A method and system for early warning of ultrafiltration device operation failure

Abstract

The application discloses an ultrafiltration device operation fault early warning method and system, belongs to the technical field of ultrafiltration device monitoring, and solves the problem that an ultrafiltration device cannot send a warning signal when encountering a fault during operation; the method comprises the following steps: summarizing cases of serious pollution blocking and flow overrun of the ultrafiltration device, collecting and storing historical operation data, collecting real-time operation data in the ultrafiltration device according to relevant parameters, constructing a mathematical model of the ultrafiltration device based on a One-Class-SVM classification algorithm, inputting the collected real-time operation data into the trained mathematical model of the ultrafiltration device, judging whether the operation state of the ultrafiltration device is abnormal, constructing a mechanism model of the ultrafiltration device and obtaining mechanism data, combining an on-off quantity sequence output by the mathematical model of the ultrafiltration device, and performing early warning information analysis; the application pushes abnormal information to relevant personnel in the initial stage of serious pollution blocking or flow overrun of the ultrafiltration device, improves system operation efficiency, and avoids affecting the normal operation of a boiler make-up water system.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of ultrafiltration device monitoring, and relates to an ultrafiltration device operation fault early warning method and system. BACKGROUND

[0002] An ultrafiltration device is a kind of pressurized membrane separation device and is an important component of a water treatment system. The ultrafiltration device uses an ultrafiltration membrane as a filtering medium. When raw water flows through the membrane surface, under the pressure difference between the two sides of the ultrafiltration membrane, the membrane surface is densely covered with many small micropores, which only allow water and small molecules to pass through to become a permeate, and large molecules in the raw water with a size greater than the membrane surface micropore diameter are retained on the water inlet side of the membrane to become a concentrate. The ultrafiltration device can effectively remove particles, colloids, bacteria, heat sources and organic matter in water, and achieve the effects of purification, separation and concentration of raw water.

[0003] In the monitoring process of the existing ultrafiltration water treatment device, a plurality of sensors need to be installed to collect the internal conditions of the device, such as monitoring data of pH value, conductivity, temperature, pressure, chemical oxygen demand and the like of water. An application patent application with the application publication number CN116226484A discloses an ultrafiltration water treatment device monitoring data management system, which uses the changes of multi-dimensional sensor data to quantify the aging degree of the corresponding ultrafiltration water treatment device, and represents the current equipment condition of the ultrafiltration water.

[0004] During the operation of the ultrafiltration system, with the interception and accumulation of pollutants on the membrane surface, the flux of the ultrafiltration membrane will gradually decrease, the treated water volume will gradually decrease, and the operating pressure difference will gradually increase, so it is necessary to regularly perform backwashing on the ultrafiltration device. Under the current technology, the operation of the ultrafiltration system mainly relies on sequential control, that is, the ultrafiltration device is stopped for backwashing only when the program runs to the set time. However, when serious pollution blocking or flow limit occurs during the operation of the ultrafiltration system, the ultrafiltration device is in an overload operation state, which is easy to cause equipment damage, but the existing ultrafiltration system cannot issue a warning signal when the ultrafiltration device encounters an abnormal situation leading to operation failure. SUMMARY

[0005] The technical scheme of the application is used to solve the problem that the ultrafiltration device cannot issue a warning signal when encountering a failure during operation.

[0006] The application solves the above technical problems through the following technical scheme:

[0007] An ultrafiltration device operation fault early warning method, comprising:

[0008] Step 1, summarize cases of serious pollution blocking and flow limit of the ultrafiltration device, collect and store historical operation data, and propose corresponding solutions and treatment measures for different types of operation failures;

[0009] Step 2, collect real-time operation data of the ultrafiltration device according to relevant parameters, and perform relevant pretreatment measures;

[0010] Step 3, construct a mathematical model of the ultrafiltration device based on the One-Class-SVM classification algorithm, including:

[0011] (1) screen the historical operation data of the ultrafiltration device in step 1, and perform relevant pretreatment measures;

[0012] (2) divide the training data set and the test data set of the ultrafiltration device;

[0013] (3) set the support vector machine classification model parameters and train the mathematical model;

[0014] (4) verify the accuracy of the mathematical model of the ultrafiltration device;

[0015] (5) evaluate and adjust the mathematical model of the ultrafiltration device;

[0016] Step 4, input the real-time operation data collected in step 2 into the mathematical model of the ultrafiltration device trained in step 3, output the model calculation result, and judge whether the operation state of the ultrafiltration device is abnormal;

[0017] Step 5, construct a mechanism model of the ultrafiltration device and obtain mechanism data, combine the on-off quantity sequence output by the mathematical model of the ultrafiltration device, and perform early warning information analysis.

[0018] Further, the relevant parameters in step 2 include the inlet flow of the ultrafiltration device, the pressure difference between the inlet and outlet of the ultrafiltration device, and the frequency feedback of the water supply pump of the ultrafiltration device.

[0019] Further, the pretreatment measures in step 2 include working condition division and principal component analysis.

