A data identification method for circulating water operation status

Abstract

The present application discloses a data identification method for circulating water operation status, comprising the following steps: step 1, collecting relevant operation data on site to form an original data set; step 2, preprocessing the data; step 3, selecting representative indicators according to the knowledge base and the actual situation; step 4, after the indicators are selected, performing k-means clustering, clustering the clustering object indicators into different distribution clusters and marking them, and different clusters represent different states; step 5, establishing a multi-core support vector machine classification prediction model with optimized weights, substituting the selected state quantities of the clusters and the cluster distribution cluster information as training values ​​into the constructed state indicator quantity prediction model for training, and obtaining a stable prediction model.

Description

Technical Field

[0001] The invention relates to a data recognition method for circulating water operation status, and in particular to a data pattern recognition method based on data mining. Background Art

[0002] Industrial circulating water system is a key subsystem in industries such as chemical, petroleum, metallurgy and textile, and is the main source of water and energy consumption. Improving the water-saving and energy-saving efficiency of industrial circulating water system has important theoretical significance and engineering application value for key industries such as chemical and petroleum. The state identification of industrial circulating water system is the first step to improve the water-saving and energy-saving of the system.

[0003] With the digital construction of enterprises, a large amount of data reflecting the operating characteristics of equipment and systems has been accumulated. These data are a mapping of the relationship between equipment. Making full use of these data and revealing their implicit laws are the key to realizing the status identification and optimization of industrial circulating water systems. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a data identification method for the operating status of circulating water in view of the above shortcomings, which solves the problems of low reuse efficiency of industrial circulating water and low water-saving and energy-saving efficiency. Data mining is used to realize the identification of the operating status of the industrial circulating water system, and the data generated during the operation of the circulating water system is fully utilized. The data is a direct mapping of the historical operation process, so the method has high accuracy in status identification.

[0005] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0006] A method for identifying data of circulating water operation status, comprising the following steps:

[0007] Step 1: Collect relevant on-site operation data to form an original data set;

[0008] Step 2: Preprocess the data;

[0009] Step 3: Select representative indicators based on the knowledge base and actual conditions;

[0010] Step 4: After the indicators are selected, k-means clustering is performed to cluster the indicators into different distribution clusters and mark them. Different clusters represent different states.

[0011] Step 5: Establish a multi-core support vector machine classification prediction model with optimized weights, substitute the selected state quantity of clustering and cluster distribution cluster information as training values ​​into the constructed state indicator quantity prediction model for training to obtain a stable prediction model.

[0012] Furthermore, in step 1, the equipment operation data, process data, environmental data, and water quality data generated during the operation of the circulating water system are collected to form an original data set, wherein each column of data represents a quantity.

[0013] Furthermore, the data preprocessing in step 2 includes missing data processing and data standardization;

[0014] Missing data processing is to fill in or delete some missing data caused by data collector failure or other reasons;

[0015] Data standardization is to eliminate the impact of data of different magnitudes on data of different attributes and to adapt data to certain algorithms. The data is standardized by min-max normalization to convert the data into a unified feature range.

[0016] The data standardization adopts min-max normalization, which can be expressed as formula (1):

[0017] Formula (1);

[0018] Where: —Original sample data The values ​​are normalized and mapped to [0,1]; —Original sample data; —The maximum value of the sample data; —Minimum value of sample data.

[0019] Furthermore, in step three, based on the knowledge base and equipment conditions, indicators that can represent the operating status of the circulating water system are selected as state indicators, and the preprocessed original data set is processed using data correlation analysis and important feature ranking to find indicators that have a strong correlation with the state indicators and record them as control indicators.

