature;

[0022] Input the noise characteristics into the multi-stage cascaded noise reduction unit. The multi-stage cascaded noise reduction unit extracts feature information through a densely connected feature extraction module, predicts the noise distribution through a noise prediction module, and achieves dynamic noise reduction in a skip connection manner to obtain noise reduction characteristics. Input the constraint condition vector into the constraint decoder to generate feature constraint conditions, and perform constraint adjustment on the noise reduction characteristics through a recursive feedback structure to obtain constraint features;

[0023] Construct a multi-head attention module to calculate the feature correlation degree of the constraint features, generate attention weights based on the feature correlation degree, enhance the constraint features through the attention weights, and introduce a residual connection to retain the original feature information to obtain fused features;

[0024] Construct a parameter evaluation module to calculate the feature quality index and the fusion effect index of the fused features, dynamically optimize the noise reduction parameters based on the feature quality index and the fusion effect index to obtain optimized parameters, input the optimized parameters into the feature decoder, and decode and reconstruct the fused features to obtain a hydrogeological feature vector.

[0025] In an alternative implementation,

[0026] Input the high-precision hydrogeological feature model into a multi-region collaborative simulation environment, perform local optimization calculations on different spatial regions through multiple decision-making units. The decision-making units share the optimization results through a communication network. The central evaluation network conducts global evaluation based on the shared results and collaboratively adjusts the optimization parameters of each decision-making unit. Combine the Monte Carlo probability search method to verify the adjusted transport strategy, and output the optimal transport parameter scheme, including:

[0027] Divide the high-precision hydrogeological feature model into multiple spatial regions according to the hydrogeological unit boundary and management partition, and configure decision-making units with a data layer, a calculation layer, and a communication layer in each spatial region;

[0028] The data layer of the decision-making unit extracts the geological structure features, hydrogeological parameter distribution, and boundary condition information of the spatial region, establishes a numerical model of the groundwater flow field, collects groundwater level, water quality index, and extraction volume data to construct a dynamic database. The calculation layer of the decision-making unit inputs the state data in the dynamic database into an optimizer, and performs local optimization calculations based on the optimization objectives of water resource balance, water quality safety, and system stability to obtain optimization results;

[0029] The decision-making units are interconnected through a communication network to construct a hierarchical caching mechanism to store the optimization results, and adopt an incremental update strategy to transmit the system state data to the central evaluation network. The central evaluation network extracts evaluation indicators of the resource regulation effect, environmental impact degree, and economic benefit level of the feature calculation, and generates a comprehensive score by highlighting the contribution of key indicators through an attention mechanism;

[0030] Based on the comprehensive score, expand the parameter search space for each decision-making unit and adjust the weights of the constraint conditions, introduce a coordination factor to balance the optimization objectives, transfer the adjustment results to the decision-making unit, generate a verification scenario using the Monte Carlo probability search method, identify typical failure modes through cluster analysis and optimize weak links to obtain verification results, screen the water volume distribution coefficient, water quality control threshold, and system regulation parameters based on the verification results, construct a dynamic adjustment rule for the transportation strategy, and output the optimal transportation parameter solution.

[0031] In an alternative implementation,

[0032] Transferring the adjustment results to the decision-making unit, generating a verification scenario using the Monte Carlo probability search method, identifying typical failure modes through cluster analysis and optimizing weak links to obtain verification results includes:

[0033] Construct a hierarchical adjustment strategy based on the evaluation results of the central evaluation network, generate adjustment results for decision-making units in different scoring intervals, transfer the adjustment results to the decision-making unit through a multi-level cache structure, extract features and identify weak links using the Monte Carlo probability search method, construct a protection mechanism for the weak links and generate evaluation results, and dynamically adjust the parameter configuration based on the evaluation results, specifically including:

[0034] Construct a hierarchical adjustment strategy based on the evaluation results of the central evaluation network, and perform differential processing on decision-making units in different scoring intervals through adjustment methods such as expanding the parameter search space and increasing the constraint weight, moderately expanding the search range and fine-tuning the constraint weight, and maintaining the existing parameter configuration to generate adjustment results;

[0035] Construct a multi-level cache structure to transfer the adjustment results to the decision-making unit, store the historical adjustment sequence through local caching, summarize the parameter information of adjacent units through regional cache nodes, transmit data using a differential update mechanism, and ensure data integrity through validity verification and timeout retransmission;

[0036] Use the Monte Carlo probability search method to extract the probability distribution characteristics of the adjustment results, obtain verification samples through a hierarchical quantization adaptive sampling strategy, orthogonalize the samples to eliminate parameter correlation and then generate a verification scenario, use cluster analysis to process the verification results of the verification scenario, determine the cluster center through dynamic neighborhood search to identify typical failure modes, construct a fault tree model to trace weak links, and establish a hierarchical early warning mechanism through system response characteristic analysis;

[0037] Adjust the buffer parameters for the weak links to construct a spatial protection circle, configure the regulation period to construct a time response mechanism, establish a collaborative optimization network to achieve regional linkage, obtain verification results by incrementally verifying the optimization effect, conduct multi-level evaluation on the verification results, determine the compliance rate of monitoring indicators through statistical analysis, evaluate the spatial distribution characteristics through the coefficient of dispersion, analyze the system stability through perturbation response, and generate evaluation results;

[0038] Based on the evaluation results, calculate the key indicators using a sliding time window, input the calculated key indicators into the hierarchical adjustment strategy, dynamically adjust the parameter search range and constraint weights of the decision-making unit, and output the optimal parameter configuration.

[0039] In an alternative implementation,

[0040] Deploy edge computing control nodes at the contaminated site, fuse the monitoring data of each region through a distributed model parameter aggregation algorithm, generate a global transport control model, and train it with the optimal transport parameter scheme to obtain a lightweight on-site control model. The lightweight on-site control model calculates the optimal transport control scheme based on real-time monitoring data and sends it to the injection device for execution, including:

[0041] Deploy edge computing control nodes at the contaminated site. The edge computing control nodes obtain water level monitoring data, water quality monitoring data, and flow monitoring data through a data acquisition module, and analyze the variation laws of the water level monitoring data, the water quality monitoring data, and the flow monitoring data using a time series feature extraction method to obtain the sampling period;

[0042] The edge computing control nodes identify outliers using an anomaly detection algorithm based on differential thresholds, repair missing data using local linear interpolation, and calculate time series features through a sliding window to obtain feature vectors;

[0043] Use a distributed model parameter aggregation algorithm to construct a transport parameter sub-model at the edge computing control node, construct a node affinity matrix based on the geographical location relationship, iteratively aggregate the transport parameter sub-models according to the node affinity matrix to form a regional model, and use an adaptive weight method to merge the regional models to generate a global transport control model;

[0044] Use a nested grid structure to encrypt the grid of the global transport control model, allocate the calculation tasks to the edge computing control nodes for parallel processing through a block iteration method, and solve them in a hierarchical progressive manner to obtain the optimal transport parameter scheme;

[0045] Extract the key features of the global transport control model through knowledge mapping, and construct a lightweight field control model using a progressive compression strategy. The lightweight field control model receives the water level monitoring data, the water quality monitoring data, and the flow monitoring data, determines the feasible solution space using a segmented optimization strategy, and searches for the optimal solution through a gradient search method and a heuristic algorithm to generate the optimal transport control plan;

[0046] Send the optimal transport control plan to the injection device for execution, adjust the injection parameters using a feedback correction mechanism, and identify potential faults and automatically switch control strategies through a state prediction algorithm.

[0047] In an alternative embodiment,

[0048] Use a distributed model parameter aggregation algorithm to construct a transport parameter sub-model at the edge computing control node, construct a node affinity matrix based on the geographical location relationship, and iteratively aggregate the transport parameter sub-models according to the node affinity matrix to form a regional model, including:

[0049] Use a distributed model parameter aggregation algorithm to construct a transport parameter sub-model at the edge computing control node. Generate a feature sequence by performing multi-scale analysis and filtering processing on the monitoring data. Calculate the spatial correlation coefficient, hydrogeological similarity coefficient, and migration flux coefficient between nodes based on the geographical location relationship to construct a node affinity matrix. Iteratively aggregate the transport parameter sub-models according to the node affinity matrix to form a regional model, specifically including:

[0050] Use a distributed model parameter aggregation algorithm to construct a transport parameter sub-model at the edge computing control node. The distributed model parameter aggregation algorithm performs multi-scale analysis on the water level monitoring data, water quality monitoring data, and flow monitoring data through wavelet decomposition, extracts high-frequency and low-frequency feature components, sets an adaptive threshold based on the energy distribution of the feature components to remove abnormal fluctuations, and performs smoothing processing on the signal using recursive filtering to generate a stable monitoring data sequence;

[0051] Extract spatio-temporal features from the stable monitoring data sequence, calculate the temporal statistical values through a sliding window, calculate the data distribution probability using kernel density estimation, determine the data mutation position based on change point detection, and construct a comprehensive feature vector by combining a multi-source data fusion algorithm;

[0052] Input the comprehensive feature vector into a gradient descent optimizer, calculate the model parameter gradient, calculate the parameter importance score using a cross-validation method, perform iterative optimization on the high-score parameters, perform linear regression on the low-score parameters, construct the transport parameter sub-model, and generate an initial parameter field;

[0053] Construct a node affinity matrix based on the geographical location relationship, calculate the Euclidean distance between nodes to obtain the spatial correlation coefficient, calculate the hydrogeological similarity coefficient using the hierarchical clustering method, calculate the migration flux coefficient based on the pollutant concentration gradient, and perform normalized weighted calculation on the spatial correlation coefficient, the hydrogeological similarity coefficient, and the migration flux coefficient to obtain the affinity weight;

[0054] Iteratively aggregate the transport parameter sub-model according to the node affinity matrix to form a regional model, preferentially fuse nodes with high affinity, perform weighted averaging on the initial parameter field using the affinity weight, calculate the fusion residual and adjust the affinity weight, and update the node affinity matrix;

[0055] Divide the regional model into multiple computational subtasks, allocate them to the edge computing control nodes, establish a message queue between nodes to transmit data, set checkpoints according to the fusion residual, and allocate node loads according to the computational amount of the initial parameter field.

