CN112464565B - Equipment fault early warning method integrating intelligent modeling and fuzzy rules - Google Patents

Abstract

The invention discloses an equipment fault early warning method, system and storage medium integrating intelligent modeling and fuzzy rules. The specific steps of the method include: (1) establishing a four-pipe leakage fault expert knowledge base; (2) building a membership degree based on a distance function (3) Obtain the symptom value of the parameter and improve the fuzzy membership function; (4) Obtain the early warning signal; (5) Calculate the membership degree of the early warning signal to judge the fault state. Under the premise of early warning, the method of the invention judges the specific failure mode according to the expert database and the fuzzy rules, thereby improving the accuracy and effectiveness of the early warning of equipment failures.

Description

An early warning method for equipment faults that integrates intelligent modeling and fuzzy rules

technical field

The invention relates to a fuzzy membership function identification method, in particular to an equipment fault early warning method integrating intelligent modeling and fuzzy rules.

Background technique

In recent years, with the continuous deepening reform of the power market, the connection between the two sides of the grid has become more and more closely, and the requirements for generating units have been continuously improved. uncertainty has risen sharply. Both power grids and power generation companies hope that the units can maintain continuous and stable operation and reduce the incidence of accidents. According to statistics, the leakage of the four-tube boilers accounts for nearly 50% of the unit accidents and more than 60% of the boiler accidents. Create very favorable conditions for the safe and stable operation of the unit. At present, the operator screen of the DCS system of the unit generally provides the functions of fixed value alarm and protection trip. However, in most cases, when an alarm occurs in the operation of the unit, the leakage fault of the four pipes has been seriously deteriorated. Generally, the on-site method can only be handled by reducing the load and applying for emergency shutdown. On the one hand, it will cause great damage to power generation enterprises In addition, the power grid is also affected to some extent. Therefore, in the early stage of failure, the use of technical means to judge the failure state in advance has become the focus of researchers in the field.

Operating data is the only source to obtain the operating status of the unit, and the hidden status information in the data is deeply excavated, which provides a feasible and effective path for the research on the early warning technology of the unit status. Zhang Wei deduced the reference value through fuzzy association rule mining to judge the status of the auxiliary equipment of the fluidized bed; some scholars extracted the representative data to establish the early warning model of the fan, and proposed the recession index for the status evaluation; some scholars used Association mining technology establishes a rule base for judging the formation process of equipment failures; some scholars have established a multivariate state estimation and early warning model for the unit milling system, and proposed a clustering method to construct a state matrix; some scholars have proposed improvements The fuzzy C-mean value of 1 is used for equipment fault diagnosis in the thermal slow-change process; some scholars have studied the modeling method of multivariate state estimation based on density clustering, and proposed the method of sliding window deviation for the determination of out-of-limit; some scholars have proposed a A fault early warning method based on power analysis and neural network is used for the diagnosis of wind turbines; in addition, there are other related early warning methods such as correlation analysis, neural network, Gaussian model and so on.

After obtaining the fault early warning signal, the invention judges the specific fault mode according to the expert database and the fuzzy rules: summarizes and summarizes the relevant literature and scientific research results of the leakage failure of the four-tube boiler, and lists the leakage of the four-tube boiler based on the distance-type fuzzy judgment rule. 's expert knowledge base for troubleshooting. At the same time, according to the standard deviation of each parameter obtained by statistical analysis of historical data, and then according to the result of comparing the regression error of the intelligent model with 5 times the standard deviation, the parameters are fuzzified, and the current state is further calculated according to the fuzzy rules. The membership degree of the mode is used to judge the current fault state. Based on a 1000MW thermal power simulation system, the method proposed in the present invention is verified by simulating normal operating conditions and related leakage faults respectively.

SUMMARY OF THE INVENTION

The technical problem to be solved by the present invention is to provide an early warning method for equipment failure, which can judge the specific failure mode according to the expert database and fuzzy rules.

