Intelligent evaluation and repair decision method for sewer pipe defect grade based on big data

Abstract

This invention discloses an intelligent assessment and repair decision-making method for drainage pipeline defects based on big data, specifically involving the field of assessment and decision-making. The method includes: constructing a five-dimensional framework; collecting various types of data such as pipe segment attributes and defect status; forming a spatiotemporally aligned data set through preprocessing, quality control, and standardization; designing a knowledge graph ontology framework based on this set; extracting entity instances and relationships to form an initial graph; and outputting high-dimensional feature vectors of pipe segments via a graph neural network; quantifying and determining defect levels; predicting the spatiotemporal evolution trend of defects through a model that integrates physical laws and data-driven approaches; quantifying risks in multiple dimensions and calculating the total risk; constructing four types of functional intelligent agents; achieving collaborative decision-making through multi-objective optimization and reinforcement learning; and generating a dynamic decision graph hierarchically prioritized. This invention achieves intelligent and precise defect assessment and maintenance decision-making, improves operation and maintenance efficiency and scientific rigor, and adapts to the dynamic changes in the pipeline network.

Description

Technical Field

[0001] This invention relates to the field of assessment and decision-making technology, and more specifically, to a method for intelligent assessment and repair decision-making of drainage pipeline defect levels based on big data. Background Technology

[0002] As a core component of urban infrastructure, drainage pipelines play a vital role in rainwater drainage and sewage transport. Their operational status directly affects urban water quality, flood control and drainage capacity, and the living environment of residents. With the advancement of urbanization, the service life of existing drainage pipelines continues to increase. Affected by various factors such as geological subsidence, media corrosion, construction defects, and external disturbances, these pipelines are prone to various structural and functional defects, requiring timely defect assessment and repair.

[0003] Currently, urban drainage pipe networks cover a wide area and have complex pipeline layouts. Under the traditional operation and maintenance management model, the investigation, assessment, and repair decisions regarding pipeline defects rely heavily on manpower and resources. To ensure the stable and efficient operation of drainage pipe network systems and improve the targetedness and timeliness of operation and maintenance management, there is an urgent need within the industry for accurate identification, scientific assessment, and reasonable repair decisions regarding drainage pipeline defects. The research and application of related technologies have become an important direction for promoting the intelligent development of drainage pipe network operation and maintenance.

[0004] However, it still has some drawbacks in practical use, such as: 1. Data collection is limited in scope and lacks standardization, relying heavily on single detection methods to obtain localized data. This makes it difficult to comprehensively cover information such as pipe segment attributes, operating status, and environmental impact. The data is severely fragmented, failing to provide complete and unified data support for assessment decisions. This results in a weak assessment foundation and affects the accuracy of subsequent processes.

[0005] 2. Defect assessments often rely on manual experience or simple model calculations, which are highly subjective and do not take into account the interrelationships of various elements in the pipeline network. This makes it difficult to accurately quantify the defect level and evolution trend, and is prone to assessment bias. Consequently, maintenance plans lack specificity and cannot efficiently solve actual pipeline problems.

[0006] 3. The risk quantification dimensions are one-sided, focusing only on the single dimension of structural safety while ignoring key factors such as functional protection and environmental impact. Furthermore, the weight allocation relies on subjective settings, which cannot objectively reflect the true risk status of the pipeline section, resulting in low credibility of risk assessment results and difficulty in guiding scientific operation and maintenance.

[0007] 4. Maintenance decisions lack dynamic adaptability and are mostly based on static solutions. They do not fully consider real-time changes in the pipeline network status and constraints such as budget and construction resources. The decision-making flexibility is insufficient, making it difficult to adapt to the dynamic operation and maintenance needs of complex pipeline networks, resulting in low operation and maintenance efficiency. Summary of the Invention

[0008] To overcome the aforementioned deficiencies of the prior art, this invention provides a method for intelligent assessment and repair decision-making of drainage pipeline defect levels based on big data, which addresses the problems mentioned in the background art through the following solutions.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a method for intelligent assessment and repair decision-making of drainage pipeline defect levels based on big data, comprising: S1: Intelligent acquisition and standardized processing of multi-source data: Construct a five-dimensional acquisition framework to collect pipeline attributes, defect status, operating parameters, environmental factors and business data. After preprocessing and quality control, a standardized data set with spatiotemporal alignment is formed through time, space and numerical standardization. S2: Pipeline Knowledge Graph Construction: Based on a standardized dataset, design a knowledge graph ontology framework and extract entity instances and relationships to form an initial knowledge graph. The graph neural network then learns and outputs high-dimensional feature vectors of pipe segments. S3: Intelligent assessment of prognostic defect levels: Based on standardized datasets and high-dimensional feature vectors of pipe segments, the defect level is quantitatively determined. By integrating physical laws and data-driven models, the spatiotemporal evolution trend of defects is predicted. The risk of pipe segments is quantified in multiple dimensions and the total risk is calculated to generate a standardized assessment report. S4: Multi-agent Co-evolutionary Decision Making: Based on defect level, evolution trend and total risk data, four types of functional agents are constructed. Through multi-objective optimization and multi-agent reinforcement learning, collaborative decision making is achieved, and repair decision schemes are output and dynamic decision graphs are generated in a priority-stratified manner.

[0010] The technical effects and advantages of this invention are as follows: 1. Construct a five-dimensional full-scenario data acquisition framework to achieve comprehensive coverage of multi-source data. Through preprocessing, quality control and standardization, form a spatiotemporally aligned data set to solve the problems of data fragmentation and insufficient standardization in traditional technologies, and lay a solid data foundation for evaluation and decision-making.