[0020] Further, the method for dividing the training data set and the test data set of the ultrafiltration device in step 3 (2) is as follows:

[0021] According to the historical operation data selected in (1), the average value of the data separated by a certain time length is calculated as the modeling data set, and the training data set and the test data set of the mathematical model of the ultrafiltration device are divided, wherein the training data set is used to train the One-Class-SVM mathematical model of the ultrafiltration device, and the test data set is used to evaluate and adjust the One-Class-SVM mathematical model of the ultrafiltration device.

[0022] Further, the calculation formula for verifying the accuracy of the mathematical model of the ultrafiltration device in step 3 (4) is as follows:

[0023]

[0024] Wherein, accuracy is the model prediction accuracy, N accuracy is the number of data samples predicted correctly by the model, N test is the total number of samples in the validation dataset.

[0025] Further, the method of step 4 (5) includes: when the accuracy of the mathematical model of the ultrafiltration device is lower than the set value in (4), returning to (3) to modify the parameters of the training model for repeated training until the required early warning accuracy of the mathematical model of the ultrafiltration device is reached.

[0026] Further, the mechanism data in step 5 includes ultrafiltration device feed pump start-stop monitoring data, ultrafiltration device start-stop monitoring data, and mathematical model abnormal state number.

[0027] Further, the method of outputting the switch sequence in step 5 is as follows:

[0028] a. If the analysis result of the mathematical model of the ultrafiltration device on real-time running data is "normal state", the mathematical model state outputs 0, and the mechanism model of the ultrafiltration device is not enabled; if the analysis result of the mathematical model of the ultrafiltration device on real-time running data is "abnormal state", the mathematical model state outputs 1;

[0029] b. If the ultrafiltration device feed pump start-stop monitoring running signal has no change, the ultrafiltration device feed pump start-stop monitoring state outputs 0; if the ultrafiltration device feed pump start-stop monitoring running signal changes from 0 to 1 or the running signal changes from 1 to 0, the ultrafiltration device feed pump start-stop monitoring state outputs 1;

[0030] c. If the ultrafiltration device water production valve running state changes from 0 to 1 or the water production valve running state changes from 1 to 0 or the water production valve keeps running state as 0, the ultrafiltration device start-stop monitoring state outputs 0; if the ultrafiltration device water production valve running state is 1, the ultrafiltration device start-stop monitoring state outputs 1;

[0031] d. If the number of consecutive abnormal states of the mathematical model does not exceed 3 times, the number of consecutive abnormal states of the mathematical model monitoring output is 0; if the number of consecutive abnormal states of the mathematical model exceeds 3 times, the number of consecutive abnormal states of the mathematical model monitoring output is 1.

[0032] Further, the method of step 5 for early warning information analysis is as follows:

[0033] a. When the mathematical model state output of the ultrafiltration device is 0, no early warning information is sent;

[0034] b. When the mathematical model state output of the ultrafiltration device is 1 and the ultrafiltration device feed pump start-stop monitoring state output is 1, no early warning information is sent;

[0035] c. When the mathematical model state output of the ultrafiltration device is 1, the water pump start-stop monitoring state output of the ultrafiltration device is 0, and the start-stop monitoring state output of the ultrafiltration device is 0, no early warning information is issued;

[0036] d. When the mathematical model state output of the ultrafiltration device is 1, the water pump start-stop monitoring state output of the ultrafiltration device is 0, the start-stop monitoring state output of the ultrafiltration device is 1, and the mathematical model continuous abnormal state number monitoring state output is 0, no early warning information is issued;

[0037] e. When the mathematical model state output of the ultrafiltration device is 1, the water pump start-stop monitoring state output of the ultrafiltration device is 0, the start-stop monitoring state output of the ultrafiltration device is 1, and the mathematical model continuous abnormal state number monitoring state output is 1, the ultrafiltration device is in an overload operation state, and early warning information is issued.

[0038] The application also provides an ultrafiltration device operation fault early warning system, comprising:

[0039] a history data storage module, a mathematical model training module, a data acquisition and processing module, an operation state diagnosis module, and an early warning information analysis module;

[0040] The history data storage module is used to summarize cases of serious pollution blocking and flow exceeding of the ultrafiltration device, collect and store historical operation data, and propose corresponding solutions and treatment measures for different types of operation faults;

[0041] The data acquisition and processing module is used to collect real-time operation data in the ultrafiltration device according to relevant parameters and perform relevant preprocessing measures;

[0042] The mathematical model training module is used to construct a mathematical model of the ultrafiltration device based on a One-Class-SVM classification algorithm, comprising:

[0043] The historical operation data of the ultrafiltration device in the history data storage module are screened and relevant preprocessing measures are performed;

[0044] The training data set and the test data set of the ultrafiltration device are divided;

[0045] The support vector machine classification model parameters are set, and the mathematical model is trained;

[0046] The accuracy of the mathematical model of the ultrafiltration device is verified;

[0047] The mathematical model of the ultrafiltration device is evaluated and adjusted;

[0048] The operation state diagnosis module is used to input the real-time operation data collected in the data acquisition and processing module into the trained mathematical model of the ultrafiltration device in the mathematical model training module, output the model calculation result, and judge whether the operation state of the ultrafiltration device is abnormal;

[0049] The early warning information analysis module is used for constructing a mechanism model of the ultrafiltration device and acquiring mechanism data, combining an output switch value sequence of the ultrafiltration device mathematical model to perform early warning information analysis.

