Water supply pipeline corrosion risk assessment method based on multi-source data and time series prediction

Abstract

The application discloses a kind of based on multi-source data and timing prediction's water supply pipeline corrosion risk evaluation method, comprising: collecting the multi-source data set of expert annotation corrosion grade;Through correlation analysis and threshold screening, extract core corrosion risk factor matrix;Introduce brand-new pipeline standard vector to correct expert label, and build corrosion risk evaluation model-adopt linear fitting and nonlinear fitting two ways, and the model of higher fitting goodness is selected;To be evaluated pipe section, according to time window collection multi-source data sequence and extract risk factor vector sequence;Risk factor vector is predicted in future time step using LSTM model;Output dynamic risk score;Based on rating generation grading early warning and maintenance plan.The application fuses multi-source data and comprehensively quantifies corrosion driving mechanism, realizes risk forward-looking evaluation by LSTM model, and outputs dynamic score in combination with high-precision interpretable model, which can achieve 0.94 fitting goodness, and can support accurate grading early warning and maintenance decision.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of water supply pipeline internal corrosion evaluation, and relates to a water supply pipeline corrosion risk evaluation method based on multi-source data and time series prediction. BACKGROUND

[0002] As a core component of urban lifeline engineering, water supply pipelines bear the key task of ensuring the normal operation of residents' life and urban economic activities. However, in the long-term service process, the inner wall of the pipeline is inevitably affected by multiple factors such as water quality chemical action, impurity deposition, fluid erosion and microbial corrosion, and there are different degrees of internal corrosion problems. Corrosion not only leads to the decrease of pipeline structural strength, the reduction of effective section and the decrease of water transmission efficiency, but also may cause leakage, secondary pollution of water quality and even pipeline burst, which threatens the safety of water supply and public health. Therefore, how to scientifically, dynamically and quantitatively evaluate the corrosion state and its evolution trend of the pipeline is a key technical problem to realize precise maintenance and ensure the safe and economic operation of the water supply network.

[0003] At present, the pipeline corrosion detection and evaluation technology mainly has the following limitations:

[0004] 1. Single data dimension, one-sided evaluation: existing methods (such as CCTV endoscope image, single water quality parameter or pipe age experience model) often rely on single or small amount of data sources. For example, patent CN114720495A mainly depends on corrosion depth or calcium carbonate thickness image features, CN112669269A focuses on the classification of structural defects in image recognition, and CN103578045A is based on static rules and expert experience scoring. These methods are difficult to fully capture the complex coupling relationship between multi-dimensional factors such as water quality chemical action, pipe scale physical and chemical properties, operating conditions and corrosion process, leading to one-sided evaluation results and difficulty in reflecting the real comprehensive corrosion risk.

[0005] 2. Static perspective, lack of foresight: existing technologies focus on the "snapshot" evaluation of current or historical state, and fail to effectively utilize the time series characteristics of corrosion-related parameters. Corrosion is a dynamic development process, and its risk is continuously affected by factors such as water quality fluctuations, flow rate and pressure changes. The lack of prediction ability for future trends makes the maintenance decision lag, and cannot realize the foresight prevention and control of risk.

[0006] 3. Insufficient model interpretability and adaptability: While some intelligent approaches (such as the convolutional neural network in CN112669269A) improve recognition efficiency, these models often exhibit "black box" characteristics, lacking clear physical meaning and interpretable quantitative mechanisms, making them difficult to guide actual maintenance actions. Furthermore, static scoring models (such as CN103578045A) struggle to adapt to the complexities of different pipe materials, environments, and the dynamic evolution of corrosion, limiting their flexibility and scalability.

[0007] 4. Weak risk quantification and early warning mechanisms: Existing technologies generally have shortcomings in converting multi-dimensional observation data into unified and comparable corrosion risk quantitative scores, and in establishing graded early warning and differentiated maintenance strategies based on this.

[0008] Therefore, the existing water supply pipeline corrosion risk assessment technology still has problems such as insufficient data utilization, limited prediction ability, lack of model interpretability, and weak risk quantification and early warning mechanisms. An innovative method that integrates multi-source data and takes into account dynamic prediction and physical interpretability is urgently needed to achieve scientific quantitative assessment and forward-looking management of water supply pipeline corrosion risks. Summary of the Invention

[0009] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a water supply pipeline corrosion risk assessment method based on multi-source data and time series prediction.

[0010] The purpose of the present invention can be achieved by the following technical solutions:

[0011] The present invention provides a water supply pipeline corrosion risk assessment method based on multi-source data and time series prediction, comprising the following steps:

[0012] Collecting a multi-source dataset of water supply pipelines with different corrosion levels annotated by experts, wherein the multi-source dataset includes multi-source data of water supply pipelines with different corrosion levels;

[0013] Correlation analysis and screening of multi-source data sets were performed to extract the corrosion risk factor matrix, and a corrosion risk assessment model was constructed based on the corrosion risk factor matrix and its corresponding corrosion levels.

[0014] For the pipe section to be evaluated, a multi-source data time series at multiple continuous time steps is obtained according to a preset time window, and a corrosion risk factor vector time series of the pipe section to be evaluated is obtained based on the multi-source data time series;

[0015] Input the corrosion risk factor vector time series into the pre-trained LSTM model to predict the corrosion risk factor vector at each future time step;

[0016] The corrosion risk factor vectors at the current moment and each future time step are input into the corrosion risk assessment model to obtain the corresponding risk scores;

[0017] Based on the risk scores of each time step, early warning levels for different time steps are generated and corresponding maintenance plans are formulated.

[0018] Specifically, the multi-source data includes inner wall image features, water quality monitoring features, pipe scale chemical and physical features, and operating condition features.