[0011] 2. By introducing knowledge graph and graph neural network technologies, the relationships between various elements of the pipeline network can be accurately represented and the high-dimensional features of the pipeline segments can be mined. Combined with a model that integrates physical laws and data-driven approaches, the defect level can be accurately determined and the evolution trend can be predicted, thereby improving the objectivity and accuracy of the assessment.

[0012] 3. Establish a multi-dimensional risk quantification system, comprehensively consider risks from structural safety, functional assurance, and environmental impact, and use the entropy weight method to achieve objective weight allocation, truly reflect the risk status of the pipeline section, provide a scientific basis for maintenance priority ranking, and improve the comprehensiveness and credibility of risk assessment.

[0013] 4. Based on multi-agent collaborative decision-making, a dynamic decision graph is generated, which can dynamically adjust maintenance plans in combination with real-time pipeline network data, taking into account multiple constraints such as budget and construction resources. It has strong adaptability and flexibility, effectively adapts to the dynamic operation and maintenance needs of complex pipeline networks, and improves the scientific nature of decision-making and execution efficiency. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the overall structure of the present invention.

[0015] Figure 2 This is a schematic diagram of the S1 process of the present invention.

[0016] Figure 3 This is a schematic diagram of the S2 process of the present invention.

[0017] Figure 4 This is a schematic diagram of the S3 process of the present invention.

[0018] Figure 5 This is a schematic diagram of the S4 process of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] refer to Figures 1-5 The big data-based intelligent assessment and repair decision-making method for drainage pipeline defect levels shown includes: S1: Intelligent acquisition and standardized processing of multi-source data: First, a five-dimensional data acquisition framework is constructed, clarifying the data acquisition methods, parameter definitions, and uses for each dimension. Then, pipe segment attributes, defect status, operating parameters, and environmental factors are collected synchronously in a unified format. Multi-level preprocessing removes outlier data and supplements missing data, with quality scoring to ensure data quality. Finally, standardization processing across time, space, and numerical dimensions forms a spatiotemporally aligned standardized data cube, providing foundational data support for subsequent knowledge graph construction, defect assessment, and decision analysis. The specific steps are as follows: S101: Establish a four-dimensional integrated data acquisition framework: Construct a five-dimensional, full-scenario data acquisition framework encompassing "sky-air-ground-internal-business systems," clearly defining the core acquisition devices and data traceability elements for each dimension. Data traceability elements: All collected data is accompanied by a timestamp. (Data acquisition time, format YYYY-MM-DDHH:MM:SS, accurate to the second), location information (Using WGS84 coordinate system latitude and longitude coordinates, formatted as [longitude, latitude], accuracy ±0.01°) and metadata of the data acquisition equipment. (Includes unique device serial number, calibration time, measurement accuracy, and installation location); S102: Synchronous Acquisition of Multi-Source Heterogeneous Data: Perform data acquisition according to a unified format. Space-based level: Regional surface subsidence data were collected via the Sentinel-1 synthetic aperture radar satellite, denoted as... (Unit: mm, accuracy ±0.1 mm); At the airborne level: Oblique photography was performed using a DJI M300RTK drone equipped with a Zenmuse P1 full-frame camera to acquire a 3D real-world model of the pipeline network area (including precise pipeline locations and topological relationships), denoted as... (Includes pipe location coordinates) (WGS84 coordinate system latitude and longitude [lon, lat], accuracy ±0.1m), topological relationships (An adjacency matrix representing the connection relationship between pipe segments, where 0 indicates no connection and 1 indicates a connection). Ground level: Data is collected through fixed monitoring stations integrating multi-sensor modules. Pipeline flow data is collected using an electromagnetic flow sensor and denoted as Q (unit: m³). 3 / s, accuracy ±0.01m 3 / s); Pipeline flow velocity data is collected by a Doppler flow velocity sensor and recorded as V (unit: m / s, accuracy ±0.01m / s). Pipeline water level data is collected using a static pressure water level sensor and denoted as H (unit: m, accuracy ±0.01m). Pipeline pressure data is collected using a diffused silicon pressure sensor and denoted as P (unit: kPa, accuracy ±1 kPa). Water quality parameter data are collected using a multi-parameter water quality sensor (integrating COD and ammonia nitrogen detection modules) and recorded as follows: (Includes chemical oxygen demand (COD) and ammonia nitrogen concentration (NH3-N), in mg / L and mg / L respectively, with an accuracy of ±0.1 mg / L). Harmful gas concentration data are collected using an electrochemical harmful gas sensor (for H2S and CH4), and denoted as... (Contains hydrogen sulfide H2S and methane CH4, in ppm, with an accuracy of ±1ppm). Construction vibration data was collected using piezoelectric vibration sensors and denoted as follows: (Unit: mm / s, accuracy ±0.1 mm / s); Ground traffic load data is collected using a resistance strain gauge pressure sensor and denoted as... (Unit: kN / m) 2 Accuracy ±0.1kN / m 2 ); Soil pH data were collected using a soil pH meter (accuracy ±0.1) and recorded as follows: (Dimensionless, range [0, 14]); Soil temperature data was collected using a platinum resistance temperature sensor (PT100) and denoted as follows: (Unit: ℃, accuracy ±0.5℃); Data was collected using a mobile inspection vehicle equipped with a pipe section property detection module. Material data of the pipe section was collected using an ultrasonic thickness gauge and recorded as follows: (Categorical variables, with values ​​such as concrete, cast iron, PVC, etc., are used in the calculation after being encoded using unique thermal codes, and the material identification accuracy is ≥95%). The diameter data of the pipe section is collected by a laser rangefinder and recorded as follows: (Unit: m, accuracy ±0.01m); The burial depth data of the pipe section was collected using a ground-penetrating radar detector and recorded as follows: (Unit: m, accuracy ±0.01m); Internal Pipeline Level: Pipeline crack length data are collected using a CCTV pipeline inspection robot (equipped with a high-definition camera and laser ranging module), and recorded as follows: (Unit: m, accuracy ±0.01m); Data on pipeline corrosion area was collected using a sonar-guided pipeline inspection robot (frequency 200kHz), and denoted as... (Unit: m) 2 , accuracy ±0.01m 2 ); Using a laser 3D scanning robot (point cloud density 100 points / mm) 2 Collect pipeline defect location data, denoted as (WGS84 coordinate system latitude and longitude [lon, lat], accuracy ±0.1m); At the business system level: Pipeline maintenance cost data is collected through the pipeline network operation and maintenance management system (ERP module) and recorded as... (Unit: Yuan, accuracy ±1 Yuan), obtain the total annual maintenance budget for the pipeline network, denoted as... And obtain the allocated budget and the remaining budget, denoted as: , ; Repair material inventory data is collected through the inventory management system and recorded as follows: (Includes material type and inventory quantity. Material type is a category variable, and inventory quantity is in pieces, with an accuracy of ±1 piece.) The construction time data for pipeline repair was collected through the construction scheduling system and recorded as follows: (Unit: h, accuracy ±0.1h).

