A method and system for intelligent monitoring and treatment of crystalline deposits in a drain

Abstract

This invention relates to the field of smart water management and pipeline operation and maintenance technology, and discloses a method and system for intelligent monitoring and treatment of crystallization deposition in drainage pipes. The method includes: collecting multimodal sensor data of drainage pipes and performing spatiotemporal alignment preprocessing; detecting and identifying crystallization deposition areas and types based on images of the pipe's inner wall; statistically analyzing the spatial distribution characteristics of crystallization areas and conducting multi-dimensional comprehensive evaluation; monitoring changes in wastewater composition and providing early warning of crystallization risks; predicting the development trend of crystallization deposition, identifying high-risk pipe sections and generating prediction reports; establishing a knowledge base of treatment methods, performing scheme matching and multi-objective optimization; establishing a database of treatment cases, retrieving similar cases to generate treatment suggestions, and providing feedback and evaluation on the effects. This invention enables all-weather intelligent monitoring, accurate identification, early warning, and optimized treatment of crystallization deposition in complex wastewater pipe networks such as those in chemical industrial parks.

Description

Technical Field

[0001] This invention relates to the field of smart water management and pipeline operation and maintenance technology, and more specifically, to a method and system for intelligent monitoring and processing of crystallization deposition in drainage pipes. Background Technology

[0002] Drainage pipeline systems are a crucial component of urban infrastructure, undertaking the key functions of wastewater collection, transportation, and discharge. In industrial wastewater discharge scenarios such as chemical industrial parks, the wastewater typically contains high concentrations of inorganic salts, metal ions, and organic acid salts. During pipeline transportation, these substances are easily affected by factors such as temperature changes, pH fluctuations, reduced flow velocity, and mixing of different wastewaters, leading to the formation of crystal deposits on the pipe walls. The continuous accumulation of these crystal deposits reduces the effective flow cross-section of the pipeline and increases hydraulic friction, potentially causing blockages and resulting in wastewater overflows and environmental pollution.

[0003] Currently, monitoring of crystallization deposits in drainage pipes mainly relies on traditional methods such as manual inspections, regular CCTV checks, and periodic cleaning. These methods have limitations, including low detection frequency, inability to monitor in real time, and difficulty in predicting crystallization trends. Recent advancements in pipe inspection technologies based on sensors and image recognition primarily target sludge deposition and structural defects in municipal pipe networks. They lack the ability to specifically identify the diverse and complex crystallization deposits unique to chemical wastewater and have failed to establish an early warning mechanism linking changes in wastewater composition to crystal formation.

[0004] Existing technologies for monitoring and treating crystallization deposition have the following main problems: accurate identification and classification of multi-type composite crystallization deposition is difficult; existing image recognition methods lack robustness in complex pipeline environments; the risk of crystallization caused by sudden changes in wastewater composition cannot be predicted in advance, and there is a lack of effective causal relationship modeling; the prediction of the spatiotemporal evolution law of crystallization deposition fails to fully integrate physical mechanism constraints, resulting in insufficient prediction accuracy and interpretability; and the treatment scheme lacks specificity and adaptive optimization capabilities, making it difficult to automatically match the optimal treatment method according to the crystallization type. Summary of the Invention

[0005] This invention provides a method and system for intelligent monitoring and treatment of crystallization deposition in drainage pipes, which solves the technical problems in related technologies that lack targeted identification capabilities for multi-type complex crystallization deposition unique to chemical wastewater, and fail to establish an early warning mechanism between changes in wastewater composition and crystal formation.

[0006] This invention provides a method for intelligent monitoring and treatment of crystallization deposition in drainage pipes, including: Multimodal sensor data of drainage pipelines are collected and spatiotemporal aligned preprocessing is performed to obtain aligned multimodal monitoring datasets; Based on the pipe inner wall images in the aligned multimodal monitoring dataset, crystallization deposition area detection and type identification are performed to obtain crystallization deposition segmentation mask and type classification results; Based on the spatial distribution characteristics of the crystallization region statistically analyzed by the crystallization deposition segmentation mask, a multi-dimensional comprehensive evaluation was performed based on the aligned multi-modal monitoring dataset to obtain the comprehensive evaluation score and state level of the crystallization deposition. Based on the aligned multimodal monitoring dataset, changes in wastewater composition are monitored and early warning of crystallization risk is carried out, resulting in component mutation detection results and crystallization risk warning information. Based on the comprehensive evaluation score of crystallization deposition, the results of component mutation detection, and the early warning information of crystallization risk, the development trend of crystallization deposition is predicted, high-risk pipe sections are identified, and a prediction report is generated. Based on the type classification results, status level and prediction report, a knowledge base of processing methods is established, and scheme matching and multi-objective optimization are performed to obtain recommended schemes. A database of processing cases is established based on the recommended solutions. Similar cases are retrieved to generate processing suggestions. Feedback and evaluation of the effects are conducted to obtain intelligent processing decision results.

[0007] In a preferred embodiment, the step of collecting multimodal sensing data of the drainage pipeline and performing spatiotemporal alignment preprocessing specifically includes: Deploy a pipeline endoscopic video acquisition unit to acquire image sequences of the pipeline's inner wall, and configure an LED supplementary lighting system with adaptive brightness adjustment; An ultrasonic thickness sensor array is deployed to acquire deposition layer thickness distribution data, and the deposition layer thickness is calculated based on the echo arrival time difference and the sound velocity in the medium. Deploy distributed conductivity monitoring nodes to acquire time-series data of wastewater conductivity; Deploy an online multi-parameter water quality analyzer to acquire time-series data of wastewater quality parameters; A unified spatial reference system is established by calibrating the positions of each sensor; a spatiotemporal alignment algorithm based on maximizing mutual information is used to eliminate spatiotemporal bias in multi-source data.

[0008] In a preferred embodiment, obtaining the crystallization deposition segmentation mask and type classification results specifically includes: Construct a multi-scale feature extraction backbone network to output feature maps at different resolutions; A channel attention module is introduced to perform global average pooling on the input feature map along the spatial dimension to obtain the channel description vector. The channel attention weight vector is obtained through the activation function, and the attention weight vector is multiplied with the original feature map channel by channel. A spatial attention module is introduced to perform max pooling and average pooling on the input feature map along the channel dimension, and the spatial attention map is obtained through the activation function. Multi-level feature fusion is achieved through a feature pyramid network; water quality parameters are combined to assist in crystallization type identification; and a multi-task learning network is constructed to achieve joint optimization of region segmentation and type identification.

[0009] In a preferred embodiment, obtaining the comprehensive evaluation score and state level of crystallization deposition specifically includes: Based on the image segmentation results, spatial distribution characteristics of crystal deposition are calculated, including crystal coverage, distribution uniformity, and aggregation degree. Based on ultrasonic thickness measurement data, the characteristic indicators of the deposition layer thickness are calculated, including average thickness, maximum thickness, thickness standard deviation, and thickness growth rate. Based on water quality parameter data, the compositional tendency of crystallization deposition is inferred. The Langerier saturation index is used to assess the scaling tendency of calcium carbonate, and the Rezner stability index is used to assess the degree of scaling or corrosion. The supersaturation is determined by calculating the ratio of the ion activity product to the solubility product constant of calcium sulfate. The impact of crystallization deposition on pipeline flow capacity was calculated based on a hydraulic model; a multi-dimensional comprehensive evaluation index system was constructed and the weights were determined using the analytic hierarchy process; and the crystallization deposition state level was determined using a fuzzy comprehensive evaluation method.

[0010] In a preferred embodiment, obtaining the component mutation detection results and crystallization risk warning information specifically includes: A dynamic baseline model of wastewater composition was established, and historical data was selected as the baseline calculation window using the sliding window method. The statistical characteristics of the parameters were calculated within the window. A multi-parameter joint mutation detection algorithm is designed to calculate the standardized deviation of the current value of the monitored parameter from the baseline, use the cumulative sum control chart method to detect the persistent shift of the parameter, and use the exponential weighted moving average method to detect the mutation of the parameter. A wastewater crystallization risk assessment model based on chemical thermodynamics was constructed to calculate the supersaturation of sparingly soluble salts under current wastewater conditions. Establish a mixing effect prediction model for confluence nodes, calculate the composition of the mixed wastewater based on the real-time flow and water quality parameters of each branch pipe using the principle of mass conservation, generate graded early warning information and push disposal suggestions.

[0011] In a preferred embodiment, generating the prediction report specifically includes: The graph structure of the drainage network is constructed, where nodes represent monitoring sections or key locations on the pipeline, edges represent pipeline connections, node feature vectors contain crystallization and deposition state indicators, water quality parameters, and pipeline attributes, and edge feature vectors contain the physical attributes of the pipeline. The time dimension feature extraction module is designed, and a gated loop unit is used to encode the historical state sequence of each node. The temporal evolution law of crystal deposition is captured by the update gate and reset gate control mechanism. The design of the spatial dimension feature aggregation module employs a graph attention network to aggregate feature information of neighboring nodes and calculate attention coefficients for neighboring nodes; it also integrates temporal and spatial dimension features for joint modeling. Combine wastewater crystallization risk assessment models to help predict the development trend of crystallization and deposition; perform multi-step time-series predictions to output future state prediction values; identify high-risk pipe sections and generate prediction reports.

[0012] In a preferred embodiment, the criteria for determining the high-risk pipe section include: The predicted crystallization coverage or deposition thickness exceeds the preset warning threshold; the predicted state level rises from the current level to a more severe level; the predicted thickness growth rate exceeds the preset rate threshold. The identified high-risk pipe sections are sorted according to their risk level. The comprehensive risk index is calculated by taking into account the severity of the predicted state, the rate of deterioration, and the time to reach the warning state. After normalizing the three factors, the comprehensive risk index is obtained by weighting and summing them according to the weight coefficients. A prediction report is then generated.

[0013] In a preferred embodiment, obtaining the recommended solution specifically includes: Establish a knowledge base for crystallization deposition treatment methods. Each record includes crystallization type, applicable treatment method, treatment efficiency parameters, cost parameters, environmental impact parameters, and applicable condition constraints. Define the objective function and constraints of the multi-objective optimization problem. The optimization objectives include maximizing the treatment effect, minimizing the treatment cost, minimizing the impact on the normal operation of the drainage system, and minimizing the environmental impact. Design a coding scheme for decision variables, which include variables for treatment method selection, treatment agent concentration, treatment time, treatment frequency, and operation sequence. A non-dominated sorting genetic algorithm is used to solve the Pareto optimal solution set; a recommended scheme is selected from the Pareto optimal solution set.

[0014] In a preferred embodiment, obtaining the intelligent processing decision result specifically includes: Establish a database of processing cases to store historical processing records. Each processing case includes the pre-processing status, the processing solution adopted, the processing cost, and the processing effect. A case-based reasoning approach is used to generate processing suggestions. The case-based reasoning includes four steps: retrieval, reuse, modification and retention. In the retrieval stage, similar historical cases are retrieved from the case database based on the current crystallization deposition state characteristics. The similarity is calculated using the weighted Euclidean distance method. The knowledge base rules are updated based on statistical analysis of the treatment effects; a treatment scheme evaluation and feedback mechanism is established, which includes quantitative evaluation and qualitative evaluation. The qualitative evaluation adopts a five-level system, and the quantitative and qualitative evaluations are combined to form a comprehensive score for the treatment scheme; thus achieving continuous optimization of treatment decisions.