[0019] Specifically, the inner wall image features are image features obtained by processing the inner wall image of the water supply pipe, including the ratio of the inner wall rust area, the maximum thickness of the corrosion nodule, the wall thickness loss, the inner wall erosion depth, the corrosion morphology type, the image grayscale value mean and the color difference change;

[0020] The water quality monitoring characteristics are characteristics obtained by testing water samples obtained by sampling the water supply pipeline, including the pH value, conductivity, turbidity, total chlorine concentration, and iron and manganese ion concentrations of the water samples;

[0021] The chemical and physical characteristics of pipe scale are indicators obtained through laboratory analysis of pipe scale samples collected on-site from water supply pipelines, including the mass fraction of iron and manganese in the pipe scale, the proportion of inorganic components, and surface roughness;

[0022] The operating condition characteristics are characteristics obtained through the water supply system operation data acquisition system, including pipeline water flow velocity, water pressure, pipe age and historical maintenance times.

[0023] Specifically, extracting the corrosion risk factor matrix includes:

[0024] Constructing original feature matrix from multi-source datasets ,in, Indicates the i The multi-source data of water supply pipelines with different corrosion levels j Features, represents the total number of corrosion grades, i.e., the total number of water supply pipe samples obtained, and n represents the total number of features in the multi-source data;

[0025] Constructing a corrosion grade label vector ,in, Indicates the i The label value of each corrosion level is obtained through expert scoring. The more serious the corrosion of the water supply pipe, the larger the label value of the corrosion level;

[0026] For the original feature matrix Each column feature in Calculate its correlation with the corrosion grade label vector Pearson correlation coefficient:

[0027]

[0028] in, Indicates the j Correlation coefficient between item characteristics and corrosion grade, Indicates the j The average value of the feature, Indicates the The average value of the label values ​​of the corrosion levels;

[0029] Set the correlation coefficient threshold , according to the correlation coefficient of each feature and the correlation coefficient threshold , filter features and extract core corrosion risk factors, which are described as:

[0030]

[0031] in, Indicates the j The absolute value of the correlation coefficient between the item feature and the corrosion grade, It is a set of characteristic indexes of corrosion risk factors screened from multi-source data;

[0032] Indexing a collection by features From the original matrix X Select the corresponding feature column to obtain the filtered corrosion risk factor matrix :

[0033]

[0034] in, Represents the number of extracted corrosion risk factors.

[0035] Specifically, the corrosion risk assessment model is constructed based on the corrosion risk factor matrix and its corresponding corrosion level, which specifically includes:

[0036] Obtain multi-source data for a new water supply pipeline, construct a corrosion risk factor vector for the new water supply pipeline based on the multi-source data, and use this vector as a standard vector. Based on this standard vector, correct the corrosion grade label values ​​for water supply pipelines with different corrosion grades to obtain corrected label values.

[0037] According to the revised label value and corrosion risk factor matrix , build a corrosion risk assessment model.

[0038] Furthermore, the correction of the corrosion grade label values ​​of water supply pipes of different corrosion grades according to the standard vector specifically includes:

[0039] Obtain multi-source data of a new water supply pipeline and extract its corresponding corrosion risk factor vector, which is recorded as the standard vector ,in is a standard vector, representing the value of each corrosion risk factor under the ideal state without corrosion, k is the number of extracted corrosion risk factors;

[0040] for each water supply pipeline in the corrosion risk factor matrix corresponding to the corrosion risk factor vector , the Euclidean distance between the corrosion risk factor vector and the standard vector is calculated as a deviation value;

[0041] All deviation values are normalized to obtain a normalized deviation vector ;

[0042] According to the normalized deviation vector, the original corrosion grade label is corrected, wherein, represents the label value of the i-th corrosion grade, and the corrected corrosion grade label value is obtained: i

[0043]

[0044] wherein, represents the corrected label value of the i-th corrosion grade, i.e., the corrosion risk evaluation score, i represents the total number of corrosion grades, is a weighting coefficient, controlling the fusion ratio between the original expert score and the objective deviation value. Further, the corrosion risk evaluation model is constructed according to the corrected label value and the corrosion risk factor matrix F, and specifically includes:

[0045] The corrosion risk factor matrix F and its corresponding corrected label value Y are taken as training data, and linear fitting transformation and nonlinear fitting transformation are used to establish the corrosion risk evaluation model:

[0046]

[0047] When linear fitting transformation is adopted, a right multiplication transformation matrix A is constructed , and the relationship between the corrosion risk factor and the corrected label value is fitted through matrix operation:

[0048]

[0049] wherein, is the corrosion risk factor matrix, is the corrected label value matrix, is the right multiplication transformation matrix obtained by least square method;

[0050] ​​​​When nonlinear fitting transformation is adopted, the XGB regression algorithm based on gradient boosting tree is preferred as the nonlinear fitting function. , fitting the corrosion risk factor matrix F and the corrected label values ​​through decision tree ensemble learning The nonlinear relationship between:

[0051]

[0052] in, is the nonlinear fitting function for the corrosion risk factor matrix The i-th corrosion risk factor vector The predicted label value of For the A regression tree function, is the regression tree space, G is the total number of regression trees;

[0053] The nonlinear fitting function Fit each corrosion risk factor vector by minimizing the following objective function With the corrected label value The nonlinear relationship between:

[0054]

[0055] in, For the t The overall objective loss function of the round iteration, is the total number of corrosion grades, i.e. the total number of water supply pipe samples obtained, is the i-th corrosion risk factor vector at the t-1th iteration The predicted label value of For the t New regression tree Corrosion risk factor vector The predicted output of Corrected label value for the sample With the current forecast value The loss function between is calculated using cross loss. Add a regression tree Regularization term of complexity;

[0056] Calculate the linear fitting function separately With nonlinear fitting function The goodness of fit R² is:

[0057]

[0058] in, is the mean of the corrected label values, is the predicted label value of the fitting function;

[0059] The fitting function with a higher goodness of fit R² was selected as the final corrosion risk assessment model.

