Thermal fault diagnosis method for stator winding

Abstract

The invention provides a stator winding thermal fault diagnosis method, and the method comprises the steps: carrying out the feature extraction of obtained historical outlet water temperature sequence data, obtaining temperature difference feature historical data and trend difference feature historical data, and forming a data sample set; marking partial samples of the data sample set and not marking the rest samples; taking the marked samples and the unmarked samples as training samples, and training based on a graph semi-supervised method to obtain a thermal fault diagnosis model; and acquiring real-time outlet water temperature sequence data of the stator winding in an operation state by using a wireless measurement device, performing feature extraction to obtain temperature difference feature real-time data and trend difference feature real-time data, and inputting the data into the thermal fault diagnosis model to obtain a thermal fault diagnosis result. A thermal fault diagnosis model can be trained based on a small number of marked samples and a large number of unmarked samples by using a graph semi-supervised method, the data marking cost is reduced, and the problem of difficulty in temperature measurement of the stator winding in an operation state is solved by using wireless temperature measurement.

Description

Technical Field

[0001] The present application belongs to the technical field of operation and maintenance management of power industry generator equipment, and specifically, relates to a stator winding thermal fault diagnosis method. Background Art

[0002] In order to ensure the safe operation of the stator windings and internal cooling water systems of water-cooled steam turbine generators, it is important to prevent stator bar blockage that may lead to insulation damage of the windings. Relevant safety regulations have put forward clear requirements: when the outlet water temperature difference of the water pipe between the stator bars reaches a certain threshold when the generator is running, the load needs to be reduced or even shut down.

[0003] At present, the main monitoring method in the industry is to use ultrasonic and hot water flow methods for measurement when the machine is shut down for maintenance. However, this method has certain limitations. For example, it is impossible to analyze the fluidity of cold water in the winding during operation, the length of the insulating water pipe is limited by on-site conditions and cannot meet the measurement requirements, resulting in poor accuracy, and it is impossible to detect abnormalities when there is slight blockage or small deviation in the sensor position.

[0004] In addition, relevant researchers have proposed more complex temperature mechanism models, and data-driven models have been gradually widely used in related fields. For example, supervised learning methods such as Gaussian process regression, support vector machine, and neural network can be used to diagnose turbine faults, which can achieve a higher accuracy rate under the condition that the labeled samples are relatively sufficient. However, a large number of labeled samples are highly dependent on the experience of relevant practitioners, and the labeling time cost and labeling difficulty are high, making it difficult to directly apply the modeling method based on traditional supervised learning to the field of stator thermal fault diagnosis. Summary of the invention

[0005] The technical problem solved by the present application is: how to provide a method for diagnosing faults of a generator stator winding using a small amount of labeled data sample sets while in operation.

[0006] The present application provides a stator winding thermal fault diagnosis method, the stator winding thermal fault diagnosis method comprising:

[0007] Perform feature extraction on the acquired historical water outlet temperature series data to obtain temperature difference feature historical data and trend difference feature historical data to form a data sample set;

[0008] Marking some samples of the data sample set and not marking the remaining samples;

[0009] The labeled samples and unlabeled samples are used as training samples, and a thermal fault diagnosis model is obtained by training based on the graph semi-supervised method.

[0010] The real-time outlet water temperature sequence data of the stator winding under the running state is obtained by using a wireless measuring device, and the features of the real-time outlet water temperature sequence data are extracted to obtain the real-time data of temperature difference features and the real-time data of trend difference features;

[0011] The temperature difference characteristic real-time data and the trend difference characteristic real-time data are input into the thermal fault diagnosis model to obtain a thermal fault diagnosis result.

[0012] Optionally, the method of extracting features from the acquired historical water outlet temperature sequence data to obtain temperature difference feature historical data includes:

[0013] Extracting data segments of a plurality of predetermined time windows from the historical water outlet temperature series data, each of the data segments comprising a plurality of water outlet temperature measurement point time series;

[0014] Calculate and obtain the median temperature time series of several outlet water temperature measurement point time series;

[0015] Calculate the difference between the time series of each outlet water temperature measurement point and the median temperature time series;

[0016] The mean of each difference sequence is calculated as the temperature difference characteristic historical data.

