Intelligent diagnosis method for process industry faults

Abstract

The application discloses a kind of process industry fault intelligent diagnosis methods, it is related to process industry technical field, solve the problem that alarm information is not accurate in prior art, and cannot early on the working condition is pre-judged, the method includes: critical heat flux density data is input to thermocouple temperature variation model, determine data set;Data set is normalized;According to normalized data set, calculate the set of femto feature vectors, and the set of femto feature vectors is input to the SVM model trained, output the position relationship of femto feature vector and two-dimensional hyperplane, determine femto data point and non-femto data point;Whether femto data point is false femto data point is judged, if it is false femto data point, then femto data point is placed as non-femto data point;According to femto data point, determine critical working condition point, and output critical working condition point, realize that system is slightly changed, flexibly and conveniently judge critical information, it is easy to operate, reduce human resource cost.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of process industry, and particularly relates to a process industry fault intelligent diagnosis method. BACKGROUND

[0002] The intelligent measurement and control system is widely applied to various industrial processes, fully utilizes sensor technology, computer technology, intelligent control technology, information technology and the like, makes the control system of the industrial process more intelligent, and further improves the automation degree of the industrial control system. The health detection of the system is a key technology for guaranteeing the safe and reliable operation of the intelligent measurement and control system in the industrial production process. The real-time health state in the production process is monitored, the fault equipment can be found and isolated early, the damage of the industrial production system caused by the chain reaction of the fault is avoided, and even the paralysis of the entire industrial process is avoided.

[0003] However, in the actual industrial process, the same fault often causes multiple alarms, so that the alarm system has a large amount of redundant information while generating effective alarm information, which interferes with the system diagnosis of the fault type, and cannot predict the working condition in advance, so that the alarm system loses accuracy and timeliness. SUMMARY

[0004] The present application provides a process industry fault intelligent diagnosis method, which solves the problem of inaccurate alarm information and inability to predict the working condition in advance in the prior art, realizes flexible and convenient judgment of critical information with only a small modification of the system, is simple to operate, reduces the labor cost, improves the process monitoring efficiency, can quickly warn the abnormal condition in the industrial process, and reduces the workload of the operator.

[0005] The present application provides a process industry fault intelligent diagnosis method, which includes:

[0006] The critical heat flux density data is input into a thermocouple temperature change model to determine a data set;

[0007] The data set is normalized to determine a normalized data set;

[0008] A femto feature vector set is calculated according to the normalized data set, and the femto feature vector set is input into a trained SVM model to output the positional relationship between the data points corresponding to the femto feature vector and a two-dimensional hyperplane, and to determine femto data points and non-femto data points; wherein the femto feature vector set includes multi-order femto feature vectors;

[0009] It is judged whether the femto data point is a false femto data point, and if it is the false femto data point, the femto data point is set as the non-femto data point;

[0010] determine a critical operating point according to the femto data point, and output the critical operating point.

[0011] In a possible implementation, the SVM model is trained, including: obtaining an objective function and a constraint condition; determining a minimum value of the objective function according to the objective function and the constraint condition, so as to determine a two-dimensional hyperplane.

[0012] In a possible implementation, the thermocouple temperature change model is expressed as:

[0013]

[0014] wherein T1 and T2 represent thermocouple inlet and outlet temperatures, units are ℃; D1 represents thermocouple inlet and outlet flow rates, units are m / s; D2 represents thermocouple outlet flow rate, units are m / s; T represents the temperature of the fluid in the thermocouple, units are ℃; and C represents the specific heat capacity of the fluid, units are J / (kg·℃). M j represents the mass of the superheated metal, units are kg; Q1 represents the total heating amount of the body, units are kJ; Q represents the total heat transfer amount in the body, units are kJ; n represents an empirical coefficient; and V represents the volume of the body, units are m 3 ; K represents a heat exchange coefficient, dimensionless; T j represents the temperature of the pipe wall in the body, units are ℃; and C j represents the specific heat of the superheater metal, units are J / (kg·℃). C w represents the specific heat capacity of the fluid, units are J / (kg·℃). ρ2 represents the density of the thermocouple outlet fluid, units are kg / m

[0015] In a possible implementation, the calculation formula of the femto feature vector is expressed as:

[0016]

[0017] wherein X i represents an i-order femto feature vector; T1 represents the first normalized body wall temperature data; T i+1 represents the i+1-order normalized body wall temperature data; t1 represents the first sampling time, units are s; t i+1 represents the i+1-order sampling time, units are s, and n represents a calculation order.