[0060] Specifically, for the pipe section to be evaluated, a multi-source data time series at multiple consecutive time steps is obtained according to a preset time window, and a corrosion risk factor vector time series of the pipe section to be evaluated is obtained based on the multi-source data time series, specifically including:

[0061] For the target water supply pipeline section, within the set sliding time window T Continuously collect each time step Multi-source raw data, building multi-source data time series ,in, Indicates the t Multi-source data vectors of time steps, Indicates the j The multi-source feature t The value of the time step;

[0062] Combined with the established feature index set , extract the corresponding corrosion risk factor sub-vector from the multi-source data vector at each time step , obtain the corrosion risk factor vector time series ,in is the current time step.

[0063] Specifically, the training process of the LSTM model is as follows:

[0064] The corrosion risk factor vector time series is used as the input training sample to predict the corrosion risk factor vector of the future target time step. According to the predicted corrosion risk factor vector and the corresponding true corrosion risk factor vector, the loss value is calculated through the loss function improved based on the correlation coefficient. The parameters of the LSTM model are updated according to the loss value to obtain the trained LSTM model.

[0065] Furthermore, the improved loss function is:

[0066]

[0067] in, 、 The predicted and true t The corrosion risk factor vector of the time step j The value of the feature, For the j Correlation coefficient between item characteristics and corrosion grade, is the number of corrosion risk factors extracted.

[0068] Compared with the prior art, the present application has the following advantages:

[0069] (1) The present application realizes comprehensive quantitative characterization of the corrosion state of the water supply pipeline by introducing multi-dimensional data sources such as inner wall image features, water quality monitoring features, pipe scale chemical and physical features, and operating condition features, overcoming the problem of the existing technology that relies on a single data source to cause the evaluation result to be one-sided. This method not only extracts indexes such as inner wall corrosion area proportion, corrosion nodule maximum thickness, wall thickness loss, corrosion depth, and corrosion morphology type from image features to intuitively reflect the corrosion morphology, but also combines water quality monitoring parameters such as pH, conductivity, turbidity, total chlorine concentration, and iron and manganese ion concentration, further introduces physical and chemical features such as pipe scale chemical composition and surface roughness, and operating condition features such as flow rate, water pressure, pipe age, and maintenance records, realizes multi-dimensional information fusion from macro to micro and from structure to environment, effectively reveals the complex coupling relationship between the corrosion process and water quality, operating conditions, and sediment characteristics, and solves the technical problem that the existing method is difficult to comprehensively reflect the actual corrosion risk. Through the joint application of the above multi-source features, the present application can form a corrosion risk assessment model with stronger interpretability and wider applicability, improve the accuracy and reliability of risk evaluation, and provide a scientific basis for subsequent dynamic prediction and differentiated maintenance, thereby realizing fine management and forward-looking prevention and control of the corrosion risk of the water supply pipeline.

[0070] (2) In the multi-source data environment, different features have different contributions to the corrosion risk of the water supply pipeline. Some features may have a low correlation with the corrosion grade, or even contain noise or redundant information. If these features are directly input into the model without screening, the model may be disturbed by irrelevant or weakly related information during the learning process, reducing the prediction accuracy and generalization ability. The present application constructs a multi-source data feature matrix and introduces expert scoring to determine the corrosion grade label, uses the Pearson correlation coefficient to quantitatively evaluate the linear correlation between each feature and the corrosion grade, and combines a preset threshold to screen out core risk factors that are strongly related to the corrosion state and eliminate invalid and noise features, thereby effectively improving the quality of the input data and the representativeness of the feature set. The feature screening realized in this way not only reduces the dimension and complexity of the model, reduces the risk of overfitting, but also enhances the discrimination ability and interpretability of the model for the corrosion risk, ultimately realizing high precision and high reliability of the corrosion risk assessment of the water supply pipeline, and providing more scientific basic data support for dynamic prediction and differentiated maintenance.

[0071] (3) The present invention introduces a standard vector to correct the corrosion grade label value, thereby solving the problem of strong subjectivity of expert scoring and insufficient differentiation of risks of different pipe sections in traditional corrosion grade assessment. Although expert scoring is based on experience, it inevitably has human bias and relativity. The objective differences in the multi-source characteristics of actual pipeline corrosion risk can often more accurately reflect its true corrosion status. By obtaining multi-source data of a new water supply pipeline to construct a standard vector under ideal conditions, and calculating the Euclidean distance between the corrosion risk factor of each pipe section and the standard vector, the degree of deviation of each pipe section from the "corrosion-free" benchmark can be quantified, and an objective representation of the corrosion risk difference can be obtained. The deviation value is fused with the expert score according to the weighted coefficient to correct the corrosion grade label so that it takes into account both expert experience judgment and objective data differences, effectively reducing the error caused by the subjectivity of the scoring and enhancing the objectivity and robustness of the corrosion risk label. In this way, the present invention achieves a more accurate and interpretable quantitative representation of corrosion risk, providing a more reliable input data basis for subsequent corrosion prediction and maintenance decision-making.

