A method for analyzing and predicting corrosion of ship pipelines

Abstract

The present invention belongs to the technical field of ship pipeline health status analysis, and in particular, relates to a ship pipeline corrosion analysis and prediction method. It includes the following steps: obtaining ship pipeline corrosion related attribute data, eliminating invalid data groups; performing cluster analysis on ship pipeline corrosion related attribute data sets, specifying a preset threshold for deviation, determining noise elements with a deviation exceeding the preset threshold and eliminating them; determining ship pipeline corrosion related attribute data sets; performing decentralized processing according to the average value of each element to complete data standardization; establishing a ship pipeline corrosion prediction model; determining the initial parameters of the model; determining the optimal model parameters of the pipeline corrosion prediction model; collecting attribute index data corresponding to the ship pipeline corrosion key attribute data set, inputting it into the pipeline corrosion prediction model, and determining the pipeline corrosion prediction result. The present application can further reduce the difficulty of the ship pipeline corrosion analysis process and the consumption of system resources.

Description

Technical Field

[0001] The invention belongs to the technical field of ship pipeline health status analysis, and in particular relates to a ship pipeline corrosion analysis and prediction method. Background Art

[0002] Ships are in a corrosive environment for a long time, which leads to serious corrosion problems in the seawater pipelines of conventional ship power systems, affecting the navigation safety of conventional ships. The early analysis of ship pipeline corrosion was mainly based on historical experience combined with various monitoring methods. However, with the increasing number of ship management, the time and cost required by traditional methods have increased rapidly. At the same time, traditional methods are difficult to achieve predictive analysis of the development process and corrosion status of ship pipeline corrosion. The perception of the pipeline corrosion problem that Keneng just occurred is low, and it is no longer suitable for future comprehensive ship maintenance and can no longer meet the needs of ship pipeline maintenance and corrosion status analysis. Summary of the invention

[0003] The purpose of the present invention is to provide a ship pipeline corrosion analysis and prediction method for realizing ship pipeline corrosion state analysis and prediction based on actual needs, for predicting ship pipeline corrosion state, and obtaining key indicators of ship pipeline corrosion.

[0004] To achieve the above purpose, the present invention adopts the following technical solution.

[0005] A method for analyzing and predicting corrosion of ship pipelines comprises the following steps:

[0006] Step 1: Raw data collection:

[0007] Obtain the relevant attribute data of ship pipeline corrosion and eliminate invalid data groups;

[0008] Step 2: Data processing:

[0009] Cluster analysis is performed on the data set of ship pipeline corrosion related attributes, and a preset threshold of deviation is specified to determine the noise elements whose deviation exceeds the preset threshold and remove them;

[0010] Step 3: Extraction of key attributes of ship pipeline corrosion:

[0011] Determine the ship pipeline corrosion related attribute data set X = {x 1 ,x 2 ...x n ...x N};

[0012] Decentralize the data according to the average value of each element to complete the data standardization; the element value x′ after processing n It can be expressed as

[0013] Based on the processed element value x′ n The establishment of a decentralized pipeline corrosion related attribute data set matrix X′=[x 1 ′,x′ 2 ...x′ n ...x′ N ];

[0014] Based on the eigenvalue decomposition method, solve the covariance matrix X 1 =X′X′ T The eigenvector of

[0015] Arrange the pipeline corrosion related attribute data based on the eigenvalue size of the eigenvector.

[0016] According to the requirements, the first S rows are selected to form the projection matrix Y, and the decentralized pipeline corrosion related attribute data set matrix X′ is projected onto the projection matrix Y to obtain the ship pipeline corrosion key attribute data set matrix;

[0017] Step 4: Establish a ship pipeline corrosion prediction model consisting of a multi-layer BP neural network; specifically based on the following steps:

[0018] Establish a ship pipeline corrosion prediction model composed of a multi-layer BP neural network, and determine the relevant parameters of the ship pipeline corrosion prediction model based on the ship pipeline corrosion key attribute data set matrix determined in step 3;