[0021] S103: Data Preprocessing and Quality Control: Perform targeted preprocessing based on the characteristics of different data types to ensure data validity. For the in-pipe detection data ( , , ): For CCTV video-derived data: perform frame-by-frame denoising (median filtering algorithm, window size 3×3), geometric correction (correcting distortion based on camera intrinsic parameter calibration results), illumination compensation (gamma correction algorithm, gamma value set to 1.5), and keyframe extraction (inter-frame difference threshold 0.1, extracting keyframes in defective areas). For laser point cloud and sonar profile derived data: perform outlier detection (3σ criterion, exceeding [...]). , Data within the specified range is considered outlier. The mean, (Standard deviation), noise filtering (Kalman filter, process noise covariance) Observation noise covariance ); For sensor time series data (Q, V, H, P, , , , , , , ): Outlier detection: The 3σ criterion is used, as above; Missing value imputation: Linear interpolation is used (missing duration ≤ 1h), and a resampling mechanism is automatically triggered when the missing duration > 1h; Noise filtering: Kalman filtering is used, with the same parameters as above; For spatial data ( , , , ): Coordinate unification and transformation: Unify to the WGS84 coordinate system, and perform projection transformation using ArcGIS software. The transformation error is ≤0.1m. Topology verification: Based on the pipe location coordinates, the distance between adjacent pipe segments is calculated. If the distance is ≤0.5m, a connection is considered to exist, and correction is made. Adjacency matrix; For business data ( , , ): Data format standardization: All data is standardized to CSV format, with fields including data identifier, value, and... , , ; Logical consistency check: maintenance costs >0. Construction Time >0. Inventory Quantity If the data does not meet the requirements, it will be considered abnormal data and a re-sampling will be triggered.

[0022] Each data output includes a quality score. The calculation method is as follows:

[0023] When Q < 0.7, the re-sampling mechanism is automatically triggered, sending a re-sampling command to the operation and maintenance system, specifying the device, parameters, and sampling time.

[0024] S104: Data Standardization and Integration Alignment Establish unified data standardization specifications to eliminate data heterogeneity and ensure spatiotemporal consistency: Time dimension: Adopt time synchronization service based on NTP protocol for all data. Uniform calibration to UTC time, using a time synchronization server to ensure time alignment accuracy of data collected by different devices. <100ms; Spatial dimension: All coordinate data are unified to the WGS84 coordinate system. The original coordinate records are retained during the conversion process, and the conversion results are accompanied by an error description (≤0.1m). Numerical dimensions: For continuous variables: perform minimum-maximum normalization, as shown in the formula: ,in This is the minimum value of this parameter over the past 3 years. This is the maximum value of this parameter over the past 3 years. ; Categorical variables ( , (Material type in the data): One-heat encoding is performed, resulting in a binary vector for each category. The vector length equals the total number of categories, with only the corresponding category position set to 1 and the rest to 0. This ultimately forms a spatiotemporally aligned, standardized data cube. The dimensions are [time dimension (divided by hour) × space dimension (divided by pipe segment ID) × parameter dimension (18 types of core parameters collected)], which supports quick data retrieval by the three-dimensional index of pipe segment-time-parameter in subsequent stages.