The technical scheme adopted in the present invention is an equipment fault early warning method integrating intelligent modeling and fuzzy rules, comprising the following steps:

(1) Establish a four-pipe leakage fault expert knowledge base;

(2) Construct membership function based on distance function;

(3) Obtain the symptom value of the parameter and improve the fuzzy membership function;

(4) Obtain early warning signals;

(5) Calculate the membership degree of the early warning signal and judge the fault state.

Specifically, the step (1) includes:

(11) Statistical parameters related to the leakage of four boiler tubes;

(12) Obtain the changing relationship between each symptom parameter and the fault type;

(13) Establish a fault expert knowledge base.

Specifically, the step (2) includes:

(21) Select the appropriate distance function as the basis;

(22) Construct membership function.

Further, described step (3) comprises:

(31) According to the regression error of the extreme learning machine, a method for obtaining the symptom value of the parameter is proposed;

(32) Carry out parameter fuzzification and perfect fuzzy membership function.

Specifically, the step (4) includes:

(41) Establish a fault early warning model based on the multiple regression estimation method;

(42) Carry out the unit model simulation to obtain the fault early warning signal.

Specifically, the step (5) includes:

(51) Receive early warning signals;

(52) Calculate the membership degree between the early warning signal and various faults according to the constructed fuzzy membership function;

(53) Determine the current fault state according to the calculation result.

Beneficial effects achieved by the present invention:

Compared with other data modeling methods, the extreme learning machine of the present invention has unique advantages in model reliability and accuracy, and the results obtained by using fuzzy identification are also consistent with the simulated failure mode, which improves the accuracy of equipment failure early warning. and effectiveness.

Description of drawings

Fig. 1 is the schematic flow chart of the method of the present invention;

Figure 2 Schematic diagram of the trend of health indicators before and after the failure.

Detailed ways

The technical solutions of the present invention will be further described below with reference to the accompanying drawings and specific embodiments.

As shown in Figure 1, an equipment fault early warning method integrating intelligent modeling and fuzzy rules includes the following contents:

Step 1: Establish an expert knowledge base for four-pipe leakage faults. The fault knowledge base is the basis of fault identification. It describes the mode relationship between various fault types and symptom parameters, summarizes the characteristic parameters related to the leakage of the four boiler tubes, and points out the relationship between each symptom parameter and the fault type through thermal calculation. alternative relation. According to the five-value symptom set description method, the expert knowledge described in each document is synthesized, and the fault expert knowledge base is obtained.

The second step is to construct a membership function based on the distance function. Choose an appropriate distance function as a basis to construct a membership function.

The fault identification is actually based on the fault diagnosis expert database, and judges and identifies the current fault state according to certain rules. The present invention proposes a new fuzzy membership degree determination method according to the regression error of the extreme learning machine, and adopts the distance function based:

为：In the formula, d _j (u ₀ , u _j ) is the distance between the fault to be identified u ₀ and the typical fault mode u _j . Obviously, the smaller the value, the greater the possibility of such faults. z _j is the fault symptom value of the ith symptom parameter, z _ij is the symptom value of the ith symptom parameter under the jth typical fault, membership function

for:

in

D=max(d _j (u ₀ ,u _j )) (3)

Obviously, as shown in Equation (2), the greater the membership degree, the closer to 1, indicating that the possibility of such failures is greater.

The third step is to obtain the symptom value of the parameter and improve the fuzzy membership function. According to the regression error of extreme learning machine, a method for obtaining the parameter symptom value is proposed, the parameters are fuzzified, and the fuzzy membership function is perfected. According to the regression error of the extreme learning machine, the present invention proposes a method for obtaining the parameter symptom value z _j as follows:

为极限学习机的回归误差，σ为对应参数的标准差，通过历史数据统计获取,

为参数的测量值，

为参数的回归估计值；in the formula

is the regression error of the extreme learning machine, σ is the standard deviation of the corresponding parameter, obtained through historical data statistics,

is the measured value of the parameter,

is the regression estimate of the parameter;

Step 4: Obtain early warning signals. The multiple regression estimation method is to establish an autoregressive model of multiple parameters based on the potential correlation between the parameters, so as to diagnose whether the potential relationship of the parameters has changed.