[0015] This invention provides an intelligent monitoring and processing system for crystallization deposition in drainage pipes, used to execute the aforementioned intelligent monitoring and processing method for crystallization deposition in drainage pipes, comprising: The multimodal data acquisition module is used to acquire multimodal sensor data of drainage pipelines and perform spatiotemporal alignment preprocessing to obtain aligned multimodal monitoring datasets. The crystallization detection and identification module, based on the pipe inner wall images in the aligned multimodal monitoring dataset, performs crystallization deposition area detection and type identification, and obtains crystallization deposition segmentation mask and type classification results; The comprehensive evaluation module statistically analyzes the spatial distribution characteristics of crystallization regions based on the crystallization deposition segmentation mask, and performs a multi-dimensional comprehensive evaluation based on the aligned multi-modal monitoring dataset to obtain the comprehensive evaluation score and state level of crystallization deposition. The risk warning module, based on the aligned multimodal monitoring dataset, monitors changes in wastewater composition and provides early warning of crystallization risk, obtaining component mutation detection results and crystallization risk warning information; The trend prediction module, based on the comprehensive evaluation score of crystal deposition, the results of component mutation detection, and the early warning information of crystallization risk, predicts the development trend of crystal deposition, identifies high-risk pipe sections, and generates a prediction report. The solution optimization module establishes a knowledge base of processing methods based on type classification results, status levels, and prediction reports, performs solution matching and multi-objective optimization, and obtains recommended solutions. The intelligent decision-making module establishes a database of processing cases based on recommended solutions, retrieves similar cases to generate processing suggestions, provides feedback and evaluation on the effects, and obtains intelligent processing decision results.

[0016] The beneficial effects of this invention are as follows: By employing multimodal sensor data collaborative acquisition and spatiotemporal alignment preprocessing, deep learning detection models, a crystallization risk early warning mechanism based on chemical thermodynamics, a spatiotemporal graph neural network prediction model, and a case-based reasoning-based continuous optimization method, intelligent monitoring and processing of crystallization deposition in drainage pipelines throughout the entire process has been achieved. This technology can accurately identify and classify multiple types of composite crystallization deposition, and provide risk warnings based on sudden changes in wastewater composition before crystallization deposition forms. It effectively solves the problems of insufficient detection accuracy and lack of early warning capabilities in existing technologies, providing reliable technical support for the intelligent operation and maintenance of complex wastewater pipe networks in chemical industrial parks and other similar locations.

[0017] By constructing a knowledge base of processing methods and a multi-objective optimization algorithm, targeted processing solutions can be automatically generated based on the specific type, degree, and location of crystallization deposition. The optimization decision is made by comprehensively considering multiple objectives such as processing effect, cost, operational impact, and environmental impact. The knowledge base and decision rules are continuously improved based on feedback from actual processing effects through a case reasoning mechanism. This realizes the transformation of processing solutions from generalized to personalized and from static to adaptive, improving processing efficiency and resource utilization, and reducing operation and maintenance costs and environmental risks. Attached Figure Description

[0018] Figure 1 This is a main flowchart of a smart monitoring and treatment method for crystal deposition in drainage pipes according to the present invention; Figure 2 This is a detailed flowchart of a method for intelligent monitoring and treatment of crystal deposition in drainage pipes according to the present invention; Figure 3 This is a block diagram of an intelligent monitoring and treatment system for crystal deposition in drainage pipes according to the present invention. Detailed Implementation

[0019] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0020] At least one embodiment of the present invention discloses an intelligent monitoring and treatment method for crystallization deposition in drainage pipes, such as... Figures 1 to 2 As shown, it includes: Step 1: Collect multimodal sensing data of drainage pipelines and perform spatiotemporal alignment preprocessing to obtain aligned multimodal monitoring datasets; Specifically, the following steps are included: Step 1.1: Deploy the pipeline endoscopic video acquisition unit to obtain the image sequence of the pipeline inner wall; Based on the monitoring needs of the drainage pipe network in the chemical industrial park, a multi-type sensor collaborative deployment method is adopted to collect pipeline operation status data.

[0021] Industrial-grade high-definition cameras are selected as image acquisition devices and installed in a circular array on pipeline inspection robots or fixed inspection stations, equipped with an adaptive brightness-adjustable LED supplementary lighting system. The cameras are evenly distributed along the circumference of the pipeline to ensure full circumferential coverage of the pipeline's inner wall.

[0022] The image acquisition strategy combines timed acquisition with event-triggered acquisition. The time interval for timed acquisition is determined based on the historical frequency and severity of crystallization in the pipe section: for high-incidence pipe sections with frequent crystallization deposition, an acquisition interval of 1 to 2 hours is used; for low-risk pipe sections with less crystallization, an acquisition interval of 3 to 6 hours is used. Event-triggered acquisition can be triggered by manual judgment or other monitoring systems (such as flow monitoring, online water quality monitoring, etc.). In this case, the acquisition interval is shortened to 10 to 30 minutes, and continuous monitoring is carried out for 2 to 6 hours to capture the formation process of crystallization deposition.

[0023] The acquired image data includes metadata information such as timestamps, pipeline location identifiers, and acquisition parameters, which are used for subsequent spatiotemporal alignment and data analysis.

[0024] Step 1.2: Deploy an ultrasonic thickness sensor array to acquire deposition layer thickness distribution data; Based on the principle of ultrasonic wave reflection at different media interfaces, multiple ultrasonic probes are evenly arranged circumferentially at key locations along the pipeline. The number of probes is determined by the pipeline diameter, typically ranging from 4 to 8. The ultrasonic probes emit pulse signals, which generate reflected echoes at the interfaces between the pipe wall and the sediment layer, and between the sediment layer and the water. The sediment layer thickness is calculated based on the echo arrival time difference and the sound velocity in the medium. The sound velocity in the sediment layer is determined through laboratory calibration or an adaptive estimation algorithm. The thickness measurement data is recorded in real time and transmitted to a monitoring system.

[0025] Step 1.3: Deploy distributed conductivity monitoring nodes to acquire time-series data of wastewater conductivity; Conductivity sensor nodes are arranged along the pipeline axis at intervals of 100 to 500 meters, with denser arrangements at key locations such as pipeline junctions and pumping stations. The sensors employ a four-electrode measurement method, are equipped with temperature compensation modules, and have a measurement cycle of 1 to 10 minutes. Each monitoring node transmits data to a data aggregation gateway via wired or wireless means. The gateway performs preliminary data processing before uploading it to the monitoring platform.

[0026] Step 1.4: Deploy an online multi-parameter water quality analyzer to acquire time-series data of wastewater quality parameters; Online water quality analyzers with multiple parameters are deployed at the main inlets, confluence points, and key pipe sections of the pipeline network. Water quality parameters closely related to the crystallization process are monitored, including pH, temperature, dissolved oxygen, total dissolved solids, calcium ion concentration, magnesium ion concentration, sulfate ion concentration, and carbonate ion concentration. Each parameter is measured using appropriate standard methods, with the measurement cycle set according to the parameter change rate, typically between 5 and 30 minutes. The analyzers upload data to the monitoring platform via a communication interface.

[0027] Step 1.5: Perform position calibration on each sensor to establish a unified spatial reference system; A cylindrical coordinate system is established with the pipeline axis as the reference, and the origin of the coordinate system is set at the beginning of the pipeline. For each sensor, its three-dimensional position parameters in the pipeline coordinate system, including axial position, radial position, and circumferential angle, are determined through field measurements or installation records. The position calibration information is stored in the sensor configuration database along with the sensor number for subsequent spatial data fusion.

[0028] Step 1.6: A spatiotemporal alignment algorithm based on maximizing mutual information is used to eliminate spatiotemporal bias in multi-source data; Due to differences in sampling clocks and installation locations among different sensors, there are time and spatial offsets between data sources. The spatiotemporal alignment algorithm optimizes the spatiotemporal offset parameters to maximize the mutual information between the aligned multi-source data. The algorithm uses grid search or gradient optimization methods to search for the optimal solution, performs fine-grained alignment of the time offset using interpolation methods, and corrects the spatial offset using coordinate transformation based on the coordinate system established in step 1.5.

[0029] Step 1.7: Perform quality verification and preprocessing on the aligned data; The multimodal dataset undergoes integrity checks, consistency checks, and outlier detection. Integrity checks verify the presence of missing data from each data source within the target time frame, filling in any missing data using methods such as linear interpolation, spline interpolation, or forward imputation. Consistency checks verify the absence of logical inconsistencies between related data sources. Outlier detection employs statistical distribution-based methods (such as the 3σ criterion and box plots) to identify measurements outside the normal range, removing or correcting confirmed outliers. The preprocessed data is organized and stored using a unified spatiotemporal index, forming a multimodal dataset available for subsequent analysis.

[0030] Step 1 outputs the aligned multimodal monitoring dataset, including pipeline inner wall image sequences, sediment thickness distribution data, electrical conductivity time series data, and water quality parameter time series data, providing a data foundation for subsequent crystallization sediment detection and condition assessment.

[0031] Step 2: Based on the pipe inner wall images in the aligned multimodal monitoring dataset, perform crystallization deposition region detection and type identification to obtain crystallization deposition segmentation mask and type classification results; Specifically, the following steps are included: Step 2.1: Preprocess the image of the inner wall of the pipe to improve image quality; To address image quality issues caused by the complex internal environment of pipelines (such as water mist obstruction, uneven lighting, and low contrast), image processing methods are employed to enhance the image.

[0032] An adaptive histogram equalization method is employed to enhance image contrast. The image is divided into several sub-regions, and histogram equalization is performed on each sub-region. Bilinear interpolation is used to smooth the boundaries between adjacent sub-regions. Gaussian filtering is used to remove image noise. Gamma correction is used to adjust image brightness, with the Gamma value adaptively determined based on the average gray value of the image. Sharpening filtering is used to enhance the edge features of crystal deposits. The preprocessed image exhibits improved contrast and sharpness, which is beneficial for subsequent crystal region detection.

[0033] Furthermore, when extremely poor lighting conditions within the pipe lead to severely inadequate image quality, an image enhancement method based on Retinex theory can be employed to improve image brightness while preserving the texture details of crystalline deposition. A multi-scale Retinex algorithm is used, employing Gaussian filters of different scales to filter the image, obtaining illumination estimates at multiple scales. The Retinex outputs at each scale are then weighted and averaged, and the enhanced image is obtained through logarithmic transformation and normalization.

[0034] Step 2.2: Construct a multi-scale feature extraction backbone network; Based on the multi-scale characteristics of crystal deposition in images, a deep convolutional neural network is used for feature extraction.

[0035] The backbone network is based on a pre-trained ResNet-50 network, which progressively extracts image features from low to high resolution through multiple convolutional layers, forming a feature map pyramid with different resolutions. Low-resolution feature maps have a larger receptive field, capturing the overall morphology and contextual information of the crystalline region; high-resolution feature maps retain more spatial details, enabling precise localization of the crystalline region's boundaries. The backbone network outputs multiple feature maps at different resolutions for use by subsequent attention and feature fusion modules. Network parameters are pre-trained on the ImageNet dataset; during transfer learning, shallow layer parameters are frozen, and only deep layer parameters are fine-tuned to adapt to the feature distribution of the pipeline image.