[0056] In the second aspect of the embodiments of the present invention, a precise transport control system for a remediation agent for in-situ groundwater remediation is provided, including:

[0057] A first unit for collecting image data and sensor measurement data corresponding to the borehole cores of the contaminated site, inputting the image data into a visual attention neural network for feature extraction to obtain soil structure feature data, inputting the sensor measurement data into a temporal attention neural network for analysis to obtain groundwater dynamic feature data, adding the soil structure feature data and the groundwater dynamic feature data to a conditional probability diffusion model for feature fusion, generating a hydrogeological feature vector and mapping it to a three-dimensional spatial grid through a spatial attention calculation module to obtain an initial hydrogeological feature model, and complementing the missing area by combining a generative probability diffusion model to obtain a high-precision hydrogeological feature model;

[0058] A second unit for inputting the high-precision hydrogeological feature model into a multi-region collaborative simulation environment, performing local optimization calculations on different spatial regions through multiple decision-making units, sharing the optimization results by the decision-making units through a communication network, globally evaluating by a central evaluation network based on the shared results and coordinately adjusting the optimization parameters of each decision-making unit, and verifying the adjusted transport strategy by combining the Monte Carlo probability search method to output an optimal transport parameter scheme;

[0059] A third unit for arranging edge computing control nodes at the contaminated site, fusing and processing the monitoring data of each region through a distributed model parameter aggregation algorithm, generating a global transport control model and training it with the optimal transport parameter scheme to obtain a lightweight on-site control model, and calculating an optimal transport control scheme based on real-time monitoring data by the lightweight on-site control model and sending it to the injection device for execution.

[0060] In the present invention, by innovatively combining a visual attention network and a temporal attention network, the precise extraction of soil structure and groundwater dynamic characteristics is achieved. By adopting a multi-region collaborative simulation environment and a distributed decision-making unit architecture, the adaptability of the optimization strategy to site heterogeneity is significantly improved. Through a distributed model parameter aggregation algorithm, the efficient fusion of multi-source monitoring data is realized. Based on a lightweight on-site control model's real-time optimization mechanism, the control scheme can be quickly adjusted according to the site state. Through a multi-level control architecture design, the stability of the system operation is significantly enhanced. The dual attention mechanism at the feature extraction level provides a reliable data basis, and the multi-region collaborative mechanism at the optimization level ensures the balance of the optimization strategy, providing an efficient and reliable precise transport control method for the in-situ groundwater remediation field, which has important engineering application value and promotion significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] Figure 1 is a schematic flow chart of the precise transport control method of the remediation agent for in-situ groundwater remediation in an embodiment of the present invention;

[0062] Figure 2 is a comparison diagram of the feature fusion effect of the precise transport control method of the remediation agent for in-situ groundwater remediation in an embodiment of the present invention;

[0063] Figure 3 is a comparison diagram of the system optimization effect of the precise transport control method of the remediation agent for in-situ groundwater remediation in an embodiment of the present invention;

[0064] Figure 4 is a comparison diagram of the data quality improvement effect of the precise transport control method of the remediation agent for in-situ groundwater remediation in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0065] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0066] The technical solutions of the present invention will be described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments.

[0067] Figure 1 is a schematic flow chart of the precise transport control method of the remediation agent for in-situ groundwater remediation in an embodiment of the present invention, as Figure 1As shown, the method includes:

[0068] Collect image data and sensor measurement data corresponding to the borehole cores of the contaminated site, input the image data into the visual attention neural network for feature extraction to obtain soil structure feature data, input the sensor measurement data into the temporal attention neural network for analysis to obtain groundwater dynamic feature data, add the soil structure feature data and the groundwater dynamic feature data to the conditional probability diffusion model for feature fusion, generate a hydrogeological feature vector and map it to a three-dimensional spatial grid through the spatial attention calculation module to obtain an initial hydrogeological feature model, and combine the generative probability diffusion model to complete the missing area to obtain a high-precision hydrogeological feature model;

[0069] In an alternative embodiment,

[0070] Collecting image data and sensor measurement data corresponding to the borehole cores of the contaminated site, inputting the image data into the visual attention neural network for feature extraction to obtain soil structure feature data, inputting the sensor measurement data into the temporal attention neural network for analysis to obtain groundwater dynamic feature data, adding the soil structure feature data and the groundwater dynamic feature data to the conditional probability diffusion model for feature fusion, generating a hydrogeological feature vector and mapping it to a three-dimensional spatial grid through the spatial attention calculation module to obtain an initial hydrogeological feature model, and combining the generative probability diffusion model to complete the missing area to obtain a high-precision hydrogeological feature model includes:

[0071] Deploy a borehole sampling system at the contaminated site to collect core image data and groundwater sensor measurement data;

[0072] Input the core image data into the adaptive histogram equalization processing module to eliminate light interference to obtain a first processed image, enhance the structural details of the first processed image through local contrast sharpening processing to obtain a second processed image, input the second processed image into the visual attention neural network, and execute convolutional operations through the encoder of the visual attention neural network to extract features and reconstruct through the decoder to obtain soil structure feature data;

[0073] Input the groundwater sensor measurement data into the outlier detection module to remove outliers to obtain first processed data, perform noise reduction on the first processed data through an adaptive sliding window to obtain second processed data, input the second processed data into the temporal attention neural network, and analyze the temporal correlation through a bidirectional long short-term memory unit to obtain groundwater dynamic feature data;

[0074] Input the soil structure characteristic data and the groundwater dynamic characteristic data into the conditional probability diffusion model, and obtain the hydrogeological feature vector through dynamic noise reduction iteration. Input the hydrogeological feature vector into the spatial attention calculation module, calculate the spatial weight according to the feature similarity and perform grid encryption, and obtain the initial hydrogeological feature model through feature interpolation;

[0075] Input the initial hydrogeological feature model into the generative probability diffusion model, perform feature generation on the missing area based on the known regional feature distribution, and evaluate and optimize through a multi-scale discriminator to obtain a high-precision hydrogeological feature model.

[0076] First, deploy a borehole sampling system inside the contaminated site. This sampling system includes a core sampling device and a groundwater sensor system. The core sampling device is equipped with high-resolution imaging equipment and can continuously collect core image data; the groundwater sensor system integrates a variety of parameter sensors and can monitor key parameter data such as groundwater level, temperature, and conductivity in real time.

[0077] For the collected core image data, input it into the adaptive histogram equalization processing module for preprocessing. This module calculates the cumulative distribution function of the local area of the image, adaptively adjusts the pixel distribution, effectively eliminates the interference caused by uneven illumination, and obtains the first processed image with uniform brightness distribution. Perform local contrast sharpening on the first processed image, calculate the gradient information of the local area of the image, enhance structural details such as edges and textures, and generate the second processed image with clearer details. Input the processed second processed image into the pre-trained visual attention neural network. The encoder of the network extracts image features step by step through multi-layer convolution operations, and the attention mechanism highlights the key area features; the decoder reconstructs through deconvolution operations to obtain the soil structure characteristic data, including key parameters such as porosity and permeability.

[0078] For the measurement data collected by the sensor, input it into the outlier detection module for preprocessing. Identify and remove data points that are significantly deviated from the normal range based on statistical methods to generate the first processed data. Apply the adaptive sliding window algorithm to the first processed data for noise reduction processing, and the window size will be dynamically adjusted according to the data fluctuation degree to obtain the smooth second processed data. Input the processed second processed data into the temporal attention neural network. The bidirectional long short-term memory unit in the network analyzes the temporal correlation of the data in the forward and backward directions, combines the attention mechanism to capture the key temporal features, and finally obtains the data characterizing the dynamic change characteristics of groundwater.

[0079] After obtaining the soil structure characteristic data and the groundwater dynamic characteristic data, the two types of data are spliced according to a preset format and used as the input of the conditional probability diffusion model. Through a multi-step noise reduction iteration process, the conditional probability diffusion model gradually extracts and fuses the correlation information of the two types of characteristics to generate a comprehensive hydrogeological characteristic vector. The obtained characteristic vector is input into the spatial attention calculation module to calculate the similarity of the characteristics at different positions in the spatial domain and generate a spatial weight matrix. Based on the generated weight matrix, the grid encryption process is guided, the grid is subdivided in the key areas, and the grid node values are filled by feature interpolation to construct an initial hydrogeological characteristic model.

[0080] The initial hydrogeological characteristic model is input into the generative probability diffusion model to analyze the characteristic distribution law of the known area. Based on the analyzed law, the missing areas in the model are generated through an iterative diffusion process. During the generation process, a multi-scale discriminator is used to evaluate the rationality of the generated characteristics from different scales and return an optimization signal. By repeatedly executing the generation-evaluation-optimization process, a complete high-precision hydrogeological characteristic model is obtained.

[0081] Exemplarily, taking a contaminated site remediation project as an example, 25 borehole sampling points are arranged within a range of 100m×100m. An industrial camera with 12 million pixels is used in each borehole, and a core image is collected every 10cm, with an image resolution of 4000×3000 pixels. The image brightness range is adjusted from the original [50, 180] to [30, 220] through adaptive histogram equalization, and the local contrast is increased by 30%. The soil structure characteristics extracted by the visual attention network include: average porosity 0.35, permeability coefficient 1.2×10⁻ 5 m / s, etc. In terms of sensor data collection, water level, temperature, and conductivity data are collected once an hour. Outlier detection eliminates data points with a deviation exceeding 3 times the standard deviation (about 0.5%). Adaptive windows of 15 data points are used for noise reduction, and the temporal attention network analyzes dynamic characteristics such as groundwater flow velocity of 0.5m / d and flow direction of 45°. In the feature fusion and modeling stage, the extracted features are organized into 256-dimensional vectors, and fused features are obtained through 1000-step noise reduction iteration. Spatial attention calculation shows that the feature similarity within adjacent 20m ranges exceeds 0.8. Accordingly, the original 1m grid is encrypted to 0.2m in the high-similarity areas. The generative model completes the filling of a total of about 20% of the missing areas in 5 places, and the multi-scale discriminator evaluates the generation quality from three scales of 0.2m, 1m, and 5m, with an average accuracy rate reaching 92%. The finally obtained hydrogeological characteristic model contains 500×500×50 grid nodes, and each node contains 25 characteristic parameters, which can be used to guide the optimization of subsequent remediation plans.

[0082] In this embodiment, by constructing a technical solution that combines multi-source data processing and deep learning, the problem of inaccurate characterization of hydrogeological features in the in-situ groundwater remediation process is effectively solved. In the prior art, in-situ groundwater remediation usually uses a single data source for geological feature analysis. For example, only relying on borehole core images for soil structure analysis, or only studying the dynamic characteristics of groundwater based on sensor data, it is difficult to comprehensively reflect the hydrogeological features of the site. In terms of data processing, traditional methods mostly use image enhancement algorithms with fixed parameters and simple data filtering methods, which cannot adapt to the differences in data quality under different site conditions. Feature extraction mainly relies on manually designed feature extraction rules, and the extracted features are often one-sided and highly subjective. To address these problems, in this embodiment, a core image acquisition and groundwater parameter monitoring system are simultaneously deployed at the data acquisition end to obtain multi-source data. In the data processing link, adaptive histogram equalization and local contrast enhancement technologies are introduced to achieve dynamic optimization of image quality; outlier detection and adaptive sliding window algorithms are used to improve the reliability of sensor data. The visual attention neural network and temporal attention neural network are innovatively applied to extract features from the spatial and temporal dimensions respectively, overcoming the limitations of traditional feature extraction methods. The deep fusion of multi-source features is achieved through the conditional probability diffusion model, and the generative probability diffusion model is used to intelligently complete the missing areas, significantly improving the integrity and accuracy of the model.

