Drain outlet off-site law enforcement method and system based on digital twinborn technology

Abstract

The invention discloses a sewage draining exit off-site law enforcement method and system based on a digital twin technology, and relates to the technical field of sewage draining exit off-site law enforcement. According to the sewage draining exit off-site law enforcement method based on the digital twinborn technology, three-dimensional point cloud data of a set sewage draining exit are obtained and input into a pre-trained structure identification model for comprehensive analysis, an attribute set of each component is obtained, a digital twinborn model is constructed, sewage draining monitoring time sequence data are continuously obtained, and the sewage draining exit off-site law enforcement method based on the digital twinborn technology is obtained. Inputting the data into a digital twin model for feature analysis to obtain a discharge behavior risk index and a discharge visual risk index of each time period, and performing comprehensive analysis to obtain a discharge abnormality evaluation index of each time period of a set drain outlet; according to the method, the preset off-site law enforcement measures are taken for the set sewage draining exit according to the discharge abnormity evaluation index of each time period, so that pre-recognition, automatic reaction and accurate response of sudden sewage draining are realized, and the instantaneity and the intelligent level of off-site supervision are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of off-site law enforcement at sewage outlets, and specifically to an off-site law enforcement method and system for sewage outlets based on digital twin technology. Background Art

[0002] At present, environmental supervision of various sewage outlets generally relies on manual inspections, fixed-point sampling, water quality index feedback and remote image observation. These methods have obvious limitations. On the one hand, the on-site law enforcement response is short-lived and the supervision force is dispersed, making it difficult to detect abnormal discharge behavior in a timely manner. On the other hand, sewage discharge data are distributed among different systems such as water quality monitoring instruments and camera equipment. The information is isolated and the processing is decentralized. There is a lack of unified modeling and linkage analysis mechanisms. Especially in complex scenarios such as the suddenness, visual uncontrollability and diversity of sewage discharge behaviors, traditional non-site law enforcement models often face problems such as low recognition accuracy, untimely warnings and high false alarm rates. However, in recent years, with the development of digital twin technology, the ability to construct digital scenes of sewage outlets based on three-dimensional modeling and virtual-reality mapping has gradually increased, making it possible to integrate spatial structure modeling, dynamic emission data analysis and image perception results into a unified risk assessment model.

[0003] The limitations of existing technologies include at least the following problems: existing technologies lack the ability to integrate and analyze emission behavior and visual representation, and it is difficult to achieve time-by-time and pre-emptive risk prediction. For example, a sewage-discharging enterprise in an industrial park suddenly discharged high-concentration oily wastewater at noon. Although the monitoring image of the discharge outlet clearly showed visual characteristics such as abnormally dark water color, large accumulation of foam, and severe liquid surface disturbance, it only relied on the ammonia nitrogen and COD values collected by the sensor end and lacked an image linkage mechanism, resulting in the pollutant concentration signal being averaged and delayed. In addition, the sampling period was long and the alarm threshold setting was delayed, thus failing to respond in the early stage of pollution. Ultimately, the emission behavior triggered a passive alarm by exceeding the concentration threshold about half an hour later. However, at this time, the abnormal emission had completed and the pollutants had diffused into the downstream area with the water flow, causing actual pollution results, thereby weakening the ability to intervene in advance in sudden emission scenarios, and then it was difficult to meet the actual needs of off-site supervision in high-frequency, high-risk, and diversified emission scenarios. Summary of the Invention

[0004] In response to the shortcomings of the existing technology, the present invention provides an off-site law enforcement method and system for sewage outlets based on digital twin technology, which solves the problem that the existing technology is difficult to integrate images and emission data, making it difficult to timely identify sudden sewage discharges, and lacks pre-emptive intelligent early warning capabilities.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a non-site enforcement method for sewage outlets based on digital twin technology, comprising the following steps: obtaining three-dimensional point cloud data of a set sewage outlet, and inputting it into a pre-trained structure recognition model for comprehensive analysis to obtain a property set of each component of the set sewage outlet, and constructing a digital twin model of the set sewage outlet; and continuously obtaining sewage discharge monitoring time series data of the set sewage outlet, wherein the sewage discharge monitoring time series data includes emission intensity time series data and emission image time series data; inputting the sewage discharge monitoring time series data of the set sewage outlet into the digital twin model for feature analysis respectively to obtain the emission behavior risk index and emission visual risk index of each time period of the set sewage outlet, and performing comprehensive analysis to obtain the emission anomaly assessment index of each time period of the set sewage outlet; taking preset non-site enforcement measures for the set sewage outlet based on the emission anomaly assessment index of each time period.

[0006] Furthermore, the attribute set includes geometric dimension attributes, spatial positioning attributes, material attributes, and topological relationship attributes; the three-dimensional point cloud data includes the voxel value, three-dimensional coordinates, and corresponding reflection intensity value of each voxel point; the structure recognition model is specifically a point cloud deep learning network, and the point cloud deep learning network includes an input layer, a feature extraction layer, a structure recognition layer, a structure feature encoding layer, and an attribute output layer.

[0007] Furthermore, the specific steps for obtaining the attribute set of each component of the set sewage outlet are as follows: in the input layer of the point cloud deep learning network, the three-dimensional point cloud data of the set sewage outlet is received and preprocessed; in the feature extraction layer of the point cloud deep learning network, the preprocessed three-dimensional point cloud data of the set sewage outlet is subjected to feature extraction processing to obtain a local structural feature vector set of each central sampling point of the set sewage outlet; in the structure recognition layer of the point cloud deep learning network, the local structural feature vector set of each central sampling point of the set sewage outlet is subjected to fusion recognition processing to obtain a component point cloud set of several components of the set sewage outlet; in the structure feature encoding layer of the point cloud deep learning network, the component point cloud set of each component of the set sewage outlet is subjected to feature encoding processing to obtain a global structural semantic feature vector of each component of the set sewage outlet; in the attribute output layer of the point cloud deep learning network, the global structural semantic feature vector of each component of the set sewage outlet is subjected to branch prediction processing to obtain the geometric size attributes, spatial positioning attributes, material attributes, and topological relationship attributes, that is, the attribute set, of each component of the set sewage outlet.

[0008] Furthermore, the specific formula for calculating the discharge anomaly assessment index for a certain period of time at a given sewage outlet is as follows: Among them, PyC is the emission anomaly assessment index of a certain period of time at the set sewage outlet, PxW is the emission behavior risk index of a certain period of time at the set sewage outlet, α1 is the behavior risk adjustment coefficient stored in the database, SxF is the emission visual risk index of a certain period of time at the set sewage outlet, α2 is the visual risk adjustment coefficient stored in the database, and α3 is the difference adjustment coefficient stored in the database.

[0009] Furthermore, the emission intensity time series data includes the total amount of suspended solids, pH value, pollution load index, sewage outlet vibration amplitude value, flow velocity fluctuation index, pollution diffusion index and pollution concentration value and concentration change rate value of each pollutant in each time period. The specific steps for obtaining the emission behavior risk index of each time period of the set sewage outlet are as follows: comprehensively analyze the emission intensity time series data of the set sewage outlet respectively to obtain the emission risk assessment index set of each time period of the set sewage outlet, including the pollution sudden change risk index and the pollution transmission risk index; and comprehensively analyze the emission risk assessment index set of each time period of the set sewage outlet to obtain the emission behavior risk index of each time period of the set sewage outlet.

