Methods for predicting corrosion and scaling in industrial circulating water
By collecting and preprocessing water quality data, and using dimensionality reduction and machine learning algorithms to establish a corrosion and scaling prediction model, the high maintenance cost problem caused by hardware equipment in existing technologies has been solved, and high-precision corrosion and scaling prediction has been achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA NAT PETROLEUM CORP
- Filing Date
- 2022-07-13
- Publication Date
- 2026-07-17
AI Technical Summary
In existing technologies, methods for monitoring corrosion and scaling in industrial circulating water require hardware equipment, resulting in high maintenance costs.
By collecting and preprocessing water quality data, a corrosion and scaling prediction model was established using dimensionality reduction and machine learning algorithms. The prediction was performed using random forests and neural networks combined with genetic algorithms, thus avoiding the use of hardware equipment.
It achieves high-precision prediction of corrosion and scaling, reduces maintenance costs, and has a high degree of agreement with online detection results.
Smart Images
Figure CN117474136B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of petrochemical technology, specifically relating to a method for predicting corrosion and scaling in industrial circulating water. Background Technology
[0002] Heat exchangers are an indispensable component in petrochemical production processes, accounting for approximately one-fifth of the total equipment investment. Heat exchangers typically use water as the cooling medium, which carries away heat from the equipment and lowers its operating temperature during its flow. The circulating cooling water is continuously reused in the system process and is highly susceptible to corrosion and scaling due to various factors such as microorganisms, impurities, water flow velocity, and the equipment environment. Corrosion thins the inner walls of pipes, while scaling produces deposits that accumulate in the pipes, affecting the heat exchanger's efficiency and leading to decreased production load, product yield, and unplanned shutdowns, resulting in significant economic losses. However, current technologies typically employ electrochemical impedance spectroscopy for online corrosion monitoring, and some monitor the adhesion rate of heat exchangers online. These methods all require hardware, leading to high maintenance costs. Summary of the Invention
[0003] The purpose of this invention is to provide a method for predicting corrosion and scaling in industrial circulating water. This method solves the problem that existing technologies, such as online corrosion monitoring using electrochemical impedance spectroscopy and online monitoring of adhesion rates using heat exchangers, all require hardware and have high maintenance costs.
[0004] The technical solution adopted in this invention is: a method for predicting corrosion and scaling in industrial circulating water, and the specific operation steps are as follows:
[0005] Step 1: Collect data on corrosion rate and adhesion rate of corrosion-resistant coupons, as well as water quality analysis data of circulating water, including but not limited to pH, COD, chloride ions, potassium ions, residual chlorine, conductivity, silica, and turbidity, and perform data preprocessing.
[0006] Step 2: Set the permissible limit PL for circulating water quality indicators according to water quality management regulations, and set the ideal limit DL for each water quality indicator based on on-site production experience, and calculate the corrected permissible limit MPL for each circulating water quality indicator.
[0007] MPL = 0.7 * PL + 0.3 * DL (1)
[0008] Step 3: For each water quality analysis data point x ji Different processing methods are obtained based on the following judgments.
[0009] If x j,i ≤DL, then
[0010] If DL < x j,i <MPL, then
[0011] If MPL ≤ x j,i <PL, then
[0012] Step 4: Arrange the result obtained in Step 3 According to the time sequence and store it in the process database, and then sum up the corrosion rate and the adhesion rate within this time interval according to the analysis frequency As the final input variable value for the subsequent corrosion and scaling prediction model;
[0013] Step 5: Based on the input variable values obtained in Step 4, perform feature engineering using a dimensionality reduction algorithm, and screen out effective water quality indicators from all water quality indicators to form a data set;
[0014] Step 6: Establish a corrosion and scaling prediction model based on the corrosion and scaling algorithm according to the effective water quality indicators screened in Step 5;
[0015] Step 7: Use the N-fold cross-validation method to divide the data set processed and screened in Step 5 into a training set and a test set according to a certain proportion. The training set is used to train the parameters of the corrosion and scaling prediction model to obtain an optimized corrosion and scaling prediction model, and the test set is used to verify the prediction ability of the established model; Finally, predict corrosion and scaling according to the established corrosion and scaling prediction model.