[0018] In a possible implementation, the femto feature vector set is input into the trained SVM model, a positional relationship between a data point corresponding to the femto feature vector and the two-dimensional hyperplane is output, the femto data point and the non-femto data point are determined, and the method includes the following steps.

[0019] According to a position relationship between each order attoptronic feature vector in the set of attoptronic feature vectors and the two-dimensional hyperplane, it is determined whether the data point corresponding to the attoptronic feature vector is an attoptronic data point.

[0020] If the attoptronic feature vector is distributed above the two-dimensional hyperplane, the data point corresponding to the attoptronic feature vector is classified as the attoptronic data point.

[0021] If the attoptronic feature vector is distributed below the two-dimensional hyperplane, the data point corresponding to the attoptronic feature vector is classified as the non-attoptronic data point.

[0022] If the attoptronic feature vector is distributed on the two-dimensional hyperplane, the data point corresponding to the attoptronic feature vector is classified as the attoptronic data point.

[0023] In a possible implementation, the judging whether the attoptronic data point is a false attoptronic data point and, if the attoptronic data point is the false attoptronic data point, classifying the attoptronic data point as the non-attoptronic data point comprises:

[0024] Observing a position relationship between the multiple orders of attoptronic feature vectors of data points in a continuous time after the attoptronic data point and the two-dimensional hyperplane, it is determined whether the attoptronic data point is a false attoptronic data point caused by local vaporization.

[0025] If yes, the attoptronic data point is classified as the false attoptronic data point, and the false attoptronic data point is further classified as the non-attoptronic data point.

[0026] If no, a first attoptronic data point in the continuous time is determined as the critical working condition point.

[0027] In a possible implementation, the objective function is expressed as:

[0028]

[0029] Wherein, W represents a weight vector, W*X+b=0, X represents the set of attoptronic feature vectors, and b represents a bias term.

[0030] In a possible implementation, the constraint condition is a Lagrange multiplier.

[0031] In a possible implementation, the determining the minimum value of the objective function according to the objective function and the constraint condition, thereby obtaining a two-dimensional hyperplane comprises:

[0032] Determining an edge length of the objective function;

[0033] According to the constraint condition, the edge length, and the objective function, an optimization objective function is determined.

[0034] By acquiring the maximum value of the optimization target function, the minimum value of the target function is determined, thereby determining the two-dimensional hyperplane.

[0035] In a possible implementation, the edge length is represented as:

[0036]

[0037] The optimization target function is represented as:

[0038]

[0039] Wherein, λ i , λ j Is represented as a set of Lagrange multiplier variables, y i , y j All represent the data class label of the normalized data set, y i , y j All belong to {-1,1}, when equal to +1, it is positive example; -1 is negative example; X i Is represented as the i-order hypersphere feature vector, X j Is represented as the j-order hypersphere feature vector, s represents the support vector set.

[0040] One or more technical solutions provided in the application have at least the following technical effects or advantages:

[0041] The application adopts a process industry fault intelligent diagnosis method, which comprises the following steps: inputting critical heat flux density data into a thermocouple temperature change model, determining a data set, the thermocouple has high measurement accuracy of temperature and obvious change perception, and the heat data in the industrial process is measured accurately; normalizing the data set to determine a normalized data set, the normalization maps the data set to a smaller range, which is convenient for observing and analyzing the data; calculating a femto feature vector set according to the normalized data set, inputting the femto feature vector set into a trained SVM model, outputting the position relationship between the data point corresponding to the femto feature vector and a two-dimensional hyperplane, and determining the femto data point and the non-femto data point; wherein the femto feature vector set comprises a multi-order femto feature vector, the state and type of the data are determined by judging the position relationship between the data point and the two-dimensional hyperplane, the operation is simple, and the dependence on experience value is reduced; judging whether the femto data point is a false femto data point, if it is a false femto data point, the femto data point is set as a non-femto data point, the data is monitored in a small range, and the monitoring time and monitoring human resources are reduced; determining a critical working condition point according to the femto data point and outputting the critical working condition point; the problems that the alarm information is inaccurate and the working condition cannot be predicted in advance in the prior art are solved, the critical information can be flexibly and conveniently judged only by making small changes to the system, the operation is simple, the human resource cost is reduced, the process monitoring efficiency is improved, the abnormal condition in the industrial process can be quickly warned, and the workload of the operator is reduced. BRIEF DESCRIPTION OF DRAWINGS