[0010] Furthermore, the specific steps for obtaining the emission risk assessment index set for each time period of the set sewage outlet are as follows: obtaining the concentration reference value of each pollutant of the set sewage outlet, and conducting a comprehensive analysis in combination with the pollution concentration value of each pollutant in each time period to obtain the pollution deviation index of each time period of the set sewage outlet; reading the total amount of suspended solids, pH value, sewage outlet vibration amplitude value and concentration change rate value of each pollutant in each time period of the set sewage outlet, and conducting a comprehensive analysis in combination with the pollution deviation index to obtain the pollution sudden change risk index of each time period of the set sewage outlet; reading the pollution load index, flow velocity fluctuation index and pollution diffusion index of each time period of the set sewage outlet, and conducting a comprehensive analysis to obtain the pollution transmission risk index of each time period of the set sewage outlet.

[0011] Furthermore, the specific formula for calculating the discharge behavior risk index of a certain period of time at a designated sewage outlet is as follows: Among them, PxW is the emission behavior risk index of the set sewage outlet in a certain period of time, WrB is the pollution shock risk index of the set sewage outlet in a certain period of time, η1 is the shock adjustment coefficient stored in the database, WcF is the pollution propagation risk index of the set sewage outlet in a certain period of time, η2 is the propagation adjustment coefficient stored in the database, and η3 is the interaction adjustment coefficient stored in the database.

[0012] Furthermore, the emission image time series data includes emission image data for each time period, and the emission image data is specifically the pixel value and two-dimensional coordinates of each pixel point in the emission image. The specific steps for obtaining the emission visual risk index for each time period of the set sewage outlet are as follows: comprehensively analyze the pixel value and two-dimensional coordinates of each pixel point in the emission image for each time period of the set sewage outlet to obtain the image perception anomaly index set for each time period of the set sewage outlet, including the effluent color deviation index, foam coverage index, and edge disturbance index; comprehensively analyze the image perception anomaly index set for each time period of the set sewage outlet to obtain the emission visual risk index for each time period of the set sewage outlet.

[0013] Furthermore, the specific steps for taking preset off-site enforcement measures for the set sewage outlet based on the emission abnormality assessment index of each time period are as follows: comprehensively analyze the emission abnormality assessment index of each time period of the set sewage outlet to obtain the emission abnormality assessment index of the next time period of the set sewage outlet; judge and analyze the emission abnormality assessment index of the next time period of the set sewage outlet and the preset emission abnormality assessment index threshold range; if the emission abnormality assessment index of the next time period of the set sewage outlet is lower than the lower limit of the preset emission abnormality assessment index threshold range, take the first enforcement measure; if the emission abnormality assessment index of the next time period of the set sewage outlet is within the preset emission abnormality assessment index threshold range, take the second enforcement measure; if the emission abnormality assessment index of the next time period of the set sewage outlet is higher than the upper limit of the preset emission abnormality assessment index threshold range, take the third enforcement measure.

[0014] The off-site enforcement system for sewage outlets based on digital twin technology includes: a data acquisition and modeling module, which is used to acquire three-dimensional point cloud data of a set sewage outlet and input it into a pre-trained structure recognition model for comprehensive analysis to obtain a property set of each component of the set sewage outlet and construct a digital twin model of the set sewage outlet; a data acquisition and analysis module, which is used to continuously acquire sewage discharge monitoring time series data of the set sewage outlet, wherein the sewage discharge monitoring time series data includes emission intensity time series data and emission image time series data; a comprehensive emission assessment module, which is used to input the sewage discharge monitoring time series data of the set sewage outlet into the digital twin model for feature analysis respectively to obtain the emission behavior risk index and emission visual risk index of the set sewage outlet in each time period, and perform comprehensive analysis to obtain the emission anomaly assessment index of the set sewage outlet in each time period; an off-site enforcement feedback module, which is used to take preset off-site enforcement measures for the set sewage outlet based on the emission anomaly assessment index of each time period.

[0015] The present invention has the following beneficial effects:

[0016] (1) The off-site law enforcement method for sewage outlets based on digital twin technology establishes a real-time binding mechanism between images and data by synchronously inputting the emission images and corresponding emission intensity data of each time period into the digital twin model. In actual operation, if the flow rate of a sewage outlet of an enterprise increases sharply in a certain period of time, and the image captures the water color changing from light gray to dark brown and the foam coverage area rapidly expands, the behavioral characteristics and visual characteristics of the period will be automatically extracted, and the fusion analysis will be carried out to quickly generate the emission anomaly assessment index and link the off-site law enforcement response. The method supports minute-level risk perception and time-level disposal, thereby realizing the pre-identification, automatic response and accurate response of sudden sewage discharge, thereby improving the immediacy and intelligence level of off-site supervision.

[0017] (2) The off-site enforcement method for sewage outlets based on digital twin technology builds a spatial structure model of the sewage outlet site based on three-dimensional point cloud data, and automatically extracts the geometric dimensions, spatial positions and attribute types of each component in the sewage outlet through a pre-trained structure recognition model, thereby realizing rapid modeling from point cloud to component semantics, thereby directly obtaining data from the site to complete the modeling without manual classification. The extracted component attribute set is not only used for geometric reconstruction of the digital scene, but also provides a spatial mapping basis for subsequent image analysis results and pollutant data, so that the entire sewage discharge process is established in a twin scene with precise structure and clear objects, thereby enhancing the spatial correlation of data processing and the reliability of results.

[0018] (3) The off-site law enforcement method for sewage outlets based on digital twin technology realizes the quantitative expression and dynamic classification judgment of the sewage discharge status by constructing an emission anomaly assessment index. In the process of continuously evaluating the emission status of each monitoring period, the corresponding law enforcement processing strategy can be automatically matched based on the change trend and numerical range of the emission anomaly assessment index of each period, so that the response law enforcement measures taken are differentiated, progressive and adjustable. In addition, this method drives law enforcement behavior based on the risk assessment results, so that the law enforcement measures and the sewage discharge risk status form a closed-loop logic, thereby improving the scheduling rationality, law enforcement targeting and response efficiency in off-site scenarios.

[0019] (4) The off-site law enforcement system for sewage outlets based on digital twin technology adopts a modular architecture design to functionally partition and parallelize the three-dimensional modeling, sewage monitoring, risk assessment and law enforcement response processes. Data interaction is carried out between the modules. While maintaining the continuity of the analysis logic, it supports asynchronous access and processing of different data streams, so that the system can realize the structural recognition of three-dimensional point cloud data and the coordinated operation of the digital twin modeling module and the sewage monitoring data analysis module, and form a unified risk judgment result in the comprehensive emission assessment module. Finally, the response mechanism is triggered by the law enforcement feedback module, thereby improving the overall operation efficiency and maintainability of the system, and thus having good task distribution capabilities, computing parallelism and expansion adaptability.

[0020] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 This is a flow chart of the off-site enforcement method for sewage outlets based on digital twin technology of the present invention.

[0022] Figure 2 This is a flowchart of the specific steps for obtaining the emission behavior risk index of each time period of a set sewage outlet in the off-site enforcement method of the sewage outlet based on digital twin technology of the present invention.

[0023] Figure 3 This is a time series diagram of the pollution sudden change risk index of a sewage outlet set in the off-site enforcement method of the sewage outlet based on digital twin technology of the present invention.

[0024] Figure 4 A time series diagram of the pollution transmission risk index of a sewage outlet is set in the off-site enforcement method of the sewage outlet based on digital twin technology of the present invention.