[0016] The characteristics of the present invention also lie in
[0017] The preprocessing of the steps includes deleting invalid data, linearly interpolating and fitting to fill in missing values, analyzing outliers through box plots to determine abnormal values and processing them according to missing values, specifically as follows:
[0018] Perform data preselection through the grey relational algorithm, so as to preselect the easily measurable data for predicting the corrosion and scaling trend; Arrange the collected data according to the time sequence and store it in the process database; Clean the data in the process database, and remove the abnormal data according to the standard deviation of the samples in the process database.
[0019] The dimensionality reduction algorithm in Step 5 selects any one of principal component analysis, locally linear embedding, partial least squares method, ridge regression, genetic algorithm, adaptive immune genetic algorithm, and mutual information.
[0020] The corrosion and scaling prediction model in Step 6 consists of a machine learning algorithm and a parameter optimization algorithm. The machine learning algorithm includes artificial neural network, random forest, and support vector machine; The parameter optimization algorithm includes genetic algorithm and particle swarm optimization algorithm.
[0021] For corrosion prediction, a random forest combined with particle swarm optimization algorithm was selected, while for scale prediction, a neural network combined with a genetic algorithm was selected.
[0022] In step 7, the ratio of the training set to the test set is 6:4, 7:3, or 8:2.
[0023] The beneficial effects of this invention are as follows: The industrial circulating water corrosion and scaling prediction method of this invention first reduces the dimensionality of the collected data, and then establishes a corrosion and scaling prediction model through a hybrid algorithm. This model algorithm considers two aspects: first, the correlation between water quality indicators and prediction targets, especially the preprocessing of water quality data; second, it uses a comprehensive hybrid modeling technique for prediction. This invention can make more accurate predictions, has a high degree of agreement with the results of online monitoring equipment, requires no hardware, and has low maintenance costs. Attached Figure Description
[0024] Figure 1 This is a comparison chart of the corrosion rate predicted and the measured value in Example 1 of the present invention;
[0025] Figure 2 This is a comparison chart of the predicted scaling rate and the measured value in Example 1 of the present invention;
[0026] Figure 3 This is a comparison chart of the corrosion rate predicted and the measured value in Example 2 of the present invention;
[0027] Figure 4 This is a comparison chart of the predicted scaling rate and the measured value in Example 2 of the present invention; Detailed Implementation
[0028] The method for predicting corrosion and scaling in industrial circulating water according to the present invention is implemented according to the following steps:
[0029] Step 1: Collect data on corrosion rate and adhesion rate of corrosion-resistant coupons, as well as water quality analysis data of circulating water, including but not limited to pH, COD, chloride ions, potassium ions, residual chlorine, conductivity, silica, and turbidity, and perform data preprocessing.
[0030] The preprocessing described in step 1 includes deleting invalid data, performing linear interpolation to fit and fill missing values, and analyzing outliers through box plots to identify and process them as missing values, as detailed below:
[0031] Data pre-selection is performed using a grey relational analysis algorithm to select easily measurable data for predicting corrosion and scaling trends; the collected data is arranged chronologically and stored in a process database; the data in the process database is cleaned, and outliers are removed based on the standard deviation of the samples in the process database.
[0032] Step 2: Set the allowable limit PL of the circulating water quality index according to the water quality management regulations, set the ideal limit DL of each water quality index based on on-site production experience, and calculate the modified allowable limit MPL of each circulating water quality index;
[0033] MPL = 0.7*PL + 0.3*DL (1)
[0034] Step 3: For each water quality analysis data x j,i Make different treatments according to the following judgments to obtain
[0035] If x j,i ≤ DL, then
[0036] If DL < x j,i < MPL, then
[0037] If MPL ≤ x j,i < PL, then
[0038] Step 4: Arrange the results obtained in Step 3 Arrange them in chronological order and store them in the process database, and then sum up the corrosion rate and adhesion rate analysis frequencies within this time interval As the final input variable value for the subsequent corrosion and scaling prediction model;
[0039] Since there is a water quality analysis value every few hours or days (the time interval is fixed), and the corrosion rate and adhesion rate are often analyzed once a month, in order to align the input and output of the data set, sum up the corrosion rate or adhesion rate analysis values corresponding to all water quality analysis indicators within the month as the final model input; the analysis frequency refers to the analysis time interval.
[0040] Step 5: Based on the input variable values obtained in Step 4, perform feature engineering using a dimensionality reduction algorithm, and select effective water quality indicators from all water quality indicators to form a data set; the purpose of feature engineering is to select effective water quality indicators;
[0041] The dimensionality reduction algorithm selects any one of principal component analysis, locally linear embedding, partial least squares method, ridge regression, genetic algorithm, adaptive immune genetic algorithm, and mutual information.