[0017] Optionally, the method of extracting features from the acquired historical water outlet temperature sequence data to obtain trend difference feature historical data includes:

[0018] The trend difference characteristic historical data is calculated based on each difference sequence and the temperature difference characteristic historical data.

[0019] Optionally, the method of marking some samples of the data sample set and not marking the remaining samples includes:

[0020] The abnormal samples in the part of samples are marked as 1, the normal samples in the part of samples are marked as 0, and the remaining samples are all marked as -1.

[0021] Optionally, the wireless measuring device includes a temperature sensor and a wireless communication module, the temperature sensor is used to extend into the water outlet pipe of the stator winding and measure real-time water outlet temperature sequence data, and the wireless communication module is used to send the real-time water outlet temperature sequence data to the outside world.

[0022] Optionally, the time length of the real-time outlet water temperature series data is greater than or equal to a predetermined time window.

[0023] Optionally, the stator winding thermal fault diagnosis method further includes:

[0024] When the thermal fault diagnosis result is abnormal, an alarm signal is issued.

[0025] The present application provides a stator winding thermal fault diagnosis method, which has the following technical effects:

[0026] Using graph semi-supervised methods, a thermal fault diagnosis model can be trained based on a small number of labeled samples and a large number of unlabeled samples, which reduces the data labeling cost. Wireless temperature measurement is used to solve the problem of difficult temperature measurement of the generator stator winding cooling water system in operation, and greatly alleviates the problem of on-site wiring difficulties. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 is a schematic diagram of a stator winding thermal fault diagnosis method according to one or more embodiments;

[0028] Figure 2 A schematic diagram of the overall distribution of temperature difference-trend difference according to one or more embodiments;

[0029] Figure 3 Schematic diagram of the overall distribution of temperature difference-trend difference after a small number of samples are marked according to one or more embodiments. DETAILED DESCRIPTION

[0030] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0031] Before describing the various embodiments of the present application in detail, the technical concept of the present application is first briefly described: the current means of monitoring thermal faults of stator windings are either unable to be monitored in the operating state, or require a large amount of data for model training, which makes monitoring difficult. To this end, the present application provides a stator winding thermal fault diagnosis method, which partially labels the data sample set and does not label most of it, and obtains a thermal fault diagnosis model based on a graph semi-supervised method. Then, a wireless measurement device is used to obtain the real-time water outlet temperature sequence data of the stator winding in the operating state, and the data is input into the thermal fault diagnosis model to obtain the diagnosis result. This method can be applied to stator windings in the operating state, and does not require a large number of labeled samples for model training. The specific principles of the stator winding thermal fault diagnosis method of the present application are described below in combination with more embodiments.

[0032] Specifically, Figure 1 As shown, the stator winding thermal fault diagnosis method provided in the first embodiment includes the following steps:

[0033] Step S10: extract features from the acquired historical water outlet temperature sequence data to obtain temperature difference feature historical data and trend difference feature historical data to form a data sample set.

[0034] Step S20: Mark some samples in the data sample set and do not mark the remaining samples.

[0035] Step S30: Using the labeled samples and the unlabeled samples as training samples, a thermal fault diagnosis model is obtained by training based on a graph semi-supervised method.

[0036] Step S40: using a wireless measuring device to obtain real-time outlet water temperature sequence data of the stator winding in the operating state, and extracting features from the real-time outlet water temperature sequence data to obtain temperature difference feature real-time data and trend difference feature real-time data.

[0037] Step S50: input the temperature difference feature real-time data and the trend difference feature real-time data into the thermal fault diagnosis model to obtain the thermal fault diagnosis result.

[0038] In one or more embodiments, step S10 includes the following sub-steps:

[0039] Step S101: extract data segments of several predetermined time windows from the historical water outlet temperature series data, each data segment includes several water outlet temperature measurement point time series. For example, the water outlet temperature series usually consists of 42-84 temperature measurement points, each measurement point has one temperature data per minute, and the data segment is extracted with 60 minutes as the predetermined time window. The data segment contains N water outlet temperature measurement point time series, which are respectively denoted as T i =(T i，1 , T i，2 , ..., T i，t ), i∈{1, 2, ..., N}.

[0040] Step S102: Calculate and obtain the median temperature time series of several outlet water temperature measurement point time series. For example, the median temperature time series is represented as T med =med(T i ), i∈{1, 2, ..., N}.