[0020] Furthermore, the data correlation analysis is a means of mining the relationship between data and data, which can analyze the dependency, correlation or influence between objects. The state index and the control index are linearly related, and the Pearson correlation coefficient γ is used to express the degree of relationship between the two. Its definition is shown in formula (2):

[0021] Formula (2);

[0022] Where: —The i-th sample ; -sample The mean of —is the i-th sample ; —For samples The mean of

[0023] When |γ|>0.95, there is a significant correlation between variables f and g;

[0024] When |γ|≥0.8, variables f and g are highly correlated;

[0025] When 0.5≤|γ|<0.8, variables f and g are moderately correlated;

[0026] When 0.3≤|γ|<0.5, variables f and g are poorly correlated;

[0027] When |γ|<0.3, the relationship is extremely weak, and variables f and g are unrelated;

[0028] When γ<0, the variables f and g change in opposite directions and are negatively correlated.

[0029] When γ=1 or γ=-1, the variables f and g are functionally related;

[0030] When -1<γ<1, the variables f and g are in a statistical relationship.

[0031] Furthermore, the important feature ranking is to rank the control index quantity by the extreme gradient algorithm. The extreme gradient algorithm is an algorithm as a weak learner improved on the basis of the decision tree. The calculation formula of the importance score of each attribute feature is shown in formula (3):

[0032] Formula (3);

[0033] Where: —The importance score of the leaf node (the jth attribute feature); — represents the sum of the first-order derivatives of all input samples mapped to leaf node j; — represents the sum of the second-order derivatives of all input samples mapped to leaf node j; — represents the L2 regularization coefficient;

[0034] They are ranked by importance using the extreme gradient algorithm and the most representative subset is selected.

[0035] Furthermore, the specific method flow of k-means clustering in step 4 is as follows:

[0036] Step 1: read the data and select the initial centers of K classes appropriately. Initially, they are generally randomly selected. In this embodiment, we set the K value to 3 based on the data situation.

[0037] Step 2: Start clustering. In each iteration of the centroid of a class, for any sample, calculate its Euclidean distance to the K centers respectively, and assign the sample to the class with the center with the shortest distance.

[0038] Step 3: re-label the category of each original sample data and re-find the position of the centroid with the smallest distance to the sample;

[0039] Step 4, repeat steps 2 and 3 until the error sum of squares converges and the centroid does not change significantly;

[0040] The calculation principle of the sum of squared errors is to calculate the mean of the sum of squares of the Euclidean distances from each type of sample data point to the nearest centroid. When this mean is the smallest, it is the final centroid of each type of sample data point.

[0041] Furthermore, the clustering process in step 2 is calculated as shown in formula (4):

[0042] Formula (4);

[0043] Where: —The category closest to sample data i among K categories, The value of is any one in [1,k]; —sample data i; j—current cluster center, ; —The mean of all samples of this category;

[0044] The centroid iteration of the class is shown in formula (5):

[0045] Formula (5);

[0046] In the formula: m—the number of sample data is m;

[0047] Then the error sum of squares is as shown in formula (6):

[0048] Formula (6);

[0049] When the K value is larger and the SSE is smaller, it means that the aggregation effect of the sample data is better;

[0050] Finally, a multi-core learning support vector machine classification prediction model with optimized weights was established. The original data of the indicators and the clustering result data were combined into a new data set, and divided into a training set and a test set in a 3:1 ratio. The training set was used to train the model to obtain a model that can predict the operating status of circulating water in real time, and the test set was used to test the accuracy of the model.

[0051] Furthermore, the specific steps of multi-core learning in step 5 are as follows:

[0052] Step 11, normalizing the data set consisting of the indicator data and the clustering result data;

[0053] Step 12, converting the normalized data into a Gaussian kernel function and a polynomial kernel function;

[0054] Step 13, determining the weights of the Gaussian kernel function and the polynomial kernel function through supervised learning, and linearly combining the weights of the Gaussian kernel function and the polynomial kernel function according to the weights to obtain a multi-kernel kernel function;

[0055] Step 14, obtaining a support vector machine model for multi-core learning according to the multi-core kernel function.

