Steam turbine fault early warning method based on Monte Carlo algorithm

Abstract

The present invention belongs to the field of thermal power generation, and in particular, relates to a method and system for early warning of steam turbine faults based on Monte Carlo algorithm, wherein the method comprises: collecting steam turbine fault case information, and classifying faults according to fault characteristic parameters contained in the case; for a specific fault, using Monte Carlo algorithm to simulate the relationship between characteristic parameters and alarm values ​​in the case one by one, and obtaining the change trend of characteristic parameters under the fault in each case; using least square method to fit the obtained characteristic parameter change trend, and obtaining the characteristic parameter change law curve under a certain fault; using the obtained turbine characteristic parameter change law curve under each fault for machine learning, to determine whether the deterioration of the turbine operating parameters will lead to faults. The method has high reliability and strong practicality, improves the accuracy of fault abnormality prediction, and can perform early detection and alarm at the same time, avoiding economic losses caused by post-detection.

Description

Technical Field

[0001] The invention belongs to the field of thermal power generation, and in particular relates to a fault early warning method for a steam turbine generator set in a thermal power plant. Background Art

[0002] Electricity plays a pivotal role in social life and the national economy, and steam turbine generator sets are one of the most important equipment in power plants. The status monitoring and fault diagnosis of steam turbine generator sets have always been of great concern. However, with the complexity of steam turbine generator sets and long-term aging, the frequency of equipment failures and sudden failures has gradually increased. Once equipment fails, a chain reaction may occur, affecting normal production work, and will inevitably cause huge economic losses. In severe cases, it may even cause catastrophic casualties. Therefore, early warning of vibration faults of steam turbine generator sets is imperative.

[0003] From the domestic situation, the steam turbine generator sets generally adopt the high and low value alarm of a single parameter of the DCS system. This monitoring method is "late to the fault" when the fault occurs. By the time the fault is discovered, there is no sufficient time to deal with it, which easily makes the operating personnel flustered and unable to take effective remedial measures in time. After the DCS system alarm is issued, there is no professional interpretation, and it is difficult to fully explore the internal connection between various parameters, so it is impossible to fully interpret the fault information. Summary of the invention

[0004] The present invention proposes a steam turbine fault early warning method based on Monte Carlo method to solve the above technical problems.

[0005] The present invention provides a steam turbine fault early warning method based on Monte Carlo algorithm, comprising the following steps:

[0006] Step 1, collect turbine fault case information, and classify the faults according to the fault characteristic parameters contained in the cases;

[0007] Step 2: For a specific fault, use the Monte Carlo algorithm to simulate the relationship between the characteristic parameters and the alarm value in each case one by one, and obtain the change trend of the characteristic parameters under the fault in each case;

[0008] Step 3, using the least square method to fit the characteristic parameter change trend obtained in step 2, to obtain a characteristic parameter change law curve under a certain fault;

[0009] Step 4: Use the obtained curves of the change rules of the characteristic parameters of the steam turbine under various faults for machine learning to determine whether the degradation of the operating parameters of the steam turbine will lead to a fault.

[0010] Furthermore, the fault characteristic parameters include the vibration through-frequency, frequency division, phase, bearing temperature, shaft displacement, expansion difference, unit speed, load, main steam temperature, reheat steam temperature, shaft seal steam supply temperature, upper and lower wall temperature of the turbine inner cylinder, interlayer temperature, and high-pressure regulating valve opening of each bearing.

[0011] Furthermore, the types of a particular fault include turbine rotor mass imbalance, dynamic and static friction, misalignment, rotor thermal bending, poor expansion, loose coupling bolts, steam flow vibration, structural resonance, and rotor cracks.

[0012] Furthermore, the machine learning in step 4 refers to supervised learning algorithms (Supervised Algorithms), mainly referring to BP (Back Propagation) neural networks, etc.

[0013] Through the above scheme, a turbine fault warning method based on the Monte Carlo algorithm is used. This method has high reliability and strong practicality, and improves the accuracy of fault abnormality prediction. At the same time, if the data is abnormal before the fault occurs, early detection and alarm can be carried out to avoid economic losses caused by subsequent detection.

[0014] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention and implement it according to the contents of the specification, the following is a detailed description of the preferred embodiments of the present invention in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a flow chart of an embodiment of a steam turbine fault early warning method based on Monte Carlo algorithm of the present invention;

[0016] Figure 2 It is a schematic diagram of simulation of the relationship between characteristic parameters and alarm values ​​in a fault case of a steam turbine fault early warning method based on Monte Carlo algorithm of the present invention;

[0017] Figure 3 The steam turbine fault early warning method based on the Monte Carlo algorithm of the present invention obtains a parameter change trend diagram when a fault occurs by curve fitting using the least squares method; DETAILED DESCRIPTION

[0018] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0019] Ginseng Figure 1 As shown, this embodiment provides a steam turbine fault early warning method based on Monte Carlo algorithm, comprising:

[0020] Step S1, collecting turbine fault case information, and classifying the faults according to the fault characteristic parameters contained in the cases;

[0021] Step S2, for a specific fault, use the Monte Carlo algorithm to simulate the relationship between the characteristic parameters and the alarm value in each case one by one, and obtain the change trend of the characteristic parameters under the fault in each case;

[0022] Step S3, using the least square method to fit the characteristic parameter change trend obtained in S2, to obtain a characteristic parameter change law curve under a certain fault;

[0023] Step S4, using the acquired curves of variation of characteristic parameters of the steam turbine under various faults for machine learning to determine whether degradation of the operating parameters of the steam turbine will lead to a fault.