[0036] Step 2.3: Introduce a channel attention module to enhance the response of crystallization-related feature channels; Based on the different contributions of different feature channels to the detection of crystal deposition, the channel attention module adaptively learns the importance weights of each channel.

[0037] The input feature map is subjected to global average pooling along the spatial dimension to obtain channel description vectors. The channel description vectors are then input into a bottleneck structure consisting of two fully connected layers. The first fully connected layer compresses the number of channels to reduce computation, while the second fully connected layer restores the number of channels. Channel attention weight vectors are obtained through the Sigmoid activation function. The attention weight vectors are then multiplied channel by channel with the original feature map to enhance feature channels related to crystal deposition and suppress irrelevant background feature channels.

[0038] Step 2.4: Introduce a spatial attention module to enhance the spatial response of the crystallization region.

[0039] Based on the fact that crystal deposits occupy specific spatial locations in an image, the spatial attention module adaptively learns the importance weights of each spatial location.

[0040] Max pooling and average pooling are performed on the input feature map along the channel dimension to obtain two single-channel spatial description maps. The two spatial description maps are concatenated along the channel dimension and input into the convolutional layer for feature fusion. A spatial attention map is obtained by applying the Sigmoid activation function. The spatial attention map is multiplied pixel by pixel with the original feature map to enhance the feature response of the crystallized region and suppress the feature response of the background region.

[0041] Step 2.5: Multi-level feature fusion is achieved through a feature pyramid network; Based on the characteristic that feature maps at different levels contain complementary information, the feature pyramid network fuses high-level semantic features with low-level detailed features.

[0042] The network employs a top-down path and lateral connection structure. The top-down path upsamples high-level low-resolution feature maps to the same size as low-level high-resolution feature maps. The lateral connections element-wise add the feature maps from each layer of the backbone network, after adjusting the channel count through convolution, to the upsampled feature maps. The fused feature map contains both high-level semantic information and preserves low-level spatial details, which is beneficial for accurate pixel-level segmentation. The output is a feature map of the same size as the original input image. After passing through a classification convolutional layer, a probability map is obtained for each pixel belonging to a crystalline deposition region. A binary segmentation mask is then obtained through thresholding.

[0043] Step 2.6: Combine water quality parameters to assist in identifying the crystallization type; Since the types of crystal deposits formed by wastewater with different chemical compositions vary, the water quality parameters collected in step 1 are used as auxiliary information to improve the accuracy of crystal type identification.

[0044] The water quality parameter vector of the current pipe section (including pH value, conductivity, calcium ion concentration, sulfate ion concentration, etc.) is normalized and then concatenated with the image feature vector before being input into the classifier. By incorporating water quality parameters, the classifier can comprehensively consider image features and chemical composition information. For example, when the pH value is alkaline and the calcium ion concentration is high, it is more likely to be classified as carbonate crystallization; when the sulfate ion concentration is high, it is more likely to be classified as sulfate crystallization. This multimodal fusion classification method improves the accuracy and robustness of crystallization type identification.

[0045] Step 2.7: Construct a multi-task learning network to achieve joint optimization of region segmentation and type recognition; Given the diverse types of crystalline deposits formed by chemical wastewater, a classification branch is added to the segmentation network to identify the types of detected crystalline regions. The classification branch extracts fixed-size feature vectors for each crystalline region from the feature map using region-of-interest pooling, and then inputs these feature vectors into the classifier, outputting the probability distribution of each crystalline type. Crystallization types include carbonate, sulfate, metal hydroxide, and organic-inorganic composite types. The classification and segmentation branches share the feature extraction part of the backbone network and are jointly trained through multi-task learning, with the segmentation and classification tasks mutually reinforcing each other to improve overall performance.

[0046] Step 2 outputs pixel-level crystal deposition segmentation masks and crystal type classification results. The segmentation mask identifies the specific location and extent of crystal deposition in the pipe inner wall image, and the type classification results indicate the chemical type of each crystal region, providing basic data for subsequent comprehensive evaluation and processing decisions.

[0047] Step 3: Based on the spatial distribution characteristics of the crystallization region statistically analyzed by the crystallization deposition segmentation mask, and based on the aligned multimodal monitoring dataset, a multi-dimensional comprehensive evaluation is performed to obtain the comprehensive evaluation score and state level of the crystallization deposition. Specifically, the following steps are included: Step 3.1: Calculate the spatial distribution characteristic index of crystallization deposition based on the image segmentation results; Based on the crystal deposition segmentation mask output in step 2, the spatial distribution characteristics of the crystallization region are statistically analyzed.

[0048] Crystallization coverage is defined as the ratio of the number of crystallized pixels in the segmented mask to the total number of pixels in the visible area of ​​the pipe wall. This indicator reflects the overall severity of crystallization deposition. Distribution uniformity is characterized by calculating the spatial distribution standard deviation of the centroid positions of the crystallized areas. The pipe wall image is divided into several sub-regions, and the number of crystallized pixels in each sub-region is counted. The coefficient of variation of the number distribution is calculated; a smaller coefficient of variation indicates a more uniform distribution. Agglomeration is characterized by connected component analysis. Connected components are marked on the segmented mask, and the number and area of ​​each connected component are counted. A small number of connected components with a large average area indicates an aggregated crystallization distribution, while a large number of connected components with a small average area indicates a dispersed crystallization distribution.

[0049] Step 3.2: Calculate the characteristic indices of the deposition layer thickness based on ultrasonic thickness measurement data; Based on the deposition layer thickness distribution data output in step 1, the thickness measurement values ​​of each measuring point on the pipe cross-section are extracted.

[0050] The thickness statistics at the current moment are calculated, including average thickness, maximum thickness, minimum thickness, and thickness standard deviation. Average thickness reflects the overall level of the sedimentary layer, maximum thickness reflects the degree of deposition in the most severely affected local areas, and thickness standard deviation reflects the uniformity of the sedimentary layer. Time series analysis is performed on historical thickness data to calculate the thickness growth rate. The least squares method is used to linearly fit the thickness data at several recent time points; the slope of the fitted line is the thickness growth rate. If the thickness data exhibits a non-linear growth trend, exponential smoothing is used to extract the trend and calculate the exponential growth rate. The thickness growth rate is an important indicator for predicting the development trend of crystallization deposition.

[0051] Step 3.3: Infer the compositional tendency of crystallization deposition based on water quality parameter data; Based on the time-series data of water quality parameters output in step 1, the scaling tendency type of wastewater is analyzed.

[0052] The Langerier saturation index was used to assess the scaling tendency of calcium carbonate. This index is defined as the difference between the actual pH value of the wastewater and the theoretical pH value at calcium carbonate saturation. A positive value indicates that the wastewater is supersaturated and prone to scaling, while a negative value indicates that the wastewater is undersaturated and prone to corrosion. The theoretical saturation pH value is calculated based on parameters such as total dissolved solids concentration, temperature, calcium ion concentration, and alkalinity. The criteria for judging the Langerier saturation index are as follows: a Langerier saturation index greater than 0.5 indicates a strong scaling tendency; between 0 and 0.5 indicates a slight scaling tendency; between -0.5 and 0 indicates a slight corrosion tendency; and less than -0.5 indicates a strong corrosion tendency.

[0053] The degree of scaling or corrosion was further assessed using the Rezna Stability Index, which is calculated based on the theoretical saturation pH and the actual pH value. The Rezna Stability Index criteria are as follows: a Rezna Stability Index less than 6.0 indicates severe scaling; between 6.0 and 6.5 indicates moderate scaling; between 6.5 and 7.0 indicates slight scaling; between 7.0 and 7.5 indicates equilibrium; between 7.5 and 8.0 indicates slight corrosion; and greater than 8.0 indicates severe corrosion.

[0054] For sulfate scaling tendency, supersaturation is determined by calculating the ratio of the activity product to the solubility product constant of calcium sulfate ions. The activity product equals the product of calcium ion activity and sulfate ion activity, and the activity of each ion equals the ion concentration multiplied by its activity coefficient. The activity coefficient can be calculated using the Debye-Hückel extended equation based on ionic strength and ionic charge. The solubility product constant is temperature-dependent and requires temperature correction. A supersaturation value greater than 1 indicates a tendency for scaling, while a value greater than 5 indicates a strong tendency for scaling.

[0055] Based on various scaling indices, the main scaling type of wastewater can be determined as follows: when the Langerier saturation index is greater than 0 and the calcium sulfate supersaturation is less than 1, it is carbonate type; when the Langerier saturation index is less than 0 and the calcium sulfate supersaturation is greater than 1, it is sulfate type; and when the Langerier saturation index is greater than 0 and the calcium sulfate supersaturation is greater than 1, it is mixed type.

[0056] Step 3.4: Calculate the impact of crystallization deposition on the flow capacity of the pipeline based on the hydraulic model; Based on the current sediment thickness data, the reduction in the effective cross-sectional area of ​​the pipe is calculated. For a circular pipe, the effective inner diameter is equal to the original inner diameter minus twice the average sediment thickness, and the effective cross-sectional area is calculated according to the formula for the area of ​​a circle.

[0057] The flow capacity of a pipe is calculated using the Manning formula, which shows that flow rate is directly proportional to the cross-sectional area, the square of the hydraulic radius, and the square of the slope, and inversely proportional to the Manning roughness coefficient. Crystallization and deposition increase pipe wall roughness; the increase in the roughness coefficient is related to the thickness of the deposition layer and the surface roughness of the deposits, and can be determined through empirical formulas or experimental calibration. The ratio of the flow capacity under the current deposition state to the flow capacity of a clean pipe is calculated to obtain the flow capacity retention rate. The flow capacity attenuation rate is equal to 1 minus the retention rate.

[0058] Step 3.5: Construct a multi-dimensional comprehensive evaluation index system and determine its weights; Based on the above-mentioned indicators, a comprehensive evaluation index system is constructed, which includes four dimensions: spatial distribution characteristics, thickness characteristics, compositional tendency, and influence of flow capacity.

[0059] The weight coefficients of each dimension and indicator were determined using the analytic hierarchy process (AHP). A hierarchical model was constructed, consisting of three levels: the target level, which is a comprehensive evaluation of crystal deposition; the criteria level, which comprises four evaluation dimensions: spatial distribution characteristics, thickness characteristics, compositional tendency, and flow capacity influence; and the indicator level, which contains specific indicators for each dimension. The spatial distribution characteristics dimension includes indicators such as crystal coverage, distribution uniformity, and aggregation degree; the thickness characteristics dimension includes indicators such as average thickness, maximum thickness, and thickness growth rate; the compositional tendency dimension includes indicators such as Langerier saturation index, Rezsner stability index, and sulfate supersaturation; and the flow capacity influence dimension includes indicators such as the flow cross-sectional area reduction rate and the flow capacity attenuation ratio.

[0060] Judgment matrices at various levels are constructed through expert consultation or historical data analysis. Each element of the judgment matrix represents a comparison of the importance of factors at the same level relative to factors at higher levels. A consistency check is performed on the judgment matrices; if the consistency ratio is less than a preset threshold, the judgment matrix is ​​considered to have satisfactory consistency; otherwise, the judgment matrix needs adjustment. The weight coefficients of each factor are obtained by calculating the eigenvector corresponding to the largest eigenvalue of the judgment matrix.