The regression model of the extreme learning machine used in the present invention, but all similar autoregressive models are applicable to the process of the present invention.

Given a training sample set X _T , the excitation output function of the hidden layer node is G(a _i , b _i , X _T ), the number of nodes in the hidden layer is L, the excitation function adopts the sigmoidal function, and the process of the extreme learning machine algorithm is :

①Randomly generate the parameter values of the hidden layer node function ( _ak , b _k ), k=1,...,L;

②Calculate the hidden layer output matrix H;

③ Calculate the output weight vector

④ Calculate the regression value X _R of the parameter:

D为矩阵H的广义逆矩阵，D＝H^T(HH^T)^-1，

为权重向量，

D is the generalized inverse matrix of matrix H, D=H ^T (HH ^T ) ^-1 ,

is the weight vector,

Under normal operating conditions, the deviation between the regression value and the measured value of the fault early warning model is within the normal random error range; if the operation of the equipment deviates from the normal operating conditions, the regression value and measured value of the fault early warning model will be If there is an obvious deviation, which exceeds the reasonable random error range, it is considered that the regression relationship between the parameters has changed, and an early warning signal is given.

Step 5: Calculate the membership degree of the early warning signal to judge the fault state.

A domestic 1000MW thermal power unit simulation system is used to simulate normal operating conditions and various fault conditions to verify the effectiveness of the method proposed in the present invention for diagnosing the leakage of the four-tube boiler.

The specific unit simulation process includes:

①Simulate the normal operating state of the unit under 50%-100% load conditions;

② Simulate four types of leakage fault states under full load conditions to verify the effectiveness of the diagnosis method.

The unit simulation system used in the present invention has been repeatedly debugged and certified, and has an approximate 1:1 relationship with the dynamic characteristics of the field unit. It has been used as a supply model for national-level industry and provincial-level industry simulation machine competitions. The simulation system is used to set different settings. Model simulation of four-pipe leakage faults of various types is carried out, and the variation curves of various typical symptom parameters are obtained.

When an early warning signal occurs, it is necessary to further diagnose the specific fault type to provide effective information for timely processing. Firstly, the symptom value of the parameter is obtained according to the regression deviation, and then the membership degree of each fault is obtained by using the expert knowledge base and the fuzzy membership function, thereby determining the specific failure mode.

An equipment fault early warning integrating intelligent modeling and fuzzy rules, including the following program modules:

Expert knowledge base module: establish a four-pipe leakage fault expert knowledge base;

Membership function module: construct membership function based on distance function;

Symptom value parameter module: obtain the parameter symptom value and improve the fuzzy membership function;

Early warning signal module: obtain early warning signals;

Fault status module: Calculate the membership degree of the early warning signal and judge the fault status.

A storage medium for an equipment failure early warning system integrating intelligent modeling and fuzzy rules, characterized in that the following program modules are stored:

Expert knowledge base module: establish a four-pipe leakage fault expert knowledge base;

Membership function module: construct membership function based on distance function;

Symptom value parameter module: obtain the parameter symptom value and improve the fuzzy membership function;

Early warning signal module: obtain early warning signals;

Fault status module: Calculate the membership degree of the early warning signal and judge the fault status.

Example 1

Model parameters:

A domestic 1000MW thermal power unit simulation system is used to simulate normal operating conditions and various fault conditions to verify the effectiveness of the method proposed in the present invention for diagnosing the leakage of the four-tube boiler. The specific unit simulation process includes: (1) simulating the normal operating state of the unit under 50%-100% load conditions to verify the validity of the regression model; (2) simulating four types of leakage fault states under full load conditions, using to verify the validity of the diagnostic method.