[0024] In this embodiment, the fault characteristic parameters include the vibration through-frequency, frequency division, phase, bearing temperature, shaft displacement, expansion difference, unit speed, load, main steam temperature, reheat steam temperature, shaft seal steam supply temperature, upper and lower wall temperature of the turbine inner cylinder, interlayer temperature, and high-pressure regulating valve opening of each bearing.

[0025] In this embodiment, the type of a specific fault includes turbine rotor mass imbalance, dynamic and static friction, misalignment, rotor thermal bending, poor expansion, loose coupling bolts, steam flow vibration, structural resonance, and rotor cracks.

[0026] In this embodiment, machine learning refers to supervised learning algorithms, mainly BP neural networks and the like.

[0027] Ginseng Figures 2 to 3 As shown, the present invention is further described in detail below.

[0028] The steam turbine fault early warning method based on Monte Carlo algorithm comprises the following steps:

[0029] Step 1: Collect turbine fault case information and classify the faults according to the fault characteristic parameters contained in the cases.

[0030] Among them, the case database is formed by the collected turbine fault case information, and the content stored in the database is some facts of the object to be processed currently. The FMEA failure mode and effect analysis method (Failure Mode and Effect Analysis) is used to distinguish the fault characteristic parameters in the case, and then the fault characteristic parameters contained in different fault types are classified differently. The classification diagram is shown in the following table.

[0031] Features X1 Features X2 Features X3 Features X4 …… Fault 1 ○ × ○ ○ …… Fault 2 ○ × × × …… Fault 3 × ○ ○ × …… …… …… …… …… …… ……

[0032] Step 2: For a specific fault, use the Monte Carlo algorithm to simulate the relationship between the characteristic parameters and the alarm value in each case one by one, and obtain the change trend of the characteristic parameters under the fault in each case.

[0033] Among them, the Monte Carlo algorithm is also called the computer random simulation method, which is a method based on probability statistics theory. The Monte Carlo algorithm uses the method of random experiment to calculate the integral, that is, the integral to be calculated is regarded as the mathematical expectation of the random variable g(r) of a certain distribution density function f(r).

[0034]

[0035] Through experiments, we get N observed values ​​r1, r2, ..., rN, and calculate the arithmetic mean of the values ​​of N random variables g(r1), g(r2), ..., g(rN)

[0036]

[0037] as an estimate of the integral.

[0038] Use computer to simulate the relationship between characteristic parameters and alarm values ​​in the case, and the implementation steps are as follows: Figure 2 shown.

[0039] After implementing the above steps, the change trend of the characteristic parameters under the current fault can be obtained.

[0040] For example, in the "steam turbine moving and static friction" fault, by simulating the change of vibration value, the deterioration trend of vibration in the "steam turbine moving and static friction" fault can be obtained, as shown in the following table, the change of vibration (unit μm) over time (unit s) in different cases.

[0041] X1 X2 X3 Alarm value X4 Case 1 53 68 105 125 175 Case 2 29 41 87 125 206 Case 3 86 99 115 125 163 …… …… …… …… …… ……

[0042] Step 3: Use the least squares method to fit the characteristic parameter change trend obtained in S2 to obtain the characteristic parameter change law curve under a certain fault.

[0043] After the previous step is implemented, a "scatter plot" of the change of characteristic parameters of a certain fault in the case can be obtained. By using the least squares method to perform curve fitting, the parameter change trend diagram when the fault occurs can be approximately obtained. For example, Figure 3 shown.

[0044] Step 4: Use the obtained curves of the change rules of the turbine characteristic parameters under various faults for machine learning to determine whether the degradation of the turbine operating parameters will lead to faults.

[0045] After the previous step is implemented, the changing pattern of output y and input x (characteristic parameter changing trend) when a certain fault occurs can be obtained, and the supervised learning algorithm (mainly BP neural network, etc.) can be used to enable the computer to "learn" the changing pattern and issue an early warning when the characteristic parameter change conforms to this pattern.

[0046] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the technical principles of the present invention, and these improvements and modifications should also be regarded as within the scope of protection of the present invention.

Claims

1. A steam turbine fault early warning method based on Monte Carlo algorithm, characterized in that: The steps include: Step 1, collect turbine fault case information, use FMEA failure mode and effect analysis method to distinguish fault characteristic parameters in the case, and classify according to the different fault characteristic parameters contained in different fault types; the fault characteristic parameters include vibration frequency, frequency division, phase, bearing temperature, shaft displacement, expansion difference, unit speed, load, main steam temperature, reheat steam temperature, shaft seal steam supply temperature, upper and lower wall temperature of turbine inner cylinder, interlayer temperature, high-pressure valve opening; Step 2: For a specific fault, use the Monte Carlo algorithm to simulate the relationship between the characteristic parameters and the alarm value in each case, including: Establish probability and statistical models; Collect data on risk variables in the model and determine the distribution function of risk factors; Determine the number of simulations based on quarterly risk analysis requirements; Establish a sampling method for random variables to generate random numbers; Randomly sample the risk variable in the probability distribution according to the random number and substitute it into the established mathematical model; Statistical analysis, estimation of mean and standard deviation; Obtain the changing trend of characteristic parameters under the fault in each case; Among them, the types of a specific fault include turbine rotor mass imbalance, dynamic and static friction, misalignment, rotor thermal bending, poor expansion, loose coupling bolts, steam flow vibration, structural resonance, and rotor cracks; Step 3, using the least square method to fit the characteristic parameter change trend obtained in step 2, to obtain a characteristic parameter change law curve under a certain fault; Step 4, using the obtained turbine characteristic parameter variation law curves under various faults for machine learning to determine whether the degradation of the turbine operating parameters will lead to a fault; the machine learning uses a supervised learning algorithm, including a BP neural network.

Images