[0025] S2: Pipeline Knowledge Graph Construction: Using the standardized data cube generated in S1 as the core input, a structured pipeline network knowledge graph is constructed and pipe segment features are extracted through a progressive process of ontology framework design, multimodal entity relationship extraction, and graph neural network feature learning; details are as follows: S201: Pipeline Knowledge Graph Ontology Design: Define five core entities, attributes, and four core relationships for a knowledge graph: Pipeline entity (i is the unique identifier ID of the pipe segment, based on) and generate): Static properties: (Pipe material) (Pipe section diameter) (Pipe section burial depth) (Location coordinates of pipe segment) (Topological relationship); Dynamic attributes: Q (flow rate), V (velocity), H (water level), P (pressure). (Crack length) (Corrosion area); Defective entity (k is the unique identifier ID for the defect, based on) (Generated with defect parameters): Attributes: (Defect location) (Crack length) (Corrosion area) (Detection time); Environmental factors entity (n is a unique identifier for environmental factors, generated based on monitoring location): property: (Surface subsidence) (Soil pH) (Soil temperature) (Water quality parameters) (Concentration of harmful gases) (Ground traffic load) (Construction vibration); Repair incident entity (m is a unique identifier for the maintenance event, generated based on maintenance records): property: (Repair costs) (Construction time) (Materials used) (Repair time); Sensor Entity (l is the sensor's unique identifier ID, based on) generate): property: (Device metadata) (Installation location) (Calibration time).

[0026] Define four types of relationships: Connection relationship: Pipe segment entity With pipe segment entity The topological connection between them, based on The adjacency matrix is ​​determined; Inclusion relationship: Pipe segment entity With defective entities The relationship of affiliation between them is based on and The spatial relationships are determined; Influencing Relationships: Environmental Factors as Entities With defective entities The interaction relationships between them are determined based on the causal correlation of time series data; Relationship: Pipe segment entity With the entity of the maintenance incident The correspondence between them is determined by matching the pipe segment ID with the identifier of the maintenance record.

[0027] S202: Multimodal Entity and Relation Extraction: From the standardized data cube of S1, entity instances and relationships are extracted using categorized and reproducible methods to ensure the objectivity and accuracy of the extraction results. Entity extraction: Structured data mapping: All numerical and categorical data collection parameters are directly mapped to the attributes of the corresponding entities according to the field identifier, generating entity instances; Spatial data clustering: For spatial data, the K-means clustering algorithm (K value adaptively determined based on the number of pipe segments in the pipeline network design drawings) is used to delineate the boundaries of pipe segment entities and extract defect entities. Attributes ensure that the physical space range matches the actual pipe segment; Relation extraction: Connection relation extraction: Based on Adjacency matrix, when matrix elements When (i, j) = 1, determine the pipe segment. and If a connection exists, establish a relation edge; Inclusion relationship extraction: Calculate pipe segment entities Spatial buffer (buffer distance is) / 2, determined based on pipe section diameter), if defect entity of If it falls within this buffer, then it is determined that... Include Establish relationship edges; Influence relationship extraction: Granger causality test (significance level α=0.05) was used to analyze the time-series correlation between environmental factor data and defect parameter data. If the test result rejects the null hypothesis of "no causal relationship", then the relationship is determined to be... and If an influence relationship exists, establish a relationship edge; Association extraction: By matching the unique ID of the pipe segment with the pipe segment identifier field of the maintenance record in the pipeline network operation and maintenance management system, if the identifiers match, then a determination is made. and If a relationship exists, establish a relationship edge.

[0028] The initial knowledge graph G=(V,E) is finally formed, where V is the entity set (including all pipe sections, defects, environmental factors, maintenance events, and sensor entity instances), and E is the relation set (including connection, inclusion, influence, and association relationship edges).

[0029] S203: Graph Neural Network Feature Learning: The initial knowledge graph G is converted into a graph data structure (nodes are entity instances, edges are relation edges), and the graph neural network is input to learn the structured features of the pipe segment entities, extracting high-dimensional feature vectors that incorporate neighborhood information. Graph Neural Network Structure: A three-layer Graph Attention Network (GAT) is used, with 128, 64, and 32 neurons in each layer, respectively. The sigmoid function is used for all layers. ; Feature update formula: ,in: The feature vector of node i in the l-th pipe segment (when the input layer l=0, (This is the attribute vector of the pipe segment entity, where the dimension equals the number of attributes). It is the set of neighboring nodes of pipe segment node i (determined based on connection relationship, i.e., pipe segment nodes directly connected to i). Let be the attention coefficient of node j in layer l to node i, used to quantify the importance of neighboring nodes, and calculated as follows: ,in The attention weight vector of the l-th layer (dimension 2) (Number of neurons in the current layer), || represents vector concatenation, and the negative slope of the LeakyReLU function is 0.2; The weight matrix of the l-th layer (dimension is the number of neurons in the current layer × the number of neurons in the next layer) is obtained through model training and optimization. Output: After calculation by the three-layer graph attention network, each pipe segment node... Output hidden feature vectors containing information about their own attributes, topology, and neighborhood relationships. (32 dimensions) This vector serves as the input feature for the subsequent defect level assessment model.