[0042] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the description of the embodiments of the present application or the prior art. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0043] Figure 1 A step flow chart of a process industry fault intelligent diagnosis method provided by the embodiment of the present application;

[0044] Figure 2 A schematic diagram of the position relationship between the data point corresponding to the femto feature vector and the two-dimensional hyperplane provided by the embodiment of the present application;

[0045] Figure 3 A flow chart of a specific application of the present application provided by the embodiment of the present application. DETAILED DESCRIPTION

[0046] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be clearly and completely described in the following. Obviously, the described embodiments are only a part of embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative effort should fall within the protection scope of the present application.

[0047] In modern process industry, the alarm and fault diagnosis system is an independent protection system. With the development of DCS, PLC and FCS technologies, the alarm and fault diagnosis system is gradually simplified, detailed and informatized. However, in actual industrial processes, a same fault often causes multiple alarms, so that the alarm system has a large amount of redundant information while generating effective alarms, which interferes with the system diagnosis of fault types, and causes the alarm system to lose accuracy.

[0048] The embodiment of the present application provides a process industry fault intelligent diagnosis method, as shown in the figure, the method comprises the following steps S101-S105. Figure 1

[0049] S101, input the critical heat flux density data to the thermocouple temperature change model, and determine the data set. Obtain the important node information of the critical heat flux density experiment from the intelligent measurement and control system, integrate the real-time information and historical information of the important nodes, use the multivariate regression data mining technology to analyze the correlation of the historical data, filter and store the required parameters, such as the physical parameters of the body outlet temperature, the body outlet pressure, the wall temperature maximum temperature and the like, convert these data into data quantities representing critical characteristics, such as saturation temperature, steam rate, wall temperature rising rate and the like, improve the model prediction accuracy of invisible data, and give the analysis classification result with a certain confidence probability (95%). Taking the body wall temperature as an example, according to the original data extracted from the critical heat flux density experiment, combining the relative relationship of the thermocouple arrangement position of the industrial loop, taking the thermocouple signal of the first wall temperature rising as the starting point, analyzing the wall temperature development process of other thermocouples around the thermocouple, and constructing a two-dimensional temperature change model of the thermocouple through data visualization, analyzing the boundary change process and change value of the wall temperature rising. Substitute the working condition node parameters, calculate the body wall temperature temperature change curve and the body wall temperature data set. The thermocouple temperature change model is represented as:

[0050]

[0051] Wherein, T1 and T2 represent the thermocouple inlet and outlet temperatures, and the unit is ℃; D1 represents the flow rate of the thermocouple inlet and outlet, and the unit is m / s; D2 represents the flow rate of the thermocouple inlet and outlet, and the unit is m / s; T represents the wall temperature of the thermocouple, and the unit is ℃; t represents the time, and the unit is s; T1, T2 and T represent the temperature of the thermocouple inlet, outlet and wall, respectively; D1 and D2 represent the flow rate of the thermocouple inlet and outlet, respectively; and t represents the time. M j ​represents the superheated metal mass, in kg; Q1 represents the total heating amount of the body, in kJ; Q represents the total heat transfer amount in the body, in kJ; n represents an empirical coefficient; V represents the volume of the body, in m 3 ; K represents a heat exchange coefficient, dimensionless; T j represents the temperature of the wall of the body, in ℃; C j represents the specific heat of the superheater metal, in C w represents the specific heat capacity of the fluid, in ρ2 represents the density of the fluid at the outlet of the thermocouple, in

[0052] S102, normalizing the data set to determine a normalized data set. The normalization processing maps all data in a uniform scale in the order of sampling time.