[0041] Step S103: Calculate the difference sequence between the time series of each outlet water temperature measurement point and the temperature median time series. For example, the difference sequence is represented by ΔT i =T i -T med .

[0042] Step S104: Calculate the mean of each difference sequence as the temperature difference characteristic historical data. i The calculation formula is as follows:

[0043]

[0044] Step S105: Calculate the trend difference characteristic historical data according to each difference sequence and the temperature difference characteristic historical data. For example, the trend difference characteristic historical data K i The calculation formula is as follows:

[0045]

[0046] Exemplarily, a sliding predetermined time window is used to capture M data segments, and the above steps S102 to S105 are repeated for each data segment to form a data sample set, which is represented by X = (H, K). Each row of samples in the data set represents the temperature difference and trend characteristics of a certain measuring point in a certain time segment, so one data segment can obtain N rows of samples, and the total number of samples is M×N. Figure 2 As shown, we can further draw a scatter plot of temperature difference and trend difference to observe the distribution of data samples.

[0047] In one or more embodiments, the method of marking some samples of the data sample set and not marking the remaining samples in step S20 includes: marking abnormal samples in some samples as 1, marking normal samples in some samples as 0, and marking the remaining samples as -1, so as to balance the number of abnormal samples and normal samples as much as possible. Figure 3 As shown, only a part of the samples need to be labeled, and a large number of black dots are unlabeled samples.

[0048] In one or more embodiments, the graph semi-supervised method maps the data set into a graph, where each node in the graph corresponds to a sample, and the connection between the nodes, i.e., the edge, corresponds to the degree of similarity between the two samples, and then spreads the label to the unlabeled samples based on the similarity propagation. When the samples present a more regular distribution, a classification model with strong generalization ability can usually be obtained, which can be used for the diagnosis of early thermal faults of stator windings. Exemplarily, the process of training the thermal fault diagnosis model based on the graph semi-supervised method in step S30 is as follows:

[0049] The labeled minority samples are denoted as S l ={(x1,y1),(x2,y2),...,(x l ,y l )},x i =(H i , K i ), and the large number of unlabeled samples is recorded as S u ={(x l+1 ,y l+1) ,(x l+2 ,y l+2 ),...,(x l+u ,y l+u)}, l<<u, l+u=s. Every two nodes have an edge, and the edge set is represented by an affinity matrix (A), as shown in formula (3).

[0050]

[0051] Where i, j∈{1, 2, ..., s}, σ>0, represents the bandwidth parameter of the Gaussian function. Define the feature mapping function f, and the classification rule is: y i =sign(f(x i )),y i ∈{-1, 1}. This problem is transformed into minimizing the energy function shown in formula (4).

[0052]

[0053] in, f l =(f(x1);f(x2);...;f(x l )), represents the mapping result of the labeled sample, f u =(f(x l+1 ); f(x l+2 );…;f(x l+u )), represents the mapping result of unlabeled samples. D=diag(d1,d2,…,d l+u ) is a diagonal matrix, where Further represented by a block matrix, The above formula is rewritten as shown in formula (5).

[0054]

[0055] make The result is shown in formula (6):

[0056] f u =(D uu -A uu ) -1 A ul f l (6)

[0057] Using f u The unlabeled samples can be classified. In practical applications, an iterative algorithm is usually used to solve the problem. The steps are:

[0058] 1) Define the non-negative matrix F s×2 ,initialization

[0059] 2) Construct the propagation matrix based on A in

[0060] 3) Update F using formula (7), where λ∈(0,1) is the compromise coefficient, which is set to 0.2 in this paper.

[0061] F(t)=λPF(t-1)+(1-λ)F(0) (7)

[0062] 4) Repeat step 3) until F(t) converges, denoted as use It is enough to classify the unlabeled samples.

[0063] In one or more embodiments, the wireless measuring device in step S40 includes a temperature sensor and a wireless communication module, the temperature sensor is used to extend into the water outlet pipe of the stator winding and measure the real-time water outlet temperature sequence data, and the wireless communication module is used to send the real-time water outlet temperature sequence data to the outside world. Exemplarily, the temperature sensor uses a DS18B20 temperature sensor, the VCC of the temperature sensor is connected to the 5V of the control mainboard Arduino, the GND is connected to the GND of the control mainboard Arduino, the data pin is connected to a digital pin of the Arduino, and is connected to 5V through a 4.7kΩ pull-up resistor. When collecting temperature, the temperature sensor needs to be encapsulated in a probe and inserted into the cold water pipe. Exemplarily, the wireless communication module uses an ESP8266WiFi module, and the connection method is: VCC is connected to 3.3V of Arduino, GND is connected to GND of Arduino, TX is connected to RX of Arduino, and RX is connected to TX of Arduino.