[0030] S3: Intelligent Assessment of Prognostic Defect Levels: The current defect level is determined using objective quantitative standards; a model integrating the physical laws of pipeline corrosion and data-driven approaches is then constructed to improve assessment accuracy; subsequently, the future spatiotemporal evolution trend of the defect is predicted; next, from three dimensions—structure, function, and environment—risk is calculated using explicit formulas and objective weights are allocated using the entropy weight method to obtain the total risk; finally, all results are integrated to generate a standardized report, providing a basis for maintenance decisions; the specific steps are as follows: S301: Defect Level Judgment Criteria: Based on the defect parameters collected by S1, a purely data-driven quantitative judgment criterion is established. when , The defect level is 1; when , The defect level is 2; when , The defect level is 3; when , The defect level is 4; If none of the above conditions are met, then there is no defect; S302: Physical Constraint and Data-Driven Fusion Modeling: Constructing a Physical Information Neural Network (PINN) model, embedding the physical laws of pipeline corrosion as hard constraints into a deep learning framework, and fusing the pipe segment feature vectors output from S2. The use of multi-source data collected by S1 improves the reliability of defect assessment and prediction. Model input: Pipe segment feature vector Static properties of pipe sections ( , , ), dynamic operating parameters (Q, V, H, P), environmental factor data ( , , , ); Physical constraint equation: Embedded pipeline corrosion rate equation, quantifying the physical laws of corrosion development, the equation is as follows: Where d is the corrosion depth (unit: mm), and the corrosion area is collected by S1. With pipe section diameter Derivation, the formula is: ; t is time (unit: year); k is the material-related corrosion constant (determined based on industrial corrosion test data of different materials, concrete k=0.02, cast iron k=0.05, PVC k=0.001); Model loss function: Balances the data fitting accuracy with the consistency of physical laws, the formula is: ,in, For data fitting loss, the cross-entropy loss function is used to calculate the difference between the model's predicted defect level and the S301 judgment level. As a physical constraint loss, mean square error loss is used to calculate the difference between the corrosion rate predicted by the model and the value calculated by the above physical equation; To balance the hyperparameters, they were determined using 5-fold cross-validation, with values ​​ranging from [0.1, 1.0]. The optimal value was selected based on minimizing the total loss of the validation set.

[0031] S303: Defect Spatiotemporal Evolution Prediction: Input data is fed into the trained PINN model, which outputs the evolution results of defects at multiple future time points, providing dynamic evidence for risk assessment. Defect level evolution prediction: Output future time( The probability distribution of defect levels (e.g., 12, 24, 36 months) is given by the formula: ,in, For pipe segment i in The level of defect at any given moment; Defect levels (1, 2, 3, 4); Input the data set for the model of pipe segment i; These are the parameters for the PINN model (including the weight matrix, bias terms, and physical constraint-related parameters). The spatiotemporal evolution function is implemented using two fully connected layers, with the input being... and The concatenated vector; W is the weight matrix of the fully connected layer (dimension is the input vector dimension × 4); b is the bias term (dimension is 4); the Softmax function converts the output into a probability distribution; Defect size evolution prediction: Based on the fusion of physical constraint equations and data-driven prediction results, outputting future... Defect critical dimension parameter variation curve at time (Crack length) (Corrosion area) The curve is smoothed using piecewise linear interpolation to ensure that the evolution trend conforms to physical laws.

[0032] S304: Multi-dimensional risk assessment: Quantifying pipe section risks from three dimensions: structural safety, functional assurance, and environmental impact. Structural risks The formula for quantifying the risk of reduced pipeline load-bearing capacity due to defects is as follows: ,in, The ultimate bearing capacity (unit: kN) of an intact pipeline is calculated based on the pipe section material, diameter, and burial depth: Concrete pipeline: Cast iron pipes: PVC pipes: ; The actual load-bearing capacity of the defective pipe (unit: kN), considering the defect size and structural load, is given by the following mathematical function: ,in: The total length of the pipe segment (unit: m) is based on data collected by S1. Coordinate calculation; The inner surface area of ​​the pipe section (unit: m²) 2 The specific calculation method is as follows: ; Maximum permissible ground traffic load (unit: kN / m) 2 (The value is taken as 100 according to industry standards). Maximum allowable construction vibration (unit: mm / s, value taken as 50 according to industry standard); A higher value indicates a higher structural risk.

[0033] Functional risks The risk of reduced drainage capacity due to quantification defects is calculated using the following formula: ,in, Pipeline design flow rate (unit: m³) 3 / s), based on pipe segment diameter and pipe slope (collected via S1). (Calculated from terrain data), derived using the Manning formula; The actual drainage capacity of the defective pipeline (unit: m) 3 / s), considering defect size and operating parameters, the formula is: ,in, The maximum allowable water level in the pipeline (unit: m, equal to the pipe section diameter). A higher value indicates a higher functional risk.

[0034] Environmental risks The specific mathematical function for the environmental pollution risks that may be caused by quantitative defects is: ,in, The maximum permissible concentration of harmful gases (unit: ppm, H2S value is 10, CH4 value is 5000, determined according to environmental protection standards). The maximum permissible concentration of COD (unit: mg / L, value taken as 50 according to environmental protection standards). The maximum allowable concentration of ammonia nitrogen (unit: mg / L, value 8 according to environmental protection standards). A higher value indicates a higher environmental risk.

[0035] Entropy weight method objective weight allocation: The entropy weight method is used to automatically calculate weights based on the risk data of all pipe sections. The steps are as follows: Constructing the evaluation matrix: Assuming the pipeline network has m pipe segments, construct the matrix... ,in, This represents the original value of the j-th risk dimension for the i-th pipe segment (j=1 corresponds to...). j=2 corresponds to j=3 corresponds to ); Matrix standardization: Forward standardization is used, and the formula is: , where min( ) represents the minimum value in the j-th column, and max( () represents the maximum value in the j-th column; Calculate information entropy: Information entropy of the j-th risk dimension for: ,in, To avoid ln(0) being meaningless; Calculate weights: weights for: satisfy , , , These are the weights for structural risk, functional risk, and environmental risk, respectively.

[0036] Total risk calculation ,in A higher value indicates a higher overall risk for the pipeline segment.