[0053] S103, calculating a hyperplane feature vector set according to the normalized data set, and inputting the hyperplane feature vector set into a trained SVM model to output the positional relationship between the data points corresponding to the hyperplane feature vector and a two-dimensional hyperplane, and determining hyperplane data points and non-hyperplane data points; wherein the hyperplane feature vector set includes multi-order hyperplane feature vectors. In one specific embodiment provided by the application, a first-order hyperplane feature vector and a second-order hyperplane feature vector are calculated according to the calculation formula of the hyperplane feature vector. The calculation formula of the hyperplane feature vector is represented as:

[0054]

[0055] wherein X i represents an i-order hyperplane feature vector; T1 represents the first normalized body wall temperature data; T i+1 represents the i+1th normalized body wall temperature data; t1 represents the first sampling time, in s; t i+1 represents the i+1th sampling time, in s, and n represents the calculation order.

[0056] The SVM model is trained, including: obtaining an objective function and a constraint condition; determining the minimum value of the objective function according to the objective function and the constraint condition, so as to determine a two-dimensional hyperplane.

[0057] Data division based on the SVM classification algorithm: the optimal division linear equation is obtained by solving the SVM classification algorithm, and the filtered data in the data set is divided into hyperplane data points and non-hyperplane data points according to the temperature value and the hyperplane feature vector set.

[0058] The hyperplane model of the SVM classification algorithm is constructed, and the mathematical expression of the hyperplane is:

[0059] W*X+b=0

[0060] where X represents the feature vector of the normalized dataset, W represents the weight vector, and b represents the bias term.

[0061] In the SVM algorithm, the goal is to find the maximum two-dimensional hyperplane that optimizes the distance function d, which is equivalent to finding the minimized objective function. The objective function is represented as:

[0062]

[0063] where W represents the weight vector, X represents the set of feature vectors in femto, and b represents the bias term.

[0064] The edge length is represented as: Adding the constraint condition as the Lagrange multiplier, the objective function is rewritten as:

[0065]

[0066] where N represents the size of the normalized dataset, λ i represents the Lagrange multiplier, y i represents the data class label of the normalized dataset, i.e.:

[0067]

[0068]

[0069] The support vector minimization Lagrange function is brought into the rewritten objective function, and the optimized objective function is represented as:

[0070]

[0071] where λ i , λ j represent a set of Lagrange multiplier variables, y i , y j represent the data class label of the normalized dataset, y i , y j all belong to {-1,1}, where +1 is the positive example; -1 is the negative example; X i represents the i-th feature vector in femto, X j represents the j-th feature vector in femto, and s represents the support vector set.

[0072] When λ i = 0, λ i has no effect on the result, where the constraint condition is:

[0073] The SMO algorithm is used to solve the optimization, and a pair of variables λi j , fix other parameters; initialize all parameters to 0, loop the following steps until convergence: select the variable to be updated λ i j , fix parameters except λ i j , update λ D i j ; find all support vector sets, update the values of Wu and b, and obtain the two-dimensional hyperplane. The specific steps for obtaining the two-dimensional hyperplane are as follows:

[0074] (1) Determine the edge length of the objective function.

[0075] (2) According to the constraint condition, the edge length and the objective function, determine the optimization objective function.

[0076] (3) Determine the minimum value of the objective function by obtaining the maximum value of the optimization objective function, so as to determine the two-dimensional hyperplane. According to the above formula, it can be obtained that finding the maximum value of the dual function is equivalent to finding the maximum value of the distance function, that is, finding the minimum value of the objective function.

[0077] According to the position relationship between each order of the flying-up characteristic vector and the two-dimensional hyperplane in the flying-up characteristic vector set, as shown in Figure 2 , determine whether the data point corresponding to the flying-up characteristic vector is a flying-up data point, including:

[0078] (1) If the flying-up characteristic vector is distributed above the two-dimensional hyperplane, the data point corresponding to the flying-up characteristic vector is classified as a flying-up data point.