[0056] Furthermore, the Gaussian kernel function in step 12 is as shown in formula (7):

[0057] Formula (7);

[0058] in:

[0059] is the Gaussian kernel function;

[0060] is the horizontal coordinate of a point in two-dimensional space;

[0061] is the ordinate of a point in two-dimensional space;

[0062] is the bandwidth of the Gaussian kernel;

[0063] The polynomial kernel function in step 12 is shown in formula (8):

[0064] Formula (8);

[0065] in:

[0066] is the polynomial kernel function;

[0067] is the horizontal coordinate of a point in two-dimensional space;

[0068] is the ordinate of a point in two-dimensional space;

[0069] T is to transpose the data;

[0070] d is the degree of the polynomial.

[0071] The present invention adopts the above technical solution, and has the following technical effects compared with the prior art:

[0072] Compared with the prior art, the present invention adopts data mining to realize the identification of the operating status of the industrial circulating water system, and makes full use of the data generated during the operation of the circulating water system. The data is a direct mapping of the historical operation process. Therefore, the method has high accuracy in status identification and provides more effective guidance for further water-saving and energy-saving operations. At the same time, the method is also conducive to intelligent monitoring and is suitable for intelligent monitoring systems for equipment operation and maintenance, laying the foundation for the intelligent development of enterprises. BRIEF DESCRIPTION OF THE DRAWINGS

[0073] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following is a brief introduction to the drawings required for the specific embodiments or the description of the prior art. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn according to the actual scale.

[0074] Figure 1 It is a flow chart for identifying the operating status of the circulating water system;

[0075] Figure 2 It is the technical roadmap of the method of the present invention;

[0076] Figure 3 It is the kmeans clustering flow chart;

[0077] Figure 4 It is a flowchart of multi-core learning support vector machine. DETAILED DESCRIPTION

[0078] Embodiment 1, as Figure 1 and Figure 2 As shown, a method for identifying data of circulating water operation status comprises the following steps:

[0079] Step 1: Collect relevant on-site operation data to form the original data set.

[0080] The equipment operation data, process data, environmental data, and water quality data generated during the operation of the circulating water system are collected to form an original data set, in which each column of data represents a quantity.

[0081] Step 2: Preprocess the data, complete missing data, and standardize the data.

[0082] Missing data processing is to fill in or delete the missing data caused by data collector failure or other reasons by mean value; data standardization is to eliminate the impact of data of different magnitudes on different attribute data and to make the data adapt to certain algorithms. Standardization uses min-max normalization to convert the data into a unified feature range.

[0083] Step 3: Select representative indicators based on the knowledge base and actual conditions.

[0084] According to the knowledge base and equipment conditions, indicators that can represent the operating status of the circulating water system are selected as state indicators. The preprocessed original data set is processed using correlation analysis and important feature ranking to find indicators that have a strong correlation with the state indicators and record them as control indicators.

[0085] The state indicators are selected manually according to the knowledge base and the enterprise's on-site conditions. The control indicators are selected by correlation analysis to select the quantities with strong correlation with the state indicators. Then, they are ranked in importance through the extreme gradient algorithm to select the most representative subset.

[0086] In this embodiment, temperature difference and water quality are selected as state indicators, and then the correlation analysis method is used to select indicators with strong correlation with the state indicators, which are recorded as control indicators. The control indicators are variables that can be artificially adjusted according to actual conditions during operation. In this embodiment, the control indicators selected include return water temperature, return water flow, chloride ion concentration, iron ion concentration and other variables.

[0087] Step 4: After the indicators are selected, k-means clustering is performed to cluster the indicators into different distribution clusters and mark them. Different clusters represent different states.

[0088] The monitoring sampling data of state indicators and control indicators are clustered to form different distribution clusters. The cluster center value of each cluster is used in combination with the circulating water operation status to determine the excellent state of the indicator. This method can be used to determine the system operation status.

[0089] Step 5: Establish a multi-core support vector machine classification prediction model with optimized weights, substitute the selected state quantity of clustering and cluster distribution cluster information as training values ​​into the constructed state indicator quantity prediction model for training to obtain a stable prediction model.