[0037] S305: Assessment Report Generation: Based on the defect level determination results of S301, the evolution prediction results of S303, and the risk quantification indicators of S304, a standardized assessment report is generated. Defect level determination result: clearly state the current defect level of the pipe section (level 1-4 or no defect), and attach the basis for the determination; Defect spatiotemporal evolution prediction results: Grade evolution: showing the probability distribution of defect grades in the next 12, 24, and 36 months; Size Evolution: Display , The change curve is marked with the prediction error range; Risk Quantification Indicators: Single-Dimensional Risk , , Specific values ​​and weights , , ; Total Risk: Specific values ​​and risk growth rate (formula is as follows) (Unit: 1 / year) Sensitivity analysis: through partial derivatives (x is an environmental factor parameter) Calculate the impact coefficient, and after ranking, identify the top three key environmental factors and their corresponding impact coefficient values.

[0038] S4: Multi-agent Co-evolutionary Decision Making: Using the defect level, evolution trend, and total risk data output by S3 as core inputs, and through a progressive process of "agent modeling - multi-objective optimization - reinforcement learning decision making - dynamic graph generation," optimal maintenance decisions are achieved without subjective intervention; the specific steps are as follows: S401: Modeling Decision-Making Environments and Intelligent Agents: Constructing Four Types of Intelligent Agents Based on Markov Decision Processes (MDPs): Pipeline segment status intelligent agent (Each pipe segment corresponds to one intelligent agent): Modeling logic: Taking the risk status of pipeline segments as the core, outputting maintenance demand priority signals to support the decision-making system in identifying high-risk pipeline segments; state space : Action space ={0, 1, 2}: 0 = no maintenance requirement, 1 = low priority maintenance requirement, 2 = high priority maintenance requirement; reward function =-0.5 -0.5 The higher the risk, the lower the reward, which drives the intelligent agent to output a high-priority maintenance signal. Its core function is to transform the risk status of pipeline segments into quantifiable maintenance needs, providing a basis for prioritization in decision-making.

[0039] Budget-constrained agents (Globally unique intelligent agent): Modeling logic: With budget allocation efficiency as the goal, under the constraint of the total budget, funds are tilted towards high-priority management segments; state space : Action space ={ }: To ensure that the maintenance funds allocated to pipe segment i meet the requirements and ; reward function , To predict the total risk after maintenance, maximizing the reward value corresponds to the highest reduction in unit financial risk. Core function: To optimize the allocation of funds under budget constraints and ensure that limited funds are used for high-value maintenance tasks.

[0040] Construction scheduling intelligent agent (Each construction team corresponds to one intelligent agent): Modeling logic: With construction efficiency as the goal, optimize the construction schedule based on the location of the pipe segment and the status of the construction team to shorten the total construction period; state space : The total number of construction teams; For the construction time of pipe segment i; Construction team status: 0 = idle, 1 = busy). Action space ={ }: For the planned time window for construction team j to repair pipe section i, meet the requirements. , The maximum available construction period for construction team j in a month; reward function Maximizing reward value corresponds to prioritizing the completion of high-risk pipeline repair and construction with optimal efficiency; Core function: Optimize the allocation and scheduling of construction resources, avoid construction conflicts, and improve maintenance execution efficiency. Material supply intelligent agent (Each type of repair material corresponds to a smart agent): Modeling logic: With the goal of matching material supply and demand, ensure timely supply of maintenance materials and avoid excessive inventory; state space : This refers to the inventory quantity of material category k. The quantity of type k materials required for repairing pipe segment i; For the supply cycle of material type k; Action space ={ }: The quantity of material of type k allocated to pipe segment i satisfies and ; Reward function: : Maximizing reward value corresponds to timely supply and optimal inventory utilization; its core function is to ensure the supply and demand of maintenance materials and avoid construction delays due to material shortages.

[0041] S402: Multi-objective cooperative optimization modeling: Based on the agent's state space and action space, define the multi-objective optimization function and constraints: Decision variables: ( =1 indicates that pipe segment i is being repaired. =0 indicates that repair will not be performed for the time being). Multi-objective optimization function: ; in, , To minimize the overall risk of the pipeline network; This represents the total maintenance cost (minimization). This indicates the longest (minimum) construction period. ; Constraints: Total budget constraint: ; Single construction team schedule constraints: ; Material inventory constraints: ; Risk threshold constraint: ; Binary constraints on decision variables: ; S403: Multi-agent reinforcement learning decision-making: A multi-agent reinforcement learning (MARL) framework with centralized training and distributed execution is adopted to achieve collaborative decision-making among agents, ensuring that the decision scheme satisfies the constraints and the optimization objective is optimal. Training phase: Environment initialization: Load the standardized data cube of S1, the knowledge graph of S2, and the risk assessment results and constraints of S3 to construct the simulation decision-making environment; Intelligent agent interaction: Each intelligent agent perceives real-time environmental information through the state space, based on... - Greedy strategy for choosing actions The value decreases linearly from 0.9 to 0.1, with a decay step size of 10,000 iterations. To explore the probability, 1- To utilize probability; Reward Calculation: The global coordinator collects action and state feedback from all agents and calculates the global reward. To balance the goals of each agent; Model optimization: Each agent is optimized using a Deep Q-Network (DQN) strategy, with the following loss function: ,in: For the current network parameters, Target network parameters (synchronized once every 1000 steps); =0.95 is the discount factor; Let be the state-action value function, representing the expected reward for performing action a in state s; Training termination condition: When the global reward... Training stops when the value stabilizes in the range [0.8, 1.0] for 1000 consecutive iterations and the constraint satisfaction rate is ≥99%.