[0050] The advantage of the present application is that the present application adopts the One-Class-SVM ultrafiltration device mathematical model in the unsupervised learning mode to judge the abnormal situation of the running state of the ultrafiltration device, acquires mechanism data by constructing the mechanism model of the ultrafiltration device, generates a switch value sequence, and combines the mathematical model to perform early warning analysis on the ultrafiltration device, so that the abnormal information is pushed to the relevant personnel in the initial stage of serious pollution blocking or flow overrun of the ultrafiltration device, time is saved for the relevant personnel to timely adjust the system operation condition, the system operation efficiency is improved, the normal operation of the boiler feed water system is avoided to be affected, and the service life of the fiber assembly of the ultrafiltration device is improved. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 is a flow chart of an ultrafiltration device running fault early warning method and system according to an embodiment of the present application;

[0052] Figure 2 is a flow chart of an ultrafiltration device mathematical model according to an embodiment of the present application;

[0053] Figure 3 is a test set confusion matrix diagram according to an embodiment of the present application;

[0054] Figure 4 is a flow chart of an ultrafiltration device mechanism model according to an embodiment of the present application. DETAILED DESCRIPTION

[0055] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme of the embodiments of the present application will be described clearly and completely below in combination with the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0056] The technical scheme of the present application will be further described below in combination with the drawings in the specification and specific embodiments:

[0057] Embodiment one

[0058] As shown in Figure 1 , specifically, an ultrafiltration device running fault early warning method is disclosed, comprising,

[0059] Step 1, summarize the cases of serious pollution blocking and flow overrun of the ultrafiltration device, collect and store historical operation data, and propose corresponding solutions and treatment measures for different types of running faults;

[0060] In the embodiment, the related faults and reasons of the ultrafiltration device of the boiler make-up water of a certain power plant are summarized; through the summary of the cases of serious pollution blocking and flow exceeding limit of the ultrafiltration device, the parameter set with the highest correlation is analyzed as the related parameters, and the historical operation data of the related parameters are collected and stored; the corresponding solutions and the best treatment measures are proposed for different fault types in the operation fault cases of the ultrafiltration device.

[0061] Step 2, collecting the real-time operation data of the ultrafiltration device according to the related parameters and performing related pretreatment measures;

[0062] In the embodiment, the related parameters include the inlet flow of the ultrafiltration device, the pressure difference between the inlet and outlet of the ultrafiltration device, the frequency feedback of the feed water pump of the ultrafiltration device, and the real-time operation state data of the ultrafiltration device selected include the inlet flow of the A ultrafiltration device, the pressure difference between the inlet and outlet of the A ultrafiltration device, the inlet flow of the B ultrafiltration device, the pressure difference between the inlet and outlet of the B ultrafiltration device, the frequency feedback of the feed water pump of the A ultrafiltration device, and the frequency feedback of the feed water pump of the B ultrafiltration device.

[0063] Among them, the pretreatment methods mainly include working condition division and principal component analysis.

[0064] Step 3, constructing the ultrafiltration device mathematical model based on the One-Class-SVM classification algorithm;

[0065] (1) screening the historical operation data of the ultrafiltration device in step 1 and performing related pretreatment measures;

[0066] Figure 2 The flow chart of the ultrafiltration device mathematical model of the embodiment, in the embodiment, through extracting one year of actual operation records of part of the measuring points in the make-up water system of a certain power plant, the actual operation data of the ultrafiltration device with serious pollution blocking and flow exceeding limit are selected as the historical operation data.

[0067] (2) dividing the training data set and the test data set of the ultrafiltration device;

[0068] In the embodiment, the selected historical operation data is selected, the average value of the data every twenty seconds during operation is calculated as the modeling data set, and the training data set and the test data set of the ultrafiltration device mathematical model are divided, wherein the training data set is used to train the One-Class-SVM ultrafiltration device mathematical model, and the test data set is used to evaluate and adjust the One-Class-SVM ultrafiltration device mathematical model;

[0069] In the embodiment, a total of 85457 ultrafiltration device samples (a group of data and a label is one sample) are generated, wherein the number of samples of the training data set of the One-Class-SVM ultrafiltration device mathematical model is 64093, and the number of samples of the test data set is 21364.

[0070] (3) Set the One-Class-SVM ultrafiltration device mathematical model parameters, and train the mathematical model;

[0071] Through analysis of the divided ultrafiltration device training set data, one or more ultrafiltration device operation fault early warning mathematical models are established; different algorithms or multiple algorithms can be used for data processing, such as one or a combination of multiple algorithms such as support vector machine algorithm, DBSCN algorithm, and random forest algorithm;

[0072] The mathematical logic of support vector machine is to map vectors to a higher-dimensional space, establish a maximum interval decision hyperplane, and build two mutually parallel hyperplanes on both sides of the hyperplane separating the data. By establishing a direction suitable for separating hyperplane, the distance between the two parallel hyperplanes is maximized. The greater the distance or gap between the parallel hyperplanes, the smaller the total error of the classifier;

[0073] In this embodiment, a One-Class-SVM classifier is used to construct an ultrafiltration device mathematical model. The logic of the One-Class-SVM classifier is to use a hypersphere, which minimizes the volume of the hypersphere, rather than using a hyperplane as in the traditional SVM algorithm. This algorithm obtains a spherical boundary around the data in the feature space, which can solve the high-dimensional problem in a large feature space. Unlike the random forest model commonly used in the art and other traditional classification regression support vector machine models based on supervised learning, the One-Class-SVM classifier uses an unsupervised learning method and does not require labeled output tags for the training set;

[0074] The One-Class-SVM classifier can require that the distance of all training data points xi to the center be strictly less than r. After solving by Lagrange duality, it is determined whether the new data point z is inside. If the distance of z to the center is less than or equal to the radius r, then it is not an abnormal point. If the distance of z to the center is outside the hypersphere, it is identified as an abnormal point.