[0090] Combined with the actual operation situation, it is easy to conclude that different distribution clusters represent different operating states, and a support vector machine classification prediction model with optimized weights is established. The indicator data set and the distribution cluster information are combined into a new data set as a training set to train the model. Finally, the real-time indicator data is input into the trained model to predict the current system operation status. If the status is not ideal, tuning guidance is given to ensure that the system operation status is always in a good state, thereby achieving the purpose of saving water and energy.

[0091] A multi-kernel learning support vector machine algorithm with optimized weights is used for state quantity classification prediction. Multi-kernel learning converts the input data into different kernel functions, and then converts different kernel functions into a synthetic kernel that can map different feature spaces, and finally uses it as a new classifier or regression function to build a model.

[0092] The final state prediction is to input the current data of the indicator into the prediction model, classify and predict the object state, and obtain the predicted state of the indicator.

[0093] The data preprocessing process in step 2 is as follows:

[0094] The selection of control indicators is to use correlation and XGBoost importance ranking to select indicators that are highly correlated with state indicators and can be adjusted during operation. In the actual system operation process, the state indicators are closely related to the control indicators. The control indicators affect the trend of the state indicators. Similarly, by adjusting the control indicators, the state indicators can be optimized, and then the system operation status can be optimized to meet the needs of water and energy saving.

[0095] Before selecting the indicator quantity, the present invention needs to preprocess the data to solve the problems of data redundancy and disorder, so that the data is standardized and more conducive to algorithm calculation. Data preprocessing includes missing data completion and data standardization, where data standardization adopts min-max normalization, which can be expressed as formula (1):

[0096] Formula (1);

[0097] Where: —Original sample data The values ​​are normalized and mapped to [0,1]; —Original sample data; —The maximum value of the sample data; —Minimum value of sample data.

[0098] The selection of the state indicator quantity in step 3 is based on the following:

[0099] The temperature difference is the temperature difference between the inlet and outlet of the cooling tower in the circulating water system. The cooling tower is an important subsystem in the circulating water operation system. The temperature difference between the inlet and outlet of the cooling tower is one of the methods for judging the operating efficiency of the cooling tower. Under the same conditions, the greater the temperature difference of the cooling tower, the higher the operating efficiency of the cooling tower. Therefore, the present invention uses the temperature difference as a state indicator to judge the operating state of the circulating water system.

[0100] The quality of water is an important factor affecting the operation of the circulating water system. During the operation of the circulating water system, it is necessary to continuously add chemicals to balance the water quality. Continuous operation and addition of chemicals will inevitably change the quality of the circulating water. The corrosive water will flush the pipe wall as the system runs, thereby causing flushing corrosion. At the same time, impurities such as salts and solid suspended matter in the cooling water will undergo a concentration reaction after circulation. In addition, when the circulating water in the cooling tower comes into contact with the atmosphere, it will also be contaminated by pollutants in the atmosphere. Both will eventually lead to scaling, corrosion and biological fouling in the industrial circulating water system. Corrosion and fouling are a key factor in water and energy consumption. Therefore, the present invention selects water quality as a state indicator to judge the operating state of the circulating water system.

[0101] When selecting control indicators, the present invention uses correlation analysis and extreme gradient boosting algorithm to select indicators that have a strong correlation with state indicators.

[0102] Data correlation analysis is a method of mining the relationship between data. It can analyze the dependency, correlation, or influence between objects. In practical applications, if the two objects under study have a linear relationship, the Pearson correlation coefficient γ can be used to express the degree of relationship between the two objects. Its definition is shown in formula (2):

[0103] Formula (2);

[0104] Where: —The i-th sample ; -sample The mean of —is the i-th sample ; —For samples The mean of .

[0105] When |γ|>0.95, there is a significant correlation between variables f and g;

[0106] When |γ|≥0.8, variables f and g are highly correlated;

[0107] When 0.5≤|γ|<0.8, variables f and g are moderately correlated;

[0108] When 0.3≤|γ|<0.5, variables f and g are poorly correlated;

[0109] When |γ|<0.3, the relationship is extremely weak, and variables f and g are unrelated;

[0110] When γ<0, the variables f and g change in opposite directions and are negatively correlated.