[0025] Figure 5 This is a block diagram of the off-site enforcement system for sewage outlets based on digital twin technology in the present invention. DETAILED DESCRIPTION

[0026] See also Figure 1, an embodiment of the present invention provides a technical solution: an off-site law enforcement method for a sewage outlet based on digital twin technology, comprising the following steps: obtaining three-dimensional point cloud data of a set sewage outlet, and inputting it into a pre-trained structure recognition model for comprehensive analysis to obtain a property set of each component of the set sewage outlet, and constructing a digital twin model of the set sewage outlet; and continuously obtaining sewage discharge monitoring time series data of the set sewage outlet, the sewage discharge monitoring time series data including emission intensity time series data and emission image time series data (i.e., the image of the last time point in the time period); inputting the sewage discharge monitoring time series data of the set sewage outlet into the digital twin model (i.e., the sewage discharge monitoring time series data and the digital twin model are associated and bound in real time, and the sewage discharge process is analyzed time period by time period), performing feature analysis respectively, obtaining the emission behavior risk index and emission visual risk index of each time period of the set sewage outlet (such as the time period interval is 3 minutes), and performing comprehensive analysis to obtain the emission anomaly assessment index of each time period of the set sewage outlet; taking preset off-site law enforcement measures for the set sewage outlet based on the emission anomaly assessment index of each time period.

[0027] Among them, the specific steps of constructing a digital twin model of a set sewage outlet are as follows: structurally initialize the component, and use the geometric dimension attributes (including length, width, thickness, and height) as the shape dimension parameters of the component entity, and the spatial positioning attributes (including the center three-dimensional coordinates and direction vector) as its position and posture parameters in three-dimensional space. Material attributes (including roughness, reflectivity, and material type) are used to set the appearance material performance of the component. Topological relationship attributes (including upstream component ID, downstream component ID, and component group affiliation) are used to establish the structural connection relationship and system grouping identification between components. After completing the component object initialization, it is further visualized and instantiated in a three-dimensional coordinate environment based on the spatial parameters in its attribute set. The central three-dimensional coordinates of each component are used as the geometric center, and based on its corresponding size parameters (such as length, width, and height), a three-dimensional geometric shape model of the component (such as a rectangle, cylinder, or curved surface) is constructed; at the same time, according to the direction vector parameters in the attribute set, a rotation transformation is performed on the model to ensure that its spatial orientation is consistent with the direction of the real component. Subsequently, according to the structure The material type, roughness and reflectivity information of the components are used to bind the corresponding material rendering parameters (such as texture, color, light response coefficient, etc.) to the components to enhance the realism of the twin scene. After the component space instantiation is completed, the logical connection relationship between the components is further constructed. According to the upstream component ID and downstream component ID annotated in the attribute set of each component, directed connections are established between the components to form a component-level topological graph structure; this graph structure is used to express the actual flow direction of the sewage channel, the structural dependency relationship of the components and the network cascade path; at the same time, according to the component group affiliation, multiple components are classified according to function or system, and the above-mentioned sewage outlet digital twin model structure is imported into the digital twin visualization platform to complete operations such as 3D modeling, component binding, state injection and interactive configuration. In the digital twin model, each component has its corresponding unique component ID and spatial positioning. Each monitoring data is mapped and bound to its corresponding component ID. For example, in explicit binding, the data contains a clear component ID field, and in implicit binding, it is bound after matching the spatial position (such as sensor coordinates) with the model component position.

[0028] The specific formula for calculating the discharge anomaly assessment index for a certain period of time at a designated sewage outlet is as follows: Among them, PyC is the emission anomaly assessment index of a certain period of time at the set sewage outlet, PxW is the emission behavior risk index of a certain period of time at the set sewage outlet, α1 is the behavior risk adjustment coefficient stored in the database, SxF is the emission visual risk index of a certain period of time at the set sewage outlet, α2 is the visual risk adjustment coefficient stored in the database, and α3 is the difference adjustment coefficient stored in the database.

[0029] It should be explained that the term [1+ln(1+α3*|PxW-SxF|)] in the formula is used to enhance the sensitivity of the inconsistency between the emission behavior risk index and the emission visual risk index, so as to avoid the emission anomaly assessment index being too high or too low.

[0030] α1, α2, and α3 can be obtained through the following steps: using historical data, combined with the emission behavior risk index and the emission visual risk index, to conduct statistical regression analysis, quantify the specific impact of each factor on the emission anomaly assessment index, and thus fit the initial weight value; secondly, using the sensitivity analysis method, adjust the value range of each coefficient, observe its impact on the emission anomaly assessment results, ensure the stability and rationality of the model, and based on the characteristics of the sewage outlet and the actual situation, correct and optimize the preliminary fitting coefficients, and finally determine the coefficient value applicable to the specific sewage outlet.

[0031] Specifically, the attribute set includes geometric dimension attributes (including length, width, thickness, and height), spatial positioning attributes (including center three-dimensional coordinates and direction vectors), material attributes (including roughness, reflectivity, and material type), and topological relationship attributes (including upstream component ID, downstream component ID, and component group affiliation). The three-dimensional point cloud data includes the voxel value, three-dimensional coordinates, and corresponding reflection intensity value of each voxel point. The structure recognition model is specifically a point cloud deep learning network, and the point cloud deep learning network includes an input layer, a feature extraction layer, a structure recognition layer, a structure feature encoding layer, and an attribute output layer.

[0032] Among them, the input layer is used to receive the pre-processed three-dimensional point cloud data of the set sewage outlet scene, providing an input basis for subsequent feature extraction.

[0033] The feature extraction layer is used to perform center sampling on the input point cloud data, construct a multi-scale spatial neighborhood, and extract the local structural feature vector set of the center sampling point to capture the geometric shape and distribution pattern of each point in the local space.

[0034] The structure recognition layer is used to perform point-level semantic classification and component instance clustering based on the local structure feature vector set of the central sampling point, obtain multiple component category labels and component IDs in the sewage outlet scene, and output the component point cloud set corresponding to each component.

[0035] The structural feature encoding layer is used to perform global feature modeling and semantic encoding on the component point cloud set of each component, generating a structural semantic feature vector containing sub-semantic dimensions such as size, posture, material, and connection tendency, providing input expression for attribute prediction.

[0036] The attribute output layer is used to output the full attribute set of the component based on the multidimensional sub-features in the structural semantic feature vector through the geometric dimension prediction branch, spatial positioning prediction branch, material attribute prediction branch and topological relationship prediction branch, which is used to drive the component construction and semantic binding of the digital twin model.