[0079] (2) If the flying-up characteristic vector is distributed below the two-dimensional hyperplane, the data point corresponding to the flying-up characteristic vector is classified as a non-flying-up data point.

[0080] (3) If the flying-up characteristic vector is distributed on the two-dimensional hyperplane, the data point corresponding to the flying-up characteristic vector is classified as a flying-up data point.

[0081] In Figure 2 , the circular data points above the plane determined by the asterisk are flying-up data points, the circular data points below the plane determined by the asterisk are non-flying-up data points, and the circular data points on the plane determined by the asterisk are flying-up data points.

[0082] S104, determine whether the flying-up data point is a false flying-up data point, if it is a false flying-up data point, then the flying-up data point is set as a non-flying-up data point.

[0083] ​​​​​The vapor fraction has a great influence on the wall temperature of the body in the critical heat flux experiment. If the problem of local boiling vaporization occurs in the experimental loop, the wall temperature of a certain monitoring point of the body will suddenly rise, and false lift-off of the monitoring point may occur, which affects the judgment of lift-off data. According to the final classification result of the SVM model, the lift-off point and the non-lift-off point are determined, and the data lift-off monitoring method is used to judge whether the working condition point appears in the lift-off stage according to the experimental body wall temperature and the multi-scale lift-off rate. That is, when a certain monitoring point appears through the SVM model output result as a lift-off point, the lift-off monitoring is also performed for other monitoring points near the monitoring point. If only the monitoring point and several monitoring points (less than 10) around it appear in the lift-off, but the other monitoring points do not appear in the lift-off, it is a false lift-off condition; if the monitoring point and the remaining monitoring points around it all appear in the lift-off point (more than 10), it is determined that the group of lift-off points is determined as the lift-off stage. In the multi-scale lift-off rate data in the lift-off stage, if more than 95% of the lift-off points are judged to belong to the lift-off stage point, the first data point classified as the lift-off stage point is taken as the critical point of the current experimental working condition, which specifically includes:

[0084] (1) The position relationship between the data points corresponding to the multi-order lift-off feature vectors of the data points in the continuous time after observing the lift-off data points and the two-dimensional hyperplane is determined to determine whether the lift-off data point is a false lift-off caused by local vaporization.

[0085] (2) If yes, the lift-off data point is set as a false lift-off data point, and then the false lift-off data point is set as a non-lift-off data point.

[0086] (3) If no, the first lift-off data point in the continuous time is determined as the critical working condition point.

[0087] In order to prevent misjudgment, 50 experimental working condition points are taken for functional test of the intelligent judgment algorithm to verify the judgment function and accuracy, whether the experimental working condition points can be accurately judged, whether there is misjudgment or misjudgment, and the algorithm accuracy is higher than 95%. Whether the working condition point reaches is judged, after the working condition point reaches, a period of time is maintained, and the average value of the experimental parameters is independently saved, the next working condition experiment is continued, the working condition critical point is continuously judged until the end. Whether all 50 working condition critical points are judged, whether there is misjudgment or misjudgment. If there is misjudgment or misjudgment, the judgment model is modified and optimized, and the test is performed.

[0088] S105, determining the critical working condition point according to the lift-off data point, and outputting the critical working condition point.

[0089] In one specific embodiment provided by the present application, as Figure 3As shown, first, the critical heat flux density data generated by the system is input into the thermocouple temperature change model to determine the data set (i.e., the bulk wall temperature data); the data set is normalized to determine the normalized data set; the first-order ejection characteristic vector and the second-order ejection characteristic vector of the data in the normalized data set are calculated, and the first-order ejection characteristic vector and the second-order ejection characteristic vector are input into the trained SVM model to determine the positional relationship between the data points corresponding to the ejection characteristic vector and the two-dimensional hyperplane, to determine the ejection data points and the non-ejection data points, and to monitor the ejection data points in continuous time to determine whether they are false ejection data points, and if they are ejection data points, the points are set as non-ejection data points; the critical working condition point is determined according to the ejection data points, and the critical working condition point is output.