[0083] In an alternative embodiment,

[0084] The soil structure feature data and the groundwater dynamic feature data are spliced and input into the conditional probability diffusion model, and the hydrogeological feature vector obtained through dynamic noise reduction iteration includes:

[0085] The soil structure feature data and the groundwater dynamic feature data are feature-spliced to obtain a mixed feature vector. The mixed feature vector is input into the dual-branch structure of the conditional probability diffusion model. The potential feature is obtained through the main branch, and the constraint condition is obtained through the auxiliary branch. Differential noise injection and noise reduction processing are performed on the features, and the fusion feature is generated by combining the constraint condition and the attention mechanism. After optimizing the parameters based on the feature quality assessment, the hydrogeological feature vector is reconstructed, specifically including:

[0086] The soil structure feature data and the groundwater dynamic feature data are feature-spliced through a hierarchical cascading structure. The same-scale feature vectors are obtained by splicing the features of the same scale, and the same-scale feature vectors are combined to form a mixed feature vector and input into the conditional probability diffusion model with a dual-branch parallel structure. The main branch of the conditional probability diffusion model maps the mixed feature vector to the latent space through a feature encoder to obtain the latent feature, and the auxiliary branch extracts the physical law constraint information, spatial continuity constraint information, and temporal consistency constraint information through a constraint encoder to obtain the constraint condition vector;

[0087] Calculate the feature importance distribution based on the mixed feature vector, perform differential noise injection on different feature regions according to the feature importance distribution, inject strong noise into the feature regions with importance lower than 0.3, inject weak noise into the feature regions with importance higher than 0.75, and realize feature degradation through multi-level noise superposition to obtain noise features;

[0088] Input the noise features into a multi-stage cascaded noise reduction unit. The multi-stage cascaded noise reduction unit extracts feature information through densely connected feature extraction modules, predicts the noise distribution through a noise prediction module, and realizes dynamic noise reduction in a skip connection manner to obtain noise-reduced features. Input the constraint condition vector into a constraint decoder to generate feature constraint conditions, and perform constraint adjustment on the noise-reduced features through a recursive feedback structure to obtain constraint features;

[0089] Construct a multi-head attention module to calculate the feature correlation degree of the constraint features, generate attention weights based on the feature correlation degree, enhance the constraint features through the attention weights, and introduce residual connections to retain the original feature information to obtain fused features;

[0090] Construct a parameter evaluation module to calculate the feature quality index and the fusion effect index of the fused features, dynamically optimize the noise reduction parameters based on the feature quality index and the fusion effect index to obtain optimized parameters, input the optimized parameters into a feature decoder, and decode and reconstruct the fused features to obtain a hydrogeological feature vector.

[0091] Perform feature splicing preprocessing. Organize the obtained soil structure feature data and groundwater dynamic feature data through a hierarchical cascaded structure, splice the features with the same resolution and scale in the two types of data to obtain multiple same-scale feature vectors. Subsequently, combine these same-scale feature vectors according to a preset rule to form a mixed feature vector containing multi-scale information. This mixed feature vector is input into the designed conditional probability diffusion model with a double-branch parallel structure, where the main branch contains a feature encoder that maps the mixed feature vector to the latent space through a multi-layer neural network to obtain a latent feature representation; the auxiliary branch contains a constraint encoder that extracts physical law constraint information (such as the relationship between permeability and porosity), spatial continuity constraint information (such as the spatial gradient of features), and temporal consistency constraint information (such as the law of groundwater dynamic change), and finally obtains a constraint condition vector.

[0092] Differentiated noise processing of features. Calculate the importance distribution of each feature region based on the mixed feature vector, and adopt a gradient-based feature importance evaluation method. According to the calculated feature importance distribution, adopt a differentiated noise injection strategy for feature regions of different importance: inject Gaussian noise with a relatively large intensity into feature regions with an importance score lower than 0.3, inject only slight noise into feature regions with an importance score higher than 0.75, and linearly adjust the noise intensity according to the importance for feature regions between the two. By superimposing noises of multiple different scales, achieve multi-level degradation of features to obtain noise features.

[0093] Input the noise features into the designed multi-stage cascaded noise reduction unit for processing. This unit first extracts feature information through a densely connected feature extraction module using a multi-layer convolutional network, and at the same time predicts the noise distribution of each region through a dedicated noise prediction module. On this basis, adopt a skip connection structure to fuse feature information at different levels to achieve a dynamic noise reduction process and obtain preliminary noise reduction features. Subsequently, input the aforementioned constraint condition vector into the constraint decoder to generate specific feature constraint conditions. Through the designed recursive feedback structure, perform multiple rounds of iterative adjustment on the constraint conditions and the noise reduction features to ensure that the features meet the physical constraints, and finally obtain the constrained features.

[0094] Construct a multi-head attention module to calculate the correlation degree between the constrained features. This module divides the features into multiple attention heads, and each head independently calculates the correlation relationship between the features to generate corresponding attention weights. Use the attention weights to selectively enhance the constrained features, and at the same time retain the original feature information through residual connections, and finally obtain a fully fused feature representation.

[0095] Construct a parameter evaluation module to quantitatively evaluate the feature fusion effect through the designed evaluation metrics. This module calculates feature quality metrics (such as signal-to-noise ratio, clarity, etc.) and fusion effect metrics (such as feature consistency, spatial continuity, etc.). Based on these evaluation metrics, adopt the gradient descent method to dynamically optimize the key parameters of the noise reduction process to obtain an optimized parameter set. Input the optimized parameters into the feature decoder to decode and reconstruct the fused features, and finally obtain a high-quality hydrogeological feature vector.

[0096] Exemplarily, take the feature fusion process of a certain polluted site as an example. The initial data includes soil structure features with a dimension of 32×32×64 and groundwater dynamic features with a dimension of 32×32×32. The features with the same spatial resolution (32×32) are spliced through a hierarchical cascade structure to obtain a 96-dimensional feature vector of the same scale. Feature importance assessment shows that approximately 20% of the feature regions have an importance lower than 0.3, mainly distributed in the boundary regions, and Gaussian noise with a standard deviation of 0.5 is injected into these regions; approximately 30% of the feature regions have an importance higher than 0.75, concentrated in the core polluted areas, and only weak noise with a standard deviation of 0.1 is injected.

[0097] During the noise reduction process, a 4-stage cascade noise reduction unit is adopted. Each stage contains 3 densely connected convolutional layers (convolution kernel size is 3×3), and features of different scales are fused through skip connections. The physical constraints extracted by the constrained encoder include: the power-law relationship between permeability and porosity (R²>0.9), the continuity constraint of the spatial gradient (gradient change <5%), and the consistency constraint of the temporal change (change between adjacent moments <10%).

[0098] The multi-head attention module uses 8 attention heads, and each head independently calculates the correlation matrix of 32×32 spatial positions. Through the attention mechanism, the weights of approximately 90% of the feature positions are enhanced to varying degrees, with an average enhancement amplitude of 35%. The final parameter evaluation shows that: the signal-to-noise ratio in the feature quality index is increased by 40%, and the clarity is increased by 25%; the feature consistency in the fusion effect index reaches 0.85, and the spatial continuity error is less than 3%. After being reconstructed by the feature decoder, a hydrogeological feature vector with a dimension of 32×32×128 is obtained. This feature vector not only retains the key information of the original features but also satisfies the physical constraint conditions.

[0099] In this embodiment, through innovative model structure design and processing strategies, the deficiencies of traditional methods in feature fusion, noise processing, constraint introduction, etc. are effectively solved, providing more reliable technical support for the hydrogeological feature characterization in the in-situ groundwater remediation process, promoting the progress of technology in this field. In the feature fusion process of traditional hydrogeological feature extraction methods, simple feature splicing or weighted average methods are usually adopted, which are difficult to effectively handle the fusion problem of multi-source heterogeneous data; in the feature optimization stage, unified denoising parameters and processing strategies are often used, without considering the importance differences of different feature regions; in terms of constraint introduction, simple penalty terms or regularization terms are mostly used, which are difficult to make full use of known physical laws and spatio-temporal constraint relationships; in the feature enhancement link, common feature selection methods lack in-depth analysis of the correlation between features, and it is easy to cause the loss of useful information. These problems lead to insufficient accuracy of the finally obtained hydrogeological feature model, making it difficult to meet the refined requirements of in-situ groundwater remediation;

[0100] This embodiment proposes a feature fusion method based on a conditional probability diffusion model, which uses a hierarchical cascade structure for feature organization. Through the precise alignment and combination of same-scale features, it lays a foundation for subsequent deep fusion. In terms of the model architecture, a dual-branch parallel structure is innovatively designed. The main branch is responsible for feature extraction and mapping, while the auxiliary branch focuses on the extraction of multi-dimensional constraint information, realizing the organic combination of feature extraction and constraint introduction. A differential noise injection strategy based on importance is introduced. By accurately calculating the feature importance distribution, differential noise processing is performed on different regions, effectively protecting the key feature information. At the same time, a multi-cascade noise reduction unit is designed. Through the structural design of dense connection and skip connection, multi-scale extraction and dynamic noise reduction of features are realized, significantly improving the quality of features.

[0101] Input the high-precision hydrogeological feature model into a multi-region collaborative simulation environment. Perform local optimization calculations on different spatial regions through multiple decision-making units. The decision-making units share the optimization results through a communication network. The central evaluation network conducts global evaluation based on the shared results and collaboratively adjusts the optimization parameters of each decision-making unit. Combine the Monte Carlo probability search method to verify the adjusted transport strategy and output the optimal transport parameter scheme.

[0102] In an alternative implementation manner,

[0103] Input the high-precision hydrogeological feature model into a multi-region collaborative simulation environment. Perform local optimization calculations on different spatial regions through multiple decision-making units. The decision-making units share the optimization results through a communication network. The central evaluation network conducts global evaluation based on the shared results and collaboratively adjusts the optimization parameters of each decision-making unit. Combine the Monte Carlo probability search method to verify the adjusted transport strategy and output the optimal transport parameter scheme, including:

[0104] Divide the high-precision hydrogeological feature model into multiple spatial regions according to the hydrogeological unit boundary and management partition, and configure decision-making units with a data layer, a calculation layer, and a communication layer in each spatial region.

[0105] The data layer of the decision-making unit extracts the geological structure features, hydrogeological parameter distribution, and boundary condition information of the spatial region, establishes a numerical model of the groundwater flow field, and collects groundwater level, water quality index, and extraction volume data to construct a dynamic database. The calculation layer of the decision-making unit inputs the state data in the dynamic database into an optimizer, and performs local optimization calculations based on the optimization objectives of water resource balance, water quality safety, and system stability to obtain the optimization results.