[0037] The specific steps of obtaining the attribute set of each component of the set sewage outlet are as follows: in the input layer of the point cloud deep learning network, the three-dimensional point cloud data of the set sewage outlet (i.e., the voxel value, three-dimensional coordinates and corresponding reflection intensity value of each voxel point) are received and preprocessed; in the feature extraction layer of the point cloud deep learning network, the preprocessed three-dimensional point cloud data of the set sewage outlet are subjected to feature extraction processing (center sampling operation, using the farthest point sampling method to select several representative center sampling points from the entire point set, and then, with each center sampling point as the benchmark, a multi-scale spatial neighborhood is constructed within a preset radius around it to form multiple local point sets. The neighborhood scale can be set to multiple different radius values, such as 0.1m, 0.3m, 0.5m, to take into account both microscopic local structural features and macroscopic geometric contour features. Then, for the point set in each neighborhood, the original geometric features such as the relative position coordinates, point density, spatial distance, etc. of each neighborhood point relative to the center sampling point are calculated, and these relative feature information is used as input and sent to the shared multi-layer perceptron for nonlinear embedding processing to obtain The primary structural expression vector of each neighborhood point is then aggregated for the structural embedding vectors of all points in each neighborhood, using operations such as maximum pooling, average pooling, or attention-weighted aggregation to fuse the structural information of multiple points into a single-scale local structural feature vector. The structural vectors extracted at multiple scales are further combined into a multi-scale structural feature vector of the central sampling point by splicing or weighted fusion to fully express the local geometric semantic features of the sampling point at different spatial scales, thereby obtaining a set of local structural feature vectors for each central sampling point of the set sewage outlet. In the structure recognition layer of the point cloud deep learning network, the set of local structural feature vectors of each central sampling point of the set sewage outlet is fused and recognized (the local structural feature vectors of all central sampling points are input into the semantic classification submodule of the structure recognition layer, and each feature vector is nonlinearly transformed using a multi-layer perceptron, and the probability of each central sampling point belonging to all preset component categories, such as sewage main pipes, water outlet boundaries, manhole covers, fence fragments, etc., is calculated using the Softmax function.According to the category with the maximum probability, the component category label of each central sampling point is determined to complete the point-level component semantic classification. Subsequently, the central sampling points with the same component category label are clustered into a candidate component point set. In combination with the similarity of their spatial position coordinates and semantic features, density clustering, region growing algorithm based on feature similarity, or adjacency connectivity clustering method based on graph structure are used to divide component instances, ensuring that multiple independent component units that are not connected to each other in space can be identified within the same category. On this basis, a unique component instance number, i.e., component ID, is assigned to each identified component, and the component is bound to its category label, and the components belonging to the same component instance are grouped together. The original neighborhood point sets corresponding to all the central sampling points of the example number are combined to restore the complete point cloud set corresponding to the component as the geometric expression data of the component), and the component point cloud sets of several components of the set sewage outlet are obtained (it should be noted here that the component has been identified as the component category label); in the structural feature encoding layer of the point cloud deep learning network, the component point cloud set of each component of the set sewage outlet is subjected to feature encoding processing (the component point cloud set is spatially standardized, specifically including centering the point cloud coordinates, translating the coordinates of all points to the position with the component center of gravity as the origin, and performing scale normalization at the same time to make different components uniform in physical size). The differences in the above are uniformly processed; this step provides a basis for geometric consistency for the subsequent extraction of size and morphological features. Subsequently, each point in the component point cloud is input into a shared multi-layer perceptron network to extract preliminary point-level embedding features. In this process, the network automatically constructs a size and morphological sub-feature vector that is closely related to the component's geometric size and boundary structure by capturing the local density changes, edge gradients, and relative position patterns of the points inside the component. Then, a spatial connection graph is constructed inside the component point cloud, and a graph structure is constructed with the Euclidean distance between points or the local area density as the edge weight. The relative position relationship between points is modeled using a graph convolution module, such as EdgeConv or GAT. At this stage, the network particularly enhances direction-related geometric morphological distribution information to form sub-feature vectors that express the component's main axis direction, rotation trend, and symmetrical structure. At the same time, the reflection intensity value and local roughness of each point in the point cloud can be estimated by the normal vector change rate and point density as additional channel inputs into the parallel material feature encoding path. A lightweight convolutional network is used to extract material-related sub-vectors to characterize the component's surface reflectivity, material texture characteristics, and roughness trends. In addition, based on the component structure diagram, a component boundary point set is further constructed, and its boundary relationship with other components is spatially adjacent to identify possible structural connection point pairs.Based on parameters such as connection direction, spatial fit and point cloud contact density, a component connection tendency sub-feature vector is formed to lay the foundation for subsequent topological relationship reasoning. Finally, the structural dimension feature vectors extracted from each sub-path are spliced and fused according to the preset encoding rules to form the global structural semantic feature vector of the component) and the global structural semantic feature vector of each component of the set sewage outlet is obtained; in the attribute output layer of the point cloud deep learning network, the global structural semantic feature vector of each component of the set sewage outlet is subjected to branch prediction processing (the global structural semantic feature vector corresponding to each component is input into the multi-layer of the attribute output layer) In the sub-branch layer, the global structural semantic feature vector has been divided into size and shape sub-vectors, posture and direction sub-vectors, material feature sub-vectors and connection relationship sub-vectors in the structural feature encoding layer. Each sub-branch selects the corresponding feature dimension for decoding operation. For geometric dimension attributes, the size and shape sub-vector part in the structural semantic feature vector is received and input into the multi-layer perceptron regression network to predict the four geometric dimension indicators of the length, width, thickness and height of the component in turn. In the prediction process, the network focuses on using the information such as the spatial ductility of the point cloud, the boundary density distribution and the main axis direction projection in the features to complete the prediction of the component. The three-dimensional size of the component is estimated. For the spatial positioning attribute, the attitude direction sub-vector part is received to calculate the three-dimensional coordinates and direction vector of the component's three-dimensional geometric center. The three-dimensional coordinate position of the component's center is first regressed through the MLP network. Then, the three components of the direction vector are further regressed and L2 normalized to represent the main axis orientation direction of the component. This is used to assist in the posture restoration in the twin model. For the material attribute, the material feature sub-vector part is received to predict the material related attributes. The MLP regression network is used to output the average surface roughness value of the component, that is, based on the normal change rate feature, the component reflection is output. The estimated value of the reflection intensity distribution pattern is then generated. Finally, the Softmax classification head is called to classify the material type, such as concrete, cast iron, or HDPE, to form a material label. For topological relationship attributes, the connection relationship subvector is received and the upstream and downstream component IDs of the component are first predicted based on spatial adjacency and directional consistency. Then, the component group ID to which the component belongs is determined by combining the preset spatial clustering labels and the identification rule library. This results in the geometric dimension attributes, spatial positioning attributes, material attributes, and topological relationship attributes of each component in the designated sewage outlet, i.e., the attribute set.

[0038] And the pre-training process of the point cloud deep learning network is as follows:

[0039] Obtain annotated point cloud sample datasets, including 3D point cloud data (coordinates, intensity, voxel value of each point, etc.), point-level or component-level labels (including component category, component ID, attribute annotation fields), topological connection information and material fields. All point cloud data must undergo unified preprocessing operations, including downsampling, normalization, coordinate alignment and boundary filtering, to ensure the consistency of the input data structure, and divide them into point cloud training set and point cloud verification set.

[0040] Initialize the point cloud deep learning network, including input dimension, number of samples, multi-scale neighborhood radius, feature embedding layer dimension, graph convolution kernel size, etc., and set training configuration parameters, such as optimizer type (such as Adam), initial learning rate, learning rate decay strategy, loss function (including semantic classification loss, instance segmentation loss, attribute regression loss, etc.).

[0041] Training is performed based on the point cloud training set, and the number of training cycles is set (such as 50-100 times). In each training cycle, forward propagation is performed in sequence (the training data is input into the point cloud deep learning network, and passes through the input layer, feature extraction layer, structure recognition layer, feature encoding layer and attribute prediction layer in sequence to generate prediction output), loss function calculation (calculation of multiple losses between the prediction results and the true labels, including point-level semantic classification loss, such as cross entropy; component instance segmentation loss, such as clustering consistency loss, component attribute regression loss, such as L1 / L2 loss, component topology consistency loss, such as structural constraint loss), backpropagation and parameter update (using the chain gradient backpropagation mechanism to backpropagate and update each parameter in the network to minimize the total loss of the current round).