Establish an expert knowledge base for four-pipe leakage failures. The fault knowledge base is the basis of fault identification. It describes the mode relationship between various fault categories and symptom parameters. According to the characteristic parameters related to the leakage of the four boiler tubes summarized in the previous literature, the symptom parameters are pointed out through thermal calculation. Variation relationship with fault type. The present invention uses a five-valued symptom set to describe the symptom parameter variation characteristics under the four-pipe leakage fault:

The five-valued symptom set description method shown in formula (10) can not only express the changes in the positive and negative directions of the parameters, but also consider the degree of change, which is more suitable for complex fault diagnosis.

According to the five-value symptom set description method, the expert knowledge described in each document is synthesized, and the fault expert knowledge base shown in Table 1 and Table 2 is obtained.

Table 1 Four-pipe leakage symptom parameter set

Table 2 Four-pipe leakage expert knowledge base

The fault identification is actually based on the fault diagnosis expert database, and judges and identifies the current fault state according to certain rules. The present invention proposes a new fuzzy membership degree determination method according to the regression error of the extreme learning machine, and adopts the distance function based:

In the formula, d _j (u ₀ , u _j ) is the distance between the fault to be identified u ₀ and the typical fault mode u _j . Obviously, the smaller the value, the greater the possibility of such faults. z _j is the fault symptom value of the ith symptom parameter, and z _ij is the symptom value of the ith symptom parameter under the jth typical fault. The membership function is:

in

D=max(d _j (u ₀ ,u _j )) (3)

Obviously, as shown in Equation (2), the greater the membership degree, the closer to 1, indicating that the possibility of such failures is greater.

According to the regression error of the extreme learning machine, the present invention proposes a method for obtaining the parameter symptom value z _j as follows:

为极限学习机的回归误差，σ为对应参数的标准差，通过历史数据统计获取。in the formula

is the regression error of the extreme learning machine, and σ is the standard deviation of the corresponding parameter, obtained from historical data statistics.

Collect data corresponding to various faults, use the established multiple regression model to perform regression estimation on each symptom parameter, and fit the health index zt representing the overall operating state of the unit. The calculation formula is:

In the formula, p is the number of symptom parameters related to leakage faults. Obviously, the closer the index is to 1, the more accurate the regression model is, and the more normal the operating state of the unit is. If the value of the health index shows a significant trend decline at a certain moment, it means that the operating state of the unit is abnormal. Figure 2 shows the change of the health index after the leakage of the A-side superheater fails.

After going through the multi-parameter regression model, a health index zt is calculated according to the regression value and the measured value using formula (6). The trend of zt reflects the running state of the modeled object. If an abnormality is detected, the fault diagnosis mechanism, that is, the fuzzy diagnosis process, is triggered at this time.

As shown in Figure 2, the solid curve represents the change trend of the health degree, and the dotted line is the limit of the health degree warning, which is generally obtained through statistical analysis under normal working conditions. It can be seen from this that under the normal working conditions of the unit, the health degree is maintained at a relatively stable level. When the fault occurs, the health degree shows a significant downward trend, reaching the warning limit at the 26th second, that is, the fault occurs 6 seconds after the occurrence. Health warning, and the over-temperature warning of the steam temperature is only given by the DCS around 90S. Obviously, the warning signal is highly sensitive to minor deterioration, and has the effect of early warning of faults.

When an early warning signal occurs, it is necessary to further diagnose the specific fault type to provide effective information for timely processing. Firstly, the symptom value of the parameter is obtained according to the regression deviation, and then the membership degree of each fault is obtained by using the expert knowledge base and the fuzzy membership function, thereby determining the specific failure mode. Table 3 shows the calculation results of the membership degrees of various faults after the fault.

Table 3 Calculation of membership degree of four-pipe leakage diagnosis

As shown in Table 3, the membership values at three time points were calculated using the four-pipe leakage expert knowledge base and fuzzy membership. It can be seen from this that with the occurrence and deterioration of the fault, the membership degree of the current fault mode to u3 increases significantly and gradually tends to 1. After the early warning signal appears, it can actually be judged by the membership degree value. If the leakage fault of the superheater is detected, as time goes on, its certainty is greater, and the calculation result of the membership degree of the 30th S further verifies the correctness of the judgment of the early warning fault.