[0075] In this embodiment, the One-Class-SVM classifier model uses a Gaussian kernel function, the proportion of abnormal value points is set to 0.005, the residual convergence condition is 0.001, and the maximum number of iterations is not limited.

[0076] (4) Verify the accuracy of the ultrafiltration device mathematical model;

[0077] The test data set is input into the trained mathematical model for verification, and the accuracy of the model is calculated. The calculation formula is as follows:

[0078]

[0079] Where accuracy is the model's prediction accuracy, and N accuracy N is the number of data samples that the model correctly predicts. test To verify the total number of samples in the dataset.

[0080] (5) Evaluate and adjust the mathematical model of the ultrafiltration device;

[0081] When (4) verifies that the accuracy of the mathematical model of the ultrafiltration device is lower than the set value, return to (3) to modify the parameters of the training model and repeat the training until the required early warning accuracy of the mathematical model of the ultrafiltration device is achieved.

[0082] In this embodiment, the accuracy of the mathematical model of the ultrafiltration device is set to 90%; when the accuracy is less than 90%, the model parameters are modified and the model is retrained.

[0083] Figure 3 This is the confusion matrix diagram of the test set in this embodiment. The ultrafiltration device test set contains two types of labels and 21,364 samples.

[0084] In the confusion matrix, 0 represents no anomalies, and 1 represents abnormal operating conditions such as contamination or overflow. Figure 3 As shown, the number of correctly predicted samples was 21,219, and the number of incorrectly predicted samples was 145. The support vector machine classification model of the ultrafiltration device achieved an accuracy of 99.32%, thus the mathematical model of the ultrafiltration device has good predictive value.

[0085] Step 4: Input the real-time operating data collected in Step 2 into the mathematical model of the ultrafiltration device trained in Step 3, output the model calculation results, and determine whether the operating status of the ultrafiltration device is abnormal.

[0086] In this embodiment, if the analysis result of the mathematical model of the ultrafiltration device on the real-time operating data is "normal state", the mathematical model output is 0; if the analysis result of the mathematical model of the ultrafiltration device on the real-time operating data is "abnormal state", the mathematical model output is 1.

[0087] Step 5: Construct a mechanism model of the ultrafiltration device and obtain mechanism data. Combine the output switch sequence of the mathematical model of the ultrafiltration device to perform early warning information analysis.

[0088] In this embodiment, due to data transmission fluctuations, the acquired real-time data has significant noise, which may lead to false alarms when input into the mathematical model of the ultrafiltration device. Therefore, it is necessary to analyze the early warning information in conjunction with the current operating status of related equipment in the ultrafiltration device. By constructing the mechanism model of the ultrafiltration device, obtaining mechanism data and generating sequential switching quantities, it is determined whether an early warning information needs to be issued.

[0089] In the embodiment, the mechanism data includes ultrafiltration device water pump start-stop monitoring data, ultrafiltration device start-stop monitoring data, and ultrafiltration device mathematical model abnormal state number.

[0090] In the embodiment, the analysis logic of the ultrafiltration device mathematical model and the mechanism model is as follows:

[0091] If the analysis result of the ultrafiltration device mathematical model on the real-time operation data is “normal state”, the mathematical model state outputs 0, and the ultrafiltration device mechanism model is not enabled; if the analysis result of the ultrafiltration device mathematical model on the real-time operation data is “abnormal state”, the mathematical model state outputs 1.

[0092] Ultrafiltration device water pump start-stop monitoring: monitoring the ultrafiltration device water pump operation signal, when the operation signal has no change, the ultrafiltration device water pump start-stop monitoring state outputs 0; when the operation signal changes from 0 to 1 or the operation signal changes from 1 to 0, the ultrafiltration device water pump start-stop monitoring state outputs 1.

[0093] Ultrafiltration device start-stop monitoring: monitoring the ultrafiltration device water production valve operation state, when the water production valve operation state changes from 0 to 1 or the water production valve operation state changes from 1 to 0 or the water production valve keeps the operation state as 0, the ultrafiltration device start-stop monitoring state outputs 0; when the water production valve operation state is 1, the state outputs 1.

[0094] Mathematical model continuous abnormal state number monitoring: monitoring the number of continuous abnormal states of the mathematical model, when the number of continuous abnormal states of the mathematical model does not exceed 3, the mathematical model continuous abnormal state number monitoring outputs 0, when the number of continuous abnormal states of the mathematical model exceeds 3, the mathematical model continuous abnormal state number monitoring outputs 1.