[0042] After each training cycle, an evaluation analysis is performed based on the point cloud validation set. That is, the point cloud validation set is input into the model parameter state obtained in the current training round, forward reasoning is performed, the output result is obtained, and the loss value between the actual result and the actual result is calculated. The loss function value and accuracy of the model on the validation set are also calculated to evaluate the performance of the model. If the performance of the model on the validation set is improved, the current parameters are retained as the best model; if the validation index does not improve or overfitting occurs (high training accuracy and low validation accuracy), the learning rate, regularization term, loss weight, etc. are adjusted.

[0043] When the training is completed and the loss and accuracy on the validation set reach the expected standards, the training process ends and a trained network model is obtained.

[0044] In this implementation plan, the structural modeling and attribute recognition steps of the point cloud deep learning network have significant advantages such as high precision, high adaptability and high automation. By preprocessing, center sampling and multi-scale spatial neighborhood construction of the three-dimensional point cloud data of the sewage outlet, it is possible to comprehensively capture the local geometric features and overall spatial form of each component, and take into account the semantic expression of micro-boundaries and macro-structures. Secondly, after feature extraction, through semantic classification and component instance clustering, it can effectively achieve accurate division and component ID identification of multiple types of components in complex scenes, avoiding component misjudgment or omission. In addition, in the structural feature encoding and attribute output stage, multi-path sub-feature learning mechanisms such as geometric size, orientation and posture, material reflection, and structural connection are introduced to ensure that the attribute expression of each component is comprehensive, multi-dimensional and semantically clear. Finally, the entire training process uses standardized data sets, loss function branch design and cross-validation mechanism to ensure that the model has good generalization ability and robustness in component identification and attribute prediction, thereby significantly improving the accuracy, integrity and automation level of digital twin model construction.

[0045] Specifically, if Figure 2 As shown, the emission intensity time series data include the total amount of suspended solids, pH value, pollution load index, sewage outlet vibration amplitude value, flow velocity fluctuation index, pollution diffusion index and pollution concentration value of each pollutant (including but not limited to organic pollutants, ammonia nitrogen, etc.), concentration change rate value in each time period. The specific steps to obtain the emission behavior risk index of the set sewage outlet in each time period are as follows: conduct a comprehensive analysis on the emission intensity time series data of the set sewage outlet (in the digital twin model, the emission intensity time series data bound to the digital twin model will be comprehensively analyzed) to obtain the emission risk assessment index set for each time period of the set sewage outlet, including the pollution surge risk index (used to measure the degree of violent fluctuation or loss of control of pollution behavior) and the pollution transmission risk index; and conduct a comprehensive analysis on the emission risk assessment index set for each time period of the set sewage outlet to obtain the emission behavior risk index for each time period of the set sewage outlet.

[0046] Among them, the total suspended matter is the concentration of undissolved particulate matter in water, reflecting the level of particle pollution, which can be obtained by an optical suspended matter sensor.

[0047] The pH value is the acidity and alkalinity level of water. Extreme values can easily affect the ecosystem or pipeline corrosion. It can be obtained through a pH sensor.

[0048] The pollution load index is the output intensity of pollutants during the period. It can be obtained by obtaining the cross-sectional area of the sewage pipe (which can be obtained through the pipeline design data stored in the database), the flow rate value at each time point (which can be obtained through the flow rate sensor), and the total concentration value (that is, the sum of the concentration values of each pollutant, which can be obtained through weighted processing, such as COD, ammonia nitrogen, etc., and the concentration value of each pollutant can be obtained through an online water quality analyzer), and multiplying them. The average processing is performed based on the multiplication result, and the result is the pollution load index.

[0049] The vibration amplitude value of the sewage outlet is the structural vibration response amplitude generated by the sewage outlet structure during the discharge process. The vibration amplitude value at each time point can be obtained through a vibration sensor and averaged. The result is the vibration amplitude value of the sewage outlet.

[0050] The flow velocity fluctuation index is the degree of change in the drainage flow rate during the period. It can be obtained by performing mean processing and standard deviation processing on the flow velocity value at each time point during the period, and then performing a ratio analysis based on the standard deviation processing result and the mean processing result, that is, standard deviation processing result / mean processing result. The result obtained is the flow velocity fluctuation index.

[0051] The pollution diffusion index is the potential intensity of the diffusion of pollutants from the sewage outlet to the sewage pool. It can be obtained by obtaining the pollution concentration gradient value, the sewage discharge distance value (the distance between the sewage outlet and the sewage pool, based on the central three-dimensional coordinates of the sewage outlet and the central three-dimensional coordinates of the sewage pool in the spatial positioning attributes, and calculated using the Euclidean distance formula), the flow rate value (that is, the average of the flow rate at each time point within the period), and performing normalization. Based on the normalization result, weighted processing is performed, and the result is the pollution diffusion index, and the pollution concentration gradient value, that is, the average of the difference between the pollution concentration of the sewage outlet and the sewage pool (the sum of the concentrations of each pollutant) at each time point within the period, and the pollution concentration of the sewage outlet and the pollution concentration of the sewage pool at each time point can be obtained through the online water quality analyzer.

[0052] The pollution concentration value is the average of the concentration values at each time point during the period, and the concentration value at each time point can be obtained through an online water quality analyzer (such as COD, NH3-N sensor).

[0053] The concentration change rate value is the rate of change of pollutant concentration, which reflects the trend of pollutant release. It is obtained by obtaining the concentration value at each time point in the period, reading the concentration values at the first time point and the last time point, and performing ratio processing (the absolute value of the difference between the concentration values at the first time point and the last time point / the concentration value at the last time point). The result is the concentration change rate value.

[0054] The specific steps for obtaining the emission risk assessment index set for each time period of the set sewage outlet are as follows: obtain the concentration reference value of each pollutant of the set sewage outlet, and conduct a comprehensive analysis based on the pollution concentration value of each pollutant in each time period to obtain the pollution deviation index of each time period of the set sewage outlet; read the total amount of suspended solids, pH value, sewage outlet vibration amplitude value and concentration change rate value of each pollutant in each time period of the set sewage outlet, and conduct a comprehensive analysis based on the pollution deviation index (i.e., first perform weighted average processing on the concentration change rate value of each pollutant to obtain the comprehensive concentration change rate value, and standardize it with the total amount of suspended solids, pH value, pollution output value, and pollution deviation index, and perform weighted processing based on the standardized processing result) to obtain the pollution sudden change risk index of each time period of the set sewage outlet; read the pollution load index, flow velocity fluctuation index, and pollution diffusion index of each time period of the set sewage outlet, and conduct a comprehensive analysis (i.e., first perform standardization processing, and perform weighted processing based on the standardized processing result) to obtain the pollution transmission risk index of each time period of the set sewage outlet.

[0055] The specific formula for calculating the discharge behavior risk index for a certain period of time at a designated sewage outlet is as follows: Among them, PxW is the emission behavior risk index of the set sewage outlet in a certain period of time, WrB is the pollution shock risk index of the set sewage outlet in a certain period of time, η1 is the shock adjustment coefficient stored in the database, WcF is the pollution propagation risk index of the set sewage outlet in a certain period of time, η2 is the propagation adjustment coefficient stored in the database, and η3 is the interaction adjustment coefficient stored in the database.

[0056] It needs to be explained that the formula This item is used to measure the behavioral coupling strength between the pollution shock risk index and the pollution transmission risk index. It can achieve nonlinear enhancement of risk perception when both types of risk indicators are at high levels and the differences are small, thereby improving the response sensitivity of the emission behavior risk index to abnormal linkage situations.