[0061] Step 3.6: Normalize each indicator; Since the dimensions and value ranges of each indicator are different, normalization processing is required to facilitate comprehensive calculation.

[0062] For positive indicators (i.e., larger values ​​indicate more severe crystallization), the maximum-minimum normalization method is used. The normalized value equals the original value minus the minimum value, then divided by the difference between the maximum and minimum values. For negative indicators (i.e., smaller values ​​indicate more severe crystallization), the inverse maximum-minimum normalization method is used. The maximum and minimum values ​​can be determined based on historical data statistics or expert experience. After normalization, the value range of each indicator is uniformly between 0 and 1, facilitating subsequent weighted calculations.

[0063] Step 3.7: Use the fuzzy comprehensive evaluation method to determine the crystallization deposition state level.

[0064] Based on the fuzziness and uncertainty of the evaluation of crystallization deposition state, a fuzzy comprehensive evaluation method is used for grade classification.

[0065] Establish an evaluation factor set, which includes the normalized evaluation indicators, namely all indicators under the four dimensions of spatial distribution characteristics, thickness characteristics, component tendency and flow capacity influence determined in step 3.5.

[0066] A rating system was established, comprising four levels: mild, moderate, severe, and extremely severe, corresponding to different degrees of crystal deposition. To facilitate the calculation of the overall evaluation score, each level was assigned a score, for example, mild corresponds to 25 points, moderate to 50 points, severe to 75 points, and extremely severe to 100 points.

[0067] Establish membership functions for each indicator to each level. These functions describe the degree to which an indicator value belongs to each level. Triangular, trapezoidal, or Gaussian membership functions can be used, with function parameters determined based on expert experience or historical data. For example, for the crystallization coverage indicator, a trapezoidal membership function can be set: coverage between 0% and 10% fully belongs to the slight level; coverage between 10% and 30% partially belongs to the slight and moderate levels; coverage between 30% and 60% partially belongs to the moderate and severe levels; and coverage above 60% fully belongs to the extremely severe level.

[0068] Calculate the membership degree of each indicator to each level based on its current value, forming a fuzzy evaluation matrix. Perform fuzzy synthesis operation on the weight vector obtained in step 3.5 and the fuzzy evaluation matrix to obtain a comprehensive evaluation vector. Each element of the vector represents the membership degree of the comprehensive evaluation result to each level. According to the principle of maximum membership degree, the level with the highest membership degree is taken as the final state level determination result. At the same time, calculate the comprehensive evaluation score, which is equal to the weighted sum of the membership degree of each level and the level score. The score ranges from 0 to 100 points, with higher scores indicating more severe crystallization deposition.

[0069] Through the above steps, step 3 outputs the comprehensive evaluation score and state level of crystallization deposition, which comprehensively reflects multi-dimensional information such as the spatial distribution, thickness characteristics, compositional tendency and flow influence of crystallization deposition, providing a basis for subsequent risk warning and treatment decisions.

[0070] Furthermore, since the traditional analytic hierarchy process (AHP) relies on subjective judgment when determining weights, which may lead to biases, a combined weighting method can be adopted. This method combines objective weighting based on entropy weighting with subjective weighting based on AHP. The aim is to integrate subjective experience and objective data information to improve the scientific rigor and rationality of weight determination. Specifically, the entropy weighting method determines weights based on the dispersion of each indicator's data. The more dispersed the indicator data, the greater the amount of information it contains, and the greater the weight it should be assigned. The information entropy of each indicator is calculated; the smaller the information entropy, the more dispersed the data. The entropy weight of each indicator is calculated based on the information entropy. The entropy weight equals 1 minus the information entropy, divided by the sum of 1 minus the information entropy of all indicators. The objective weights obtained by the entropy weighting method and the subjective weights obtained by AHP are linearly combined. The combination coefficient can be optimized by minimizing the deviation between the comprehensive evaluation result and the actual state.

[0071] Step 4: Based on the aligned multimodal monitoring dataset, monitor changes in wastewater composition and conduct early warning of crystallization risk to obtain compositional mutation detection results and crystallization risk warning information; Specifically, the following steps are included: Step 4.1: Establish a dynamic baseline model of wastewater composition; Based on the characteristic that the composition of chemical wastewater exhibits normal fluctuations, a dynamic baseline model is established using the sliding window method to adapt to the slow drift of the composition.

[0072] For each monitoring parameter, historical data from the most recent sampling periods are selected as the baseline calculation window. The window length is determined based on the parameter's variation characteristics: parameters that change rapidly (such as conductivity and temperature) use a 12- to 24-hour window, while parameters that change slowly (such as ion concentration) use a 48- to 72-hour window. Within the window, the parameter's statistical characteristics are calculated, including the mean, standard deviation, maximum, minimum, and rate of change. The mean is used as the baseline center value, and the standard deviation is used to determine the normal fluctuation range. As new data arrives, the window slides forward, and the baseline model is updated accordingly to adapt to the slow drift changes in wastewater composition. Robust statistical methods are used for baseline updates, using the median instead of the mean and the interquartile range instead of the standard deviation to reduce the impact of extreme values ​​on the baseline.

[0073] Step 4.2: Design a multi-parameter joint mutation detection algorithm; Sudden changes in wastewater composition are usually manifested by simultaneous abnormal changes in multiple parameters. A multi-parameter joint detection algorithm is designed to improve the accuracy and reliability of detection.

[0074] For each monitored parameter, its standardized deviation relative to the baseline is calculated. The standardized deviation is equal to the current value minus the baseline center value, then divided by the baseline dispersion. A cumulative sum control chart method is used to detect persistent parameter shifts. The cumulative sum statistic accumulates the difference between the standardized deviation and the reference value at each time point. A shift is determined to have occurred when the cumulative sum exceeds a preset control limit. An exponentially weighted moving average method is used to detect abrupt changes in the parameter. The exponentially weighted moving average assigns greater weight to recent data, enabling rapid response to abrupt changes.

[0075] The mutation detection employs a tiered threshold strategy: a component mutation alarm is triggered when the deviation of a single key parameter exceeds the first threshold (preferably set to 3 times the standard deviation) or when the deviation of two or more parameters simultaneously exceeds the second threshold (preferably set to 2 times the standard deviation). The threshold setting comprehensively considers the statistical distribution characteristics of historical data and the acceptable false alarm rate, and is generally controlled within 5%.

[0076] Step 4.3: Construct a wastewater crystallization risk assessment model based on chemical thermodynamics; Based on the chemical thermodynamics of crystal deposition, a quantitative relationship model between wastewater composition and crystallization risk is established.

[0077] For various types of sparingly soluble salts that may form, their supersaturation under current wastewater conditions is calculated. Supersaturation is defined as the ratio of the ion activity product to the solubility product constant. The calculation steps are as follows: Calculate the activity of each ion based on its concentration and activity coefficient. The activity coefficient is calculated based on the ionic strength of the solution using the Debye-Hückel equation or its extended form. Calculate the ion activity product, which is equal to the product of the activities of the ions constituting the sparingly soluble salt. Divide the ion activity product by the solubility product constant to obtain the supersaturation. The solubility product constant is temperature-dependent and is corrected for temperature using the van der Hoff equation. When the supersaturation is greater than 1, the wastewater is in a supersaturated state and has a thermodynamic driving force for crystallization and precipitation; the higher the supersaturation, the higher the crystallization risk. The supersaturation of common sparingly soluble salts such as calcium carbonate, calcium sulfate, and magnesium hydroxide is calculated comprehensively, and the maximum value is taken as the comprehensive crystallization risk index.

[0078] Step 4.4: Establish a prediction model for the mixing effect of the confluence node; The chemical industrial park's pipeline network has multiple branch pipes converging. A wastewater mixing prediction model is established for these converging nodes. Based on the real-time flow rate and water quality parameters of each branch pipe, the composition of the mixed wastewater is calculated using the principle of mass conservation. The concentration of a certain ion after mixing is equal to the sum of the products of the ion concentration and flow rate in each branch pipe, divided by the total flow rate. The pH calculation after mixing needs to consider the influence of the acid-base buffer system, as pH directly affects the equilibrium and supersaturation of the carbonate system. This can be solved by simultaneously applying the charge balance equation and the mass balance equation.

[0079] Based on the composition of the mixed wastewater, assess the supersaturation of each sparingly soluble salt using the method in step 4.3 to determine whether mixing will increase the risk of crystallization. Pay special attention to potential chemical reactions that may occur after mixing wastewater from different enterprises. For example, mixing calcium-containing wastewater with carbonate-containing wastewater will produce calcium carbonate precipitate, and mixing heavy metal-containing wastewater with alkaline wastewater will produce metal hydroxide precipitate.

[0080] Step 4.5: Generate tiered early warning information and push out handling suggestions; Based on the results of component mutation detection and crystallization risk assessment, a graded early warning system is generated. The early warning levels are divided into three levels: Attention Level, Warning Level, and Severe Level.

[0081] The "Attention" level corresponds to a standardized deviation between 2 and 3 standard deviations or an oversaturation between 1 and 2. Abnormal events are recorded, and the monitoring frequency is increased to twice the normal frequency; no manual intervention is required at this time. The "Warning" level corresponds to a standardized deviation between 3 and 4 standard deviations or an oversaturation between 2 and 5. Early warning information is sent to maintenance personnel, and preventative measures are recommended, such as adjusting wastewater discharge plans, optimizing runoff ratios, or adding scale inhibitors. The "Severe" level corresponds to a standardized deviation exceeding 4 standard deviations or an oversaturation exceeding 5. An emergency alarm is issued, and immediate intervention measures are recommended, such as suspending wastewater discharge from the relevant enterprise or initiating emergency response procedures.

[0082] The early warning information includes the name and value of the abnormal parameter, the degree of deviation, the calculation result of the supersaturation, the risk type (carbonate type, sulfate type or mixed type), the expected range of pipe sections affected, and the recommended measures. It is pushed to relevant personnel through various channels such as the monitoring platform interface, SMS, and mobile applications.

[0083] Furthermore, since a single supersaturation index cannot fully reflect the actual formation risk of crystallization deposits, and the crystallization process is influenced by both thermodynamic and kinetic factors, a comprehensive risk assessment method combining thermodynamic supersaturation and kinetic induction time can be adopted. The aim is to more accurately predict the timing and severity of crystallization deposit formation. Specifically, induction time refers to the time interval from when wastewater reaches a supersaturated state to when detectable crystals begin to appear. Induction time is related to factors such as supersaturation, temperature, stirring intensity, and impurity content. Induction time under different conditions is estimated through experimental measurements or empirical formulas, and then compared with the residence time of wastewater in the pipeline. If the residence time is greater than the induction time, crystallization is likely to form within the pipeline; if the residence time is less than the induction time, crystallization may form downstream or in the treatment facility. By combining the two factors of supersaturation and induction time, a two-dimensional risk matrix is ​​constructed to more finely classify risk levels: high supersaturation and short induction time correspond to high risk, while low supersaturation or long induction time corresponds to low risk.