[0019] Step 5: Extract the data in the key attribute data set matrix of ship pipeline corrosion as the original data set, optimize the ship pipeline corrosion prediction model through genetic algorithm, and determine the initial parameters of the model; specifically:

[0020] a1. Determine the initial weights and weight thresholds of each layer of the ship pipeline corrosion prediction model composed of a multi-layer BP neural network; establish an initial population composed of initial weights and weight thresholds as individuals, and encode the individuals in the initial population;

[0021] a2. Generate new initial weights and weight thresholds using the memory genetic algorithm, and substitute them into the ship pipeline corrosion prediction model. Use the data in the training data set to verify the prediction error of the ship pipeline corrosion prediction model corresponding to each sub-individual. where ω i refers to the weight corresponding to the i-th subpopulation and the prediction error under the weight threshold, where T refers to the number of samples, x n Refers to the sample value of the nth output ship pipeline corrosion index, x′ n Refers to the sample prediction value of the nth output ship pipeline corrosion index; based on the output error ω i Determine the probability of each child individual being selected as the parent of the next new population Where I refers to the total number of sub-individuals;

[0022] a3, generating new sub-individuals based on the genetic variation method, and determining the optimal sub-individual based on step a2 for iterative cycle; including:

[0023] New sub-individuals are generated based on the genetic algorithm, and the new sub-individuals can be expressed as

[0024] Among them 1,s and l 2,s represents the s-th attribute index of the key attribute data of ship pipeline corrosion corresponding to the new sub-individual, L 1,s and L 2,s represents the s-th attribute index of the key attribute data of ship pipeline corrosion corresponding to the previous generation individual, β∈[0,1];

[0025] Based on the mutation algorithm, the number of new individual populations is obtained by expanding the current individuals. The expanded sub-individuals can be expressed as:

[0026]

[0027] Among them l s represents the sth attribute index of the key attribute data of ship pipeline corrosion corresponding to the new individual, L s represents the sth attribute index of the key attribute data of ship pipeline corrosion corresponding to the current individual, λ 1 and λ 2 is the coefficient of variation, d refers to the number of iterations, and D is the preset maximum number of iterations;

[0028] a4, loop the above steps a2 and a3 to generate a new individual population, and use the prediction error ω i Determine the fitness of the sub-individuals and select the sub-individual with the highest fitness to enter the next cycle until the optimal population that meets the requirements is obtained or until the iteration ends;

[0029] Step 6: Generate the final subpopulation individuals according to step 5, extract the corresponding weights and weight thresholds, and determine the optimal parameters of the ship pipeline corrosion prediction model; divide the original data set into a training data set and a validation data set, use the training data set and the validation data set to train the ship pipeline corrosion prediction model, and determine the optimal model parameters of the pipeline corrosion prediction model;

[0030] Step 7: Collect the attribute index data corresponding to the key attribute data set of ship pipeline corrosion, input it into the pipeline corrosion prediction model, and determine the pipeline corrosion prediction result.

[0031] A further improvement or preferred implementation scheme of the aforementioned ship pipeline corrosion analysis and prediction method, wherein step 2 specifically comprises:

[0032] Based on the ship pipeline corrosion related attribute dataset X established in step 1, 1 ,x 2 ...x i ...}, x i ={x i1 ,x i2 ...x ip ...} represents the i-th group of data, x ip represents the attribute value of the pth corrosion-related attribute in the i-th group of data; establishes a clustering set C of ship pipeline corrosion-related attributes containing K cluster centers = {c 1 ,c 2 ,...c k ...c K}, based on the K-means clustering principle, determine the number of clusters K that minimizes the error, and its objective function can be expressed as:

[0033]

[0034] Where N k It refers to the total number of samples in the kth cluster;

[0035] Based on the determined number of clusters K, each cluster subset c is determined separately k The similarity r between element c and the cluster center k,j , r k,j Represents the similarity index from the jth element in the kth cluster subset to the cluster center; normalize each cluster subset c k Similarity index of elements in ;

[0036] Determine the deviation of elements in each subset Specify a preset deviation threshold, identify noise elements whose deviation exceeds the preset threshold and remove them.