[0095] Figure 4 The flow chart of the ultrafiltration device mechanism model of the embodiment is as follows: after the state output of the ultrafiltration device mathematical model, the ultrafiltration device needs to analyze the operation state by means of the mechanism model, and combined with the on-off sequence output by the mathematical model and the mechanism model, it is judged whether the early warning information needs to be sent, and the on-off sequence judgment method is as follows:

[0096] When the state output of the ultrafiltration device mathematical model is 0, the output result of the ultrafiltration device mathematical model is “normal state”, and no early warning information is sent.

[0097] When the state output of the ultrafiltration device mathematical model is 1 and the state output of the ultrafiltration device water pump start-stop monitoring is 1, no early warning information is sent.

[0098] When the state output of the ultrafiltration device mathematical model is 1, the state output of the ultrafiltration device water pump start-stop monitoring is 0, and the state output of the ultrafiltration device start-stop monitoring is 0, no early warning information is sent.

[0099] When the mathematical model state output of the ultrafiltration device is 1, the start-stop monitoring state output of the water supply pump of the ultrafiltration device is 0, the start-stop monitoring state output of the ultrafiltration device is 1, and the mathematical model continuous abnormal state number monitoring state output is 0, no early warning information is issued.

[0100] When the mathematical model state output of the ultrafiltration device is 1, the start-stop monitoring state output of the water supply pump of the ultrafiltration device is 0, the start-stop monitoring state output of the ultrafiltration device is 1, and the mathematical model continuous abnormal state number monitoring state output is 1, the ultrafiltration device is in an overload operation state, early warning information is issued to remind personnel to adjust the operation parameters as soon as possible.

[0101] When the ultrafiltration device is seriously clogged or the flow is out of limit, according to the collected real-time operation data, combined with the on-off sequence composed of the mathematical model and the mechanism model of the ultrafiltration device, it is judged whether the ultrafiltration device needs to issue early warning information, and the early warning information and adjustment measures are timely pushed to the relevant personnel.

[0102] Embodiment two

[0103] The embodiment discloses an ultrafiltration device operation fault early warning system, comprising: a historical data storage module, a data acquisition and processing module, a mathematical model construction module, an operation state diagnosis module, and an early warning information analysis module.

[0104] The historical data storage module is used to summarize cases of serious clogging and flow out of limit of the ultrafiltration device, collect and store historical operation data, and propose corresponding solutions and treatment measures for different types of operation faults.

[0105] In the embodiment, the related faults and reasons of the ultrafiltration device of the boiler of a certain thermal power plant are summarized; by summarizing the cases of serious clogging and flow out of limit of the ultrafiltration device, the parameter set with the highest correlation is analyzed as the relevant parameters, and the historical operation data of the relevant parameters are collected and stored; corresponding solutions and best treatment measures are proposed for different fault types in the ultrafiltration device operation fault cases.

[0106] The data acquisition and processing module is used to collect real-time operation data in the ultrafiltration device according to the relevant parameters and perform relevant preprocessing measures.

[0107] In the embodiment, the relevant parameters include the inlet flow of the ultrafiltration device, the inlet and outlet pressure difference of the ultrafiltration device, and the water supply pump frequency feedback of the ultrafiltration device, and the real-time operation state data of the ultrafiltration device selected include the inlet flow of the A ultrafiltration device, the inlet and outlet pressure difference of the A ultrafiltration device, the inlet flow of the B ultrafiltration device, the inlet and outlet pressure difference of the B ultrafiltration device, the water supply pump frequency feedback of the A ultrafiltration device, and the water supply pump frequency feedback of the B ultrafiltration device.

[0108] The preprocessing methods mainly include working condition division and principal component analysis.

[0109] The mathematical model training module is used to construct a mathematical model of the ultrafiltration device based on a One-Class-SVM classification algorithm;

[0110] The historical operation data of the ultrafiltration device in the screening historical data storage module are preprocessed;

[0111] Figure 2 For the mathematical model flowchart of the ultrafiltration device of the embodiment, in the embodiment, the actual operation records of some measuring points in a make-up water system of a power plant in one year are extracted, and the actual operation data of the ultrafiltration device in which serious pollution blocking and flow exceeding occur are screened out as the historical operation data.

[0112] The training data set and the test data set of the ultrafiltration device are divided;

[0113] In the embodiment, the historical operation data screened out are selected, the average value of the data every twenty seconds during operation is calculated as the modeling data set, and the training data set and the test data set of the mathematical model of the ultrafiltration device are divided, wherein the training data set is used to train the One-Class-SVM mathematical model of the ultrafiltration device, and the test data set is used to evaluate and adjust the One-Class-SVM mathematical model of the ultrafiltration device;

[0114] In the embodiment, a total of 85457 samples (one group of data and one label are one sample) of the ultrafiltration device are generated, wherein the number of samples of the training data set of the One-Class-SVM mathematical model of the ultrafiltration device is 64093, and the number of samples of the test data set is 21364.

[0115] The parameters of the support vector machine classification model are set, and the mathematical model is trained;

[0116] By analyzing the divided training set data of the ultrafiltration device, one or more early warning mathematical models of the operation fault of the ultrafiltration device are established; different algorithms or multiple algorithms can be used for data processing, for example, one or more of the support vector machine algorithm, the DBSCN algorithm, the random forest algorithm and the like.