[0084] Step 5: Based on the comprehensive evaluation score of crystallization deposition, the results of component mutation detection, and the early warning information of crystallization risk, predict the development trend of crystallization deposition, identify high-risk pipe sections, and generate a prediction report; Specifically, the following steps are included: Step 5.1: Construct a graphical representation of the drainage network; Based on the topological connections of the pipe network, the drainage pipe network is abstracted into a directed graph data structure. The nodes in the graph represent monitoring sections or key locations on the pipes, including sensor installation locations, pipe junctions, pump station locations, etc.; the edges in the graph represent pipe connections, and the direction of the edges is consistent with the direction of water flow, pointing from upstream nodes to downstream nodes.

[0085] The node feature vector contains information such as crystallization deposition state indicators, water quality parameters, and pipeline attributes at that location, specifically including crystallization coverage, average deposition thickness, thickness growth rate, conductivity, pH value, temperature, pipeline diameter, pipeline material code, and pipeline service life. The edge feature vector contains the physical attributes of the pipeline, including pipeline length, pipeline slope, design flow rate, and actual flow rate. The graph structure data is updated according to a preset time step to form a time-series graph data sequence.

[0086] Step 5.2: Design a time-dimensional feature extraction module to capture the temporal evolution of crystallization deposition; Based on the time-varying characteristics of crystallization deposition states, a gated recurrent unit (GRU) is used to encode the historical state sequence of each node. The GRU is an improved recurrent neural network structure that effectively solves the gradient vanishing problem in long-term sequence training through two control mechanisms: update gates and reset gates.

[0087] The update gate controls the degree of influence of the historical state from the previous time step on the current time step. Its control value continuously varies between 0 and 1 and is automatically learned and determined by the model. When the update gate's control value approaches 1, more historical information is retained to capture long-term dependencies; when the control value approaches 0, more reliance is placed on the current input to respond to short-term changes. The reset gate controls which information from the historical state from the previous time step needs to be forgotten. Its control value also continuously varies between 0 and 1. When the reset gate's control value approaches 0, historical information is ignored; when the control value approaches 1, historical information is retained. Through these two gating mechanisms, the gated recurrent unit can adaptively learn the long-term trends and short-term fluctuations in the crystallization and deposition evolution process.

[0088] For each node, the sequence of feature vectors within its historical time window is sequentially input into a gated recurrent unit. The time window length is typically set to 24 time steps, corresponding to 24 hours of historical data, with sampling occurring every hour. The final state vector output by the unit serves as the temporal dimension feature representation for that node, encoding the historical evolution and trends of crystallization deposition. Model parameters are trained using the backpropagation algorithm, employing an adaptive learning rate optimization method and a gradient pruning mechanism to prevent gradient explosion during training.

[0089] Step 5.3: Design a spatial dimension feature aggregation module to learn the spatial correlation between adjacent nodes; Based on the spatial correlation of crystallization deposition states between adjacent pipe segments in a pipeline network, a graph attention network is used to aggregate the feature information of neighboring nodes. The graph attention network calculates an attention coefficient for each pair of adjacent nodes, which reflects the degree of influence of neighboring nodes on the current node.

[0090] The attention coefficient is calculated as follows: the feature vectors of the current node and its neighbors are mapped to the same feature space through linear transformations; the two transformed feature vectors are concatenated and input into a single-layer neural network, and after processing by a nonlinear activation function, the original attention score is obtained; the attention scores of all neighbors of the current node are normalized so that their sum is one, resulting in the final attention coefficient. The spatial feature representation of the current node is equal to the weighted sum of the feature vectors of all neighboring nodes according to the attention coefficient.

[0091] To enhance the model's expressive power, a multi-head attention mechanism is adopted, which involves computing multiple sets of attention coefficients and weighted features in parallel, concatenating or averaging the results to capture the spatial correlation between nodes from multiple perspectives.

[0092] Step 5.4: Integrate features from the time and space dimensions for joint modeling; Based on the spatiotemporal coupling characteristics of crystallization deposition, temporal and spatial features are fused. The fusion method involves feature concatenation followed by a nonlinear transformation through a fully connected layer, which learns the interaction between the temporal and spatial features. The fused feature vector simultaneously contains both temporal evolution and spatial propagation information of the crystallization deposition, enabling a more comprehensive characterization of its spatiotemporal dynamics. This fused feature vector serves as input to the prediction module, used to generate state predictions for future time steps.

[0093] Step 5.5: Combine wastewater crystallization risk assessment model to help predict the development trend of crystal deposition; The formation of crystal deposits in chemical wastewater pipelines is affected by chemical reactions. The calculation results of the wastewater crystallization risk assessment model constructed in step 4.3 are used as auxiliary features to input into the prediction model to improve the accuracy of the prediction.

[0094] The specific method is as follows: For each node, the supersaturation of common sparingly soluble salts such as calcium carbonate, calcium sulfate, and magnesium hydroxide is calculated based on the current water quality parameters, and the supersaturation value is used as an additional node feature. The supersaturation feature vector is concatenated with the original node feature vector to obtain an enhanced node feature vector, which is then input into a spatiotemporal graph neural network for prediction. By introducing the supersaturation feature, the model can perceive the driving force of the current chemical environment on the formation of crystal deposition. For example, when the supersaturation of calcium carbonate is high, the model will predict that the future crystal deposition growth rate of that node will be faster. This method, which combines data-driven and mechanism-driven approaches, retains the learning ability of the neural network while incorporating prior knowledge of chemical thermodynamics, thus improving the accuracy and interpretability of the prediction. For confluence nodes, the change in supersaturation after mixing wastewater from different branch pipes also needs to be considered. The supersaturation after mixing is calculated based on the mixing effect prediction model in step 4.4 and used as the feature input for the confluence node.

[0095] Step 5.6: Perform multi-step time series prediction and output the predicted future state value; Based on a trained spatiotemporal graph neural network model, the crystallization deposition state at multiple future time steps is predicted.

[0096] The prediction uses an autoregressive approach, predicting the state at the next time step based on historical data. The prediction results are then used as input to continue predicting the state at even further time steps, iterating until the preset prediction time window length is reached.

[0097] To improve the stability of multi-step prediction, a combination of teacher-mandated and free-running strategies is adopted during the training phase. The teacher-mandated strategy uses the actual observations as the input for the next step to accelerate model convergence, while the free-running strategy uses the model's predicted values ​​as the input for the next step to improve the actual prediction capability. The two strategies are used in combination in a certain proportion.

[0098] The prediction output includes the predicted crystallization deposition state of each node at each future time step, as well as the prediction confidence interval. The confidence interval is estimated using a random deactivation method, which involves randomly shutting down some neuron connections multiple times during prediction. The mean and standard deviation of the multiple prediction results are calculated, with the mean serving as the final prediction value and the standard deviation reflecting the uncertainty of the prediction, thus providing a range of confidence for the prediction results.

[0099] Step 5.7: Identify high-risk pipe sections and generate a prediction report; Based on multi-step prediction results, high-risk pipe sections whose crystallization deposition state may deteriorate within future time windows are identified.

[0100] The criteria for determining high-risk pipe sections include the following three aspects: First, the predicted crystallization coverage or deposition thickness exceeds a preset warning threshold. These thresholds are determined based on historical data statistical analysis. The warning threshold for crystallization coverage is typically set at 40% to 60%, and the warning threshold for deposition thickness is determined based on the pipe diameter, generally set as the thickness value corresponding to a 10% to 15% reduction in the effective inner diameter. Second, the predicted condition level rises from the current level to a more severe level. The condition level uses the four-level classification standard of slight, moderate, severe, and extremely severe determined in step 3.7. A predicted condition level is considered high-risk when it is one level or more higher than the current condition level. Third, the predicted thickness growth rate exceeds a preset rate threshold. This rate threshold is determined based on the historical growth rate statistics of the pipe section, typically set at 1.5 to 2 times the historical average growth rate. Pipe sections meeting any of the above criteria are identified as high-risk pipe sections.

[0101] High-risk pipeline sections are identified and ranked according to their risk level. Risk level is quantified using a comprehensive risk index, which considers three factors: the severity of the predicted state, the rate of deterioration, and the time to reach the alert state. The severity of the predicted state is represented by the comprehensive evaluation score output in step 3; a higher score indicates greater risk. The rate of deterioration is calculated by dividing the difference between the predicted state score and the current state score by the predicted time window length; a faster rate of deterioration indicates greater risk. The time to reach the alert state is calculated using linear interpolation or trend extrapolation based on the predicted state change curve; the time when the predicted state first exceeds the alert threshold is considered the time to reach the alert state, with earlier occurrences indicating greater risk. The three factors are normalized and then weighted and summed according to weight coefficients to obtain the comprehensive risk index. These weight coefficients are set based on actual operational needs; typically, the weight for the severity of the predicted state is 0.4, the weight for the rate of deterioration is 0.3, and the weight for the reciprocal of the time to reach the alert state is 0.3. High-risk pipeline sections are ranked from highest to lowest according to the comprehensive risk index, with priority given to the sections with the highest risk index.

[0102] The system generates a forecast report, which includes the current status, forecast status, status change trends, estimated time to reach the warning state for each pipe segment, a list and ranking of high-risk pipe segments, and the comprehensive risk index and its constituent factors for each high-risk pipe segment. This information provides a basis for preventative maintenance decisions. The report visually displays the status forecast curves and risk distribution for each pipe segment in chart format, facilitating maintenance personnel to quickly identify pipe segments requiring special attention.

[0103] Furthermore, since standard graph attention networks only consider static connections between nodes and fail to fully utilize edge feature information, edge-enhanced graph attention networks can be adopted. The aim is to integrate pipeline attribute information into the spatial feature aggregation process, improving the accuracy of spatial correlation modeling. Specifically, when calculating the attention coefficient, in addition to considering the features of adjacent nodes, the edge features connecting the two nodes are also included in the calculation. The edge feature vector is linearly transformed and concatenated with the node feature vector, then input into the attention calculation network. In this way, edge attributes such as pipeline length, slope, and flow rate can affect the calculation of the attention coefficient, enabling the model to learn the differences in the spatial propagation patterns of crystallization deposition under different pipeline attribute conditions.

[0104] Step 6: Based on the type classification results, status level and prediction report, establish a processing method knowledge base, perform scheme matching and multi-objective optimization, and obtain recommended schemes; Specifically, the following steps are included: Step 6.1: Establish a knowledge base for crystallization deposition processing methods; Based on the physicochemical properties of different types of crystal deposition, a knowledge base is established to map the relationship between processing methods and crystal types.

[0105] The knowledge base uses a structured storage method, and each record contains fields such as crystallization type, applicable processing method, processing efficiency parameters, cost parameters, environmental impact parameters, and applicable condition constraints.

[0106] Suitable treatment methods for carbonate crystals include acidic solvent treatment and mechanical scraping. Acidic solvents can include hydrochloric acid, citric acid, acetic acid, etc., with different acids having different dissolution efficiencies and corrosiveness. Suitable treatment methods for sulfate crystals include chelating agent treatment and high-pressure water jet cleaning. Chelating agents can include ethylenediaminetetraacetic acid and its salts. Suitable treatment methods for metal hydroxide crystals include pH adjustment dissolution and chemical reduction, requiring the selection of appropriate treatment agents based on the specific metal type. Organic-inorganic composite crystals usually require a combined treatment approach, first using an oxidant to decompose the organic matter, and then using an acid or chelating agent to dissolve the inorganic components.