[0037] As a further improvement or preferred implementation of the aforementioned ship pipeline corrosion analysis and prediction method, step 2 also includes a step for filling blank data and invalid data.

[0038] A further improvement or preferred implementation scheme of the aforementioned ship pipeline corrosion analysis and prediction method, wherein the relevant parameters determined in step 3 at least include:

[0039] The number of neurons in the input layer is S, where S is the total number of rows in the matrix of the key attribute data set of ship pipeline corrosion;

[0040] Input layer input {v 1 ,v 2 ...v a ...v A}, where v aThe key attribute data of row a in the key attribute data set matrix of ship pipeline corrosion;

[0041] The number of neurons in the hidden layer is M, where α∈[1,2S] refers to the adjustment coefficient;

[0042] The number of neurons in the output layer is N, and the output of the output layer is the indicators related to ship pipeline corrosion;

[0043] Model activation function: Sigmoid function.

[0044] The ship pipeline corrosion analysis and prediction method of the present application can further reduce the difficulty of the ship pipeline corrosion analysis process and the consumption of system resources. Through iterative updates and adaptive corrections based on historical experience data, it can continuously correct and improve the prediction accuracy of the ship pipeline corrosion status. At the same time, there is no need to perform overly complex large-capacity data analysis, thereby reducing the system load. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 This is a schematic diagram of the corrosion status of ship pipelines;

[0046] Figure 2 It is a schematic diagram of the corrosion status of ellipsoidal ship pipelines;

[0047] Figure 3 It is a schematic diagram of the corrosion status of cylindrical ship pipelines;

[0048] Figure 4 This is a schematic diagram of the corrosion state of uniform wall thickness ship pipelines; DETAILED DESCRIPTION

[0049] The problem of ship pipeline corrosion is an abnormal pipeline phenomenon often found in the use and maintenance of various types of ships, especially seawater auxiliary cooling pipelines in ships with great depth and long distance navigation, ship inlet and outlet water pipelines and other ship pipelines. Analysis shows that seawater pipelines are subjected to the impact of seawater for a long time, and the metal surface of the inner wall is easily damaged, exposing the metal matrix to seawater, which accelerates the electrical corrosion process. The flow field of seawater flowing through the corrosion section is more likely to produce turbulence, causing turbulent corrosion and cavitation corrosion. Among them, the arch back of the bend and the area with reduced wall thickness will further lead to aggravated corrosion due to stress concentration and reduced wall strength.

[0050] It can be seen that on actual ships, there are many types of factors that may affect the specific corrosion state of ship pipelines. The index values ​​of these influencing factors are an intuitive expression of the corrosion state of ship pipelines. By analyzing these influencing factor values, we can analyze the specific corrosion state of ship pipelines and predict the future corrosion state of ship pipelines based on the current values ​​and change laws of influencing factors. Based on this, this application proposes the following Figure 1The ship pipeline corrosion analysis and prediction method shown in the figure includes the following specific steps:

[0051] Step 1: Raw data collection:

[0052] Obtain the relevant attribute data of ship pipeline corrosion and eliminate invalid data groups;

[0053] The relevant attribute data in this application refers to the attribute data used to characterize the corrosion state of ship pipelines or affect the corrosion state of ship pipelines. For example, the commonly used materials for conventional ship seawater pipelines are red copper, alloy steel, copper-nickel alloy, and duplex stainless steel. The attributes of the ship pipeline corrosion state involved in the material attributes themselves include the corrosion rate of the material, the electrochemical corrosion rate of the metal material, the strength of the material, etc. At the same time, the medium environment in which the ship is sailing, including seawater flow rate, seawater dissolved oxygen content, seawater temperature, etc., in the actual analysis process, the initial data should be selected according to the actual working environment of the ship pipeline and its own attribute characteristics.