[0117] The mathematical logic of the support vector machine is to map the vector to a higher-dimensional space, establish a maximum interval decision hyperplane, build two mutually parallel hyperplanes on both sides of the hyperplane separating the data, maximize the distance between the two parallel hyperplanes by establishing a direction suitable separating hyperplane, and the greater the distance or gap between the parallel hyperplanes, the smaller the total error of the classifier;

[0118] In the embodiment, a One-Class-SVM classifier is used to construct a mathematical model of the ultrafiltration device; the logic of the One-Class-SVM classifier is to use a hypersphere, and the volume of the hypersphere is expected to be minimized, instead of using a hyperplane to divide, as in the traditional SVM algorithm; the algorithm obtains a spherical boundary around the data in the feature space, which can solve the high-dimensional problem in a large feature space; compared with the random forest model commonly used in the art and other traditional classification regression support vector machine models based on supervised learning, the One-Class-SVM classifier uses an unsupervised learning method and does not need to mark the output label of the training set;

[0119] The One-Class-SVM classifier can require that the distance from all training data points xi to the center be strictly less than r. After solving by Lagrange dual, it is judged whether a new data point z is inside. If the distance from z to the center is less than or equal to the radius r, then z is not an abnormal point. If the distance from z to the center is outside the hypersphere, then z is identified as an abnormal point.

[0120] In the embodiment, the One-Class-SVM classifier model uses a Gaussian kernel function, the proportion of abnormal value points is set to 0.005, the residual convergence condition is 0.001, and the maximum number of iterations is not limited.

[0121] Verify the accuracy of the mathematical model of the ultrafiltration device;

[0122] The test data set is input into the trained mathematical model for verification, and the accuracy of the model is calculated. The calculation formula is as follows:

[0123]

[0124] Wherein, accuracy is the prediction accuracy of the model, N accuracy is the number of data samples predicted correctly by the model, N test is the total number of samples in the verification data set.

[0125] Evaluate and adjust the mathematical model of the ultrafiltration device;

[0126] When the accuracy of the mathematical model of the ultrafiltration device is lower than the set value, the parameters of the training model in the mathematical model of the ultrafiltration device are modified and retrained until the required warning accuracy of the mathematical model of the ultrafiltration device is reached.

[0127] In the embodiment, the accuracy of the mathematical model of the ultrafiltration device is set to 90%; when the accuracy is less than 90%, the model parameters are modified and retrained;

[0128] Figure 3For the test set of this embodiment Confusion Matrix chart, the test set of the ultrafiltration device contains 2 kinds of label categories, and the number of samples is 21364;

[0129] 0 in the confusion matrix represents no abnormality, and 1 represents abnormal working conditions such as fouling or super flow, as shown in the table. Figure 3 The number of samples predicted correctly is 21219, and the number of samples predicted incorrectly is 145. The test accuracy of the ultrafiltration device support vector machine classification model can reach 99.32%, so the mathematical model of the ultrafiltration device has good prediction value.

[0130] The running state diagnosis module is used to input the real-time running data collected by the data acquisition and processing module into the ultrafiltration device mathematical model trained by the mathematical model training module, and output the model calculation result to determine whether the running state of the ultrafiltration device is abnormal.

[0131] In this embodiment, if the analysis result of the ultrafiltration device mathematical model on the real-time running data is "normal state", the mathematical model state outputs 0, and if the analysis result of the ultrafiltration device mathematical model on the real-time running data is "abnormal state", the mathematical model state outputs 1.

[0132] The early warning information analysis module is used to build an ultrafiltration device mechanism model and obtain mechanism data, combine the on-off quantity sequence output by the ultrafiltration device mathematical model, and analyze the early warning information.

[0133] In this embodiment, due to data transmission fluctuations, the real-time data collected has large noise, and there is a possibility of false positives when transmitted into the ultrafiltration device mathematical model, so it is necessary to combine the running state of the related equipment in the current ultrafiltration device to analyze the early warning information, build an ultrafiltration device mechanism model, obtain mechanism data and generate sequence on-off quantity, and determine whether to issue early warning information.

[0134] In this embodiment, the mechanism data includes ultrafiltration device feed pump start-stop monitoring data, ultrafiltration device start-stop monitoring data, and ultrafiltration device mathematical model abnormal state times.

[0135] In this embodiment, the analysis logic of the ultrafiltration device mathematical model and the mechanism model is as follows:

[0136] If the analysis result of the ultrafiltration device mathematical model on the real-time running data is "normal state", the mathematical model state outputs 0, and the ultrafiltration device mechanism model is not enabled; if the analysis result of the ultrafiltration device mathematical model on the real-time running data is "abnormal state", the mathematical model state outputs 1.

[0137] The ultrafiltration device water pump start-stop monitoring monitors the running signal of the ultrafiltration device water pump. When the running signal has no change, the ultrafiltration device water pump start-stop monitoring state outputs 0. When the running signal changes from 0 to 1 or the running signal changes from 1 to 0, the ultrafiltration device water pump start-stop monitoring state outputs 1.