[0107] The knowledge base also stores information such as the operating parameter range, safety precautions, and waste liquid treatment requirements for each treatment method.

[0108] Step 6.2: Define the objective function and constraints for the multi-objective optimization problem; Given that the processing scheme needs to take into account multiple factors, the optimization of the processing scheme is modeled as a multi-objective optimization problem.

[0109] The optimization objectives include four aspects: maximizing treatment effectiveness, expressed as the crystal removal rate, which is equal to the ratio of the amount of crystals reduced after treatment to the amount of crystals before treatment; a higher removal rate is better. Minimizing treatment costs, including reagent costs, labor costs, equipment usage costs, and wastewater treatment costs, are calculated based on the treatment method and volume. Minimizing the impact on the normal operation of the drainage system, expressed as downtime and flow loss; some treatment methods require suspending pipeline operation, affecting the normal function of the drainage system. Minimizing environmental impact, including reagent residue, secondary pollution risk, and wastewater discharge volume. Constraints include reagent concentration not exceeding the safety limit, treatment time not exceeding the allowable downtime, and post-treatment water quality meeting discharge standards.

[0110] Step 6.3: Design a coding scheme for decision variables; Based on the characteristic that the processing scheme contains multiple types of decision variables, a hybrid coding scheme is designed to uniformly represent different types of variables.

[0111] The decision variables include: treatment method selection variable, which is a discrete variable and takes the value of the number of each treatment method in the knowledge base; treatment agent concentration variable, which is a continuous variable and takes the value range determined by the parameter range in the knowledge base; treatment time variable, which is a continuous variable and represents the duration of a single treatment; treatment frequency variable, which is an integer variable and represents the number of treatments within the planning period; and operation sequence variable, which is an arrangement variable and represents the order in which multiple pipe sections are treated.

[0112] Real number encoding is used to uniformly represent various variables. Discrete and integer variables are converted through rounding operations, and permutation variables are represented by random key encoding. That is, each pipe segment is assigned a random real number, and the operation order is determined by sorting the real numbers.

[0113] Step 6.4: The improved non-dominated sorting genetic algorithm is used to solve for the Pareto optimal solution set; Multi-objective optimization problems do not have a single optimal solution but rather a set of Pareto optimal solutions, which are solved using a non-dominated sorting genetic algorithm. The non-dominated sorting genetic algorithm encodes each treatment scheme as an individual, and the encoding length of an individual is equal to the sum of the dimensions of all decision variables.

[0114] During algorithm initialization, an initial population of a preset size is randomly generated. Each individual represents a processing scheme according to the encoding scheme in step 6.3. In the fitness evaluation phase, the processing scheme is decoded for each individual, and the target values ​​are calculated according to the objective function. In the non-dominated sorting phase, the population is divided into multiple non-dominated layers according to the Pareto dominance relationship. The first layer is the Pareto front of the current population, i.e., the set of non-dominated solutions. The second layer is the Pareto front after removing the first layer, and so on. In the crowding calculation phase, the crowding distance is calculated for individuals within the same non-dominated layer. The crowding distance reflects the distance between an individual and its neighboring individuals in the target space. The larger the distance, the sparser the solutions around the individual. Retaining the individual is beneficial to maintaining the diversity of solutions.

[0115] The selection operation prioritizes individuals with lower non-dominated hierarchies, and within the same hierarchy, prioritizes individuals with larger crowding distances to maintain solution diversity. The crossover operation employs a simulated binary crossover method, which mimics the single-point crossover behavior of a binary-coded genetic algorithm, combining the encoded values ​​of two parent individuals to generate offspring individuals.

[0116] The mutation operation employs a multinomial mutation method, which perturbs the encoded values ​​of individuals based on a multinomial probability distribution, with the perturbation magnitude decreasing as the number of generations increases. The parameters of the crossover and mutation operators are adaptively adjusted according to the number of generations, with larger perturbations used initially to enhance global search capabilities and smaller perturbations used later to enhance local search capabilities. Iterative evolution continues until a preset maximum number of generations is reached or a convergence condition is met. The convergence condition is that the Pareto front no longer undergoes significant changes over several consecutive generations.

[0117] Step 6.5: Select a recommended solution from the Pareto optimal solution set; Given that decision-makers may have different preferences, multiple options are offered as a selection strategy.

[0118] The ideal point method selects the solution closest to the ideal point, a virtual point where all objectives take optimal values. The distance is calculated using normalized Euclidean distance. The approximation-to-ideal-solution ranking method comprehensively considers both the closeness of the solution to the ideal point and its distance from the negative ideal point, a virtual point where all objectives take worst values. The solution with the highest relative closeness is selected, balancing the merits of each objective. The weighted sum method transforms multiple objectives into a single objective. A weighted objective value is calculated based on the objective weights set by the decision-maker, and the solution with the optimal weighted objective value is selected. This method is suitable when the decision-maker has a clear preference for the importance of each objective.

[0119] If the decision-maker does not specify a preference, a compromise solution selection strategy is adopted by default, choosing the equilibrium solution located in the middle of the Pareto front, which achieves a good compromise effect on all objectives. The output provides detailed information on the recommended solution, including the treatment method, agent type and concentration, treatment time, operation sequence for each pipe section, as well as expected treatment effects, cost estimates, and operational impact assessments.

[0120] Furthermore, since standard non-dominated sorting genetic algorithms may suffer from insufficient selection pressure when dealing with high-dimensional multi-objective problems, a reference-point-based non-dominated sorting genetic algorithm can be employed. The aim is to guide the population towards a uniform distribution across different regions of the Pareto front by introducing pre-defined reference points. Specifically, a set of uniformly distributed reference points is pre-set in the objective space. These reference points are generated using the systematic methods of Das and Dennis, obtained through uniform sampling on the simplex. In the selection operation, in addition to considering the non-dominated hierarchy, the association between individuals and reference points is also considered. Each individual is associated with the nearest reference point, with priority given to individuals associated with reference points having a smaller number of associated individuals, thus ensuring a uniform distribution of the population on the Pareto front. This method can better maintain solution diversity when the number of objectives is large.

[0121] Step 7: Establish a case database based on the recommended solutions, retrieve similar cases to generate processing suggestions, provide feedback and evaluation on the effects, and obtain intelligent processing decision results; Specifically, the following steps are included: Step 7.1: Establish a case database to store historical processing records; Based on the need to accumulate processing experience, a processing case database was designed to store complete information for each processing session.

[0122] Each treatment case includes the following fields: Case Number; Treatment Time; Pipeline Location; Pre-treatment Status, including crystallization coverage, average thickness, crystal type, and water quality parameters; Treatment Scheme Used, including treatment method, type and concentration of treatment agent, and treatment time; Treatment Costs, including reagent costs, labor costs, and equipment costs; Treatment Results, including crystal removal rate, post-treatment thickness, and post-treatment coverage; Operational Impacts, including water outage time and flow loss; Environmental Impacts, including wastewater discharge and treatment agent residue; and Personnel Evaluation, including subjective ratings. The case database is stored using a relational database and supports queries and searches based on various criteria.

[0123] After each processing operation, maintenance personnel enter the processing effect data, and the system automatically calculates the processing effect indicators and stores them in the case database. The accumulation of the case database provides a data foundation for subsequent case reasoning and knowledge extraction.

[0124] Step 7.2: Generate processing suggestions using a case-based reasoning method; Based on the principle that similar problems should be addressed with similar solutions, a case-based reasoning method is used to retrieve similar cases from historical cases and generate processing suggestions.

[0125] Case-based reasoning comprises four steps: retrieval, reuse, refinement, and retention. In the retrieval phase, similar historical cases are retrieved from the case database based on the current crystallization deposition state characteristics. Similarity is calculated using a weighted Euclidean distance method; the distance value equals the square root of the sum of the squares of the differences between each feature and their weights. The smaller the differences between the current state and historical cases on each feature, the higher the similarity. Features include crystal coverage, average thickness, crystal type, pH value, and conductivity. Weights are determined based on the influence of each feature on the selection of the treatment scheme. Crystal type and coverage have relatively high weights of 0.3 and 0.25, respectively, while other features have relatively low weights between 0.05 and 0.15. The top 5 cases with the highest similarity are retrieved, and their weighted average similarity is calculated. If the average similarity is greater than 0.8, a highly similar case is considered to have been found.

[0126] In the reuse phase, the treatment plan from the best-performing case among similar cases is extracted as the initial solution to the current problem. In the correction phase, the treatment plan is adjusted appropriately based on the differences between the current state and the states of similar cases; for example, if the current crystal thickness is thicker than in similar cases, the concentration of the treatment agent may be increased or the treatment time extended. In the retention phase, the current case and its treatment results are stored in a case database for later retrieval.

[0127] Step 7.3: Update the knowledge base rules based on statistical analysis of the processing results; Based on a large number of processing records accumulated in the case database, statistical analysis methods are used to extract processing patterns and update the knowledge base rules.

[0128] Statistical analysis includes the following: analyzing the optimal treatment method for different crystallization types, statistically analyzing the average removal rate and cost of different treatment methods for each type of crystallization, and selecting the treatment method with high removal rate and low cost as the recommended method for that type. For example, statistics show that the average removal rate of carbonate crystallization using citric acid dissolution is 92%, with a cost of 150 yuan per meter of pipe, which is better than other methods. Therefore, citric acid dissolution is set as the preferred treatment method for carbonate crystallization. Analyzing the relationship between treatment parameters and treatment effect, using regression analysis to establish an empirical formula between treatment agent concentration, treatment time, and removal rate. For example, for carbonate crystallization, there is a power function relationship between removal rate and citric acid concentration and treatment time; the removal rate increases with increasing concentration and time. The parameters in the empirical formula are determined by fitting historical data using the least squares method. Analyzing the reasons for treatment failures, identifying key factors leading to poor treatment effect, such as insufficient treatment agent concentration, too short treatment time, and incorrect crystallization type judgment, summarizing these lessons learned into rules and storing them in a knowledge base to avoid repeating mistakes, analyzing the differences in crystallization characteristics of different pipe sections, identifying high-incidence and special pipe sections, and developing targeted treatment strategies for these sections. A statistical analysis is conducted quarterly, and the knowledge base is updated with recommended processing methods, empirical formulas for parameters, precautions, and other relevant information based on the analysis results.

[0129] Step 7.4: Establish a mechanism for evaluating and providing feedback on the treatment plan; Based on the need for continuous improvement, an evaluation and feedback mechanism for processing solutions will be established to collect feedback from operations and maintenance personnel. The evaluation mechanism includes both quantitative and qualitative evaluations.

[0130] Quantitative evaluation is based on objective treatment effect data, including indicators such as crystallization removal rate, treatment cost, and water outage time. It automatically calculates the values ​​of each indicator and compares them with expected values ​​to generate a treatment effect score. The quantitative score calculation method is as follows: the deviation rate between the actual and expected values ​​of each indicator is normalized. The crystallization removal rate is calculated as a positive indicator (actual removal rate divided by expected removal rate), while treatment cost and water outage time are calculated as negative indicators (expected value divided by actual value). The normalized values ​​of each indicator are weighted and summed according to weight coefficients to obtain the quantitative score. The weight coefficients are determined based on the importance of the indicator: crystallization removal rate has a weight of 0.5, treatment cost has a weight of 0.3, and water outage time has a weight of 0.2. The quantitative score ranges from 0 to 100 points.