[0054] Step 2: Data processing:

[0055] Due to the complex and changeable working environment of the ship pipeline system, temporary disposal measures for various events, and the operation errors of the ship pipeline corrosion detection system itself, various abnormal noise data will inevitably appear in the process of monitoring and analyzing the corrosion status of ship pipelines. In order to ensure the validity of the data and improve the accuracy of the prediction results of the ship pipeline corrosion status analysis, the collected raw data must be preprocessed. It should be noted that in the actual processing process, since the static parameter attribute data remains unchanged, in the subsequent processing processes, these static attribute data are usually stripped off and the dynamic attribute data is retained for analysis, so as to reduce the data dimension, reduce the cost of data processing, and improve processing efficiency. The main contents of data processing include:

[0056] Based on the ship pipeline corrosion related attribute dataset X established in step 1, 1 ,x 2 ...x i ...}, x i ={x i1 ,x i2 ...x ip ...} represents the i-th group of data, x ip represents the attribute value of the pth corrosion-related attribute in the i-th group of data; establishes a clustering set C of ship pipeline corrosion-related attributes containing K cluster centers = {c 1 ,c 2 ,...c k ...c K};

[0057] Based on the K-means clustering principle, the number of clusters K that minimizes the error is determined, and its objective function can be expressed as:

[0058]

[0059] Where N k It refers to the total number of samples in the kth cluster;

[0060] Based on the determined number of clusters K, each cluster subset c is determined separately k The similarity r between element c and the cluster center k,j , r k,j Represents the similarity index from the jth element in the kth cluster subset to the cluster center; normalize each cluster subset c k Similarity index of elements in ;

[0061] Determine the deviation of elements in each subset Specify a preset threshold for deviation, determine the noise elements whose deviation exceeds the preset threshold and remove them;

[0062] Based on the above steps, the influence of corrosion abnormal attribute data that deviates from the actual situation on the prediction of ship pipeline corrosion can be prevented by eliminating noise elements. The data after the noise data processing generally needs to be screened, and several groups of data with higher data integrity are extracted as effective data for subsequent analysis and sorting; however, in actual situations, due to the complex working state of the pipeline and the abnormalities of the monitoring system and signal transmission, there may be a certain amount of missing data in the original data. The existence of these real data will directly affect the analysis results of ship pipeline corrosion. At the same time, since the corrosion state of ship pipelines continues to accumulate and change over time, the various indicators usually have the characteristics of time series continuity, and the missing value filling method can be used to avoid or limit the influence of missing data on the prediction of ship pipeline corrosion; specifically:

[0063] For the ship pipeline corrosion related attribute data X, the subsets are divided according to the data integrity, and the complete subset set C with complete data and the incomplete subset set C′ with missing data are obtained;

[0064] Step 3: Extraction of key attributes of ship pipeline corrosion:

[0065] Further analysis shows that there are a large number of factors among the aforementioned influencing factors that are correlated with each other or even directly cause and effect, that is, there are one or more secondary factors that are directly affected by the main factors. In order to avoid repeated calculation and analysis of the main factors, which may increase the influence of a certain main factor, these secondary factors should be eliminated. Specifically:

[0066] Based on the above steps, the final ship pipeline corrosion related attribute data set X = {x1 ,x 2 ...x n ...x N}; Decentralize according to the average value of each element to complete the data standardization; the element value x′ after processing n It can be expressed as

[0067] Based on the processed element value x′ n The establishment of a decentralized pipeline corrosion related attribute data set matrix X′=[x 1 ′,x′ 2 ...x′ n ...x′ N ];

[0068] Based on the eigenvalue decomposition method, solve the covariance matrix X 1 =X′X′ T The eigenvector of

[0069] Arrange the pipeline corrosion related attribute data based on the eigenvalue size of the eigenvector.