[0111] When γ=1 or γ=-1, the variables f and g are functionally related;

[0112] When -1<γ<1, the variables f and g are in a statistical relationship.

[0113] The eXtreme Gradient Boosting (XGBoost) algorithm is an improved weak learner algorithm based on the decision tree. The calculation formula for the importance score of each attribute feature is shown in formula (3):

[0114] Formula (3);

[0115] Where: —The importance score of the leaf node (the jth attribute feature); — represents the sum of the first-order derivatives of all input samples mapped to leaf node j; — represents the sum of the second-order derivatives of all input samples mapped to leaf node j; — represents the L2 regularization coefficient.

[0116] Finally, through correlation analysis and extreme gradient ranking, this example selects the following control indicators: the control indicators corresponding to the state indicator temperature difference are: outlet temperature, total return water temperature, total sewage temperature, and the control indicators corresponding to the state indicator water quality are: refrigeration station flow, total water supply flow, and conductivity. By controlling these indicators, the state indicator can be adjusted, and then the operation state can be optimized to achieve the purpose of water and energy saving.

[0117] like Figure 3 As shown, the specific method flow of k-means clustering in step 4 is as follows:

[0118] Step 1: read the data and select the initial centers of K classes appropriately. Initially, they are generally randomly selected. In this embodiment, we set the K value to 3 based on the data situation.

[0119] Step 2: Start clustering. In each iteration of the centroid of a class, for any sample, calculate its Euclidean distance to the K centers respectively, and assign the sample to the class with the center with the shortest distance.

[0120] Step 3: re-label the category of each original sample data and re-find the position of the centroid with the smallest distance to the sample;

[0121] Step 4: Repeat steps 2 and 3 until the error sum of squares converges and the centroid does not change significantly.

[0122] The calculation principle of the sum of squared errors is to calculate the mean of the sum of squares of the Euclidean distances from each type of sample data point to the nearest centroid. When this mean is the smallest, it is the final centroid of each type of sample data point.

[0123] The clustering process in step 2 is calculated as shown in formula (4):

[0124] Formula (4);

[0125] Where: —The category closest to sample data i among K categories, The value of is any one in [1,k]; —sample data i; j—current cluster center, ; —The mean of all samples in this category.

[0126] The centroid iteration of the class is shown in formula (5):

[0127] Formula (5);

[0128] In the formula: m—the number of sample data is m.

[0129] Then the error sum of squares is as shown in formula (6):

[0130] Formula (6);

[0131] When the K value is larger and the SSE is smaller, it means that the aggregation effect of the sample data is better.

[0132] Finally, a multi-core learning support vector machine classification prediction model with optimized weights was established. The original data of the indicators and the clustering result data were combined into a new data set, and divided into a training set and a test set in a 3:1 ratio. The training set was used to train the model to obtain a model that can predict the operating status of circulating water in real time, and the test set was used to test the accuracy of the model.

[0133] like Figure 4 As shown, the process of multi-core learning in step five is to first convert the input data into different kernel functions, then convert the different kernel functions into a synthetic kernel that can map different feature spaces, and finally use it as a new classifier or regression function to build a model, and finally output the result through model training. In the entire multi-core learning process, it is necessary to optimize the weights and reintegrate the single-core kernel functions to obtain a multi-core kernel function with the maximum interval surface. The kernel functions used in this embodiment are Gaussian kernel functions and polynomial kernel functions.

[0134] The Gaussian kernel function is shown in formula (7):

[0135] Formula (7);

[0136] in:

[0137] is the Gaussian kernel function;

[0138] is the horizontal coordinate of a point in two-dimensional space;

[0139] is the ordinate of a point in two-dimensional space;

[0140] is the bandwidth of the Gaussian kernel.