[0072] (4) When constructing the corrosion risk assessment model, the present invention introduces two modeling methods, linear fitting and nonlinear fitting, in order to solve the problems of singleness and uncertainty in model selection of traditional methods. For multi-source data, the relationship between features and corrosion levels may be linear or there may be complex nonlinear coupling. Relying solely on a single linear or nonlinear model will lead to insufficient fitting or oversimplification, making it difficult to fully reflect the corrosion mechanism. By constructing a linear fitting model based on the least squares method and a nonlinear fitting model based on the gradient boosting tree (XGB regression), and using the goodness of fit as the evaluation index, the optimal model is automatically selected, achieving an accurate characterization of the relationship between the corrosion risk factor and the corrected label value. This method not only improves the adaptability and robustness of the model under different data distributions, but also takes into account the model interpretability and prediction accuracy, thereby achieving a more scientific and reliable dynamic quantitative evaluation of the corrosion risk of water supply pipelines, and providing solid data support for subsequent maintenance decisions and risk warnings.

[0073] (5) The present invention predicts the corrosion risk factor vectors at each future time step by inputting the time series of the corrosion risk factor vector into the pre-trained LSTM model, and inputs the predicted value and the risk factor at the current moment into the corrosion risk assessment model to obtain the risk scores at different time steps. Based on the score results, the warning level is generated and the maintenance plan is formulated. The aim is to solve the problem that the existing corrosion risk assessment can only reflect the current status and lacks the prediction of future risk evolution trend and forward-looking maintenance guidance. The method uses the LSTM model to model the time series characteristics of multi-source data, which can capture the dynamic law of corrosion risk changing over time. Combined with the evaluation model, it can realize the quantitative prediction of future corrosion risk, thereby identifying potential high-risk pipe sections in advance and issuing graded warnings, providing a basis for formulating differentiated and dynamic maintenance plans for the water supply network, realizing the transition from passive repair to active prevention and control, and significantly improving the safety and economy of pipe network operation.

[0074] (6) The present invention introduces an improved loss function based on correlation coefficient weighting during the LSTM model training process, aiming to solve the problem that traditional time series prediction models fail to distinguish the importance of different features on corrosion risk. Conventional LSTM training only minimizes the error between the predicted value and the true value, and cannot reflect the difference in the importance of each corrosion risk factor to the overall corrosion level, resulting in insufficient attention to key features during the model learning process, reducing the reliability and interpretability of the prediction. By introducing the Pearson correlation coefficient between each feature and the corrosion level as a weighting factor, the model gives higher weight to features that are highly correlated with the corrosion level during training, thereby improving the model's sensitivity and learning ability to key corrosion drivers. This improvement not only enhances the accuracy and stability of corrosion risk prediction, but also better captures the core driving factors of corrosion evolution, providing more reliable technical support for forward-looking risk assessment and precise maintenance of water supply networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] Figure 1 This is a flow chart of a water supply pipeline corrosion risk assessment method according to an embodiment of the present invention;

[0076] Figure 2 Schematic diagram of linear model and nonlinear model fitting according to an embodiment of the present invention. DETAILED DESCRIPTION

[0077] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts should fall within the scope of protection of the present invention.

[0078] Example 1:

[0079] The embodiment discloses a water supply pipeline corrosion risk evaluation method based on multi-source data and time series prediction, as shown in the figure, and the specific steps are as follows: Figure 1

[0080] (1) Collecting multi-source data sets of water supply pipelines of different corrosion grades marked by experts, wherein the multi-source data sets include multi-source data of water supply pipelines of different corrosion grades;

[0081] The multi-source data includes inner wall image features, water quality monitoring features, pipe scale chemical and physical features, and operating condition features.

[0082] The inner wall image features are image features obtained by image processing of the obtained inner wall images of the water supply pipeline, including inner wall corrosion area ratio, maximum thickness of corrosion nodule, wall thickness loss, inner wall erosion depth, corrosion morphology type, image gray value mean and color difference change; wherein, for the inner wall image features, the inner wall images of the pipeline are obtained by using high-definition CCTV or robot inspection equipment, and multi-dimensional indexes such as inner wall corrosion area ratio, maximum thickness of corrosion nodule, wall thickness loss, and inner wall erosion depth are extracted by image preprocessing (denoising, enhancement, segmentation) and image analysis algorithm, and the image gray value mean and color difference change are calculated by combining color histogram and gray matrix analysis, and the corrosion morphology type is determined by morphological feature recognition.

[0083] The water quality monitoring features are features obtained by detecting the water samples obtained by sampling the water supply pipeline, including pH value, conductivity, turbidity, total chlorine concentration, and concentrations of iron ions and manganese ions; wherein, for the water quality monitoring features, the water samples are collected at the specified positions of the pipeline, and the pH value, conductivity, turbidity, and total chlorine concentration are detected by using portable devices such as pH meter, conductivity meter, and turbidity meter, and the concentrations of iron ions and manganese ions are determined by spectrophotometry or inductively coupled plasma emission spectroscopy (ICP-OES).

[0084] The pipe scale chemical and physical features are indexes obtained by laboratory analysis of pipe scale samples collected on site from the water supply pipeline, including mass fraction of iron and manganese in the pipe scale, inorganic component proportion, and surface roughness; wherein, for the pipe scale chemical and physical features, the samples of the pipe scale attached to the inner wall of the pipeline are collected during water stop maintenance or excavation, and the mass fraction of iron and manganese and the inorganic component proportion are analyzed by using X-ray fluorescence (XRF) and X-ray diffraction (XRD) methods in the laboratory, and the surface roughness is determined by using a three-dimensional profilometer or a scanning electron microscope (SEM).