[0057] η1, η2, and η3 can be obtained through the following steps: Based on historical data, determine the initial impact weights of each variable (pollution shock risk index and pollution transmission risk index) on the emission behavior risk index through statistical regression analysis. Then, use the sensitivity analysis method to adjust the value range of the coefficient to evaluate the stability and applicability of these parameters to the formula output. Next, further fit the weights through model optimization (such as machine learning algorithms or) to ensure that the formula can accurately reflect the status of actual emission behavior risks.

[0058] The specific implementation example of calculating the emission behavior risk index of a certain period of time at a set sewage outlet is as follows. The existing data are as follows: including the pollution shock risk index and pollution transmission risk index of the five periods of time at the set sewage outlet, as shown in Table 1 and Figure 3-4 :

[0059] Table 1 Example of time series data of emission risk assessment index set for sewage outlets:

[0060] Pollution shock risk index Pollution Transmission Risk Index Period 1 0.316 0.274 Period 2 0.327 0.278 Period 3 0.319 0.283 Period 4 0.334 0.275 Period 5 0.329 0.288

[0061] The catastrophic adjustment coefficient η1 stored in the database is approximately: 0.648;

[0062] The propagation adjustment coefficient η2 stored in the database is approximately: 0.716;

[0063] The interaction adjustment coefficient η3 stored in the database is approximately: 0.584;

[0064] Substituting the data in Table 1 and the above adjustment coefficient into the specific formula for calculating the discharge behavior risk index for a certain period of time at a given sewage outlet, we obtain:

[0065] The discharge behavior risk index of the first period of the sewage outlet is set as ln(1+√(0.316 0.648 ×0.274 0.716 ))×(1+0.548×(exp(-0.316×0.274)) / (1+|0.316-0.274|))≈0.545;

[0066] The discharge behavior risk index of the second period of the sewage outlet is set to be ln(1+√(0.327 0.648 ×0.278 0.716 ))×(1+0.548×(exp(-0.327×0.278)) / (1+|0.327-0.278|))≈0.550;

[0067] The discharge behavior risk index of the third period of the sewage outlet is set as ln(1+√(0.319 0.648 ×0.283 0.716 ))×(1+0.548×(exp(-0.319×0.283)) / (1+|0.319-0.283|))≈0.551;

[0068] The discharge behavior risk index of the fourth period of the sewage outlet is set as ln(1+√(0.334 0.648 ×0.275 0.716 ))×(1+0.548×(exp(-0.334×0.275)) / (1+|0.334-0.275|))≈0.549;

[0069] The discharge behavior risk index of the fifth period of the sewage outlet is set as ln(1+√(0.329 0.648 ×0.2880.716 ))×(1+0.548×(exp(-0.329×0.288)) / (1+|0.329-0.288|))≈0.557.

[0070] In this implementation plan, through the structural decomposition and coupling modeling of the pollution shock risk index and the pollution transmission risk index, a two-dimensional accurate characterization of the abnormal trend and spatial diffusion trend of pollution discharge behavior is achieved, which has the advantages of fast response speed, fine recognition granularity, and accurate risk perception. Secondly, in the process of index construction, a number of time series parameters with actual physical significance are introduced, such as vibration amplitude, concentration change rate, pollution concentration gradient, etc., to avoid the distortion caused by a single concentration indicator. At the same time, the nonlinear interaction terms of the pollution shock and transmission index are integrated into the formula design, which can produce an exponential amplification effect under the linkage situation where the two increase at the same time but the difference is not large, so as to improve the identification ability of hidden behaviors such as sudden strong discharge and continuous stealth discharge. Finally, through the regression weighting and sensitivity analysis mechanism based on historical data, the adjustment coefficient is made adjustable and controllable, thereby improving the adaptability and robustness of the formula in different scenarios, and then providing a highly reliable mathematical basis and decision-making support for the quantification of pollution discharge behavior risks.

[0071] Specifically, the emission image time series data includes the emission image data of each time period. The emission image data is specifically the pixel value and two-dimensional coordinates of each pixel point in the emission image. The specific steps for obtaining the emission visual risk index of each time period of the set sewage outlet are as follows: comprehensively analyze the pixel value and two-dimensional coordinates of each pixel point in the emission image of each time period of the set sewage outlet to obtain the image perception anomaly index set of each time period of the set sewage outlet, including the effluent color deviation index, foam coverage index, and edge disturbance index; comprehensively analyze the image perception anomaly index set of each time period of the set sewage outlet to obtain the emission visual risk index of each time period of the set sewage outlet.

[0072] Among them, the specific steps of obtaining the image perception anomaly index set of each time period of the set sewage outlet are as follows: read the pixel value of each pixel in the emission image of each time period of the set sewage outlet, and extract the corresponding several color components, namely hue (H), saturation (S) and brightness (V) components, and perform mean processing on them respectively to obtain the mean value of each color component of each pixel in the emission image of each time period of the set sewage outlet, and obtain the reference value of each color component of each pixel in the emission image of the set sewage outlet (by obtaining the historical mean value of each color component of each pixel in the emission image of the set sewage outlet at several historical time points, and performing mean processing), and combining the mean value of each pixel in the emission image of each time period. The mean value of each color component is comprehensively analyzed (calculated and analyzed based on the Euclidean distance formula) to obtain the effluent color shift index of each period of the set sewage outlet; the pixel value of each pixel point in the discharge image of each period of the set sewage outlet is read and grayscale processing is performed to obtain the grayscale pixel value of each pixel point in the discharge image of each period of the set sewage outlet, and edge detection processing is performed (based on image gradient algorithms, such as Sobel, Laplacian or Canny edge detection operators, edge detection processing is performed on the grayscale pixel values in the image, and the edge lines of the areas with drastic brightness changes are extracted, and then the areas with obvious contours or brightness boundaries in the image are located, and the set brightness threshold is combined with the regional connectivity). The foam coverage index of each time period of the set sewage outlet is obtained by performing a comprehensive analysis (performing a statistical analysis on each pixel point in each foam area, and performing a ratio processing based on the statistical analysis result and the total pixel points of the emission image). The grayscale pixel value of each pixel point in the emission image of each time period of the set sewage outlet is read, and edge detection processing is performed (after graying the emission image, the edge extraction operation is performed by the Canny operator, including Gaussian filtering denoising, gradient calculation, non-maximum suppression, etc.). The edge map of the image is obtained by performing processing steps such as multi-threshold tracking and double threshold tracking; and the edge image is regionally screened in combination with the preset water area mask, and finally the edge contour line of the corresponding water area in the image is extracted to obtain the edge contour line of the water area in the discharge image of each time period of the set sewage outlet, and the two-dimensional coordinates of each pixel point on the edge contour line of the water area in the discharge image of each time period of the set sewage outlet are read, and the horizontal coordinates in the two-dimensional coordinates are sorted to construct a water body edge curve, and the vertical coordinates in the two-dimensional coordinates of each pixel point on the water body edge curve (the pixel points that have been sorted by the horizontal coordinates) are read, and the standard deviation processing is performed to obtain the edge disturbance index of each time period of the set sewage outlet.