[0042] Step 6: Establish a corrosion and scaling prediction model based on the corrosion and scaling algorithm according to the effective water quality indicators selected in Step 5;
[0043] The corrosion and scaling prediction model in step 6 consists of machine learning algorithms and parameter optimization algorithms. The machine learning algorithms include artificial neural networks, random forests, and support vector machines; the parameter optimization algorithms include genetic algorithms and particle swarm optimization algorithms.
[0044] For corrosion prediction, a random forest combined with particle swarm optimization algorithm was selected, while for scale prediction, a neural network combined with a genetic algorithm was selected.
[0045] Step 7: Using N-fold cross-validation, the dataset processed and filtered in Step 5 is divided into a training set and a test set according to a certain ratio, which is 6:4, 7:3, or 8:2. The training set is used to train the parameters of the corrosion and scaling prediction model to obtain the optimized corrosion and scaling prediction model, and the test set is used to verify the predictive ability of the established model. Finally, corrosion and scaling are predicted based on the established corrosion and scaling prediction model.
[0046] Example 1
[0047] This case study uses circulating water quality data from a petrochemical company in Northwest China as a basis.
[0048] Water quality analysis data of circulating water from four water treatment plants were collected from October 2016 to October 2020. Fourteen water quality indicators were selected, including pH, turbidity, orthophosphate, nitrate, residual chlorine, potassium ion, calcium hardness, concentration factor, total hardness, conductivity, total iron, chloride ion, heterotrophic bacteria, and suspended solids for subsequent treatment. The analysis frequency for scaling adhesion rate and corrosion rate was once per month. Other water quality indicators were divided into four categories according to their analysis frequency: three times per day, once per day, three times per week, and once per week. Then, ideal upper and lower limits and permissible upper and lower limits for water quality indicators were set according to actual management systems and technical requirements. The values are shown in Table 1.
[0049] Table 1. Setting of Water Quality Parameters for Circulating Water
[0050]
[0051]
[0052] After data preprocessing and water quality data calculation, using monthly data as the benchmark, the water quality index was summed based on the analysis frequency to unify the time dimension attributes of the water quality index, corrosion rate, and scaling rate, ultimately resulting in a dataset of 138 water quality data points. In modeling, the datasets were randomly partitioned in a 6:4 ratio, with a training data matrix of 14×83 and a test data matrix of 14×55. An adaptive immune genetic algorithm (algorithm parameters are shown in Table 2) was used for dimensionality reduction, resulting in five water quality indicators: potassium ions, phosphorus, conductivity, total iron, and heterotrophic bacteria. A random forest combined with particle swarm optimization algorithm was used to train the corrosion rate model, and a neural network combined with a genetic algorithm was used to predict the scaling rate model. The model parameter values obtained during training are shown in Tables 3 and 4, respectively. Finally, model prediction was performed, and the results are shown in Table 3. Figure 1 , 2 As shown, it can be observed that the predicted results are closely distributed along the diagonal, which is in good agreement with the actual values.
[0053] Table 2. Parameter settings for the adaptive immune genetic algorithm
[0054]
[0055] Table 3. Corrosion rate model parameter values
[0056]
[0057] Table 4. Parameter values of the scaling rate model
[0058]
[0059] Example 2
[0060] This case study uses circulating water quality data from a petrochemical company in Northwest China as a basis.
[0061] Water quality analysis data of circulating water from eight water treatment plants were collected from November 2018 to March 2021. Twelve water quality indicators closely related to corrosion rate were selected, including pH, turbidity, residual chlorine, potassium ions, calcium hardness, concentration factor, total hardness, conductivity, total iron, chloride ions, heterotrophic bacteria, and suspended solids, for subsequent treatment. The corrosion rate was analyzed once a month. Other water quality indicators were divided into four categories according to their analysis frequency: three times a day, once a day, three times a week, and once a week. Then, ideal upper and lower limits and allowable upper and lower limits of water quality indicators were set according to actual management system and technical requirements. The values are shown in Table 5.