[0085] ​The operating condition characteristics are the characteristics obtained through the water supply system operation data acquisition system, including pipeline water flow velocity, water pressure, pipe age and historical maintenance times; among them, for the operating condition characteristics, based on the water supply network SCADA system and historical archive data, the pipeline water flow velocity, water pressure and other real-time operating data are extracted, and the pipe age and historical maintenance times are obtained in combination with the pipeline completion archives and maintenance records.

[0086] (2) Correlation analysis and screening of multi-source data sets were performed to extract the corrosion risk factor matrix. A corrosion risk assessment model was constructed based on the corrosion risk factor matrix and its corresponding corrosion level, including:

[0087] Constructing original feature matrix from multi-source datasets ,in, Indicates the i The multi-source data of water supply pipelines with different corrosion levels j Features, including the features corresponding to the various multi-source data mentioned above, represents the total number of corrosion grades, i.e., the total number of water supply pipe samples obtained, and n represents the total number of features in the multi-source data;

[0088] Constructing a corrosion grade label vector ,in, Indicates the i The label value of each corrosion level is obtained through expert scoring. The more serious the corrosion of the water supply pipe, the larger the label value of the corrosion level;

[0089] For the original feature matrix Each column feature in Calculate its correlation with the corrosion grade label vector Pearson correlation coefficient:

[0090]

[0091] in, Indicates the j Correlation coefficient between item characteristics and corrosion grade, Indicates the j The average value of the feature, Indicates the The average value of the label values ​​of the corrosion levels;

[0092] Set the correlation coefficient threshold , according to the correlation coefficient of each feature and the correlation coefficient threshold , filter features and extract core corrosion risk factors, which are described as:

[0093]

[0094] in, Indicates the j The absolute value of the correlation coefficient between the item feature and the corrosion grade, It is a set of characteristic indexes of corrosion risk factors screened from multi-source data;

[0095] Indexing a collection by features From the original matrix X Select the corresponding feature column to obtain the filtered corrosion risk factor matrix :

[0096]

[0097] in, Represents the number of extracted corrosion risk factors.

[0098] Obtain multi-source data for a new water supply pipeline and construct a corrosion risk factor vector for the new water supply pipeline based on the multi-source data. This vector is used as a standard vector. The corrosion grade label values ​​of water supply pipelines with different corrosion grades are then corrected based on the standard vector to obtain the corrected label values. Specifically, the following steps are performed:

[0099] Obtain multi-source data of a new water supply pipeline and extract its corresponding corrosion risk factor vector, which is recorded as the standard vector ,in is a standard vector, which represents the values ​​of each corrosion risk factor under the ideal state without corrosion. k is the number of extracted corrosion risk factors;

[0100] Corrosion risk factor matrix The corrosion risk factor vector corresponding to each water supply pipeline , calculate the Euclidean distance between it and the standard vector , as the deviation value;

[0101] Normalize all deviation values ​​to obtain the normalized deviation vector ;

[0102] The original corrosion grade label is based on the normalized deviation vector Make corrections, where Indicates the i The label value of each corrosion level is calculated to obtain the corrected label value of the corrosion level:

[0103]

[0104] in, Indicates the corrected i The label value of each corrosion level, that is, the corrosion risk assessment score, Indicates the total number of corrosion levels, is the weighting coefficient that controls the fusion ratio between the original expert score and the objective deviation value.

[0105] According to the revised label value and corrosion risk factor matrix , build a corrosion risk assessment model, specifically including:

[0106] Corrosion risk factor matrix The corresponding corrected label value As training data, the corrosion risk assessment model is established using linear fitting transformation and nonlinear fitting transformation:

[0107] When taking linear fitting transformation, construct the right multiplication transformation matrix , the relationship between the corrosion risk factor and the corrected label value is fitted through matrix operations:

[0108]

[0109] in, is the corrosion risk factor matrix, is the corrected label value matrix, is the right-multiplied transformation matrix obtained by the least squares method;

[0110] When nonlinear fitting transformation is adopted, the XGB regression algorithm based on gradient boosting tree is preferred as the nonlinear fitting function. , fitting the corrosion risk factor matrix F and the corrected label values ​​through decision tree ensemble learning The nonlinear relationship between:

[0111]

[0112] in, is the nonlinear fitting function for the corrosion risk factor matrix The i-th corrosion risk factor vector The predicted label value of For the A regression tree function, is the regression tree space, G is the total number of regression trees;

[0113] Nonlinear fitting function Fit each corrosion risk factor vector by minimizing the following objective function With the corrected label value The nonlinear relationship between:

[0114]

[0115] in, For the tThe overall objective loss function of the round iteration, is the total number of corrosion grades, i.e. the total number of water supply pipe samples obtained, is the i-th corrosion risk factor vector at the t-1th iteration The predicted label value of For the t New regression tree Corrosion risk factor vector The predicted output of Corrected label value for the sample With the current forecast value The loss function between is calculated using cross loss. Add a regression tree Regularization term of complexity;

[0116] Calculate the linear fitting function separately With nonlinear fitting function The goodness of fit R² is:

[0117]

[0118] in, is the mean of the corrected label values, is the predicted label value of the fitting function;

[0119] The fitting function with a higher goodness of fit R² was selected as the final corrosion risk assessment model.

[0120] (3) For the pipe section to be evaluated, obtain its multi-source data time series at multiple continuous time steps according to the preset time window, and obtain the corrosion risk factor vector time series of the pipe section to be evaluated based on the multi-source data time series, specifically including:

[0121] For each water supply pipeline segment to be assessed, multi-source data is continuously collected over a pre-set time window (e.g., daily, weekly, or monthly) to generate time series data covering multiple consecutive time steps. This multi-source data includes internal wall image features, water quality monitoring characteristics, chemical and physical characteristics of pipe scale, and operating condition characteristics, ensuring that the time series accurately reflects the dynamic evolution of the pipeline corrosion state.