[0073] The specific formula for calculating the emission visual risk index of a given sewage outlet during a certain period of time is as follows: Among them, SxF is the discharge visual risk index of the set sewage outlet in a certain period of time, YsP is the effluent color shift index of the set sewage outlet in a certain period of time, λ1 is the color shift adjustment coefficient stored in the database, PmF is the foam coverage index of the set sewage outlet in a certain period of time, λ2 is the foam coverage adjustment coefficient stored in the database, ByR is the edge disturbance index of the set sewage outlet in a certain period of time, λ3 is the edge disturbance adjustment coefficient stored in the database, and λ4 is the superposition adjustment coefficient stored in the database.

[0074] It should be explained that λ1, λ2, λ3, and λ4 can be obtained through the following steps: Based on historical data, the initial impact weights of each variable (effluent color shift index, foam coverage index, edge disturbance index) on the emission visual risk index are determined through statistical regression analysis. Then, the value range of the coefficient is adjusted using the sensitivity analysis method to evaluate the stability and applicability of these parameters to the formula output. Next, the weights are further fitted through model optimization (such as machine learning algorithms or) to ensure that the formula can accurately reflect the status of actual emission visual risk, and the coefficients are fine-tuned based on the characteristics of different time periods to ensure that they are suitable for specific emission visual risk assessment needs.

[0075] In this implementation scheme, based on the pixel-level image analysis mechanism, multiple visual perception dimensions such as color offset, foam coverage and edge disturbance are integrated to achieve high-resolution, low-latency, non-contact risk monitoring of the sewage discharge process. Secondly, by extracting color parameters such as hue, saturation and brightness and combining them with the historical image mean, slight changes in water color can be accurately identified, and abnormal water conditions can be quickly responded to. In addition, a foam extraction algorithm based on image gradient and connected area recognition is introduced to improve the recognition ability of high-foam sewage discharge behavior, and the edge disturbance standard deviation calculation is used to effectively capture the water surface disturbance trend, thereby assisting in the identification of flow rate anomalies or intermittent discharge phenomena. Finally, in the index fusion layer, adjustment coefficients and interaction enhancement factors are introduced to ensure enhanced response capabilities when multiple feature anomalies occur simultaneously in the image, and all adjustment coefficients are obtained through historical data regression analysis, sensitivity testing and model optimization, thereby effectively improving the accuracy, sensitivity and time-varying adaptability of the visual risk index.

[0076] Specifically, the specific steps for taking preset non-site enforcement measures for the set sewage outlets based on the emission anomaly assessment index of each time period are as follows: a comprehensive analysis is conducted on the emission anomaly assessment index of each time period of the set sewage outlet, that is, based on the emission anomaly assessment index of the previous time periods (such as 3 time periods), a time series weighted vector is constructed, and a weighted sliding average is performed on it to obtain the abnormal trend base value of the set sewage outlet, and the standard deviation value of the emission anomaly assessment index of each time period of the set sewage outlet is calculated to characterize the short-term fluctuation degree of the risk index, and the emission anomaly assessment index of the current time period is compared with the emission anomaly assessment index of the previous time period. The change analysis is performed on the data (i.e., the difference between the emission anomaly assessment index of the current period and the emission anomaly assessment index of the previous period / the emission anomaly assessment index of the previous period), and then the change analysis and the standard deviation are weighted, and the weighted processing result of the abnormal trend base value is coupled, i.e., the abnormal trend base value × (1 + weighted processing result), to form the predicted output value of the emission anomaly assessment index of the next period, and regard it as the emission anomaly assessment index of the next period to obtain the emission anomaly assessment index of the set sewage outlet for the next period; the emission anomaly assessment index of the set sewage outlet for the next period is compared with the preset emission anomaly assessment index. The threshold interval is used for judgment and analysis; if the emission anomaly assessment index of the next time period of the sewage outlet is set to be lower than the lower limit of the preset emission anomaly assessment index threshold interval, the first law enforcement measure is taken, which is to mark the next time period as a low-risk state, and keep the current monitoring frequency and data acquisition cycle unchanged, without the need to allocate additional computing resources and law enforcement warning channels, and only record the predicted value for trend tracking and subsequent model updates; if the emission anomaly assessment index of the next time period of the sewage outlet is within the preset emission anomaly assessment index threshold interval, the second law enforcement measure is taken, which is to include the next time period in the key focus period and deploy in advance Computing resources are used to improve the frequency of image recognition and data analysis, and at the same time, prepare to enter the event warm-up mode, that is, push intermediate risk prediction prompts to the supervision platform so that supervisors can pay attention to the emission performance of this period in advance; if the emission anomaly assessment index of the next period of the sewage outlet is set to be higher than the upper limit of the preset emission anomaly assessment index threshold range, the third law enforcement measure will be taken, which is to start the early warning level response mechanism, that is, generate a high-risk early warning work order in advance, dispatch the remote camera to perform the field picture collection task of the next period, and pre-lock the monitoring data channel of this period to ensure that once the actual emission is abnormal, a closed-loop response chain can be formed in the first time.

[0077] In this implementation plan, by constructing a time series prediction model based on the emission anomaly assessment index, the early perception of the pollution risk in the next period and the pre-configuration of graded responses are achieved, which has significant advantages such as strong predictability, pre-response, and strategy adaptation. Secondly, by constructing a weighted sliding vector for the emission anomaly assessment index of the first three periods, the trend base value and the fluctuation intensity are integrated to form a dynamic coupling prediction, so that the potential level of the risk can be accurately judged before the actual risk occurs, and the corresponding resource scheduling strategy can be matched, thereby avoiding the problem of delayed response in supervision. Finally, through differentiated law enforcement preparation paths, the transformation of non-site law enforcement from passive perception to active pre-control is achieved, thereby improving the intelligent supervision capability and resource allocation efficiency.

[0078] See also Figure 5 , an embodiment of the present invention provides a technical solution: an off-site enforcement system for sewage outlets based on digital twin technology, including: a data acquisition and modeling module, used to acquire three-dimensional point cloud data of a set sewage outlet, and input it into a pre-trained structure recognition model for comprehensive analysis, to obtain a property set of each component of the set sewage outlet, and to construct a digital twin model of the set sewage outlet; a data acquisition and analysis module, used to continuously acquire sewage discharge monitoring time series data of the set sewage outlet, the sewage discharge monitoring time series data including emission intensity time series data and emission image time series data; a comprehensive emission assessment module, used to input the sewage discharge monitoring time series data of the set sewage outlet into the digital twin model for feature analysis respectively, to obtain the emission behavior risk index and emission visual risk index of the set sewage outlet in each time period, and to perform comprehensive analysis to obtain the emission anomaly assessment index of the set sewage outlet in each time period; an off-site enforcement feedback module, used to take preset off-site enforcement measures for the set sewage outlet based on the emission anomaly assessment index of each time period.

[0079] Although the preferred embodiments of the present invention have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0080] Obviously, those skilled in the art may make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if such changes and modifications fall within the scope of the claims and their equivalents, the present invention is intended to include such changes and modifications.

Claims

1. The off-site enforcement method for sewage outlets based on digital twin technology is characterized by: The following steps are involved: Obtain the 3D point cloud data of the designated sewage outlet and input it into the pre-trained structure recognition model for comprehensive analysis to obtain the attribute set of each component of the designated sewage outlet and construct a digital twin model of the designated sewage outlet; and continuously acquiring sewage discharge monitoring time series data of the set sewage outlet, wherein the sewage discharge monitoring time series data includes discharge intensity time series data and discharge image time series data; Input the sewage discharge monitoring time series data of the set sewage outlet into the digital twin model for feature analysis, and obtain the discharge behavior risk index and discharge visual risk index of the set sewage outlet in each time period. A comprehensive analysis is then performed to obtain the discharge anomaly assessment index of the set sewage outlet in each time period. Based on the emission anomaly assessment index for each time period, preset off-site enforcement measures are taken for the designated sewage outlets.