[0042] Execution phase: Real-time perception: Each agent loads the trained policy model and perceives environmental changes based on the real-time state data collected by S1 (such as inventory updates and changes in the status of construction teams) and the risk data dynamically updated by S3. Autonomous decision-making: The pipeline segment status agent outputs the priority of maintenance needs, the budget constraint agent allocates funds based on the priority, the construction scheduling agent matches construction resources, and the material supply agent allocates materials; Conflict resolution: The global coordinator detects action conflicts (such as insufficient funding, material shortages, and overlapping construction times) and resolves conflicts by dynamically adjusting the action space (such as prioritizing high-priority pipe sections and adjusting construction time windows). Solution output: Output feasible repair decision solutions, including a list of pipe sections to be repaired, funding allocation amount, construction schedule, and material allocation list.

[0043] S404: Adaptive Dynamic Decision Graph Generation: Based on multi-agent collaborative decision-making results, combined with S3's risk assessment and evolutionary prediction data, a dynamic decision graph hierarchically arranged by execution priority is generated. Decision threshold definition: High risk threshold Medium risk threshold A total risk greater than or equal to 0.7 is classified as high risk, while a total risk within the range of [0.5, 0.7] is classified as medium risk. Benefit-cost ratio threshold Based on historical maintenance data statistics: (1 / yuan); Risk growth rate threshold : A risk increase of ≥0.2 within 12 months is considered rapid growth.

[0044] Hierarchical Decision Graph: Immediately Executed Layer: Lists those that meet the requirements and The pipeline section, along with a detailed implementation plan: Fund allocation: The maintenance funds will be allocated in full. Construction scheduling: Prioritize matching the construction team that is closest to the pipeline section and is currently available. The construction time will be strictly arranged according to the estimated repair time of the pipeline section to avoid delays in the construction period. Material allocation: Allocate sufficient quantities of all types of materials required for the repair of this pipe section to ensure that the supply of materials matches the needs of the repair and construction.

[0045] Monitoring preparation layer: List those that meet the following criteria: total risk ∈ [medium risk threshold, high risk threshold] or risk growth rate > threshold. Pipe section, with attached: Monitoring plan: Collect pipeline crack length, corrosion area and total risk data once a month using the data acquisition equipment specified by S1, and track defect development and risk changes in real time; Contingency Plan: Estimate the maintenance costs for this pipe section in advance (based on actual maintenance costs), the quantity of various materials required (determined according to repair technical requirements), and the estimated construction period (based on the repair time of similar pipe sections in the past).

[0046] Strategy Deferred Layer: Lists pipe sections that meet the criteria of total risk < medium risk threshold and can be repaired together with pipe sections within a 1km radius, with the following attached: Merging conditions: When the surrounding pipeline sections enter the immediate execution layer or the monitoring preparation layer, the merged maintenance process is initiated; Optimal timing: Based on the existing schedule of the construction team and the inventory status of maintenance materials, a comprehensive assessment suggests a window of opportunity for concentrated maintenance at the end of the quarter or half-year.

[0047] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent assessment and repair decision-making of drainage pipeline defect levels based on big data, characterized in that, include: S1: Intelligent acquisition and standardized processing of multi-source data: Construct a five-dimensional acquisition framework to collect pipeline attributes, defect status, operating parameters, environmental factors and business data. After preprocessing and quality control, a standardized data set with spatiotemporal alignment is formed through time, space and numerical standardization. S2: Pipeline Knowledge Graph Construction: Based on a standardized dataset, design a knowledge graph ontology framework and extract entity instances and relationships to form an initial knowledge graph. The graph neural network then learns and outputs high-dimensional feature vectors of pipe segments. S3: Intelligent assessment of prognostic defect levels: Based on standardized datasets and high-dimensional feature vectors of pipe segments, the defect level is quantitatively determined. By integrating physical laws and data-driven models, the spatiotemporal evolution trend of defects is predicted. The risk of pipe segments is quantified in multiple dimensions and the total risk is calculated to generate a standardized assessment report. S4: Multi-agent Co-evolutionary Decision Making: Based on defect level, evolution trend and total risk data, four types of functional agents are constructed. Through multi-objective optimization and multi-agent reinforcement learning, collaborative decision making is achieved, and repair decision schemes are output and dynamic decision graphs are generated in a priority-stratified manner.

2. The intelligent assessment and repair decision-making method for drainage pipeline defect levels based on big data as described in claim 1, characterized in that: The intelligent acquisition and standardization processing of multi-source data includes: A five-dimensional, full-scenario data acquisition framework is constructed, encompassing air, ground, internal systems, and business systems. This framework clarifies the core acquisition equipment and data traceability elements for each dimension. Data traceability elements include the timestamp of the data acquisition time, coordinates of the acquisition location, and metadata of the acquisition equipment. Data on pipe segment attributes, defect status, operating parameters, environmental factors, and business-related data are accurately collected in a unified format. Pipe segment attributes include material, diameter, and burial depth; defect status includes crack length and corrosion area; operating parameters include flow rate, velocity, water level, and pressure; environmental factors include surface subsidence, soil pH, traffic load, and construction vibration; and business data includes maintenance data. Cost, material inventory, and construction time; targeted preprocessing of collected multi-source heterogeneous data, including outlier removal, missing value supplementation, and noise reduction for time-series data, unified coordinate system and topological verification for spatial data, and standardized format and logical consistency verification for business data; establishment of a data quality control mechanism, calculating data quality scores based on the proportion of effective data volume and data accuracy compliance rate, triggering a re-collection mechanism when the quality score falls below a preset threshold; and fusion of preprocessed qualified data into a spatiotemporally aligned standardized data set through time synchronization calibration, spatial coordinate unification, and numerical normalization.