[0106] The decision-making units are interconnected through a communication network to construct a hierarchical caching mechanism for storing the optimization results, and an incremental update strategy is adopted to transmit the system state data to the central evaluation network. The central evaluation network extracts features to calculate the evaluation indicators of the resource regulation effect, environmental impact degree, and economic benefit level, and generates a comprehensive score by highlighting the contributions of key indicators through an attention mechanism;

[0107] Based on the comprehensive score, the parameter search space of each decision-making unit is expanded and the constraint condition weights are adjusted. A coordination factor is introduced to balance the optimization objectives, and the adjustment results are transmitted to the decision-making units. The Monte Carlo probability search method is used to generate verification scenarios, and typical failure modes are identified through cluster analysis and weak links are optimized to obtain verification results. Based on the verification results, the water volume distribution coefficient, water quality control threshold, and system regulation parameters are screened, and the dynamic adjustment rules of the transport strategy are constructed to output the optimal transport parameter scheme.

[0108] The high-precision hydrogeological feature model is regionally divided. According to the boundary characteristics of the hydrogeological units of the site and the requirements of the management partition, the entire site is divided into multiple relatively independent spatial regions. In each divided spatial region, decision-making units with a three-layer architecture are configured, including a data layer, a calculation layer, and a communication layer, forming a distributed decision-making network.

[0109] In the data layer of the decision-making unit, the geological structure feature data of the spatial region are systematically extracted, including stratigraphic distribution, fault structures, etc.; at the same time, the distribution information of hydrogeological parameters, such as permeability coefficient, storage coefficient, etc., is obtained; and the boundary condition information of the region is determined, including head boundary, flow boundary, etc. Based on these basic data, a numerical model of the groundwater flow field in the region is established. At the same time, dynamic data such as the groundwater level, water quality indicators (such as pH value, conductivity, pollutant concentration, etc.), and groundwater extraction volume in the region are continuously collected to construct a dynamically updated dynamic database.

[0110] In the calculation layer, the decision-making unit inputs the state data in the dynamic database into the optimizer for processing. The optimizer performs calculations based on three core optimization objectives: the water resource balance objective ensures the regional water volume balance, the water quality safety objective guarantees the water quality compliance, and the system stability objective maintains the dynamic balance of the groundwater system. Through the multi-objective optimization algorithm, the local optimization results that meet the current regional characteristics are obtained.

[0111] Each decision-making unit is interconnected through a communication network, and a hierarchical caching mechanism is constructed to store the optimization results. An incremental update strategy is adopted to regularly transmit the system state data to the central evaluation network. The central evaluation network is responsible for extracting features and calculating multiple evaluation indicators: the resource regulation effect indicator reflects the water resource utilization efficiency, the environmental impact degree indicator evaluates the ecological environment impact, and the economic benefit level indicator measures the economic rationality. These indicators are weighted through an attention mechanism to highlight the contributions of key indicators, and finally a comprehensive score is generated.

[0112] Based on the comprehensive score results, the system adjusts the optimization process of each decision-making unit: expands the parameter search space to find a better solution, adjusts the constraint condition weights to balance multiple objectives, and introduces a coordination factor to ensure the coordination of regional decisions. These adjustment results are transmitted back to each decision-making unit, and then a Monte Carlo probability search method is used to generate multiple verification scenarios. Typical failure modes are identified through cluster analysis, and the weak links of the system are optimized accordingly to obtain the verification results. Finally, based on the verification results, key control parameters are determined, including the water volume distribution coefficient, water quality control threshold, and system regulation parameters, and a dynamic adjustment rule for the transportation strategy is constructed, and finally the optimal transportation parameter scheme is output.

[0113] Exemplarily, taking a contaminated site remediation project as an example, the site area is 1 square kilometer, and it is divided into 5 spatial regions according to the hydrogeological conditions, and each region is configured with a decision-making unit.

[0114] At the data layer, taking Decision-making Unit 1 as an example: the extracted geological structure features show that the area is mainly a gravel aquifer with an average thickness of 15 meters; the hydrogeological parameters include a permeability coefficient of 1.2×10⁻ 4 m / s and a storage coefficient of 0.15; the boundary conditions are a constant head boundary on the west side (head 20 meters) and a constant flow boundary on the east side (flow 100 m³ / d). The grid size of the established groundwater flow field model is 5 meters × 5 meters, and the time step is 1 day. The dynamic database is updated every hour and includes the water level data of 10 monitoring wells, the water quality data (COD, ammonia nitrogen, etc.) of 5 sampling points, and the pumping volume data of 3 pumping wells.

[0115] At the calculation layer, the objective function weights set by the optimizer are: water resource balance 0.4, water quality safety 0.4, and system stability 0.2. The local optimization results show that the optimal pumping volume in this area is 150 m³ / d and the injection volume is 120 m³ / d.

[0116] The communication layer adopts a three - level caching mechanism: the local cache stores data for the past 24 hours, the regional cache stores data for the past 7 days, and the central cache stores data for the past 30 days. The status data transmits incremental updates to the central evaluation network every 30 minutes. The metric values calculated by the central evaluation network are: the resource regulation effect is 0.85 (full score 1.0), the environmental impact degree is 0.15 (full score 0.3), and the economic benefit level is 0.75 (full score 1.0). The comprehensive score calculated through the attention mechanism is 0.82.

[0117] Based on this score, the system expands the parameter search range of decision - making unit 1 by 20%, and adjusts the constraint condition weights to: water quality constraint 0.5, water volume constraint 0.3, and cost constraint 0.2. The Monte Carlo method is used to generate 1000 verification scenarios, and cluster analysis identifies 3 typical failure modes: water level over - limit, pollutant concentration exceeding the standard, and system response delay. In response to these problems, the key parameters determined after optimization and adjustment include: water volume distribution coefficient 0.8, COD control threshold 40mg / L, and system response time threshold 4 hours. The final output transport parameter scheme includes: injection well layout spacing 30 meters, injection pressure 0.2MPa, injection cycle 6 hours, etc.

[0118] In this embodiment, the distributed decision - making architecture significantly improves the data processing speed and optimization calculation efficiency of the system, significantly reduces the response delay, realizes a rapid response to changes in site conditions. The multi - objective collaborative optimization framework makes water resource regulation more balanced, significantly improves the water quality compliance rate, makes the system operation more stable, and significantly improves the overall remediation efficiency. The probability verification method significantly improves the adaptability and robustness of the optimization scheme, significantly reduces the system operation failure rate, and greatly reduces the maintenance cost. Existing in - situ groundwater remediation transport control technologies usually adopt a centralized architecture in decision - making system design, where all regional data and calculation tasks are concentrated in a single control center for processing, resulting in slow system response speed and large calculation load; in terms of optimization goal setting, they often overly focus on the optimization of a single indicator, such as only considering water volume balance or water quality compliance, lacking a systematic consideration of multi - objective collaborative optimization; in the decision - making verification link, they usually adopt deterministic verification methods, making it difficult to cope with uncertain factors in the actual site; in terms of parameter adjustment, they mostly adopt fixed regulation rules, lacking dynamic adaptability and being difficult to respond to changes in site conditions in a timely manner. These problems seriously restrict the effect of in - situ groundwater remediation.

[0119] This embodiment adopts a distributed decision-making architecture, divides the site into multiple regions according to the characteristics of hydrogeological units, configures an independent three-layer architecture decision-making unit for each region, significantly improves the parallel processing ability and response speed of the system, reduces the computational load of a single decision-making unit at the same time, and constructs a multi-level data acquisition and processing system. Information such as geological structure, hydrogeological parameters, and boundary conditions is collected through the data layer system of the decision-making unit, a refined groundwater flow field model is established, and real-time data update is realized through a dynamic database. This design provides comprehensive and reliable data support for subsequent optimization decisions and designs a multi-objective collaborative optimization framework. By simultaneously considering three core objectives of water resource balance, water quality safety, and system stability at the calculation layer, the overall optimization of the repair process is achieved.

[0120] In an alternative embodiment,

[0121] The adjustment result is transmitted to the decision-making unit, the Monte Carlo probability search method is used to generate a verification scenario, and typical failure modes are identified through cluster analysis and weak links are optimized to obtain verification results including:

[0122] Based on the evaluation results of the central evaluation network, a hierarchical adjustment strategy is constructed, adjustment results are generated for decision-making units in different scoring intervals, the adjustment results are transmitted to the decision-making unit through a multi-level cache structure, the Monte Carlo probability search method is used to extract features and identify weak links, a protection mechanism is constructed for the weak links and evaluation results are generated, and the parameter configuration is dynamically adjusted based on the evaluation results, specifically including:

[0123] Based on the evaluation results of the central evaluation network, a hierarchical adjustment strategy is constructed, and different adjustment methods such as expanding the parameter search space and increasing the constraint weight, moderately expanding the search range and fine-tuning the constraint weight, and maintaining the existing parameter configuration are used to differentially process decision-making units in different scoring intervals and generate adjustment results;

[0124] A multi-level cache structure is constructed to transmit the adjustment results to the decision-making unit, the historical adjustment sequence is stored through local caching, the parameter information of adjacent units is summarized through regional cache nodes, a differential update mechanism is used to transmit data, and data integrity is ensured through validity verification and timeout retransmission;

[0125] The Monte Carlo probability search method is used to extract the probability distribution characteristics of the adjustment results, verification samples are obtained through a hierarchical quantization adaptive sampling strategy, the samples are orthogonalized to eliminate parameter correlation and then a verification scenario is generated, the verification results of the verification scenario are processed through cluster analysis, the cluster center is determined through dynamic neighborhood search to identify typical failure modes, a fault tree model is constructed to trace weak links, and a hierarchical early warning mechanism is established through system response characteristic analysis;

[0126] Adjust the buffer parameters for the weak links to construct a spatial protection circle, configure the regulation period to construct a time response mechanism, establish a collaborative optimization network to achieve regional linkage, evaluate the optimization effect through incremental verification to obtain verification results, conduct multi-level evaluation on the verification results, determine the compliance rate of monitoring indicators through statistical analysis, evaluate the spatial distribution characteristics through the coefficient of dispersion, analyze the system stability through perturbation response, and generate evaluation results;

[0127] Based on the evaluation results, use a sliding time window to calculate key indicators, input the calculated key indicators into the hierarchical adjustment strategy, dynamically adjust the parameter search range and constraint weights of the decision-making unit, and output the optimal parameter configuration.

[0128] Construct a hierarchical adjustment strategy according to the evaluation results of the central evaluation network, and adopt different adjustment methods for decision-making units in different scoring intervals: for decision-making units with lower scores, adopt the method of expanding the parameter search space and increasing the constraint weight for significant adjustment; for decision-making units with medium scores, adopt the method of moderately expanding the search range and slightly adjusting the constraint weight for gentle adjustment; for decision-making units with higher scores, maintain the current parameter configuration. Through this hierarchical adjustment strategy, generate targeted adjustment results.