[0070] According to the requirements, the first S rows are selected to form the projection matrix Y, and the decentralized pipeline corrosion related attribute data set matrix X′ is projected onto the projection matrix Y to obtain the ship pipeline corrosion key attribute data set matrix;

[0071] Step 4: Establish a ship pipeline corrosion prediction model consisting of a multi-layer BP neural network; specifically based on the following steps:

[0072] Establish a ship pipeline corrosion prediction model composed of a multi-layer BP neural network, and determine the relevant parameters of the ship pipeline corrosion prediction model based on the ship pipeline corrosion key attribute data set matrix determined in step 3;

[0073] The relevant parameters include at least:

[0074] The number of neurons in the input layer is S, where S is the total number of rows in the matrix of the key attribute data set of ship pipeline corrosion;

[0075] Input layer input {v 1 ,v 2 ...v a ...v A}, where v a The key attribute data of row a in the key attribute data set matrix of ship pipeline corrosion;

[0076] The number of neurons in the hidden layer is M, where α∈[1,2S] refers to the adjustment coefficient;

[0077] The number of neurons in the output layer is N, and the output of the output layer is the indicators related to ship pipeline corrosion;

[0078] Model activation function: Sigmoid function;

[0079] Step 5: Extract the data in the key attribute data set matrix of ship pipeline corrosion as the original data set, optimize the ship pipeline corrosion prediction model through genetic algorithm, and determine the initial parameters of the model; specifically:

[0080] a1. Determine the initial weights and weight thresholds of each layer of the ship pipeline corrosion prediction model composed of a multi-layer BP neural network; establish an initial population composed of initial weights and weight thresholds as individuals, and encode the individuals in the initial population;

[0081] a2. Generate new initial weights and weight thresholds using the memory genetic algorithm, and substitute them into the ship pipeline corrosion prediction model. Use the data in the training data set to verify the prediction error of the ship pipeline corrosion prediction model corresponding to each sub-individual. where ω i refers to the weight corresponding to the i-th subpopulation and the prediction error under the weight threshold, where T refers to the number of samples, x n Refers to the sample value of the nth output ship pipeline corrosion index, x′ n Refers to the sample prediction value of the nth output ship pipeline corrosion index; based on the output error ω i Determine the probability of each child individual being selected as the parent of the next new population Where I refers to the total number of sub-individuals;

[0082] a3, generating new sub-individuals based on the genetic variation method, and determining the optimal sub-individual based on step a2 for iterative cycle; including:

[0083] New sub-individuals are generated based on the genetic algorithm, and the new sub-individuals can be expressed as

[0084] Among them 1,s and l 2,s represents the s-th attribute index of the key attribute data of ship pipeline corrosion corresponding to the new sub-individual, L 1,s and L 2,s represents the s-th attribute index of the key attribute data of ship pipeline corrosion corresponding to the previous generation individual, β∈[0,1];

[0085] Based on the mutation algorithm, the number of new individual populations is obtained by expanding the current individuals. The expanded sub-individuals can be expressed as:

[0086]

[0087] Among them ls represents the sth attribute index of the key attribute data of ship pipeline corrosion corresponding to the new individual, L s represents the sth attribute index of the key attribute data of ship pipeline corrosion corresponding to the current individual, λ 1 and λ 2 is the coefficient of variation, d refers to the number of iterations, and D is the preset maximum number of iterations;

[0088] a4, loop the above steps a2 and a3 to generate a new individual population, and use the prediction error ω i Determine the fitness of the sub-individuals and select the sub-individual with the highest fitness to enter the next cycle until the optimal population that meets the requirements is obtained or until the iteration ends;

[0089] Step 6: Generate the final subpopulation individuals according to step 5, extract the corresponding weights and weight thresholds, and determine the optimal parameters of the ship pipeline corrosion prediction model; divide the original data set into a training data set and a validation data set, use the training data set and the validation data set to train the ship pipeline corrosion prediction model, and determine the optimal model parameters of the pipeline corrosion prediction model;

[0090] Step 7: Collect the attribute index data corresponding to the key attribute data set of ship pipeline corrosion, input it into the pipeline corrosion prediction model, and determine the pipeline corrosion prediction result.