Under the premise of early warning, the invention judges the specific failure mode according to the expert database and fuzzy rules: summarizes the relevant literature and scientific research results of the leakage failure of the four boiler tubes, and lists the four boiler leakage failures based on the distance type fuzzy judgment rule. Troubleshooting expert knowledge base. At the same time, according to the standard deviation of each parameter obtained by statistical analysis of historical data, and then according to the result of comparing the regression error of the intelligent model with 5 times the standard deviation, the parameters are fuzzified, and the current state is further calculated according to the fuzzy rules. The membership degree of the mode is used to judge the current fault state. A 1000MW unit simulation system is used to simulate the fault characteristics of the unit's operating conditions and four-pipe leakage, and the method proposed in the present invention is verified. The correctness and effectiveness of the proposed method.

Claims (5)

1. An equipment fault early warning method integrating intelligent modeling and fuzzy rules is characterized by comprising the following steps:

(1) Establishing a four-tube leakage fault expert knowledge base, which specifically comprises the following steps:

(11) Counting related characteristic parameters of the leakage of four pipes of the boiler;

(12) Obtaining the change relation between each symptom parameter and the fault type;

(13) Establishing a fault expert knowledge base;

(2) Constructing a membership function based on the distance function;

(3) Obtaining a parameter symptom value, perfecting a fuzzy membership function, and specifically comprising the following steps:

(31) According to the regression error of the extreme learning machine, a parameter symptom value solving method is provided;

(32) Fuzzification of parameters is carried out to perfect fuzzy membership function,

parameter symptom value z _i The method for obtaining comprises the following steps:

in the formula

Is the regression error of the extreme learning machine, sigma is the standard deviation of the corresponding parameters, is obtained through historical data statistics,

is a measured value of a parameter that is,

is a regression estimate of the parameter;

collecting data corresponding to various faults, carrying out regression estimation on various symptom parameters by using the established multivariate regression model, and fitting a health degree index calculation formula representing the whole running state of the unit into the formula:

wherein p is the number of symptom parameters related to the leakage fault;

(4) Acquiring an early warning signal, specifically comprising:

(41) Establishing a fault early warning model based on a multiple regression estimation method;

(42) Performing unit model simulation to acquire a fault early warning signal, wherein the specific process is as follows:

under the normal operation condition, the deviation between the regression value and the measured value of the fault early warning model is within a normal random error range; if the operation of the equipment deviates from the normal operation condition, the regression value and the measured value of the fault early warning model have obvious deviation and exceed the reasonable random error range, the regression relationship among the parameters is considered to be changed, and an early warning signal is given;

(5) Calculating the membership degree of the early warning signal, and judging the fault state, wherein the specific process comprises the following steps:

when the early warning signal appears, the specific fault type is further diagnosed, firstly, a symptom value of a parameter is obtained according to the regression deviation, then, the membership degree of each fault is obtained by utilizing an expert knowledge base and a fuzzy membership function, and therefore, the specific fault mode is determined.

2. The equipment fault early warning method integrating intelligent modeling and fuzzy rules according to claim 1, wherein the method comprises the following steps: the step (2) specifically comprises:

(21) Selecting a suitable distance function as a basis;

(22) And constructing a membership function.

3. The equipment fault early warning method fusing intelligent modeling and fuzzy rules according to claim 2, characterized in that: the distance function is:

in the formula d _j (u ₀ ,u _j ) For a fault u to be identified ₀ And typical failure mode u _j Distance between, z _ij The symptom value of the ith symptom parameter under the jth typical fault;

membership function

Comprises the following steps:

wherein D = max (D) _j (u ₀ ,u _j ))(3)。

4. The utility model provides an equipment trouble early warning system who fuses intelligent modeling and fuzzy rule which characterized in that: the method comprises the following program modules:

the expert knowledge base module: establishing a four-tube leakage fault expert knowledge base, which specifically comprises the following steps:

(11) Counting related characteristic parameters of the leakage of four pipes of the boiler;

(12) Obtaining the change relation between each symptom parameter and the fault type;