[0131] Qualitative evaluation is conducted subjectively by operations and maintenance personnel based on their actual operational experience. The evaluation content includes the feasibility of the solution, security, and environmental impact, using a five-level scoring system. The five-level scoring criteria and their corresponding numerical scores are as follows: A score of 100 corresponds to an excellent rating. The criteria for this rating are: the treatment plan fully achieves the expected results; the operation is simple, smooth, and without difficulty; the safety measures are complete and there are no safety hazards; there is no adverse impact on the environment and the residue of the treatment agent is within the allowable range; and the overall execution process is smooth and requires no additional adjustments. A good grade corresponds to 80 points. The criteria for judgment are that the treatment plan basically achieves the expected results, there are minor inconveniences in the operation process but they do not affect the overall execution, the safety measures are basically in place with only a few details that need attention, the environmental impact is small and within an acceptable range, and the overall execution process is relatively smooth with only minor adjustments required. The general level corresponds to 60 points. The judgment criteria are: the solution barely achieves the expected effect, there are certain difficulties in the operation process that require additional coordination, there are insufficient safety measures that require on-site supplementary protection, there is a certain impact on the environment that requires additional control measures, and the overall execution process requires multiple adjustments to complete. The poor grade corresponds to 40 points. The criteria for judgment are: the treatment plan fails to achieve the expected effect or the effect is significantly lower than expected; there are many difficulties in the operation process, which seriously affects the execution efficiency; there are insufficient safety measures and obvious safety hazards; there is a significant impact on the environment, which requires additional pollution control measures; and the overall execution process is not smooth and requires frequent adjustments or interruptions. A very poor rating corresponds to 20 points. The criteria for judgment are that the treatment plan has not achieved the expected results or has a negative impact, the operation process is extremely difficult or even impossible to execute normally, safety measures are seriously lacking and there are major safety risks, it causes obvious pollution or secondary pollution to the environment, and the overall implementation fails and a new plan needs to be formulated.

[0132] After the maintenance personnel select a five-level rating system in the evaluation interface, the rating will be automatically converted into a corresponding numerical score as a qualitative score.

[0133] The quantitative and qualitative evaluations are combined to form a comprehensive score for the treatment plan. The comprehensive score is calculated as a weighted average of the quantitative and qualitative scores, with a weight of 0.7 for the quantitative score and 0.3 for the qualitative score. The comprehensive score ranges from 0 to 100 points. The scoring results are stored in the case database, and detailed data for each sub-indicator are recorded for subsequent analysis.

[0134] For solutions with a comprehensive score below 50, they are automatically classified as low-scoring solutions, and operations personnel are prompted to fill in a problem description and improvement suggestions. The problem description should detail the specific reasons for the low score, and the improvement suggestions should propose targeted optimization measures. This feedback information is used for subsequent knowledge base optimization and rule updates. For solutions with a comprehensive score above 85, they are marked as excellent cases, given priority in case reasoning, and their processing parameters and operation procedures are used as benchmark cases for reference in other similar situations.

[0135] The feedback mechanism is implemented through a user interface, which provides a concise evaluation form. The form includes a five-level rating selection button, satisfaction ratings for each sub-indicator, a problem description text box, and an improvement suggestion text box. Maintenance personnel can quickly complete the evaluation after handling the issue. The required field is the five-level rating selection; other fields are optional. Evaluation data is synchronized to the case database in real time. Regular evaluation statistical reports are generated, displaying the average score for various handling solutions, score distribution, and the proportion of high-scoring and low-scoring cases, providing data support for continuous optimization of handling decisions.

[0136] Step 7.5: Achieve continuous optimization of processing decisions; With the continuous accumulation and updating of case databases and knowledge bases, decision processing performance is continuously improved.

[0137] Optimizations include: expanding the case database—as the number of processing runs increases, the number of cases in the database grows, covering a more comprehensive state space, enabling case reasoning to find more similar historical cases and improving the accuracy of recommended solutions; refining knowledge rules—through regular statistical analysis, the processing method recommendations and parameter empirical formulas in the knowledge base are continuously optimized to reflect the latest processing experience and patterns; identifying special cases—by analyzing failed and abnormal cases, situations requiring special handling are identified, such as certain pipe sections requiring unconventional processing methods due to special chemical environments, and these special rules are added to the knowledge base; and improving processing efficiency—by optimizing processing parameters and workflows, processing costs and time are reduced while ensuring processing effectiveness.

[0138] Regularly generate optimization reports to show the trend of changes in processing decision performance, including indicators such as the improvement of average removal rate, the reduction of average cost, and the improvement of solution recommendation accuracy, providing a basis for continuous improvement.

[0139] Step 7.6: Output the optimized processing decision results and an evaluation of the improvement effect; Based on the accumulation of case databases and knowledge bases, the output of decision-making results is continuously improved.

[0140] Based on the current crystallization deposition state, similar historical cases are quickly retrieved using a case-based reasoning method. Combined with the processing rules in the knowledge base, processing suggestions are automatically generated, including recommended processing methods, treatment agent type and concentration, processing time, expected effects and costs.

[0141] It also provides multiple alternative solutions for operations and maintenance personnel to choose from, with each solution's advantages, disadvantages, and applicable conditions noted. An improvement effectiveness evaluation report is output, including indicators such as the growth trend of the case database, the number of times knowledge base rules are updated, changes in the accuracy of processing solution recommendations, improvements in average processing effectiveness, and reductions in average processing costs.

[0142] The evaluation metrics are calculated as follows: Recommendation accuracy equals the number of times maintenance personnel adopt recommended solutions divided by the total number of recommendations; Average processing effectiveness improvement rate equals the average removal rate in the current period minus the average removal rate in the initial period, then divided by the average removal rate in the initial period; The average cost reduction rate is calculated in a similar way.

[0143] The report also includes analyses of typical success and failure cases, summarizing lessons learned. It automatically generates treatment suggestions when a crystalline deposition state requiring intervention is detected, assisting maintenance personnel in decision-making and improving treatment efficiency and effectiveness.

[0144] Furthermore, when the case database has a limited number of cases, case-based reasoning may not find enough similar historical cases. Rule-based expert reasoning can be used as a supplement to provide reasonable processing suggestions even with insufficient cases. Specifically, expert reasoning includes a set of IF-THEN rules, summarized by domain experts based on their processing experience. For example: IF crystal type is carbonate AND crystal coverage is greater than 50% THEN recommends citric acid dissolution treatment at a concentration of 5% to 10%, with a processing time of 2 to 4 hours. The rule base initially contains 20 to 30 basic rules, covering common crystal types and state combinations. When case-based reasoning cannot find similar cases (average similarity less than 0.6), it automatically switches to rule-based reasoning mode, matching applicable rules based on the current state and generating processing suggestions. As the case database accumulates, the frequency of case-based reasoning gradually increases, while the frequency of rule-based reasoning gradually decreases, eventually forming a hybrid reasoning mode with case-based reasoning as the primary method and rule-based reasoning as a supplement.

[0145] A smart monitoring and treatment system for crystallization deposition in drainage pipes is provided to execute the aforementioned smart monitoring and treatment method for crystallization deposition in drainage pipes. Figure 3 As shown, it includes: The multimodal data acquisition module is used to acquire multimodal sensor data of drainage pipelines and perform spatiotemporal alignment preprocessing to obtain aligned multimodal monitoring datasets. The crystallization detection and identification module, based on the pipe inner wall images in the aligned multimodal monitoring dataset, performs crystallization deposition area detection and type identification, and obtains crystallization deposition segmentation mask and type classification results; The comprehensive evaluation module statistically analyzes the spatial distribution characteristics of crystallization regions based on the crystallization deposition segmentation mask, and performs a multi-dimensional comprehensive evaluation based on the aligned multi-modal monitoring dataset to obtain the comprehensive evaluation score and state level of crystallization deposition. The risk warning module, based on the aligned multimodal monitoring dataset, monitors changes in wastewater composition and provides early warning of crystallization risk, obtaining component mutation detection results and crystallization risk warning information; The trend prediction module, based on the comprehensive evaluation score of crystal deposition, the results of component mutation detection, and the early warning information of crystallization risk, predicts the development trend of crystal deposition, identifies high-risk pipe sections, and generates a prediction report. The solution optimization module establishes a knowledge base of processing methods based on type classification results, status levels, and prediction reports, performs solution matching and multi-objective optimization, and obtains recommended solutions. The intelligent decision-making module establishes a database of processing cases based on recommended solutions, retrieves similar cases to generate processing suggestions, provides feedback and evaluation on the effects, and obtains intelligent processing decision results.

[0146] In one embodiment of the present invention, a specific example is provided: The wastewater treatment plant in the chemical industrial park has an influent main pipeline system. The network consists of branch pipes from five chemical companies converging into a single main pipe. This main pipe has a diameter of DN800 and a total length of approximately three kilometers, constructed of fiberglass. The wastewater transported by the branch pipes exhibits significant differences in composition, including pharmaceutical wastewater with high concentrations of calcium and magnesium ions, dye wastewater containing sulfates, and electroplating wastewater containing heavy metals. During operation, the different wastewaters frequently crystallize and deposit at the confluence point, severely impacting drainage capacity.

[0147] The method of this invention was used for intelligent monitoring and processing of the pipeline network. A 30-day field test was conducted in the main pipeline and branch pipe confluence areas. During the test, three sets of pipeline endoscopic video acquisition units, twelve sets of ultrasonic thickness sensor arrays, eight conductivity monitoring nodes, and two multi-parameter water quality analyzers were deployed, monitoring a total pipeline length of approximately 1.5 kilometers, covering five key monitoring sections. The sensing system collected data according to a preset strategy, and after spatiotemporal alignment preprocessing, a unified multimodal dataset was formed. An example of multimodal monitoring data collected at a certain moment is shown in Table 1: Table 1: Example of multimodal monitoring data collected at a certain moment; Based on the acquired image data, a deep learning model was used for crystal deposition detection and type identification. The model identified carbonate-type crystal deposition at the main pipe station K0+500, a composite crystal deposition of calcium carbonate and magnesium hydroxide at the inlet point of branch pipe A, and calcium sulfate-type crystal deposition at the inlet point of branch pipe B.

[0148] Based on a comprehensive assessment method using multi-source data fusion, the crystallization deposition state level at each monitoring location was calculated. The main pipe station K0+500 and the branch pipe A confluence point were assessed as moderate and require attention; other locations were assessed as slight and require no immediate action.

[0149] The wastewater composition mutation detection module continuously monitors changes in water quality parameters. On a certain day, it detected a sudden increase in the calcium ion concentration in branch pipe A to 520 mg / L, while the carbonate concentration in the main pipe also rose to 180 mg / L. The calculated supersaturation of calcium carbonate was 3.8, triggering a warning level alert, and it was recommended to take preventive measures.