[0090] According to the critical working condition point information provided by the application, the system first automatically triggers fault-tolerant control, adopts intelligent feedforward control method, and calculates the current efficiency ratio based on the information amount of the monitoring point through online learning of the RBF neural network, and corrects the control quantity in real time in combination with the energy efficiency ratio stored in the database.

[0091] In one specific embodiment provided by the application, taking the bulk wall temperature rising rate of the critical heat flux density experiment as an example, after the critical working condition point is obtained in advance, the processing mainly includes three steps: the feedforward control is adopted to maintain a faster rising rate, so that the rising rate is a constant close to 1 at the inlet and outlet of different preheaters and valves; if the rising rate is out of limit, an alarm is triggered, and the power and flow are adjusted according to the current inlet and outlet temperature and flow; if it is still out of limit after adjustment, a shutdown signal is sent, and manual further processing is waited.

[0092] If the flow and pressure of a monitoring point suddenly drop sharply, or a serious emergency such as pipe rupture occurs, the system immediately triggers the emergency shutdown control and the loop protection. The emergency shutdown subsystem is used to monitor the bulk, the preheater, the main pump, the pressure stabilizer, the measurement and control system and various pipes and other equipment, and when an abnormal emergency occurs during operation, the emergency shutdown program is automatically run to trigger the interlocking operation, i.e., the temperature and pressure are lowered, the power is urgently cut off, and the operator is notified. The overall emergency shutdown control algorithm adopts sequential control, monitors the entire critical heat flux density experiment process, and the emergency shutdown instruction has the highest priority.

[0093] If there is a working condition fault, the system automatically triggers the loop protection measure, and the loop protection subsystem includes loop leakage protection, insulation protection, temperature protection and flow protection.

[0094] The loop leakage protection monitors the pressure change of each monitoring point in the experimental working condition, and judges whether there is leakage of gas, liquid or the like according to the pressure drop speed. First, it is automatically judged whether the system is in the process of adjusting the working condition pressure, and if it is judged that it is not in the adjustment process, the experiment has leakage, and the system immediately alarms and automatically executes emergency shutdown.

[0095] The insulation protection is realized by monitoring the leakage current in the experimental condition, the system automatically detects the insulation condition, and the leakage current is monitored in real time. If the leakage current exceeds the limit value, the system will immediately alarm and automatically execute emergency shutdown.

[0096] The temperature protection is realized by monitoring the temperature change of each measuring point in the experimental condition. If the wall temperature rising rate exceeds the upper limit value, the heating power of the body is automatically cut off by percentage. If the temperature change exceeds the upper limit, the system will immediately alarm and automatically execute emergency shutdown.

[0097] The flow protection includes preheater inlet flow rate rising rate protection and body inlet flow rate rising rate protection. If the preheater inlet flow rate rising rate and the body inlet flow rate rising rate exceed the upper limit value, the system will immediately alarm and automatically execute emergency shutdown.

[0098] The method has flexible configuration and simple operation, reduces the cost of human resources, improves the process monitoring efficiency, can quickly warn the abnormal condition in the industrial process, reduces the workload of the operator, and solves the problems of low alarm efficiency, complex fault diagnosis and large operation difficulty of the existing system.

[0099] Each embodiment in the specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other. Each embodiment focuses on the difference from other embodiments. The whole or part of the present application can be used in many general or special computer system environments or configurations. For example: personal computer, server computer, handheld device or portable device, tablet device, mobile communication terminal, multi-processor system, microprocessor-based system, programmable electronic device, network PC, small computer, large computer, distributed computing environment including any of the above systems or devices, etc.

[0100] The above examples are only used to illustrate the technical solutions of the present application, and are not limited to the present application. 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 or all of the technical features can be replaced by equivalent alternatives. The modification or replacement does not change the essence of the corresponding technical solution out of the scope of the technical solutions of the present application.