[0091] The above scheme is described in detail below with reference to specific embodiments:

[0092] The main corrosion forms of conventional ship seawater pipelines include cavitation corrosion, erosion corrosion and uniform corrosion. Different types of corrosion will form defects with different morphologies, and the degree of influence on the structural strength of seawater pipelines is also different. According to the morphology of corrosion defects, ellipsoidal defect model, cylindrical defect model and uniform wall thickness defect model are used to simulate cavitation corrosion, erosion corrosion and uniform corrosion respectively.

[0093] As the final stage of pipeline corrosion, ship pipelines will fail because they cannot withstand internal pressure. Usually, when the von Mises equivalent stress on the wall surface in the corrosion defect area reaches the specified non-proportional extension strength R of the seawater pipeline material, the ship pipeline will fail. p0.2 When the seawater pipeline fails due to corrosion.

[0094] This is used as the corrosion failure criterion to determine the state of the seawater pipeline when it fails due to corrosion, including cavitation corrosion, impact corrosion and uniform corrosion defects. The simulation results show that when the seawater pipeline fails due to corrosion, the state of the defective parts of the corroded seawater pipeline Figure 2 , 3 , Figure 4 shown.

[0095] For the above different pipeline corrosion conditions, the pipeline corrosion reaches the specified non-proportional extension strength R determined by simulation. p0.2 The attribute values ​​of each pipeline corrosion element at that time were compared with the pipeline corrosion state prediction results predicted by the aforementioned scheme, and the corrosion state indicators such as the depth of corrosion defects, axial length, and area of ​​corrosion defects were analyzed respectively. The statistical results showed that the minimum relative error between the simulated analysis value and the scheme prediction value on the above three corrosion state indicators was 0.9%, and the maximum relative error was 11.2%. Among them, the average relative error of the area of ​​corrosion defects was only 6.5%, which can meet the application needs of actual ship pipeline corrosion prediction.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention, rather than to limit the scope of protection of the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the essence and scope of the technical solution of the present invention.

Claims

1. A method for analyzing and predicting corrosion of ship pipelines, characterized in that: The steps include: Step 1: Raw data collection: Obtain the relevant attribute data of ship pipeline corrosion and eliminate invalid data groups; Step 2: Data processing: Cluster analysis is performed on the data set of ship pipeline corrosion related attributes, and a preset threshold of deviation is specified to determine the noise elements whose deviation exceeds the preset threshold and remove them; Step 3: Extraction of key attributes of ship pipeline corrosion: Determine the ship pipeline corrosion related attribute data set X = {x1, x2...x n ...x N }; Decentralize the data according to the average value of each element to complete the data standardization; the element value x′ after processing n It can be expressed as Based on the processed element value x′ n The establishment of a decentralized pipeline corrosion related attribute data set matrix X′=[x1′,x′2...x′ n ...x′ N ]; Based on the eigenvalue decomposition method, solve the covariance matrix X1=X′X′ T The eigenvector of Arrange the pipeline corrosion related attribute data based on the eigenvalue size of the eigenvector. According to the requirements, the first S rows are selected to form the projection matrix Y, and the decentralized pipeline corrosion related attribute data set matrix X′ is projected onto the projection matrix Y to obtain the ship pipeline corrosion key attribute data set matrix; Step 4: Establish a ship pipeline corrosion prediction model consisting of a multi-layer BP neural network; specifically based on the following steps: Establish a ship pipeline corrosion prediction model composed of a multi-layer BP neural network, and determine the relevant parameters of the ship pipeline corrosion prediction model based on the ship pipeline corrosion key attribute data set matrix determined in step 3; Step 5: Extract the data in the key attribute data set matrix of ship pipeline corrosion as the original data set, optimize the ship pipeline corrosion prediction model through genetic algorithm, and determine the initial parameters of the model; specifically: a1. Determine the initial weights and weight thresholds of each layer of the ship pipeline corrosion prediction model composed of a multi-layer BP neural network; establish an initial population composed of initial weights and weight thresholds as individuals, and encode the individuals in the initial population; a2. Generate new initial weights and weight thresholds using the memory genetic algorithm, and substitute them into the ship pipeline corrosion prediction model. Use the data in the training data set to verify the prediction error of the ship pipeline corrosion prediction model corresponding to each sub-individual. where ω i refers to the weight corresponding to the i-th subpopulation and the prediction error under the weight threshold, where T refers to the number of samples, x n Refers to the sample value of the nth output ship pipeline corrosion index, x′ n Refers to the sample prediction value of the nth output ship pipeline corrosion index; based on the output error ω i Determine the probability of each child individual being selected as the parent of the next new population Where I refers to the total number of sub-individuals; a3, generate new sub-individuals based on the genetic variation method, and determine the optimal sub-individual based on step a2 for iterative cycle; include: New sub-individuals are generated based on the genetic algorithm, and the new sub-individuals can be expressed as Among them 1,s and l 2,s represents the s-th attribute index of the key attribute data of ship pipeline corrosion corresponding to the new sub-individual, L 1,s and L 2,s represents the s-th attribute index of the key attribute data of ship pipeline corrosion corresponding to the previous generation individual, β∈[0,1]; Based on the mutation algorithm, the number of new individual populations is obtained by expanding the current individuals. The expanded sub-individuals can be expressed as: Among them l s represents the sth attribute index of the key attribute data of ship pipeline corrosion corresponding to the new individual, L s represents the s-th attribute index of the key attribute data of ship pipeline corrosion corresponding to the current individual, λ1 and λ2 are the coefficients of variation, d refers to the number of iterations, and D is the preset maximum number of iterations; a4, loop the above steps a2 and a3 to generate a new individual population, and use the prediction error ω i Determine the fitness of the sub-individuals and select the sub-individual with the highest fitness to enter the next cycle until the optimal population that meets the requirements is obtained or until the iteration ends; Step 6: Generate the final subpopulation individuals according to step 5, extract the corresponding weights and weight thresholds, and determine the optimal parameters of the ship pipeline corrosion prediction model; divide the original data set into a training data set and a validation data set, use the training data set and the validation data set to train the ship pipeline corrosion prediction model, and determine the optimal model parameters of the pipeline corrosion prediction model; Step 7: Collect the attribute index data corresponding to the key attribute data set of ship pipeline corrosion, input it into the pipeline corrosion prediction model, and determine the pipeline corrosion prediction result.