[0150] Based on historical data and the current status, the spatiotemporal neural network prediction model predicts that the deposition thickness at the main pipeline station K0+500 will continue to increase to more than 15mm within the next week, which may deteriorate from the moderate level to the severe level and be identified as a high-risk pipeline section.

[0151] The multi-objective optimization algorithm generates treatment plans based on crystallization type and state level. For carbonate-type crystals at main pipeline station K0+500, citric acid solvent is recommended for treatment, with the treatment scheduled during low-flow hours at night. This approach is expected to achieve a high crystal removal rate and minimize impact on the drainage system operation.

[0152] The generated processing solutions and implementation effect data are shown in Table 2: Table 2: Examples of generated processing solutions and implementation results; After processing and execution, feedback data is collected, and the case reasoning module stores the processed cases in the knowledge base. Through multiple processing cycles, the knowledge base is gradually improved, enabling it to more accurately recommend the optimal processing solution for different states.

[0153] By applying the method of this invention, the crystallization and deposition problem in the pipeline network of the chemical industrial park has been effectively controlled, realizing the transformation from passive response to active prevention. The pipeline flow capacity has been maintained at a good level, reducing the occurrence of emergency repair incidents.

[0154] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A method for intelligent monitoring and treatment of crystallization deposition in drainage pipes, characterized in that, Includes the following steps: Multimodal sensor data of drainage pipelines are collected and spatiotemporal aligned preprocessing is performed to obtain aligned multimodal monitoring datasets; Based on the pipe inner wall images in the aligned multimodal monitoring dataset, crystallization deposition area detection and type identification are performed to obtain crystallization deposition segmentation mask and type classification results; Based on the spatial distribution characteristics of the crystallization region statistically analyzed by the crystallization deposition segmentation mask, a multi-dimensional comprehensive evaluation was performed based on the aligned multi-modal monitoring dataset to obtain the comprehensive evaluation score and state level of the crystallization deposition. Based on the aligned multimodal monitoring dataset, changes in wastewater composition are monitored and early warning of crystallization risk is carried out, resulting in component mutation detection results and crystallization risk warning information. Based on the comprehensive evaluation score of crystallization deposition, the results of component mutation detection, and the early warning information of crystallization risk, the development trend of crystallization deposition is predicted, high-risk pipe sections are identified, and a prediction report is generated. Based on the type classification results, status level and prediction report, a knowledge base of processing methods is established, and scheme matching and multi-objective optimization are performed to obtain recommended schemes. A database of processing cases is established based on the recommended solutions. Similar cases are retrieved to generate processing suggestions. Feedback and evaluation of the effects are conducted to obtain intelligent processing decision results.

2. The intelligent monitoring and treatment method for crystallization deposition in drainage pipes according to claim 1, characterized in that, The process of collecting multimodal sensor data from drainage pipelines and performing spatiotemporal alignment preprocessing specifically includes: Deploy a pipeline endoscopic video acquisition unit to acquire image sequences of the pipeline's inner wall, and configure an LED supplementary lighting system with adaptive brightness adjustment; An ultrasonic thickness sensor array is deployed to acquire deposition layer thickness distribution data, and the deposition layer thickness is calculated based on the echo arrival time difference and the sound velocity in the medium. Deploy distributed conductivity monitoring nodes to acquire time-series data of wastewater conductivity; Deploy an online multi-parameter water quality analyzer to acquire time-series data of wastewater quality parameters; A unified spatial reference system is established by calibrating the positions of each sensor; a spatiotemporal alignment algorithm based on maximizing mutual information is used to eliminate spatiotemporal bias in multi-source data.

3. The intelligent monitoring and treatment method for crystallization deposition in drainage pipes according to claim 1, characterized in that, The obtained crystallization deposition segmentation mask and type classification results specifically include: Construct a multi-scale feature extraction backbone network to output feature maps at different resolutions; A channel attention module is introduced to perform global average pooling on the input feature map along the spatial dimension to obtain the channel description vector. The channel attention weight vector is obtained through the activation function, and the attention weight vector is multiplied with the original feature map channel by channel. A spatial attention module is introduced to perform max pooling and average pooling on the input feature map along the channel dimension, and the spatial attention map is obtained through the activation function. Multi-level feature fusion is achieved through a feature pyramid network; water quality parameters are combined to assist in crystallization type identification; and a multi-task learning network is constructed to achieve joint optimization of region segmentation and type identification.

4. The intelligent monitoring and treatment method for crystallization deposition in drainage pipes according to claim 1, characterized in that, The obtained comprehensive evaluation score and state level of crystallization deposition specifically include: Based on the image segmentation results, spatial distribution characteristics of crystal deposition are calculated, including crystal coverage, distribution uniformity, and aggregation degree. Based on ultrasonic thickness measurement data, the characteristic indicators of the deposition layer thickness are calculated, including average thickness, maximum thickness, thickness standard deviation, and thickness growth rate. Based on water quality parameter data, the compositional tendency of crystallization deposition is inferred. The Langerier saturation index is used to assess the scaling tendency of calcium carbonate, and the Rezner stability index is used to assess the degree of scaling or corrosion. The supersaturation is determined by calculating the ratio of the ion activity product to the solubility product constant of calcium sulfate. The impact of crystallization deposition on pipeline flow capacity was calculated based on a hydraulic model; a multi-dimensional comprehensive evaluation index system was constructed and the weights were determined using the analytic hierarchy process; and the crystallization deposition state level was determined using a fuzzy comprehensive evaluation method.

5. The intelligent monitoring and treatment method for crystallization deposition in drainage pipes according to claim 1, characterized in that, The obtained component mutation detection results and crystallization risk warning information specifically include: A dynamic baseline model of wastewater composition was established, and historical data was selected as the baseline calculation window using the sliding window method. The statistical characteristics of the parameters were calculated within the window. A multi-parameter joint mutation detection algorithm is designed to calculate the standardized deviation of the current value of the monitored parameter from the baseline, use the cumulative sum control chart method to detect the persistent shift of the parameter, and use the exponential weighted moving average method to detect the mutation of the parameter. A wastewater crystallization risk assessment model based on chemical thermodynamics was constructed to calculate the supersaturation of sparingly soluble salts under current wastewater conditions. Establish a mixing effect prediction model for confluence nodes, calculate the composition of the mixed wastewater based on the real-time flow and water quality parameters of each branch pipe using the principle of mass conservation, generate graded early warning information and push disposal suggestions.

6. The intelligent monitoring and treatment method for crystallization deposition in drainage pipes according to claim 1, characterized in that, The generation of the prediction report specifically includes: The graph structure of the drainage network is constructed, where nodes represent monitoring sections or key locations on the pipeline, edges represent pipeline connections, node feature vectors contain crystallization and deposition state indicators, water quality parameters, and pipeline attributes, and edge feature vectors contain the physical attributes of the pipeline. The time dimension feature extraction module is designed, and a gated loop unit is used to encode the historical state sequence of each node. The temporal evolution law of crystal deposition is captured by the update gate and reset gate control mechanism. The design of the spatial dimension feature aggregation module employs a graph attention network to aggregate feature information of neighboring nodes and calculate attention coefficients for neighboring nodes; it also integrates temporal and spatial dimension features for joint modeling. Combine wastewater crystallization risk assessment models to help predict the development trend of crystallization and deposition; perform multi-step time-series predictions to output future state prediction values; identify high-risk pipe sections and generate prediction reports.

7. The intelligent monitoring and treatment method for crystallization deposition in drainage pipes according to claim 6, characterized in that, The criteria for determining high-risk pipeline sections include: The predicted crystallization coverage or deposition thickness exceeds the preset warning threshold; the predicted state level rises from the current level to a more severe level; the predicted thickness growth rate exceeds the preset rate threshold. The identified high-risk pipe sections are sorted according to their risk level. The comprehensive risk index is calculated by taking into account the severity of the predicted state, the rate of deterioration, and the time to reach the warning state. After normalizing the three factors, the comprehensive risk index is obtained by weighting and summing them according to the weight coefficients. A prediction report is then generated.

8. The intelligent monitoring and treatment method for crystallization deposition in drainage pipes according to claim 1, characterized in that, The recommended solution specifically includes: Establish a knowledge base for crystallization deposition treatment methods. Each record includes crystallization type, applicable treatment method, treatment efficiency parameters, cost parameters, environmental impact parameters, and applicable condition constraints. Define the objective function and constraints of the multi-objective optimization problem. The optimization objectives include maximizing the treatment effect, minimizing the treatment cost, minimizing the impact on the normal operation of the drainage system, and minimizing the environmental impact. Design a coding scheme for decision variables, which include variables for treatment method selection, treatment agent concentration, treatment time, treatment frequency, and operation sequence. A non-dominated sorting genetic algorithm is used to solve the Pareto optimal solution set; a recommended scheme is selected from the Pareto optimal solution set.

9. The intelligent monitoring and treatment method for crystallization deposition in drainage pipes according to claim 1, characterized in that, The obtained intelligent processing decision results specifically include: Establish a database of processing cases to store historical processing records. Each processing case includes the pre-processing status, the processing solution adopted, the processing cost, and the processing effect. A case-based reasoning approach is used to generate processing suggestions. The case-based reasoning includes four steps: retrieval, reuse, modification and retention. In the retrieval stage, similar historical cases are retrieved from the case database based on the current crystallization deposition state characteristics. The similarity is calculated using the weighted Euclidean distance method. The knowledge base rules are updated based on statistical analysis of the treatment effects; a treatment scheme evaluation and feedback mechanism is established, which includes quantitative evaluation and qualitative evaluation. The qualitative evaluation adopts a five-level system, and the quantitative and qualitative evaluations are combined to form a comprehensive score for the treatment scheme; thus achieving continuous optimization of treatment decisions.

10. A smart monitoring and treatment system for crystallization deposition in drainage pipes, characterized in that, A method for intelligent monitoring and treatment of crystallization deposition in drainage pipes as described in any one of claims 1-9 includes: The multimodal data acquisition module is used to acquire multimodal sensor data of drainage pipelines and perform spatiotemporal alignment preprocessing to obtain aligned multimodal monitoring datasets. The crystallization detection and identification module, based on the pipe inner wall images in the aligned multimodal monitoring dataset, performs crystallization deposition area detection and type identification, and obtains crystallization deposition segmentation mask and type classification results; The comprehensive evaluation module statistically analyzes the spatial distribution characteristics of crystallization regions based on the crystallization deposition segmentation mask, and performs a multi-dimensional comprehensive evaluation based on the aligned multi-modal monitoring dataset to obtain the comprehensive evaluation score and state level of crystallization deposition. The risk warning module, based on the aligned multimodal monitoring dataset, monitors changes in wastewater composition and provides early warning of crystallization risk, obtaining component mutation detection results and crystallization risk warning information; The trend prediction module, based on the comprehensive evaluation score of crystal deposition, the results of component mutation detection, and the early warning information of crystallization risk, predicts the development trend of crystal deposition, identifies high-risk pipe sections, and generates a prediction report. The solution optimization module establishes a knowledge base of processing methods based on type classification results, status levels, and prediction reports, performs solution matching and multi-objective optimization, and obtains recommended solutions. The intelligent decision-making module establishes a database of processing cases based on recommended solutions, retrieves similar cases to generate processing suggestions, provides feedback and evaluation on the effects, and obtains intelligent processing decision results.

Images