[0129] To ensure the efficient transmission and storage of adjustment results, construct a multi-level cache structure. Store the historical adjustment sequences of each decision-making unit at the local cache level for tracking the parameter change trend; summarize the parameter information of adjacent decision-making units at the regional cache node for facilitating regional collaborative optimization. The data transmission adopts a differential update mechanism, only transmitting the changed parameter information, and ensuring the integrity and reliability of data transmission through the validity check and timeout retransmission mechanisms.

[0130] Use the Monte Carlo probability search method to deeply analyze the adjustment results, extract the probability distribution characteristics of parameters, obtain verification samples through a hierarchical quantization adaptive sampling strategy, and dynamically adjust the sampling density according to the parameter importance. Orthogonalize the obtained samples to eliminate the correlation between parameters and generate independent verification scenarios. Use cluster analysis to process the results of verification scenarios, determine the cluster center through dynamic neighborhood search, identify the typical failure modes of the system, construct a fault tree model, systematically track and analyze each weak link, and establish a hierarchical early warning mechanism by analyzing the system response characteristics.

[0131] For the identified weak links, multi-level optimization measures are taken: a spatial protection circle is constructed by adjusting buffer parameters to prevent the spread of adverse effects; a time response mechanism is constructed by configuring the regulation period to ensure that the system can respond to changes in a timely manner; a regional linkage is achieved by establishing a collaborative optimization network to improve the overall regulation effect. The optimization effect is evaluated through incremental verification to obtain the verification results. The verification results are evaluated at multiple levels: the compliance rate of monitoring indicators is determined through statistical analysis to evaluate the optimization effect; the spatial distribution characteristics of parameters are evaluated through the coefficient of variation to test the uniformity of regulation; the system stability is evaluated through perturbation response analysis to verify the reliability of the optimization scheme.

[0132] Based on the evaluation results, the sliding time window method is used to calculate key indicators such as the compliance rate, uniformity, and stability. These indicators are input into the hierarchical adjustment strategy to dynamically adjust the parameter search range and constraint weights of the decision-making unit, and finally the optimal parameter configuration scheme is output.

[0133] Exemplarily, taking a contaminated site remediation project as an example, the site contains 5 decision-making units. According to the evaluation results of the central evaluation network, the score of Unit 1 is lower than the threshold, the scores of Units 2 and 3 are medium, and the scores of Units 4 and 5 are higher. A significant adjustment strategy is adopted for Unit 1: the parameter search space is doubled, and the water quality constraint weight is increased from 0.3 to 0.5; a mild adjustment is adopted for Units 2 and 3: the search range is expanded by 20%, and the constraint weight is slightly adjusted; Units 4 and 5 maintain the existing configuration unchanged. In the multi-level cache structure, the local cache stores the recent 100 sets of adjustment records, and the regional cache node summarizes the parameter information of the adjacent 5 units. The data transmission adopts differential update every 30 minutes, and the transmission success rate reaches 99.9%.

[0134] The Monte Carlo method generates 1000 verification scenarios, and 3 types of typical failure modes are identified through cluster analysis: water quality exceeding the standard, abnormal water level, and system response delay. The constructed fault tree shows that the main weak links include: injection pressure control, water quality monitoring frequency, and system response time. For these weak links, a 50-meter spatial protection circle is set, the regulation period is set to 4 hours, and a collaborative optimization network covering adjacent units is established. The incremental verification evaluation shows that: the compliance rate of monitoring indicators has been significantly improved, the spatial distribution tends to be uniform, and the system's response to perturbations remains stable. The 24-hour sliding time window is used to calculate key indicators, and the parameter configuration is dynamically optimized according to the indicator changes to determine the optimal parameters including key control parameters such as injection pressure, injection period, and monitoring frequency.

[0135] In this embodiment, a hierarchical adjustment strategy is constructed, and different decision-making units are differentially processed according to the evaluation results. A substantial adjustment strategy is adopted for the units with poor performance, a mild adjustment is carried out for the medium units, and the excellent units are kept stable. This differential adjustment method significantly improves the accuracy and efficiency of optimization. A multi-level cache structure is designed. The historical adjustment sequences are stored in the local cache, the adjacent unit information is summarized by the regional cache nodes, and the differential update mechanism is used to transmit data. This hierarchical data management method significantly improves the data processing efficiency and reliability of the system. The verification samples are obtained through a hierarchical quantization adaptive sampling strategy, the parameter correlation is eliminated by orthogonalization processing, and the typical failure modes are identified by combining clustering analysis, which significantly improves the comprehensiveness and reliability of the verification results;

[0136] In some in-situ groundwater remediation optimization and control technologies, in terms of the parameter adjustment strategy, a unified adjustment method is usually adopted, and differential processing cannot be implemented according to the characteristics of different regions, resulting in unsatisfactory optimization effects; in terms of data transmission and storage, a simple centralized storage structure is mostly used, which is likely to cause data transmission bottlenecks and storage pressure; in the scheme verification link, deterministic verification methods are commonly used, and it is difficult to comprehensively evaluate the performance of the system under complex conditions; in terms of system optimization, there is a lack of systematic analysis and targeted optimization of weak links, resulting in insufficient system reliability. These problems seriously affect the overall effect of the remediation project.

[0137] Edge computing control nodes are deployed in the polluted site. The monitoring data of each region are fused through a distributed model parameter aggregation algorithm to generate a global transport control model, and a lightweight on-site control model is trained with the optimal transport parameter scheme. The lightweight on-site control model calculates the optimal transport control scheme according to the real-time monitoring data and sends it to the injection device for execution.

[0138] In an alternative embodiment,

[0139] Deploying edge computing control nodes in the polluted site, fusing the monitoring data of each region through a distributed model parameter aggregation algorithm to generate a global transport control model, and training a lightweight on-site control model with the optimal transport parameter scheme, and the lightweight on-site control model calculating the optimal transport control scheme according to the real-time monitoring data and sending it to the injection device for execution includes:

[0140] Edge computing control nodes are deployed in the polluted site. The edge computing control node obtains water level monitoring data, water quality monitoring data, and flow monitoring data through a data acquisition module, and analyzes the variation laws of the water level monitoring data, the water quality monitoring data, and the flow monitoring data by using a time series feature extraction method to obtain the sampling period;

[0141] The edge computing control node uses an outlier detection algorithm based on differential thresholds to identify outliers, uses local linear interpolation to repair missing data, and calculates time series features through a sliding window to obtain feature vectors;

[0142] The distributed model parameter aggregation algorithm is used to construct a transport parameter sub-model in the edge computing control node, a node affinity matrix is constructed based on the geographical location relationship, and the transport parameter sub-model is iteratively aggregated according to the node affinity matrix to form a regional model. An adaptive weight method is used to merge the regional model to generate a global transport control model;

[0143] The nested grid structure is used to encrypt the grid of the global transport control model, the computing tasks are allocated to the edge computing control nodes for parallel processing through the block iteration method, and the optimal transport parameter solution is obtained by using a hierarchical progressive method;

[0144] The key features of the global transport control model are extracted by means of knowledge mapping, and a lightweight field control model is constructed by using a progressive compression strategy. The lightweight field control model receives the water level monitoring data, the water quality monitoring data and the flow monitoring data, determines the feasible solution space by using a segmented optimization strategy, and performs optimization through a gradient search method and a heuristic algorithm to generate an optimal transport control scheme;

[0145] The optimal transport control scheme is sent to the injection device for execution, a feedback correction mechanism is used to adjust the injection parameters, and potential faults are identified through a state prediction algorithm and the control strategy is automatically switched.

[0146] Edge computing control nodes are deployed at key locations in the polluted site. Each node is equipped with a data acquisition module to continuously collect water level, water quality and flow monitoring data. The time series feature extraction method is used to analyze the variation laws of these monitoring data, including periodic variation, trend variation and mutation characteristics, etc., and the optimal sampling period is determined accordingly to realize the dynamic adjustment of data acquisition.

[0147] In the data preprocessing stage, the edge computing control node first performs outlier detection. An algorithm based on differential thresholds is used to identify possible outliers by calculating the difference between adjacent data points and comparing it with a preset threshold. For the identified abnormal data, the system will mark and temporarily remove it. Subsequently, for the missing values in the dataset, the local linear interpolation method is used for repair, which considers the change trend of adjacent data points to ensure the rationality of the interpolation result. After the data repair is completed, the time series features are calculated through the sliding window method, including statistical features such as mean, variance, and trend, to form feature vectors.

[0148] In the model construction stage, a distributed model parameter aggregation algorithm is adopted. First, a transport parameter sub-model is constructed at each edge computing control node, and this sub-model reflects the characteristics of the local area. Then, based on the geographical location relationship of the nodes, a node affinity matrix is constructed, which describes the degree of association between different nodes. According to the affinity matrix, the system iteratively aggregates the transport parameter sub-models to form a regional model that can represent the characteristics of a larger area. Finally, an adaptive weight method is used to merge these regional models into a global transport control model according to the reliability and representativeness of each regional model.

[0149] To improve the computing efficiency, the system uses a nested grid structure to encrypt the grid of the global transport control model. Finer grids are used in key areas, and coarser grids are maintained in general areas. The large-scale computing tasks are distributed to each edge computing control node through a block iterative method to achieve parallel processing. A hierarchical progressive solution strategy is adopted. First, the coarse grid is solved to obtain an initial solution, and then it is gradually refined to obtain a high-precision solution, and finally the optimal transport parameter scheme is obtained.

[0150] To achieve fast on-site control, the system extracts the key features of the global transport control model through knowledge mapping, including the main parameter relationships and control rules. A progressive compression strategy is adopted to remove the secondary features and retain the core control logic to construct a lightweight on-site control model. This lightweight model receives real-time monitoring data, uses a segmented optimization strategy to determine the feasible solution space under the current conditions, and then combines gradient search and heuristic algorithms to perform fast optimization to generate the optimal transport control scheme.

[0151] The system sends the optimal transport control scheme to the injection device for execution. During the execution process, a feedback correction mechanism is used to adjust the injection parameters in real time to ensure the control effect. At the same time, the system running state is continuously analyzed through a state prediction algorithm. When a potential fault is identified, it automatically switches to a backup control strategy to ensure the continuous and stable operation of the system.

[0152] Exemplarily, take the remediation project of a polluted site of a chemical plant as an example. 12 edge computing control nodes are arranged in the site, and each node is equipped with water level, water quality (pH, conductivity, COD, etc.) and flow monitoring equipment. Through time series analysis, it is determined that the water level data is sampled once per hour, the water quality data is sampled once every four hours, and the flow data is sampled once every half hour.