2. A method for analyzing and predicting corrosion of ship pipelines according to claim 1, characterized in that: The step 2 specifically includes: Based on the ship pipeline corrosion related attribute data set X = {x1, x2...x i ...}, x i ={x i1 ,x i2 ...x ip ...} represents the i-th group of data, x ip represents the attribute value of the pth corrosion-related attribute in the i-th group of data; establishes a clustering set C of ship pipeline corrosion-related attributes containing K cluster centers = {c1, c2, ...c k ...c K }, based on the K-means clustering principle, determine the number of clusters K that minimizes the error, and its objective function can be expressed as: Where N k It refers to the total number of samples in the kth cluster; Based on the determined number of clusters K, each cluster subset c is determined separately k The similarity r between element c and the cluster center k,j , r k,j Represents the similarity index from the jth element in the kth cluster subset to the cluster center; normalize each cluster subset c k Similarity index of elements in ; Determine the deviation of elements in each subset Specify a preset deviation threshold, identify noise elements whose deviation exceeds the preset threshold and remove them.

3. A method for analyzing and predicting corrosion of ship pipelines according to claim 2, characterized in that: The step 2 also includes a step for implementing blank data and invalid data filling.

4. A method for analyzing and predicting corrosion of ship pipelines according to claim 1, characterized in that: The relevant parameters determined in step 3 include at least: The number of neurons in the input layer is S, where S is the total number of rows in the matrix of the key attribute data set of ship pipeline corrosion; Input layer input {v1,v2...v a ...v A }, where v a The key attribute data of row a in the key attribute data set matrix of ship pipeline corrosion; The number of neurons in the hidden layer is M, where α∈[1,2S] refers to the adjustment coefficient; The number of neurons in the output layer is N, and the output of the output layer is the indicators related to ship pipeline corrosion; Model activation function: Sigmoid function.

Images