[0141] The polynomial kernel function is shown in formula (8):

[0142] Formula (8);

[0143] in:

[0144] is the polynomial kernel function;

[0145] is the horizontal coordinate of a point in two-dimensional space;

[0146] is the ordinate of a point in two-dimensional space;

[0147] T is to transpose the data;

[0148] d is the degree of the polynomial.

[0149] The specific steps of multi-core learning in this embodiment are as follows:

[0150] Step 11, normalizing the data set consisting of the indicator data and the clustering result data;

[0151] Step 12, converting the normalized data into a Gaussian kernel function and a polynomial kernel function;

[0152] Step 13, determining the weights of the Gaussian kernel function and the polynomial kernel function through supervised learning, and linearly combining the weights of the Gaussian kernel function and the polynomial kernel function according to the weights to obtain a multi-kernel kernel function;

[0153] Step 14, obtaining a support vector machine model for multi-core learning according to the multi-core kernel function.

[0154] After obtaining the support vector machine model, this embodiment can input the indicator data collected in real time into the model, combine historical data to make a prediction on the operating status of the circulating water, and give corresponding operating instructions to optimize the operating status of the circulating water system, thereby achieving the needs of water and energy saving.

[0155] The method of the present invention utilizes a data mining method to realize the recognition of the operating status of the circulating water system, fully utilizes the process data generated by the system itself, can effectively recognize the operating status of the circulating water system, and improve the operating efficiency of the system.

[0156] The description of the present invention is given for the purpose of illustration and description, and is not intended to be exhaustive or to limit the invention to the disclosed form. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments are selected and described in order to better illustrate the principles and practical applications of the present invention and to enable those of ordinary skill in the art to understand the present invention and thereby design various embodiments with various modifications suitable for particular uses.

Claims

1. A method for identifying data of circulating water operation status, characterized in that: The following steps are involved: Step 1: Collect relevant on-site operation data to form an original data set; Step 2: Preprocess the data; Step 3: Select representative indicators based on the knowledge base and actual conditions; Step 4: After the indicators are selected, k-means clustering is performed to cluster the indicators into different distribution clusters and mark them. Different clusters represent different states. Step 5: Establish a multi-core support vector machine classification prediction model with optimized weights, substitute the selected state quantity of clustering and cluster distribution cluster information as training values ​​into the constructed state indicator quantity prediction model for training, and obtain a stable prediction model; In the step 3, according to the knowledge base and the equipment conditions, an indicator that can represent the operating status of the circulating water system is selected as the state indicator, and the pre-processed original data set is processed by using data correlation analysis and important feature ranking to find out the indicator that has a strong correlation with the state indicator and record it as the control indicator; The data correlation analysis is a means of mining the relationship between data and data. It can analyze the dependency, correlation or influence between objects. The state index and the control index are linearly related. The Pearson correlation coefficient γ is used to express the degree of relationship between the two. Its definition is shown in formula (2): Formula (2); Where: —The i-th sample ; -sample The mean of —is the i-th sample ; —For samples The mean of When |γ|>0.95, there is a significant correlation between variables f and g; When |γ|≥0.8, variables f and g are highly correlated; When 0.5≤|γ|<0.8, variables f and g are moderately correlated; When 0.3≤|γ|<0.5, variables f and g are poorly correlated; When |γ|<0.3, the relationship is extremely weak, and variables f and g are unrelated; When γ<0, the variables f and g change in opposite directions and are negatively correlated. When γ=1 or γ=-1, the variables f and g are functionally related; When -1<γ<1, the variables f and g are in a statistical relationship; The important feature ranking is to rank the control index quantity by the extreme gradient algorithm. The extreme gradient algorithm is an algorithm as a weak learner improved on the basis of the decision tree. The calculation formula of the importance score of each attribute feature is shown in formula (3): Formula (3); Where: —The importance score of the leaf node (the jth attribute feature); — represents the sum of the first-order derivatives of all input samples mapped to leaf node j; — represents the sum of the second-order derivatives of all input samples mapped to leaf node j; — represents the L2 regularization coefficient; They are ranked by importance using the extreme gradient algorithm and the most representative subset is selected.