[0064] Further, in step S40, the time length of the real-time outlet water temperature sequence data obtained by the wireless measuring device is greater than or equal to the predetermined time window. After the real-time outlet water temperature sequence data is feature extracted, the extracted temperature difference feature real-time data and trend difference feature real-time data are input into the thermal fault diagnosis model to obtain the thermal fault diagnosis result. If the result is 1, it is diagnosed as abnormal, and an alarm needs to be issued and an alarm signal is sent. Otherwise, it is normal and real-time diagnosis continues.

[0065] The stator winding thermal fault diagnosis method provided in this embodiment uses a graph semi-supervised method to train a thermal fault diagnosis model based on a small number of labeled samples and a large number of unlabeled samples, thereby reducing the data labeling cost, and using wireless temperature measurement to solve the problem of difficulty in measuring the temperature of the generator stator winding cooling water system in operation, and greatly alleviates the problem of difficult on-site wiring.

[0066] The specific implementation methods of the present application are described in detail above. Although some embodiments have been shown and described, those skilled in the art should understand that these embodiments can be modified and improved without departing from the principles and spirit of the present application whose scope is defined by the claims and their equivalents. These modifications and improvements should also be within the scope of protection of the present application.

Claims

1. A method for diagnosing thermal faults of stator windings, characterized in that: The stator winding thermal fault diagnosis method comprises: Perform feature extraction on the acquired historical water outlet temperature series data to obtain temperature difference feature historical data and trend difference feature historical data to form a data sample set; Marking some samples of the data sample set and not marking the remaining samples; The labeled samples and unlabeled samples are used as training samples, and a thermal fault diagnosis model is obtained by training based on the graph semi-supervised method. The real-time outlet water temperature sequence data of the stator winding under the running state is obtained by using a wireless measuring device, and the features of the real-time outlet water temperature sequence data are extracted to obtain the real-time data of temperature difference features and the real-time data of trend difference features; The temperature difference characteristic real-time data and the trend difference characteristic real-time data are input into the thermal fault diagnosis model to obtain a thermal fault diagnosis result.

2. The stator winding thermal fault diagnosis method according to claim 1, characterized in that: The method of extracting features from the acquired historical water outlet temperature sequence data to obtain temperature difference feature historical data includes: Extracting data segments of a plurality of predetermined time windows from the historical water outlet temperature series data, each of the data segments comprising a plurality of water outlet temperature measurement point time series; Calculate and obtain the median temperature time series of several outlet water temperature measurement point time series; Calculate the difference between the time series of each outlet water temperature measurement point and the median temperature time series; The mean of each difference sequence is calculated as the temperature difference characteristic historical data.

3. The stator winding thermal fault diagnosis method according to claim 2, characterized in that: The method of extracting features from the acquired historical water outlet temperature sequence data to obtain trend difference feature historical data includes: The trend difference characteristic historical data is calculated based on each difference sequence and the temperature difference characteristic historical data.

4. The stator winding thermal fault diagnosis method according to claim 1, characterized in that: The method of marking some samples of the data sample set and not marking the remaining samples comprises: The abnormal samples in the part of samples are marked as 1, the normal samples in the part of samples are marked as 0, and the remaining samples are all marked as -1.

5. The stator winding thermal fault diagnosis method according to claim 1, characterized in that: The wireless measuring device includes a temperature sensor and a wireless communication module. The temperature sensor is used to extend into the water outlet pipe of the stator winding and measure real-time water outlet temperature sequence data. The wireless communication module is used to send the real-time water outlet temperature sequence data to the outside world.

6. The stator winding thermal fault diagnosis method according to claim 2, characterized in that: The time length of the real-time outlet water temperature sequence data is greater than or equal to a predetermined time window.

7. The stator winding thermal fault diagnosis method according to claim 1, characterized in that: The stator winding thermal fault diagnosis method further includes: When the thermal fault diagnosis result is abnormal, an alarm signal is issued.

Images