[0153] In data preprocessing, the system sets the difference threshold to three times the standard deviation and successfully identifies the outliers accounting for 2% of the total data volume. About 5% of the missing data is repaired by local linear interpolation. A 20-dimensional feature vector including mean, variance, trend, etc. is calculated through a sliding window of 30 data points.

[0154] During the model construction process, each node first constructs a local sub-model that includes parameters such as permeability coefficient and storage coefficient. An affinity matrix is constructed based on the distance relationship between nodes, and the 12 sub-models are aggregated through 5 rounds of iteration to form 3 regional models, which are finally merged into a global model covering the entire site.

[0155] In the calculation optimization stage, a three-layer nested grid structure is adopted, with the finest grid spacing being 1 meter and the coarsest grid spacing being 10 meters. The computing tasks are assigned to 12 nodes for parallel processing, and the optimal solution is obtained through three rounds of progressive solution.

[0156] The lightweight model retains 30% of the characteristic parameters of the original model, and the response speed is increased by three times. The model adopts a five-segment optimization strategy to divide the solution space and combines an improved particle swarm algorithm for optimization, and can generate a control scheme within a few seconds on average.

[0157] During the implementation of the plan, the feedback correction mechanism adjusts the injection parameters once per minute, and the state prediction algorithm gives early warnings of potential failures. The system can complete the control strategy switch within seconds when detecting abnormalities.

[0158] In this embodiment, edge computing control nodes are deployed in the contaminated site. By analyzing the change rules of monitoring data through the time series feature extraction method, the dynamic adjustment of the sampling period is realized, which significantly improves the pertinence and efficiency of data collection, ensures the data quality, adopts an anomaly detection algorithm based on differential threshold and a local linear interpolation method to realize the accurate identification of abnormal data and the reasonable repair of missing data, constructs a vector representation that comprehensively reflects the data characteristics through sliding window calculation of time series features, provides a reliable basis for subsequent modeling, and forms a global transport control model by constructing sub-models at the edge nodes and performing iterative aggregation based on geographical location relationships, improving the efficiency and accuracy of model construction. In particular, the adoption of a nested grid structure and a block iteration method realizes the efficient parallel processing of computing tasks;

[0159] In the existing in-situ groundwater remediation control technology, in terms of data collection and processing, fixed sampling periods and unified data processing methods are usually adopted, which cannot adapt to the dynamic changes of site conditions; in terms of anomaly handling, traditional methods for identifying outliers and repairing missing data are relatively simple, prone to data distortion; in terms of model construction, centralized computing architectures are mostly used, with low computing efficiency and difficulty in processing large-scale data; in terms of on-site control, the commonly used control models have complex structures and slow response speeds, making it difficult to meet the real-time control requirements. These problems seriously restrict the improvement of the remediation effect. In this embodiment, through an innovative edge computing architecture, intelligent data processing methods, distributed modeling strategies, and lightweight control models, the deficiencies of traditional technologies in aspects such as data processing, computing efficiency, control performance, and system reliability are effectively solved. The control accuracy and efficiency in the in-situ groundwater remediation process are significantly improved, the system operation cost is reduced, the remediation effect is enhanced, providing a new technical path for the development of in-situ groundwater remediation technology and having important engineering application value.

[0160] In an alternative embodiment,

[0161] Using a distributed model parameter aggregation algorithm to construct a transport parameter sub-model at the edge computing control node, constructing a node affinity matrix based on the geographical location relationship, and iteratively aggregating the transport parameter sub-models according to the node affinity matrix to form a regional model includes:

[0162] Using a distributed model parameter aggregation algorithm to construct a transport parameter sub-model at the edge computing control node, generating a feature sequence by multi-scale analysis and filtering of monitoring data, calculating spatial correlation coefficients, hydrogeological similarity coefficients, and migration flux coefficients between nodes based on the geographical location relationship to construct a node affinity matrix, and iteratively aggregating the transport parameter sub-models according to the node affinity matrix to form a regional model, specifically including:

[0163] Using a distributed model parameter aggregation algorithm to construct a transport parameter sub-model at the edge computing control node, the distributed model parameter aggregation algorithm performs multi-scale analysis on water level monitoring data, water quality monitoring data, and flow monitoring data through wavelet decomposition, extracts high-frequency and low-frequency feature components, sets an adaptive threshold according to the energy distribution of the feature components to remove abnormal fluctuations, and performs smoothing processing on the signal using recursive filtering to generate a stable monitoring data sequence;

[0164] Performing spatio-temporal feature extraction on the stable monitoring data sequence, calculating temporal statistical values through a sliding window, calculating the data distribution probability using kernel density estimation, determining the data mutation position based on change point detection, and constructing a comprehensive feature vector by combining multi-source data fusion algorithms;

[0165] Input the comprehensive feature vector into the gradient descent optimizer to calculate the model parameter gradients. Use the cross-validation method to calculate the parameter importance scores, perform iterative optimization on the high-score parameters, perform linear regression on the low-score parameters, construct the transport parameter sub-model, and generate the initial parameter field.

[0166] Construct a node affinity matrix based on the geographical location relationship, calculate the Euclidean distance between nodes to obtain the spatial correlation coefficient, use the hierarchical clustering method to calculate the hydrogeological similarity coefficient, calculate the migration flux coefficient based on the pollutant concentration gradient, and perform normalized weighted calculation on the spatial correlation coefficient, the hydrogeological similarity coefficient, and the migration flux coefficient to obtain the affinity weight.

[0167] Iteratively aggregate the transport parameter sub-model according to the node affinity matrix to form a regional model. Prioritize the fusion of nodes with high affinity. Use the affinity weight to perform weighted averaging on the initial parameter field, calculate the fusion residual, and adjust the affinity weight to update the node affinity matrix.

[0168] Divide the regional model into multiple computational subtasks, allocate them to the edge computing control nodes, establish a message queue between nodes to transmit data, set checkpoints according to the fusion residual, and allocate node loads according to the computational amount of the initial parameter field.

[0169] Use a distributed model parameter aggregation algorithm to construct a transport parameter sub-model at the edge computing control node. This algorithm performs wavelet decomposition on the collected water level monitoring data, water quality monitoring data, and flow monitoring data to achieve multi-scale analysis. The original signal is decomposed into feature components of different frequencies through wavelet transform, and the high-frequency components (reflecting short-term fluctuations) and low-frequency components (reflecting long-term trends) are extracted respectively. According to the energy distribution of each feature component, an adaptive threshold is set to identify and remove abnormal fluctuations. Subsequently, a recursive filtering method is used to smooth the signal to eliminate residual noise and generate a stable monitoring data sequence.

[0170] Extract spatio-temporal features from the processed stable monitoring data sequence. First, calculate the temporal statistical values through the sliding window method, including statistical features such as mean, variance, skewness, and kurtosis. Use the kernel density estimation method to calculate the distribution probability of the data and obtain the probability density feature of the data. Determine the mutation position of the data through the change point detection algorithm to identify key change points. Finally, combine the multi-source data fusion algorithm to integrate different types of features and construct a comprehensive feature vector that comprehensively reflects the system characteristics.

[0171] Input the constructed comprehensive feature vector into the gradient descent optimizer to calculate the gradient information of each parameter of the model. Use the cross-validation method to evaluate the importance of different parameters and generate parameter importance scores. For parameters with high importance scores, use the iterative optimization method for fine-tuning, and for parameters with low importance scores, use the linear regression method for rapid estimation. Based on the optimized parameters, construct a transport parameter sub-model and generate an initial parameter field describing the characteristics of the current region.

[0172] Construct a node affinity matrix based on the geographical location relationship, calculate the Euclidean distance between each node to obtain the spatial correlation coefficient characterizing the degree of spatial association. Use the hierarchical clustering method to analyze the hydrogeological characteristics of each node region and calculate the hydrogeological similarity coefficient. Calculate the migration flux coefficient based on the spatial distribution of pollutant concentrations to characterize the relevance of pollutant migration. After normalizing these three types of coefficients, calculate the weighted sum to obtain the comprehensive affinity weight.

[0173] Iteratively aggregate the transport parameter sub-model according to the node affinity matrix to form a regional model. During the aggregation process, preferentially fuse nodes with higher affinity, and use the calculated affinity weight to perform weighted averaging on the initial parameter field. Calculate the residuals generated during the fusion process, dynamically adjust the affinity weight according to the size of the residuals, and update the node affinity matrix accordingly to ensure the accuracy of the aggregation result.

[0174] Divide the regional model into multiple computational subtasks and allocate them to each edge computing control node for parallel processing. Establish a message queue system between nodes to achieve efficient data transmission. Set checkpoints according to the fusion residuals to monitor the convergence of the calculation process. At the same time, reasonably allocate the node load according to the computational amount of the initial parameter field to ensure the balanced utilization of computing resources.

[0175] Exemplarily, take a certain groundwater pollution site as an example. There are 15 edge computing control nodes deployed in the site. The original data collected by each node includes: water level data (once per hour), water quality data (pH, COD, etc., once every four hours), and flow data (once every half hour).

[0176] Through 4-layer wavelet decomposition, decompose various types of monitoring data into components with different frequencies. Energy distribution analysis shows that the high-frequency component accounts for about 20%. Set an adaptive threshold to remove abnormal fluctuations whose amplitudes exceed 3 times the mean. Use the Kalman filter for signal smoothing to obtain a stable monitoring data sequence. Calculate the time-series statistical features using a 24-hour sliding window, perform kernel density estimation using a Gaussian kernel function, and detect data mutation points through the CUSUM algorithm. After fusing multi-source data, obtain a 128-dimensional comprehensive feature vector.

[0177] During the gradient descent optimization process, the parameter importance is calculated through 5-fold cross-validation. For parameters with importance scores exceeding 0.7, 100 rounds of iterative optimization are performed, and for other parameters, linear regression estimation is used to construct a sub-model containing 20 key parameters. In the calculation of node affinity, the weight of the spatial correlation coefficient is 0.4, the weight of the hydrogeological similarity coefficient is 0.3, and the weight of the migration flux coefficient is 0.3. Through 5 rounds of iterative aggregation, 15 sub-models are combined into 3 regional models. Finally, the calculation task is divided into 60 sub-tasks and assigned to 15 nodes for parallel processing. A checkpoint is set every 20% of the data volume processed, and 3-5 sub-tasks are assigned to each node according to the calculation complexity.

[0178] In this embodiment, wavelet decomposition is used to realize multi-scale data analysis. By extracting high-frequency and low-frequency feature components and combining adaptive thresholding and recursive filtering, refined processing of the monitoring data is achieved, significantly improving the data quality and providing a reliable basis for subsequent modeling. Combining various methods such as time series statistics, kernel density estimation, and change point detection, a comprehensive feature extraction system is constructed. Multi-scale analysis and adaptive filtering significantly improve the signal-to-noise ratio of the monitoring data, effectively eliminate abnormal fluctuations, and ensure the reliability of the data. The multi-dimensional feature extraction and fusion mechanism enables the model to more comprehensively depict the site characteristics, significantly enhancing the richness and accuracy of feature expression. The differential optimization strategy significantly improves the calculation efficiency while ensuring the optimization accuracy of key parameters, achieving a balance between optimization efficiency and accuracy.