2. A method for identifying data of circulating water operation status according to claim 1, characterized in that: In the step 1, the equipment operation data, process data, environmental data, and water quality data generated during the operation of the circulating water system are collected to form an original data set, wherein each column of data represents a quantity.

3. A method for identifying data of circulating water operation status according to claim 1, characterized in that: The data preprocessing in step 2 includes missing data processing and data standardization; Missing data processing is to fill in or delete some missing data caused by data collector failure or other reasons; Data standardization is to eliminate the impact of data of different magnitudes on data of different attributes and to adapt data to certain algorithms. The data is standardized by min-max normalization to convert the data into a unified feature range. The data standardization adopts min-max normalization, which can be expressed as formula (1): Formula (1); Where: —Original sample data The values ​​are normalized and mapped to [0,1]; —Original sample data; —The maximum value of the sample data; —Minimum value of sample data.

4. A method for identifying data of circulating water operation status according to claim 1, characterized in that: The specific method flow of k-means clustering in step 4 is as follows: Step 1: read the data and select the initial centers of K classes appropriately. Initially, they are generally randomly selected. In this embodiment, we set the K value to 3 based on the data situation. Step 2: Start clustering. In each iteration of the centroid of a class, for any sample, calculate its Euclidean distance to the K centers respectively, and assign the sample to the class with the center with the shortest distance. Step 3: re-label the category of each original sample data and re-find the position of the centroid with the smallest distance to the sample; Step 4, repeat steps 2 and 3 until the error sum of squares converges and the centroid does not change significantly; The calculation principle of the sum of squared errors is to calculate the mean of the sum of squares of the Euclidean distances from each type of sample data point to the nearest centroid. When this mean is the smallest, it is the final centroid of each type of sample data point.

5. A method for identifying data of circulating water operation status according to claim 4, characterized in that: The clustering process in step 2 is calculated as shown in formula (4): Formula (4); Where: —The category closest to sample data i among K categories, The value of is any one in [1,k]; —sample data i; j—current cluster center, ; —The mean of all samples of this category; The centroid iteration of the class is shown in formula (5): Formula (5); In the formula: m—the number of sample data is m; Then the error sum of squares is as shown in formula (6): Formula (6); When the K value is larger and the SSE is smaller, it means that the aggregation effect of the sample data is better; Finally, a multi-core learning support vector machine classification prediction model with optimized weights was established. The original data of the indicators and the clustering result data were combined into a new data set, and divided into a training set and a test set in a 3:1 ratio. The training set was used to train the model to obtain a model that can predict the operating status of circulating water in real time, and the test set was used to test the accuracy of the model.

6. A method for identifying data of circulating water operation status according to claim 1, characterized in that: The specific steps of multi-core learning in step 5 are as follows: Step 11, normalizing the data set consisting of the indicator data and the clustering result data; Step 12, converting the normalized data into a Gaussian kernel function and a polynomial kernel function; Step 13, determining the weights of the Gaussian kernel function and the polynomial kernel function through supervised learning, and linearly combining the weights of the Gaussian kernel function and the polynomial kernel function according to the weights to obtain a multi-kernel kernel function; Step 14, obtaining a support vector machine model for multi-core learning according to the multi-core kernel function.

7. A method for identifying data of circulating water operation status according to claim 6, characterized in that: The Gaussian kernel function in step 12 is shown in formula (7): Formula (7); in: is the Gaussian kernel function; is the horizontal coordinate of a point in two-dimensional space; is the ordinate of a point in two-dimensional space; is the bandwidth of the Gaussian kernel; The polynomial kernel function in step 12 is shown in formula (8): Formula (8); in: is the polynomial kernel function; is the horizontal coordinate of a point in two-dimensional space; is the ordinate of a point in two-dimensional space; T is to transpose the data; d is the degree of the polynomial.

Images