3. The intelligent assessment and repair decision-making method for drainage pipeline defect levels based on big data as described in claim 1, characterized in that: The formation of the initial knowledge graph includes: The knowledge graph ontology framework is designed, clearly defining five types of entities and their corresponding attributes: pipe segments, defects, environmental factors, maintenance events, and sensors. Four types of relationships are defined: connections between pipe segments, inclusion of pipe segments and defects, influence of environmental factors and defects, and association between pipe segments and maintenance events. Standardized data set fields are directly mapped to entity attributes through structured data mapping. Spatial data clustering is used to delineate entity spatial boundaries, generating a set of entity instances with unique identifiers. A data-driven approach is employed to extract relationships: connection relationships are determined based on topological data, inclusion relationships based on spatial affiliation, influence relationships based on temporal correlation, and association relationships based on identifier matching. The entity instance set and the relationship set are integrated to construct the initial knowledge graph.

4. The intelligent assessment and repair decision-making method for drainage pipeline defect levels based on big data as described in claim 1, characterized in that: The high-dimensional feature vector of the pipe segment includes: The initial knowledge graph is converted into a graph data structure, and a pipeline network graph model is constructed with pipeline segment entities as nodes and relationships as edges. A graph attention network is used to learn features of the pipeline network graph model. Attention weights are assigned by quantifying the importance of neighboring nodes, and the attributes of the pipeline segment entities themselves, the topological relationships between pipeline segments, and the relationship information of neighboring entities are integrated. After multi-layer network calculation, a high-dimensional feature vector of the pipeline segment with fixed dimensions is output.

5. The intelligent assessment and repair decision-making method for drainage pipeline defect levels based on big data as described in claim 1, characterized in that: The model that integrates physical laws and data-driven approaches includes: Model framework construction: The basic framework of the fusion model is built using a physical information neural network, and the network hierarchical structure and the number of neurons in each layer are determined. Input data determination: The model input data are defined as high-dimensional feature vectors of pipe segments, static attributes of pipe segments, dynamic operating parameters and environmental factor data, among which the static attributes of pipe segments, dynamic operating parameters and environmental factor data are all derived from standardized datasets; Physical constraint embedding: The physical constraint equations related to pipeline corrosion rate are embedded into the above neural network framework to establish the relationship between physical laws and model structure; Loss function design: Construct a composite loss function that includes a data fitting loss term and a physical constraint loss term, and set a balancing hyperparameter for the two losses; Model parameter optimization: Based on the preset training dataset, the model's weight matrix and bias parameters are iteratively optimized by minimizing the composite loss function to complete the model construction.

6. The intelligent assessment and repair decision-making method for drainage pipeline defect levels based on big data as described in claim 1, characterized in that: The multi-dimensional quantification of pipeline segment risk and calculation of total risk include: The risks of the pipeline segment are quantified from three dimensions: structural safety, functional assurance, and environmental impact. The quantification of each dimension is based on standardized data sets and high-dimensional feature vectors of the pipeline segment. Risk quantification formulas for each dimension are constructed. Based on pipeline segment attributes, defect status, operating parameters, and environmental factors, structural safety risk value, functional assurance risk value, and environmental impact risk value are calculated sequentially. The entropy weight method is used to construct a risk assessment matrix for all pipeline segments. After matrix standardization and information entropy calculation, the objective weights of each dimension are obtained, and the sum of the weights is 1. Then, the total risk value of the pipeline segment is calculated by weighted summation based on the risk values ​​of each dimension and the corresponding objective weights.

7. The intelligent assessment and repair decision-making method for drainage pipeline defect levels based on big data as described in claim 1, characterized in that: The construction involves four types of functional intelligent agents: pipeline segment status agent, budget constraint agent, construction scheduling agent, and material supply agent. The modeling logic for each agent is built according to Markov decision processes. The pipeline segment status agent focuses on the risk status of the pipeline segment. Its state space includes defect status and total risk data, and its action space includes three types of maintenance needs. The reward function is related to the current and future risks of the pipeline segment. The budget constraint agent aims at budget allocation efficiency. Its state space includes total budget, maintenance costs, and revenue and expenditure budget data. Its action space is the allocation of maintenance funds. The reward function is related to the reduction of financial risk and budget utilization rate. The construction scheduling agent aims at construction efficiency. Its state space includes the number of construction teams, pipeline segment construction time, location, and construction team status. Its action space is the construction plan time window. The reward function is related to the maintenance completion rate of high-priority pipeline segments and the project schedule utilization rate. The material supply agent aims at supply and demand matching. Its state space includes material inventory, supply cycle, and required material quantity. Its action space is the material allocation quantity. The reward function is related to the supply timeliness rate and inventory utilization rate.

8. The intelligent assessment and repair decision-making method for drainage pipeline defect levels based on big data as described in claim 1, characterized in that: The dynamic decision graph includes: This includes constructing a system based on defect levels, spatiotemporal evolution trends, total risk data, and multi-agent collaborative decision-making results. It clarifies objective judgment criteria for risk thresholds, cost-benefit ratios, and risk growth rates. The system is divided into three layers according to execution priority: an immediate execution layer, a monitoring and preparation layer, and a strategy postponement layer. The immediate execution layer includes the allocation of maintenance funds, construction schedules, and material allocation plans for high-risk pipe sections. The monitoring and preparation layer includes regular monitoring plans and contingency maintenance plans for medium-risk or rapidly growing risk pipe sections. The strategy postponement layer includes the conditions for combined maintenance of low-risk pipe sections and recommended maintenance time windows.

Images