Claims

1. A method for intelligent diagnosis of faults in a process industry, characterized in that, The application relates to a method for determining a critical condition point of a heat transfer system, comprising the following steps: inputting critical heat flux density data into a thermocouple temperature change model to determine a data set; normalizing the data set to determine a normalized data set; According to the normalization data set, a set of femto feature vectors is calculated, the set of femto feature vectors is calculated using a femto feature vector calculation formula, and the set of femto feature vectors is input into a trained SVM model to output the positional relationship between a data point corresponding to the femto feature vector and a two-dimensional hyperplane, thereby determining femto data points and non-femto data points; wherein the set of femto feature vectors includes multi-order femto feature vectors; wherein the femto feature vector calculation formula is represented as: wherein, is represented as is an order femto feature vector. is represented as the first normalized body wall temperature data; is represented as the first normalized body wall temperature data. is represented as the first sampling time, with the unit of . is represented as the first sampling time, with the unit of , is represented as the calculation order. determining whether the femtosecond data point is a false femtosecond data point, and if the femtosecond data point is the false femtosecond data point, the femtosecond data point is set as the non-femtosecond data point; the method comprises the following steps: observing the position relationship between the multi-order femtosecond feature vector of the data points in the continuous time after the femtosecond data point and the two-dimensional hyperplane, determining whether the femtosecond data point is a false femtosecond data point caused by local vaporization, if yes, the femtosecond data point is set as the false femtosecond data point, and then the false femtosecond data point is set as the non-femtosecond data point, and if no, the first femtosecond data point in the continuous time is determined as a critical condition point; determining a critical condition point according to the femtosecond data point and outputting the critical condition point.

2. The method of claim 1, wherein, The method for training the SVM model comprises the following steps: obtaining a target function and a constraint condition; determining the minimum value of the target function according to the target function and the constraint condition, so as to determine a two-dimensional hyperplane.

3. The method of claim 1, wherein, The thermocouple temperature change model is expressed as: wherein, , represents the temperature of the thermocouple inlet, outlet, in °C; represents the flow rate of the thermocouple inlet, outlet, in m / s; ; represents the mass of the superheated metal, in kg; ; represents the total heating of the body, in MW; ; represents the total heat transfer in the body, in MW; ; represents the empirical coefficient; represents the volume of the body, in m3; ; represents the heat exchange coefficient, dimensionless; represents the temperature of the tube wall in the body, in °C; represents the specific heat of the superheater metal, in kJ / kg°C; ; represents the specific heat capacity of the fluid, in kJ / kg°C; , represents the density of the thermocouple outlet fluid, in kg / m3; .

4. The method of claim 1, wherein, The method for inputting the femtosecond feature vector set into the trained SVM model to output the position relationship between the data point corresponding to the femtosecond feature vector and the two-dimensional hyperplane, and determining the femtosecond data point and the non-femtosecond data point comprises the following steps: determining whether the data point corresponding to the femtosecond feature vector is a femtosecond data point according to the position relationship between each-order femtosecond feature vector in the femtosecond feature vector set and the two-dimensional hyperplane; if the femtosecond feature vector is distributed above the two-dimensional hyperplane, the data point corresponding to the femtosecond feature vector is classified as the femtosecond data point; if the femtosecond feature vector is distributed below the two-dimensional hyperplane, the data point corresponding to the femtosecond feature vector is classified as the non-femtosecond data point; if the femtosecond feature vector is distributed on the two-dimensional hyperplane, the data point corresponding to the femtosecond feature vector is classified as the femtosecond data point.

5. The method of claim 2, wherein, The target function is expressed as: wherein, denotes a weight vector, , denotes the set of said femto feature vectors, denotes a bias term.

6. The method of claim 5, wherein, The constraint condition is a Lagrange multiplier.

7. The method of claim 6, wherein, The method for determining the minimum value of the target function according to the target function and the constraint condition, so as to obtain a two-dimensional hyperplane, comprises the following steps: determining the edge length of the target function; determining an optimization target function according to the constraint condition, the edge length and the target function; determining the minimum value of the target function by obtaining the maximum value of the optimization target function, so as to determine a two-dimensional hyperplane.

8. The method of claim 7, wherein, The edge length is expressed as: The optimization target function is expressed as: wherein, , are denoted as a set of Lagrange multiplier variables, , are denoted as data class labels of the normalized data set, , are both in {-1, 1}, positive example when equal to +1; -1 when negative example; are denoted as order femto feature vectors, are denoted as order femto feature vectors, are denoted as a set of support vectors.

Images