[0122] The corrosion risk factor vector of the corresponding time step is extracted from the multi-source data collected at each time step using the same method as step (2).

[0123] The corrosion risk factor vectors for all consecutive time steps are arranged in chronological order to construct a time series of the corrosion risk factor vectors for the pipeline segment to be assessed. This time series accurately reflects the dynamic evolution of pipeline corrosion-related characteristics over time, facilitating subsequent corrosion risk trend analysis and future state prediction using a time series prediction model.

[0124] (4) Input the corrosion risk factor vector time series into the pre-trained LSTM model to predict the corrosion risk factor vector at each time step in the future;

[0125] Among them, the LSTM model training process is:

[0126] The corrosion risk factor vector time series is used as the input training sample to predict the corrosion risk factor vector of the future target time step. According to the predicted corrosion risk factor vector and the corresponding true corrosion risk factor vector, the loss value is calculated through the loss function improved based on the correlation coefficient. The parameters of the LSTM model are updated according to the loss value to obtain the trained LSTM model.

[0127] The improved loss function is:

[0128]

[0129] in, 、 The predicted and true t The corrosion risk factor vector of the time step j The value of the feature, For the j Correlation coefficient between item characteristics and corrosion grade, is the number of corrosion risk factors extracted.

[0130] (5) Input the corrosion risk factor vectors at the current moment and each future time step into the corrosion risk assessment model to obtain the corresponding risk scores;

[0131] (6) Based on the risk scores of each time step, generate warning levels for different time steps and formulate corresponding maintenance plans, including:

[0132] 6.1 Risk score time series analysis:

[0133] Perform trend analysis on the risk score series of the assessed pipeline section in multiple future time steps, and use a sliding window to calculate the short-term risk change rate (such as the difference or weighted moving average of the risk score).

[0134] The risk level and risk development speed are captured through the dual indicators of the absolute value and change rate of the risk score.

[0135] 6.2 Multi-dimensional risk trend clustering:

[0136] The risk trend time series of the pipe section is divided into different mode categories, such as "stable low risk", "slowly rising risk", "rapidly rising risk", "fluctuating risk", etc. using unsupervised learning methods (such as K-means based on dynamic time warping (DTW) or spectral clustering).

[0137] This clustering can identify different trajectories of risk evolution, thereby supporting more accurate early warning classification.

[0138] 6.3, dynamic assignment of early warning levels:

[0139] According to the clustering results and the threshold setting of the risk score at each time step, the corresponding early warning level of each time step is assigned.

[0140] For example, the "rapidly rising risk" category can upgrade the early warning level even if the current risk score is not high, to warn potential deterioration in advance.

[0141] On the contrary, the "stable low risk" category maintains a lower early warning level to avoid unnecessary frequent maintenance.

[0142] 6.4, dynamic maintenance plan development:

[0143] Combined with the early warning level and risk trend category, the maintenance priority ranking and specific maintenance measure suggestions are automatically generated.

[0144] For example, pipe sections with high early warning levels and rapidly rising risks are given priority for emergency inspection and repair, while pipe sections with low early warning levels and stable risks are given routine inspection.

[0145] The maintenance plan can be dynamically adjusted according to the pipe network resources and the availability of maintenance personnel, to achieve optimal allocation of resources.

[0146] 6.5, closed-loop feedback optimization:

[0147] The actual maintenance effect and subsequent risk score are included in the feedback mechanism to dynamically adjust the clustering model and threshold setting, improving the accuracy and economy of early warning and maintenance.

[0148] Embodiment 2:

[0149] This embodiment takes a water supply pipe network in Southeast China as the implementation background, and uses expert annotated corrosion level data collected from pipe internal images during pipe maintenance as a data source based on the method established in the specific implementation manner, extracts a multi-source data set to construct an original feature matrix X:

[0150]

[0151] wherein, represents the multi-source data of the i th corrosion level water supply pipe. jFeatures, represents the total number of corrosion levels, that is, the total number of water supply pipe samples obtained, and n represents the total number of features in the multi-source data. In this embodiment, m = 20, n = 4.

[0152] Constructing a corrosion grade label vector y :

[0153]

[0154] in, Indicates the i The label value of the corrosion level, in this embodiment, m = 20.

[0155] For the original feature matrix Each column feature in Calculate its correlation with the corrosion grade label vector Pearson correlation coefficient:

[0156]

[0157] in, Indicates the j Correlation coefficient between item characteristics and corrosion grade, Indicates the j The average value of the feature, Indicates the The average of the label values ​​for the three corrosion levels.

[0158] Set the correlation coefficient threshold = 0.8, based on the correlation coefficient of each feature and the correlation coefficient threshold , filter features and extract core corrosion risk factors, which are described as:

[0159]

[0160] in, Indicates the j The absolute value of the correlation coefficient between the item feature and the corrosion grade, It is a set of characteristic indexes of corrosion risk factors screened from multi-source data.

[0161] Indexing a collection by features From the original matrix X Select the corresponding feature column to obtain the filtered corrosion risk factor matrix :

[0162]

[0163] in, Indicates the number of extracted corrosion risk factors. In this embodiment, after screening k = 3.

[0164] Based on the new water supply pipeline data, the corresponding corrosion risk factor vector is extracted and recorded as the standard vector :

[0165]

[0166] in is a standard vector, which represents the values ​​of each corrosion risk factor under the ideal state without corrosion. k is the number of extracted corrosion risk factors, k = 3.

[0167] Corrosion risk factor matrix The corrosion risk factor vector corresponding to each water supply pipeline :

[0168]

[0169] Calculate the Euclidean distance between it and the standard vector , as the deviation value.