2. The off-site enforcement method for sewage outlets based on digital twin technology according to claim 1 is characterized in that: The attribute set includes geometric size attributes, spatial positioning attributes, material attributes, and topological relationship attributes. The three-dimensional point cloud data includes the voxel value, three-dimensional coordinates, and corresponding reflection intensity value of each voxel point. The structure recognition model is specifically a point cloud deep learning network, and the point cloud deep learning network includes an input layer, a feature extraction layer, a structure recognition layer, a structure feature encoding layer, and an attribute output layer.

3. The off-site enforcement method for sewage outlets based on digital twin technology according to claim 2 is characterized in that: The specific steps to obtain the attribute set of each component of the sewage outlet are as follows: In the input layer of the point cloud deep learning network, the three-dimensional point cloud data of the set sewage outlet is received and preprocessed; In the feature extraction layer of the point cloud deep learning network, feature extraction is performed on the pre-processed three-dimensional point cloud data of the set sewage outlet to obtain a local structural feature vector set for each central sampling point of the set sewage outlet; In the structure recognition layer of the point cloud deep learning network, the local structure feature vector set of each central sampling point of the set sewage outlet is fused and recognized to obtain a component point cloud set of several components of the set sewage outlet; In the structural feature encoding layer of the point cloud deep learning network, feature encoding processing is performed on the component point cloud set of each component of the set sewage outlet to obtain the global structural semantic feature vector of each component of the set sewage outlet; In the attribute output layer of the point cloud deep learning network, branch prediction processing is performed on the global structural semantic feature vector of each component of the set sewage outlet to obtain the geometric size attributes, spatial positioning attributes, material attributes, and topological relationship attributes of each component of the set sewage outlet, that is, the attribute set.

4. The off-site enforcement method for sewage outlets based on digital twin technology according to claim 1 is characterized in that: The specific formula for calculating the discharge anomaly assessment index for a certain period of time at a designated sewage outlet is as follows: Among them, PyC, PxW, and SxF are the emission anomaly assessment index, emission behavior risk index, and emission visual risk index of a certain period of time at the set sewage outlet, respectively; α1, α2, and α3 are the behavior risk adjustment coefficient, visual risk adjustment coefficient, and difference adjustment coefficient stored in the database, respectively.

5. The off-site enforcement method for sewage outlets based on digital twin technology according to claim 1 is characterized in that: The emission intensity time series data includes the total amount of suspended solids, pH value, pollution load index, sewage outlet vibration amplitude value, flow rate fluctuation index, pollution diffusion index, and pollution concentration value and concentration change rate value of each pollutant in each time period. The specific steps for obtaining the emission behavior risk index of each time period for the set sewage outlet are as follows: Comprehensively analyze the emission intensity time series data of the designated sewage outlets to obtain the emission risk assessment index set for each period of the designated sewage outlets, including the pollution shock risk index and the pollution transmission risk index; A comprehensive analysis is conducted on the emission risk assessment index set for each time period of the set sewage outlet to obtain the emission behavior risk index for each time period of the set sewage outlet.

6. The off-site enforcement method for sewage outlets based on digital twin technology according to claim 5 is characterized in that: The specific steps to obtain the emission risk assessment index set for each period of the set sewage outlet are as follows: Obtain the concentration reference value of each pollutant at the set sewage outlet, and conduct a comprehensive analysis based on the pollution concentration value of each pollutant in each time period to obtain the pollution deviation index for each time period at the set sewage outlet; Read the total amount of suspended solids, pH value, vibration amplitude value of the sewage outlet and concentration change rate value of each pollutant in each period of the set sewage outlet, and conduct a comprehensive analysis in combination with the pollution deviation index to obtain the pollution sudden change risk index of each period of the set sewage outlet; The pollution load index, flow velocity fluctuation index and pollution diffusion index of each time period of the set sewage outlet are read, and a comprehensive analysis is performed to obtain the pollution transmission risk index of each time period of the set sewage outlet.

7. The off-site enforcement method for sewage outlets based on digital twin technology according to claim 5 is characterized in that: The specific formula for calculating the discharge behavior risk index for a certain period of time at a designated sewage outlet is as follows: Among them, PxW, WrB, and WcF are the emission behavior risk index, pollution shock risk index, and pollution transmission risk index of a certain period of time at the set sewage outlet, respectively; η1, η2, and η3 are the shock adjustment coefficient, transmission adjustment coefficient, and interaction adjustment coefficient stored in the database, respectively.

8. The off-site enforcement method for sewage outlets based on digital twin technology according to claim 1 is characterized in that: The emission image time series data includes emission image data for each time period. The emission image data is specifically the pixel value and two-dimensional coordinates of each pixel point in the emission image. The specific steps for obtaining the emission visual risk index for each time period of the set sewage outlet are as follows: Comprehensively analyze the pixel value and two-dimensional coordinates of each pixel point in the discharge image of each time period of the set sewage outlet, and obtain the image perception anomaly index set of each time period of the set sewage outlet, including the effluent color deviation index, foam coverage index, and edge disturbance index; A comprehensive analysis is performed on the image perception anomaly index set of each time period of the set sewage outlet to obtain the emission visual risk index of each time period of the set sewage outlet.

9. The off-site enforcement method for sewage outlets based on digital twin technology according to claim 1 is characterized in that: The specific steps for taking pre-set off-site enforcement measures for designated sewage outlets based on the discharge anomaly assessment index for each period are as follows: Comprehensively analyze the discharge anomaly assessment index of each period of the set sewage outlet to obtain the discharge anomaly assessment index of the next period of the set sewage outlet; Conduct judgment and analysis on the discharge anomaly assessment index of the next period of the set sewage outlet and the preset discharge anomaly assessment index threshold range; If the discharge anomaly assessment index of the next period of the designated sewage outlet is lower than the lower limit of the preset discharge anomaly assessment index threshold range, the first enforcement measure is taken; If the discharge anomaly assessment index of the next period of the designated sewage outlet is within the preset discharge anomaly assessment index threshold range, the second enforcement measure will be taken; If the discharge anomaly assessment index of the next period of the sewage outlet is set to be higher than the upper limit of the preset discharge anomaly assessment index threshold range, the third law enforcement measure will be taken.

10. A sewage outlet off-site enforcement system based on digital twin technology, applying the sewage outlet off-site enforcement method based on digital twin technology according to any one of claims 1 to 9, characterized in that: include: The data acquisition and modeling module is used to obtain the three-dimensional point cloud data of the set sewage outlet and input it into the pre-trained structure recognition model for comprehensive analysis to obtain the attribute set of each component of the set sewage outlet and build a digital twin model of the set sewage outlet; A data acquisition and analysis module is used to continuously acquire the sewage discharge monitoring time series data of the set sewage discharge outlet, wherein the sewage discharge monitoring time series data includes the discharge intensity time series data and the discharge image time series data; The comprehensive emission assessment module is used to input the sewage monitoring time series data of the set sewage outlet into the digital twin model for feature analysis, obtain the emission behavior risk index and emission visual risk index of each time period of the set sewage outlet, and conduct a comprehensive analysis to obtain the emission anomaly assessment index of each time period of the set sewage outlet; The off-site enforcement feedback module is used to take preset off-site enforcement measures for the set sewage outlets based on the emission anomaly assessment index for each time period.

Images