[0179] In the construction technology of existing in-situ groundwater remediation models, in terms of data processing, usually a single filtering method is used to process monitoring data, which cannot effectively identify and eliminate multi-scale noise; in terms of feature extraction, it is often limited to simple statistical features and lacks in-depth analysis of the spatio-temporal evolution characteristics of the data; in terms of parameter optimization, a unified optimization strategy is used to process all parameters, ignoring the differences in parameter importance; in terms of model aggregation, simple spatial distance is often used as the fusion basis, failing to fully consider the influence of hydrogeological characteristics and pollutant migration characteristics. These problems lead to insufficient model accuracy and difficulty in accurately describing site characteristics. Through innovative data processing methods, a comprehensive feature extraction system, a differential optimization strategy, and a multi-factor coupling aggregation mechanism, this embodiment effectively solves the deficiencies of traditional technologies in data processing, feature extraction, parameter optimization, and model aggregation. It significantly improves the construction efficiency and accuracy of in-situ groundwater remediation models, providing more reliable support for the optimization of remediation plans.

[0180] Figure 2This is a comparison chart of the feature fusion effects of the precise transport control method for the remediation agent in the in-situ groundwater remediation of the embodiments of the present invention, which shows the change trend of feature consistency with the number of iterations during the feature fusion process of different technical solutions. Through the innovative double-branch parallel structure and multi-stage cascaded noise reduction mechanism in the embodiments of the present invention, a relatively high fusion accuracy (0.68) is achieved at the initial stage of iteration (5 times), and it continues to improve in subsequent iterations, finally reaching a high-precision level of 0.87 after 35 iterations. In contrast, the fusion accuracies of the traditional feature fusion method and the single-constraint feature fusion method at the initial stage of iteration are 0.58 and 0.62 respectively, and they can only reach 0.70 and 0.76 finally. Especially in the critical stage of 10-20 iterations, the accuracy improvement rate of the embodiments of the present invention is significantly higher than that of other solutions, thanks to the effective cooperation of the differential noise processing strategy and the multi-head attention mechanism. After 25 iterations, the accuracy curve of the embodiments of the present invention tends to be smooth, indicating that the fusion process has reached a stable state, and the smoothness of the curve is also better than that of other solutions, reflecting the stability and reliability of the fusion process. In addition, the original feature information retained through the residual connection enables the embodiments of the present invention to well maintain the physical meaning of the features while maintaining a high fusion accuracy. These data fully prove the significant advantages of the embodiments of the present invention in terms of feature fusion effect and efficiency.

[0181] Figure 3 This is a comparison chart of the system optimization effects of the precise transport control method for the remediation agent in the in-situ groundwater remediation of the embodiments of the present invention, which shows the change trend of the system compliance rate with the optimization cycle under different optimization strategies. Through the hierarchical adjustment strategy and the multi-stage cache structure in the embodiments of the present invention, a relatively high compliance rate (82%) is achieved at the initial stage of optimization (4h), and it continues to improve in the subsequent optimization process, finally reaching a high compliance rate of 94% at 28h. In contrast, the compliance rates of the traditional fixed strategy and the simple adaptive strategy at the initial stage of optimization are 75% and 78% respectively, and they can only reach 85% and 88% finally. Especially in the critical optimization stage of 8-16h, the improvement speed of the compliance rate of the embodiments of the present invention is significantly faster than that of other solutions. After 20h, the compliance rate curve of the embodiments of the present invention tends to be stable, indicating that the system has reached a stable optimization state. At the same time, the smoothness of the curve is also better than that of other solutions, reflecting the stability and reliability of the system optimization process. These data fully prove the significant advantages of the embodiments of the present invention in terms of system optimization effect and efficiency.

[0182] Figure 4This is a comparison chart of the data quality improvement effect of the precise transport control method of the remediation agent for in-situ groundwater remediation in the embodiments of the present invention, showing the change trend of the signal-to-noise ratio in the data processing process of different technical solutions. In the embodiments of the present invention, through multi-scale wavelet decomposition and adaptive threshold filtering, a relatively high signal-to-noise ratio (38 dB) is achieved at the initial stage of processing (20 min), and it continues to increase with the increase of the processing time, and finally reaches a high signal-to-noise ratio of 85 dB at 100 min. In contrast, the signal-to-noise ratios of traditional wavelet threshold denoising and empirical mode decomposition at the initial stage of processing are 31 dB and 33 dB respectively, and they can only reach 55 dB and 60 dB finally. Especially in the key processing stage of 40 - 60 min, the signal-to-noise ratio improvement rate of the embodiments of the present invention is significantly higher than that of other solutions, which benefits from the effective elimination of residual noise by the recursive filtering method. At the same time, the signal-to-noise ratio curve of this technical solution is smoother, indicating that the processing process is more stable and reliable. These data fully prove the significant advantages of this technical solution in improving data quality.

[0183] The present invention can be a method, device, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions for performing various aspects of the present invention loaded thereon.

[0184] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for precisely controlling the transport of a remediation agent for in-situ groundwater remediation, characterized in that, include: Image data and sensor measurement data corresponding to the drill core of the contaminated site are collected. The image data is input into the visual attention neural network for feature extraction to obtain soil structure feature data. The sensor measurement data is input into the temporal attention neural network for analysis to obtain groundwater dynamic feature data. The soil structure feature data and groundwater dynamic feature data are added to the conditional probability diffusion model for feature fusion. The hydrogeological feature vector is generated and mapped to the three-dimensional spatial grid through the spatial attention calculation module to obtain the initial hydrogeological feature model. The generative probability diffusion model is combined to complete the missing areas to obtain a high-precision hydrogeological feature model. The high-precision hydrogeological characteristic model is input into a multi-region collaborative simulation environment. Local optimization calculations are performed on different spatial regions through multiple decision-making units. The decision-making units share the optimization results through a communication network. The central evaluation network performs a global evaluation based on the shared results and collaboratively adjusts the optimization parameters of each decision-making unit. The adjusted transport strategy is verified by combining the Monte Carlo probability search method to output the optimal transport parameter solution. Edge computing control nodes are deployed at the contaminated site, and the monitoring data of each area are fused and processed through a distributed model parameter aggregation algorithm to generate a global transport control model. This model is trained with the optimal transport parameter scheme to obtain a lightweight field control model. The lightweight field control model calculates the optimal transport control scheme based on the real-time monitoring data and sends it to the injection device for execution.

2. The method according to claim 1, characterized in that, Image data and sensor measurement data corresponding to the borehole core of the contaminated site are collected. The image data is input into the visual attention neural network for feature extraction to obtain soil structure feature data. The sensor measurement data is input into the temporal attention neural network for analysis to obtain groundwater dynamic feature data. The soil structure feature data and groundwater dynamic feature data are added to the conditional probability diffusion model for feature fusion. The hydrogeological feature vector is generated and mapped to the three-dimensional spatial grid through the spatial attention calculation module to obtain the initial hydrogeological feature model. The missing areas are supplemented by the generative probability diffusion model to obtain a high-precision hydrogeological feature model, including: Deploy a borehole sampling system at the contaminated site to collect core image data and groundwater sensor measurement data; Inputting the core image data into an adaptive histogram equalization processing module to eliminate illumination interference to obtain a first processed image, performing local contrast sharpening on the first processed image to enhance structural details to obtain a second processed image, inputting the second processed image into a visual attention neural network, performing convolution operation on an encoder of the visual attention neural network to extract features, and reconstructing the soil structural feature data through a decoder; Inputting the groundwater sensor measurement data into an outlier detection module to remove outliers to obtain first processed data, performing denoising on the first processed data through an adaptive sliding window to obtain second processed data, inputting the second processed data into a temporal attention neural network, and analyzing temporal associations through a bidirectional long short-term memory unit to obtain groundwater dynamic characteristic data; Input the soil structure feature data and the groundwater dynamic feature data into the conditional probability diffusion model in a spliced manner, obtain the hydrogeological feature vector through dynamic noise reduction iteration, input the hydrogeological feature vector into the spatial attention calculation module, calculate the spatial weight according to the feature similarity and perform grid encryption, and obtain the initial hydrogeological feature model through feature interpolation; Input the initial hydrogeological feature model into the generative probability diffusion model, perform feature generation on the missing area based on the known regional feature distribution, and evaluate and optimize through a multi-scale discriminator to obtain a high-precision hydrogeological feature model.

3. The method according to claim 2, characterized in that, Input the soil structure feature data and the groundwater dynamic feature data into the conditional probability diffusion model in a spliced manner, and the hydrogeological feature vector obtained through dynamic noise reduction iteration includes: Perform feature splicing on the soil structure feature data and the groundwater dynamic feature data to obtain a mixed feature vector, input the mixed feature vector into the dual-branch structure of the conditional probability diffusion model, obtain potential features through the main branch, obtain constraint conditions through the auxiliary branch, perform differential noise injection and noise reduction processing on the features, generate fused features by combining the constraint conditions and the attention mechanism, and reconstruct the hydrogeological feature vector after optimizing the parameters based on the feature quality evaluation, specifically including: Perform feature splicing on the soil structure feature data and the groundwater dynamic feature data through a hierarchical cascading structure, splice the features of the same scale to obtain a same-scale feature vector, combine the same-scale feature vectors to form a mixed feature vector and input it into the conditional probability diffusion model of the dual-branch parallel structure. The main branch of the conditional probability diffusion model maps the mixed feature vector to the latent space through a feature encoder to obtain potential features, and the auxiliary branch extracts physical law constraint information, spatial continuity constraint information and temporal consistency constraint information through a constraint encoder to obtain a constraint condition vector; Calculate the feature importance distribution based on the mixed feature vector, perform differential noise injection on different feature regions according to the feature importance distribution, inject strong noise into the feature regions with importance lower than 0.3, inject weak noise into the feature regions with importance higher than 0.75, and realize feature degradation through multi-level noise superposition to obtain noise features; Input the noise features into a multi-cascaded noise reduction unit. The multi-cascaded noise reduction unit extracts feature information through a densely connected feature extraction module, predicts the noise distribution through a noise prediction module, and realizes dynamic noise reduction by adopting a skip connection method to obtain noise-reduced features. Input the constraint condition vector into a constraint decoder to generate feature constraint conditions, and perform constraint adjustment on the noise-reduced features through a recursive feedback structure to obtain constraint features; Construct a multi-head attention module to calculate the feature correlation degree of the constraint features, generate attention weights based on the feature correlation degree, enhance the constraint features through the attention weights and introduce a residual connection to retain the original feature information to obtain fused features; The construction parameter evaluation module calculates the feature quality index and the fusion effect index of the fused features, dynamically optimizes the noise reduction parameters based on the feature quality index and the fusion effect index to obtain optimized parameters, inputs the optimized parameters into the feature decoder, and decodes and reconstructs the fused features to obtain a hydrogeological feature vector.