[0062] Table 5. Setting of Water Quality Indicators for Circulating Water
[0063]
[0064]
[0065] After data preprocessing and water quality data calculation, using monthly data as the benchmark, the water quality index was summed based on the analysis frequency to unify the time dimension attributes of the water quality index, corrosion rate, and scaling rate, ultimately resulting in a dataset of 185 water quality data points. In modeling, the datasets were randomly partitioned in a 6:4 ratio, with a training data matrix of 12×111 and a test data matrix of 12×74. An adaptive immune genetic algorithm (algorithm parameters are shown in Table 6) was used for dimensionality reduction, resulting in four water quality indicators: potassium ions, phosphorus, conductivity, and heterotrophic bacteria. A random forest combined with particle swarm optimization was used to train the corrosion rate model, and a neural network combined with a genetic algorithm was used to predict the scaling rate model. The model parameter values obtained during training are shown in Tables 7 and 8, respectively. Finally, model prediction was performed, and the results are shown in Table 7. Figure 3 , 4 As shown, it can be observed that the predicted results are closely distributed along the diagonal, which is in good agreement with the actual values.
[0066] Table 6. Parameter Settings for Adaptive Immune Genetic Algorithm
[0067]
[0068]
[0069] Table 7. Corrosion Rate Model Parameter Values
[0070]
[0071] Table 8. Parameter values for the scaling rate model
[0072]
Claims
1. A method for predicting corrosion and scaling in industrial circulating water, characterized in that, The specific operating steps are as follows: Step 1: Collect data on corrosion rate and adhesion rate of corrosion-resistant coupons, as well as water quality analysis data of circulating water, including but not limited to pH, COD, chloride ions, potassium ions, residual chlorine, conductivity, silica, and turbidity, and perform data preprocessing. Step 2: Set the permissible limit PL for circulating water quality indicators according to water quality management regulations, and set the ideal limit DL for each water quality indicator based on on-site production experience, and calculate the corrected permissible limit MPL for each circulating water quality indicator. MPL = 0.7 * PL + 0.3 * DL (1) Step 3: For each water quality analysis data point x j,i Different processing methods are obtained based on the following judgments. If x j,i ≤DL, then If DL < x j,i <MPL, then If MPL ≤ x j,i <PL, then Step 4: Take the result from Step 3 The data is stored in the process database in chronological order, and then summed according to the analysis frequency of corrosion rate and adhesion rate within that time interval. As the final input variable values for subsequent corrosion and scaling prediction models; Step 5: Based on the input variable values obtained in Step 4, feature engineering is performed using a dimensionality reduction algorithm to select effective water quality indicators from all water quality indicators to form a dataset; Step 6: Establish a corrosion and scaling prediction model based on the effective water quality indicators selected in Step 5; Step 7: Using N-fold cross-validation, the dataset processed and filtered in Step 5 is divided into a training set and a test set according to a certain ratio. The training set is used to train the parameters of the corrosion and scaling prediction model to obtain the optimized corrosion and scaling prediction model. The test set is used to verify the predictive ability of the established model. Finally, corrosion and scaling are predicted based on the established corrosion and scaling prediction model.
2. The method for predicting corrosion and scaling in industrial circulating water according to claim 1, characterized in that, The preprocessing described in step 1 includes deleting invalid data, performing linear interpolation to fit and fill missing values, and analyzing outliers through box plots to identify and process them as missing values, as detailed below: Data pre-selection is performed using a grey relational analysis algorithm to identify easily measurable data for predicting corrosion and scaling trends; the collected data is then stored in the process database in chronological order. The data in the process database is cleaned, and outlier data is removed based on the standard deviation of the samples in the process database.
3. The method for predicting corrosion and scaling in industrial circulating water according to claim 1, characterized in that, The dimensionality reduction algorithm described in step 5 can be any one of principal component analysis, local linear embedding, partial least squares, ridge regression, genetic algorithm, adaptive immune genetic algorithm, or mutual information.
4. The method for predicting corrosion and scaling in industrial circulating water according to claim 1, characterized in that, The corrosion and scaling prediction model described in step 6 consists of machine learning algorithms and parameter optimization algorithms. The machine learning algorithms include artificial neural networks, random forests, and support vector machines; the parameter optimization algorithms include genetic algorithms and particle swarm optimization algorithms. For corrosion prediction, a combination of random forests and particle swarm optimization algorithms is used, while for scaling prediction, a combination of neural networks and genetic algorithms is used.
5. The method for predicting corrosion and scaling in industrial circulating water according to claim 1, characterized in that, The ratio of the training set to the test set in step 7 is 6:4, 7:3, or 8:2.