[0170] Normalize all deviation values ​​to obtain the normalized deviation vector:

[0171]

[0172] The original corrosion grade label is based on the normalized deviation vector Make corrections, where Indicates the i The label value of each corrosion level is calculated to obtain the corrected label value of the corrosion level:

[0173]

[0174] in, Indicates the corrected i The label value of each corrosion level, that is, the corrosion risk assessment score, Indicates the total number of corrosion levels, m = 20, is the weighting coefficient that controls the fusion ratio between the original expert score and the objective deviation value.

[0175] Corrosion risk factor matrix The corresponding corrected label value As training data, the corrosion risk assessment model is established using linear fitting transformation and nonlinear fitting transformation:

[0176] When taking linear fitting transformation, construct the right multiplication transformation matrix , the relationship between the corrosion risk factor and the corrected label value is fitted through matrix operations:

[0177]

[0178] in, is the corrosion risk factor matrix, is the corrected label value matrix, is the right-multiplied transformation matrix obtained by the least squares method;

[0179] When nonlinear fitting transformation is adopted, the XGB regression algorithm based on gradient boosting tree is preferred as the nonlinear fitting function. , fitting the corrosion risk factor matrix F and the corrected label values ​​through decision tree ensemble learning The nonlinear relationship between:

[0180]

[0181] in, is the nonlinear fitting function for the corrosion risk factor matrix The i-th corrosion risk factor vector The predicted label value of For the A regression tree function, is the regression tree space, G is the total number of regression trees;

[0182] The nonlinear fitting function Fit each corrosion risk factor vector by minimizing the following objective function With the corrected label value The nonlinear relationship between:

[0183]

[0184] in, For the t The overall objective loss function of the round iteration, is the total number of corrosion grades, i.e. the total number of water supply pipe samples obtained, is the i-th corrosion risk factor vector at the t-1th iteration The predicted label value of For the t New regression tree Corrosion risk factor vector The predicted output of Corrected label value for the sample With the current forecast value The loss function between is calculated using cross loss. Add a regression tree Regularization term of complexity;

[0185] Calculate the linear fitting function separately With nonlinear fitting function The goodness of fit R² is:

[0186]

[0187] in, is the mean of the corrected label values, is the predicted label value of the fitting function;

[0188] The effects of linear and nonlinear models are as follows Figure 2 As shown, the linear fitting function The goodness of fit R²=0.885, nonlinear fitting function The goodness of fit R²=0.940, and the fitting function with a higher goodness of fit R² was selected as the final corrosion risk assessment model.

[0189] Thus, this embodiment has completed the construction of a corrosion risk assessment model for a water supply pipeline corrosion risk assessment method based on multi-source data and time series prediction, which can achieve a goodness of fit of 0.94.

[0190] If the above functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0191] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and such modifications or substitutions are intended to be within the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the scope of protection of the claims.

Claims

1. A water supply pipeline corrosion risk assessment method based on multi-source data and time series prediction, characterized by: The following steps are involved: Collecting a multi-source dataset of water supply pipelines with different corrosion levels annotated by experts, wherein the multi-source dataset includes multi-source data of water supply pipelines with different corrosion levels; Correlation analysis and screening of multi-source data sets were performed to extract the corrosion risk factor matrix, and a corrosion risk assessment model was constructed based on the corrosion risk factor matrix and its corresponding corrosion levels. For the pipe section to be evaluated, a multi-source data time series at multiple continuous time steps is obtained according to a preset time window, and a corrosion risk factor vector time series of the pipe section to be evaluated is obtained based on the multi-source data time series; Input the corrosion risk factor vector time series into the pre-trained LSTM model to predict the corrosion risk factor vector at each future time step; The corrosion risk factor vectors at the current moment and each future time step are input into the corrosion risk assessment model to obtain the corresponding risk scores; Based on the risk scores of each time step, generate warning levels for different time steps and formulate corresponding maintenance plans; The multi-source data includes inner wall image features, water quality monitoring features, pipe scale chemical and physical features, and operating condition features; The corrosion risk assessment model is constructed based on the corrosion risk factor matrix and its corresponding corrosion level, specifically including: Obtain multi-source data for a new water supply pipeline, construct a corrosion risk factor vector for the new water supply pipeline based on the multi-source data, and use this vector as a standard vector. Based on this standard vector, correct the corrosion grade label values ​​for water supply pipelines with different corrosion grades to obtain corrected label values. According to the revised label value and corrosion risk factor matrix , build a corrosion risk assessment model; The correction of the corrosion grade label values ​​of water supply pipes of different corrosion grades according to the standard vector specifically includes: Obtain multi-source data of a new water supply pipeline and extract its corresponding corrosion risk factor vector, which is recorded as the standard vector ,in is a standard vector, which represents the values ​​of each corrosion risk factor under the ideal state without corrosion. k is the number of extracted corrosion risk factors; Corrosion risk factor matrix The corrosion risk factor vector corresponding to each water supply pipeline , calculate the Euclidean distance between it and the standard vector , as the deviation value; Normalize all deviation values ​​to obtain the normalized deviation vector ; The original corrosion grade label is based on the normalized deviation vector Make corrections, where Indicates the m The label value of each corrosion level is calculated to obtain the corrected label value of the corrosion level: in, Indicates the corrected i The label value of each corrosion level, that is, the corrosion risk assessment score, Indicates the total number of corrosion levels, is the weighting coefficient, which controls the fusion ratio between the original expert score and the objective deviation value; The corrosion risk assessment model is constructed based on the corrected label value and the corrosion risk factor matrix F, specifically including: Corrosion risk factor matrix The corresponding corrected label value As training data, the corrosion risk assessment model is established using linear fitting transformation and nonlinear fitting transformation: When taking linear fitting transformation, construct the right multiplication transformation matrix , the relationship between the corrosion risk factor and the corrected label value is fitted through matrix operations: in, is the corrosion risk factor matrix, is the corrected label value matrix, is the right-multiplied transformation matrix obtained by the least squares method; When nonlinear fitting transformation is adopted, the XGB regression algorithm based on gradient boosting tree is used as the nonlinear fitting function , fitting the corrosion risk factor matrix F and the corrected label values ​​through decision tree ensemble learning The nonlinear relationship between: in, is the nonlinear fitting function for the corrosion risk factor matrix The i-th corrosion risk factor vector The predicted label value of For the A regression tree function, is the regression tree space, G is the total number of regression trees; The nonlinear fitting function Fit each corrosion risk factor vector by minimizing the following objective function With the corrected label value The nonlinear relationship between: in, For the t The overall objective loss function of the round iteration, is the total number of corrosion grades, i.e. the total number of water supply pipe samples obtained, is the i-th corrosion risk factor vector at the t-1th iteration The predicted label value of For the t New regression tree Corrosion risk factor vector The predicted output of Corrected label value for the sample With the current forecast value The loss function between is calculated using cross loss. Add a regression tree Regularization term of complexity; Calculate the linear fitting function separately With nonlinear fitting function The goodness of fit R² is: in, is the mean of the corrected label values, is the predicted label value of the fitting function; The fitting function with a higher goodness of fit R² was selected as the final corrosion risk assessment model.