4. The method according to claim 1, characterized in that, Input the high-precision hydrogeological feature model into the multi-region collaborative simulation environment, perform local optimization calculations on different spatial regions through multiple decision-making units, share the optimization results through a communication network among the decision-making units, and the central evaluation network conducts global evaluation based on the shared results and collaboratively adjusts the optimization parameters of each decision-making unit. Combine the Monte Carlo probability search method to verify the adjusted transport strategy, and the output optimal transport parameter scheme includes: Divide the high-precision hydrogeological feature model into multiple spatial regions according to the hydrogeological unit boundary and management partition, and configure decision-making units with a data layer, a calculation layer, and a communication layer in each spatial region; The data layer of the decision-making unit extracts the geological structure features, hydrogeological parameter distributions, and boundary condition information of the spatial region, establishes a numerical model of the groundwater flow field, collects groundwater level, water quality index, and extraction volume data to construct a dynamic database. The calculation layer of the decision-making unit inputs the state data in the dynamic database into the optimizer, and performs local optimization calculations based on the optimization goals of water resource balance, water quality safety, and system stability to obtain optimization results; The decision-making units are interconnected through a communication network to construct a hierarchical caching mechanism to store the optimization results, and adopt an incremental update strategy to transmit the system state data to the central evaluation network. The central evaluation network extracts the evaluation indicators of the feature calculation resource regulation effect, environmental impact degree, and economic benefit level, and generates a comprehensive score by highlighting the contribution of key indicators through an attention mechanism; Based on the comprehensive score, expand the parameter search space and adjust the constraint condition weights for each decision-making unit, introduce a coordination factor to balance the optimization goals, transmit the adjustment results to the decision-making units, generate verification scenarios using the Monte Carlo probability search method, identify typical failure modes through cluster analysis and optimize weak links to obtain verification results. Based on the verification results, screen the water volume distribution coefficient, water quality control threshold, and system regulation parameters, construct a dynamic adjustment rule for the transport strategy, and output the optimal transport parameter scheme.

5. The method according to claim 1, wherein Transmit the adjustment results to the decision-making units, generate verification scenarios using the Monte Carlo probability search method, identify typical failure modes through cluster analysis and optimize weak links to obtain verification results including: Construct a hierarchical adjustment strategy based on the evaluation results of the central evaluation network, generate adjustment results for decision-making units in different score intervals, transmit the adjustment results to the decision-making units through a multi-level caching structure, extract features and identify weak links using the Monte Carlo probability search method, construct a protection mechanism for the weak links and generate evaluation results, and dynamically adjust the parameter configuration based on the evaluation results, specifically including: Construct a hierarchical adjustment strategy based on the evaluation results of the central evaluation network, and perform differential processing on decision-making units in different scoring intervals through adjustment methods such as expanding the parameter search space and increasing the constraint weight, moderately expanding the search range and fine-tuning the constraint weight, and maintaining the existing parameter configuration to generate adjustment results; Construct a multi-level cache structure to transfer the adjustment results to the decision-making unit, store the historical adjustment sequence through local caching, summarize the parameter information of adjacent units through regional cache nodes, use a differential update mechanism to transmit data, and ensure data integrity through validity verification and timeout retransmission; Adopt the Monte Carlo probability search method to extract the probability distribution characteristics of the adjustment results, obtain verification samples through a hierarchical quantization adaptive sampling strategy, generate verification scenarios after orthogonalizing the samples to eliminate parameter correlation, use cluster analysis to process the verification results of the verification scenarios, determine the cluster center through dynamic neighborhood search to identify typical failure modes, construct a fault tree model to trace weak links, and establish a hierarchical early warning mechanism through system response characteristic analysis; Adjust the buffer parameters for the weak links to construct a spatial protection circle, configure a control period to construct a time response mechanism, establish a collaborative optimization network to achieve regional linkage, obtain verification results by evaluating the optimization effect through incremental verification, perform multi-level evaluation on the verification results, determine the compliance rate of monitoring indicators through statistical analysis, evaluate the spatial distribution characteristics through the coefficient of variation, and analyze the system stability through perturbation response to generate evaluation results; Based on the evaluation results, calculate key indicators using a sliding time window, input the calculated key indicators into the hierarchical adjustment strategy, dynamically adjust the parameter search range and constraint weight of the decision-making unit, and output the optimal parameter configuration.

6. The method according to claim 1, wherein Deploy edge computing control nodes at the contaminated site, fuse the monitoring data of each region through a distributed model parameter aggregation algorithm to generate a global transport control model, and train it with the optimal transport parameter scheme to obtain a lightweight on-site control model. The lightweight on-site control model calculates the optimal transport control scheme based on real-time monitoring data and sends it to the injection device for execution, including: Deploy edge computing control nodes at the contaminated site. The edge computing control nodes obtain water level monitoring data, water quality monitoring data, and flow monitoring data through a data acquisition module, and analyze the variation laws of the water level monitoring data, the water quality monitoring data, and the flow monitoring data using a time series feature extraction method to obtain the sampling period; The edge computing control nodes use an outlier detection algorithm based on differential thresholds to identify outliers, repair missing data using local linear interpolation, and calculate time series features through a sliding window to obtain feature vectors; Use a distributed model parameter aggregation algorithm to construct a transport parameter sub-model at the edge computing control node, construct a node affinity matrix based on the geographical location relationship, iteratively aggregate the transport parameter sub-models according to the node affinity matrix to form a regional model, and use an adaptive weight method to merge the regional models to generate a global transport control model; The global transport control model is encrypted with a nested grid structure, and the computing tasks are assigned to the edge computing control nodes for parallel processing by a block iteration method. The optimal transport parameter scheme is obtained by solving in a hierarchical progressive manner; The key features of the global transport control model are extracted by a knowledge mapping method, and a lightweight field control model is constructed by a progressive compression strategy. The lightweight field control model receives the water level monitoring data, the water quality monitoring data, and the flow monitoring data, determines the feasible solution space by a segmented optimization strategy, and searches for the optimal transport control scheme through a gradient search method and a heuristic algorithm; The optimal transport control scheme is sent to the injection device for execution, the injection parameters are adjusted by a feedback correction mechanism, and potential faults are identified by a state prediction algorithm and the control strategy is automatically switched.

7. The method according to claim 6, wherein A distributed model parameter aggregation algorithm is used to construct a transport parameter sub-model at the edge computing control node, a node affinity matrix is constructed based on the geographical location relationship, and the transport parameter sub-models are iteratively aggregated according to the node affinity matrix to form a regional model, including: A distributed model parameter aggregation algorithm is used to construct a transport parameter sub-model at the edge computing control node. The monitoring data is processed by multi-scale analysis and filtering to generate a feature sequence. The spatial correlation coefficient, the hydrogeological similarity coefficient, and the migration flux coefficient between nodes are calculated based on the geographical location relationship to construct a node affinity matrix. The transport parameter sub-models are iteratively aggregated according to the node affinity matrix to form a regional model, specifically including: A distributed model parameter aggregation algorithm is used to construct a transport parameter sub-model at the edge computing control node. The distributed model parameter aggregation algorithm performs multi-scale analysis on the water level monitoring data, the water quality monitoring data, and the flow monitoring data by wavelet decomposition, extracts high-frequency and low-frequency feature components, sets an adaptive threshold according to the energy distribution of the feature components to remove abnormal fluctuations, and performs smoothing processing on the signal by recursive filtering to generate a stable monitoring data sequence; The spatio-temporal features of the stable monitoring data sequence are extracted, the temporal statistical values are calculated by a sliding window, the data distribution probability is calculated by kernel density estimation, the data mutation position is determined based on change point detection, and a comprehensive feature vector is constructed by combining a multi-source data fusion algorithm; The comprehensive feature vector is input into a gradient descent optimizer to calculate the model parameter gradient. The cross-validation method is used to calculate the parameter importance score. The high-score parameters are iteratively optimized, and the low-score parameters are linearly regressed to construct the transport parameter sub-model and generate an initial parameter field; A node affinity matrix is constructed based on the geographical location relationship, the Euclidean distance between nodes is calculated to obtain the spatial correlation coefficient, the hierarchical clustering method is used to calculate the hydrogeological similarity coefficient, and the migration flux coefficient is calculated based on the pollutant concentration gradient. The spatial correlation coefficient, the hydrogeological similarity coefficient, and the migration flux coefficient are normalized and weighted to calculate the affinity weight; Iteratively aggregate the transport parameter sub-model according to the node affinity matrix to form a regional model, preferentially fuse nodes with high affinity, perform weighted averaging on the initial parameter field using the affinity weights, calculate the fusion residual and adjust the affinity weights, and update the node affinity matrix; Divide the regional model into multiple computational subtasks, allocate them to the edge computing control nodes, establish message queues between nodes to transmit data, set checkpoints according to the fusion residual, and allocate node loads according to the computational amount of the initial parameter field.

8. A precise transport control system for in-situ remediation agents of groundwater, which is used to implement the method described in any one of the foregoing claims 1-7, characterized in that, Including: The first unit is used to collect image data and sensor measurement data corresponding to the borehole cores of the contaminated site, input the image data into the visual attention neural network for feature extraction to obtain soil structure feature data, input the sensor measurement data into the temporal attention neural network for analysis to obtain groundwater dynamic feature data, add the soil structure feature data and groundwater dynamic feature data to the conditional probability diffusion model for feature fusion, generate a hydrogeological feature vector and map it to a three-dimensional spatial grid through the spatial attention calculation module to obtain an initial hydrogeological feature model, and complete the missing area by combining the generative probability diffusion model to obtain a high-precision hydrogeological feature model; The second unit is used to input the high-precision hydrogeological feature model into the multi-region collaborative simulation environment, perform local optimization calculations on different spatial regions through multiple decision-making units, the decision-making units share the optimization results through the communication network, the central evaluation network conducts global evaluation based on the shared results and collaboratively adjusts the optimization parameters of each decision-making unit, and combines the Monte Carlo probability search method to verify the adjusted transport strategy and output the optimal transport parameter scheme; The third unit is used to deploy edge computing control nodes at the contaminated site, fuse and process the monitoring data of each region through the distributed model parameter aggregation algorithm, generate a global transport control model and train it with the optimal transport parameter scheme to obtain a lightweight on-site control model, and the lightweight on-site control model calculates the optimal transport control scheme according to the real-time monitoring data and sends it to the injection device for execution.

Images