(13) Establishing a fault expert knowledge base;

a membership function module: constructing a membership function based on the distance function;

a symptom parameter module: obtaining a parameter symptom value, perfecting a fuzzy membership function, and specifically comprising the following steps:

(31) According to the regression error of the extreme learning machine, a parameter symptom value solving method is provided;

(32) Fuzzification of parameters is carried out to perfect fuzzy membership function,

parameter symptom value z _j The method comprises the following steps:

in the formula

Is the regression error of the extreme learning machine, sigma is the standard deviation of the corresponding parameters, is obtained through historical data statistics,

is a measured value of a parameter that is,

is a regression estimate of the parameter;

collecting data corresponding to various faults, carrying out regression estimation on various symptom parameters by using the established multivariate regression model, and fitting a health degree index calculation formula representing the whole running state of the unit into the formula:

wherein p is the number of symptom parameters related to the leakage fault;

the early warning signal module: acquiring an early warning signal, specifically comprising:

(41) Establishing a fault early warning model based on a multiple regression estimation method;

(42) Performing unit model simulation to acquire a fault early warning signal, wherein the specific process is as follows:

under the normal operation condition, the deviation between the regression value and the measured value of the fault early warning model is within the normal random error range; if the operation of the equipment deviates from the normal operation condition, the regression value and the measured value of the fault early warning model have obvious deviation and exceed the reasonable random error range, the regression relationship among the parameters is considered to be changed, and an early warning signal is given;

a fault state module: calculating the membership degree of the early warning signal, and judging the fault state, wherein the specific process comprises the following steps:

when the early warning signal appears, a specific fault type is further diagnosed, firstly, a symptom value of a parameter is obtained according to the regression deviation, then, the membership degree of each fault is obtained by utilizing an expert knowledge base and a fuzzy membership function, and therefore a specific fault mode is determined.

5. A storage medium of an equipment fault early warning system fusing intelligent modeling and fuzzy rules is characterized by storing the following program modules:

expert knowledge base module: establishing a four-tube leakage fault expert knowledge base, which specifically comprises the following steps:

(11) Counting related characteristic parameters of the leakage of four pipes of the boiler;

(12) Obtaining the change relation between each symptom parameter and the fault type;

(13) Establishing a fault expert knowledge base;

a membership function module: constructing a membership function based on the distance function;

a symptom parameter module: obtaining a parameter symptom value, perfecting a fuzzy membership function, and specifically comprising the following steps:

(31) According to the regression error of the extreme learning machine, a parameter symptom value solving method is provided;

(32) Fuzzification of parameters is carried out to perfect fuzzy membership function,

parameter symptom value z _i The method comprises the following steps:

in the formula

Is the regression error of the extreme learning machine, sigma is the standard deviation of the corresponding parameters, is obtained through historical data statistics,

is a measured value of a parameter that is,

is a regression estimate of the parameter;

collecting data corresponding to various faults, carrying out regression estimation on various symptom parameters by using the established multivariate regression model, and fitting a health degree index calculation formula representing the whole running state of the unit into the formula:

wherein p is the number of symptom parameters related to the leakage fault;

the early warning signal module: acquiring an early warning signal, specifically comprising:

(41) Establishing a fault early warning model based on a multiple regression estimation method;

(42) Performing unit model simulation to acquire a fault early warning signal, wherein the specific process is as follows:

under the normal operation condition, the deviation between the regression value and the measured value of the fault early warning model is within the normal random error range; if the operation of the equipment deviates from the normal operation condition, the regression value and the measured value of the fault early warning model have obvious deviation and exceed the reasonable random error range, the regression relationship among the parameters is considered to be changed, and an early warning signal is given;

a fault state module: calculating the membership degree of the early warning signal, and judging the fault state, wherein the specific process comprises the following steps:

when the early warning signal appears, the specific fault type is further diagnosed, firstly, a symptom value of a parameter is obtained according to the regression deviation, then, the membership degree of each fault is obtained by utilizing an expert knowledge base and a fuzzy membership function, and therefore, the specific fault mode is determined.

Images