[0138] The ultrafiltration device start-stop monitoring monitors the running state of the water production valve of the ultrafiltration device. When the running state of the water production valve changes from 0 to 1 or the running state of the water production valve changes from 1 to 0 or the running state of the water production valve remains 0, the ultrafiltration device start-stop monitoring state outputs 0. When the running state of the water production valve is 1, the state outputs 1.

[0139] The mathematical model continuous abnormal state number monitoring monitors the number of abnormal states of the mathematical model. When the number of continuous abnormal states of the mathematical model does not exceed 3, the mathematical model continuous abnormal state number monitoring outputs 0. When the number of continuous abnormal states of the mathematical model exceeds 3, the mathematical model continuous abnormal state number monitoring outputs 1.

[0140] Figure 4 For the flow chart of the mechanism model of the ultrafiltration device of the present embodiment, after the state output of the mathematical model of the ultrafiltration device, the ultrafiltration device needs to analyze the running state by means of the mechanism model, and combines the on-off quantity sequence output by the mathematical model and the mechanism model to judge whether the early warning information needs to be sent. The on-off quantity sequence judgment method is as follows:

[0141] When the state output of the mathematical model of the ultrafiltration device is 0, the output result of the mathematical model of the ultrafiltration device is “normal state”, and no early warning information is sent.

[0142] When the state output of the mathematical model of the ultrafiltration device is 1 and the state output of the ultrafiltration device water pump start-stop monitoring is 1, no early warning information is sent.

[0143] When the state output of the mathematical model of the ultrafiltration device is 1, the state output of the ultrafiltration device water pump start-stop monitoring is 0, and the state output of the ultrafiltration device start-stop monitoring is 0, no early warning information is sent.

[0144] When the state output of the mathematical model of the ultrafiltration device is 1, the state output of the ultrafiltration device water pump start-stop monitoring is 0, the state output of the ultrafiltration device start-stop monitoring is 1, and the state output of the mathematical model continuous abnormal state number monitoring is 0, no early warning information is sent.

[0145] When the state output of the mathematical model of the ultrafiltration device is 1, the state output of the ultrafiltration device water pump start-stop monitoring is 0, the state output of the ultrafiltration device start-stop monitoring is 1, and the state output of the mathematical model continuous abnormal state number monitoring is 1, the ultrafiltration device is in an overload running state, early warning information is sent, and personnel are reminded to adjust the running parameters as soon as possible.

[0146] When the ultrafiltration device is seriously polluted or the flow exceeds the limit, according to the collected real-time operation data, combined with the on-off sequence composed of the mathematical model and the mechanism model of the ultrafiltration device, it is judged whether the ultrafiltration device needs to send a warning information, and the warning information and adjustment measures are timely pushed to the relevant personnel.

[0147] The above examples are only used to illustrate the technical solutions of the present application, but not to limit it; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that the technical solutions recorded in the foregoing examples can be modified, or some technical features can be replaced by equivalents; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for predicting an operation failure of an ultrafiltration device, characterized by, The method comprises: Step 1, summarize the cases of serious fouling and flow exceeding of the ultrafiltration device, collect and store historical operation data, and propose corresponding solutions and treatment measures for different types of operation faults; Step 2, collect real-time operation data of the ultrafiltration device according to relevant parameters, and perform relevant pretreatment measures; Step 3, construct a mathematical model of the ultrafiltration device based on the One-Class-SVM classification algorithm, including: (1) screening the historical operation data of the ultrafiltration device in step 1, and performing relevant pretreatment measures; (2) dividing the training data set and the test data set of the ultrafiltration device; (3) setting the support vector machine classification model parameters and training the mathematical model; (4) verifying the accuracy of the mathematical model of the ultrafiltration device; (5) evaluating and adjusting the mathematical model of the ultrafiltration device; Step 4, input the real-time operation data collected in step 2 into the trained mathematical model of the ultrafiltration device in step 3, output the model calculation results, and judge whether the operation state of the ultrafiltration device is abnormal; Step 5, construct a mechanism model of the ultrafiltration device and obtain mechanism data, combine the on-off quantity sequence output by the mathematical model of the ultrafiltration device, and perform early warning information analysis; The relevant parameters in step 2 include the inlet flow of the ultrafiltration device, the pressure difference between the inlet and outlet of the ultrafiltration device, and the frequency feedback of the water supply pump of the ultrafiltration device; The pretreatment measures in step 2 include working condition division and principal component analysis; The method for dividing the training data set and the test data set of the ultrafiltration device in step 3 (2) is as follows: According to the historical operation data selected in (1), the average value of the data at a certain time interval during operation is calculated as the modeling data set, and the training data set and the test data set of the mathematical model of the ultrafiltration device are divided, wherein the training data set is used to train the One-Class-SVM mathematical model of the ultrafiltration device, and the test data set is used to evaluate and adjust the One-Class-SVM mathematical model of the ultrafiltration device.

2. The method according to claim 1, wherein The calculation formula for verifying the accuracy of the mathematical model of the ultrafiltration device in step 3 (4) is as follows: wherein, is the accuracy of the model prediction, is the number of data samples that the model predicted correctly, is the total number of samples of the validation dataset.

3. The method according to claim 2, wherein The method for evaluating and adjusting the mathematical model of the ultrafiltration device in step 4 (5) includes: when the accuracy of the mathematical model of the ultrafiltration device in (4) is lower than the set value, return to (3) to modify the parameters of the training model for repeated training until the required early warning accuracy of the mathematical model of the ultrafiltration device is reached.