2. The water supply pipeline corrosion risk assessment method based on multi-source data and time series prediction according to claim 1 is characterized in that: The inner wall image features are image features obtained by image processing of the inner wall image of the water supply pipe, including the ratio of the inner wall rust area, the maximum thickness of the corrosion nodule, the wall thickness loss, the inner wall erosion depth, the corrosion morphology type, the image grayscale value mean and the color difference change; The water quality monitoring characteristics are characteristics obtained by testing water samples obtained by sampling the water supply pipeline, including the pH value, conductivity, turbidity, total chlorine concentration, and iron and manganese ion concentrations of the water samples; The chemical and physical characteristics of pipe scale are indicators obtained through laboratory analysis of pipe scale samples collected on-site from water supply pipelines, including the mass fraction of iron and manganese in the pipe scale, the proportion of inorganic components, and surface roughness; The operating condition characteristics are characteristics obtained through the water supply system operation data acquisition system, including pipeline water flow velocity, water pressure, pipe age and historical maintenance times.

3. The water supply pipeline corrosion risk assessment method based on multi-source data and time series prediction according to claim 1 is characterized in that: The extraction of the corrosion risk factor matrix specifically includes: Constructing original feature matrix from multi-source datasets ,in, Indicates the i The multi-source data of water supply pipelines with different corrosion levels j Features, represents the total number of corrosion grades, i.e., the total number of water supply pipe samples obtained, and n represents the total number of features in the multi-source data; Constructing a corrosion grade label vector ,in, Indicates the m The label value of each corrosion level is obtained through expert scoring. The more serious the corrosion of the water supply pipe, the larger the label value of the corrosion level; For the original feature matrix Each column feature in Calculate its correlation with the corrosion grade label vector Pearson correlation coefficient: in, Indicates the j Correlation coefficient between item characteristics and corrosion grade, Indicates the j The average value of the feature, Indicates the The average value of the label values ​​of the corrosion levels; Set the correlation coefficient threshold , according to the correlation coefficient of each feature and the correlation coefficient threshold , filter features and extract core corrosion risk factors, which are described as: in, Indicates the j The absolute value of the correlation coefficient between the item feature and the corrosion grade, It is a set of characteristic indexes of corrosion risk factors screened from multi-source data; Indexing a collection by features From the original matrix X Select the corresponding feature column to obtain the filtered corrosion risk factor matrix : in, Represents the number of extracted corrosion risk factors.

4. The water supply pipeline corrosion risk assessment method based on multi-source data and time series prediction according to claim 1 is characterized in that: The method of obtaining a time series of multi-source data of the pipe section to be evaluated at multiple consecutive time steps according to a preset time window and obtaining a time series of a corrosion risk factor vector of the pipe section to be evaluated based on the multi-source data time series specifically includes: For the target water supply pipeline section, within the set sliding time window T Continuously collect each time step Multi-source raw data, building multi-source data time series ,in, Indicates the t Multi-source data vectors of time steps, Indicates the j The multi-source feature t The value of the time step; Combined with the established feature index set , extract the corresponding corrosion risk factor sub-vector from the multi-source data vector at each time step , obtain the corrosion risk factor vector time series ,in is the current time step.

5. The water supply pipeline corrosion risk assessment method based on multi-source data and time series prediction according to claim 1 is characterized in that: The training process of the LSTM model is as follows: The corrosion risk factor vector time series is used as the input training sample to predict the corrosion risk factor vector of the future target time step. According to the predicted corrosion risk factor vector and the corresponding true corrosion risk factor vector, the loss value is calculated through the loss function improved based on the correlation coefficient. The parameters of the LSTM model are updated according to the loss value to obtain the trained LSTM model.

6. The water supply pipeline corrosion risk assessment method based on multi-source data and time series prediction according to claim 5 is characterized in that: The improved loss function is: in, 、 The predicted and true t The corrosion risk factor vector of the time step j The value of the feature, For the j Correlation coefficient between item characteristics and corrosion grade, is the number of corrosion risk factors extracted.

Images