4. The method according to claim 3, wherein The mechanism data in step 5 include the start-stop monitoring data of the water supply pump of the ultrafiltration device, the start-stop monitoring data of the ultrafiltration device, and the number of abnormal states of the mathematical model.

5. The method of claim 4, wherein the step of determining the operating state of the ultrafiltration device comprises: determining whether the ultrafiltration device is operating in a normal state, a low flow state, or a high flow state. The method for outputting the on-off quantity sequence in step 5 is as follows: a. If the analysis result of the real-time operation data by the mathematical model of the ultrafiltration device is "normal state", the mathematical model state outputs 0, and the mechanism model of the ultrafiltration device is not enabled; if the analysis result of the real-time operation data by the mathematical model of the ultrafiltration device is "abnormal state", the mathematical model state outputs 1; b. If the start-stop monitoring operation signal of the water supply pump of the ultrafiltration device has no change, the start-stop monitoring state of the water supply pump of the ultrafiltration device outputs 0; if the start-stop monitoring operation signal of the water supply pump of the ultrafiltration device changes from 0 to 1 or the operation signal changes from 1 to 0, the start-stop monitoring state of the water supply pump of the ultrafiltration device outputs 1; c. If the water production valve operating state of the ultrafiltration device changes from 0 to 1 or the water production valve operating state changes from 1 to 0 or the water production valve operating state remains 0, the ultrafiltration device start-stop monitoring state outputs 0; if the water production valve operating state of the ultrafiltration device is 1, the ultrafiltration device start-stop monitoring state outputs 1; d. If the number of continuous abnormal states of the mathematical model is not more than 3 times, the number of continuous abnormal states of the mathematical model monitoring output is 0; if the number of continuous abnormal states of the mathematical model is more than 3 times, the number of continuous abnormal states of the mathematical model monitoring output is 1.

6. The method for early warning of operational faults in an ultrafiltration device according to claim 5, characterized in that, The method for analyzing the early warning information in step 5 is as follows: a. When the ultrafiltration device mathematical model state output is 0, no early warning information is sent out; b. When the ultrafiltration device mathematical model state output is 1 and the ultrafiltration device water pump start-stop monitoring state output is 1, no early warning information is sent out; c. When the ultrafiltration device mathematical model state output is 1, the ultrafiltration device water pump start-stop monitoring state output is 0, and the ultrafiltration device start-stop monitoring state output is 0, no early warning information is sent out; d. When the ultrafiltration device mathematical model state output is 1, the ultrafiltration device water pump start-stop monitoring state output is 0, the ultrafiltration device start-stop monitoring state output is 1, and the number of continuous abnormal states of the mathematical model monitoring state output is 0, no early warning information is sent out; e. When the ultrafiltration device mathematical model state output is 1, the ultrafiltration device water pump start-stop monitoring state output is 0, the ultrafiltration device start-stop monitoring state output is 1, and the number of continuous abnormal states of the mathematical model monitoring state output is 1, the ultrafiltration device is in an overload operating state, and early warning information is sent out.

7. An ultrafiltration device operational failure early warning system, comprising: It comprises: a historical data storage module, a mathematical model training module, a data acquisition and processing module, an operating state diagnosis module, and an early warning information analysis module; The historical data storage module is used to summarize cases of serious fouling and flow exceeding of the ultrafiltration device, collect and store historical operating data, and propose corresponding solutions and treatment measures for different types of operating faults; The data acquisition and processing module is used to collect real-time operating data in the ultrafiltration device according to relevant parameters and perform relevant preprocessing measures; The mathematical model training module is used to construct an ultrafiltration device mathematical model based on a One-Class-SVM classification algorithm, comprising: screening the historical operating data of the ultrafiltration device in the historical data storage module and performing relevant preprocessing measures; dividing the training data set and the test data set of the ultrafiltration device; setting the support vector machine classification model parameters and training the mathematical model; verifying the accuracy of the ultrafiltration device mathematical model; evaluating and adjusting the ultrafiltration device mathematical model; The operating state diagnosis module is used to input the real-time operating data collected by the data acquisition and processing module into the trained ultrafiltration device mathematical model in the mathematical model training module, output the model calculation results, and determine whether the operating state of the ultrafiltration device is abnormal; The early warning information analysis module is used to construct an ultrafiltration device mechanism model, obtain mechanism data, combine the on-off quantity sequence output by the ultrafiltration device mathematical model, and perform early warning information analysis; The relevant parameters include the inlet flow of the ultrafiltration device, the inlet-outlet pressure difference of the ultrafiltration device, and the frequency feedback of the water pump of the ultrafiltration device; The preprocessing measures include working condition division and principal component analysis; The method for dividing the training and testing datasets for the ultrafiltration device is as follows: Based on the selected historical operating data, the average value of the data at certain intervals during operation is calculated as the modeling dataset. The training dataset and test dataset of the mathematical model of the ultrafiltration device are divided. The training dataset is used to train the One-Class-SVM mathematical model of the ultrafiltration device, and the test dataset is used to evaluate and adjust the One-Class-SVM mathematical model of the ultrafiltration device.

Images