Water wheel generator thrust bearing bush temperature rise abnormity detection method based on One-Class SVM

Abstract

The invention discloses a hydraulic generator thrust bearing bush temperature rise anomaly detection method based on One-Class SVM. The method comprises the following steps that S1, normal temperature rise historical data of a hydraulic generator thrust bearing bush and a hydraulic turbine set start-stop time table are obtained; s2, missing value processing and abnormal value processing are carried out on the temperature rise data of the thrust bearing bush to obtain normal temperature rise historical data of the thrust bearing bush, and a temperature feature engineering library is established for the processed temperature data of the thrust bearing bush; s3, inputting the temperature characteristic indexes established in the step S2 into a One-Class SVM (Support Vector Machine) model to train an anomaly detection model; s4, detecting the abnormal condition of the new data of the temperature rise of the thrust bearing bush by using the trained anomaly detection model, and judging whether the new data of the temperature rise of the thrust bearing bush is abnormal or not according to the One-Class SVM model decision function value; according to the method, on the premise that a large amount of historical normal temperature rise data is known, the temperature rise data feature engineering library is established, and the model for detecting the temperature rise anomaly of the thrust bearing bush is trained.

Description

Technical Field

[0001] The invention relates to the technical field of temperature rise detection of a thrust bearing of a hydro-turbine generator set, and in particular to a method for detecting abnormal temperature rise of a thrust bearing of a hydro-turbine generator set based on a One-Class SVM. Background Art

[0002] The thrust bearing of a turbine is one of the important components of a turbine, and its normal operation has a significant impact on the performance and life of the turbine. Abnormal temperature rise may be caused by factors such as lubrication problems, bearing wear, and workload changes, which may lead to equipment failure or performance degradation; real-time monitoring of abnormal temperature rise of the thrust bearing can help operation and maintenance personnel take maintenance measures in time to prevent further damage and improve the reliability of the turbine. Therefore, it is particularly important to monitor the operating status of the thrust bearing of the turbine in real time and detect abnormalities.

[0003] At present, there are more and more studies on anomaly detection algorithms, such as anomaly detection methods based on multi-domain feature extraction, anomaly data detection algorithms based on improved ant colony algorithms, etc. Among them, the anomaly detection method based on multi-domain feature extraction uses the relevant vector machine to select and reduce the dimension of data features, and constructs a classifier to detect abnormal data, but this detection method can only provide a single prediction value in the prediction, which makes the anomaly detection result error large; and the anomaly data detection algorithm based on the improved ant colony algorithm, it extracts and clusters the abnormal data feature quantity of the improved ant colony algorithm, and determines the abnormal data detection value, but the data feature extraction effect of this method is poor, which leads to low accuracy of data anomaly detection. Summary of the invention

[0004] The purpose of the present invention is to overcome the above-mentioned shortcomings and provide a method for detecting temperature rise anomalies of thrust bearings of hydro-turbine generators based on One-Class SVM. Based on a large amount of known historical normal temperature rise data, a temperature rise data feature engineering library is established to train a model for detecting temperature rise anomalies of thrust bearings.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: a method for detecting abnormal temperature rise of thrust bearing of a hydro-turbine generator based on One-Class SVM, comprising the following steps:

[0006] S1. Obtain the historical data of normal temperature rise of the thrust bearing of the hydro-turbine generator and the start and stop schedule of the hydro-turbine unit;

[0007] S2. Perform missing value processing and abnormal value processing operations on the thrust bearing temperature rise data to obtain the normal temperature rise history data of the thrust bearing, and establish a temperature feature engineering library for the processed thrust bearing temperature data;

[0008] S3, input the temperature characteristic index established in S2 into the One-Class SVM model to train the anomaly detection model;

[0009] S4. Use the above-trained anomaly detection model to detect the anomaly of the new data of thrust bearing temperature rise, and judge whether the new data of thrust bearing temperature rise is abnormal based on the decision function value of the One-Class SVM model.

[0010] Furthermore, the step S1 specifically includes:

[0011] The real-time temperature rise data of the thrust bearing is obtained through the temperature sensor installed on the thrust bearing seat of the turbine, and the temperature rise data is recorded and stored using the data acquisition system. After the data is acquired, the historical data is processed and cleaned to ensure that the acquired temperature rise data are all normal temperature rise change data of the thrust bearing. The data processing and cleaning include the steps of removing outliers and filling in missing data. Finally, data archiving and backup are established to preserve the data for a long time.

[0012] Furthermore, the step S2 specifically includes:

[0013] (1) Taking each startup time of the hydro-turbine generator as the starting time of the thrust bearing temperature rise, and the working time required for the thrust bearing temperature rise to be stable as the total segmentation length, the temperature rise data of the thrust bearing of the hydro-turbine generator is segmented according to a preset window length and a preset moving length to obtain a plurality of temperature rise data windows;

[0014] (2) Establish a temperature feature engineering library for each data window to obtain the temperature mean, standard deviation, maximum and minimum values, and average change rate of each data window.

[0015] Furthermore, the temperature characteristic index is expressed as:

[0016] Average temperature of data window:

[0017]

[0018] Among them, Z i (Ave_temp) is the average temperature change of the i-th data window, n is the number of data windows, (x1, x2, ..., x m ) is the temperature data in each data window, Z i =(x1,x2,...,x m ) i ;

[0019] Data window temperature standard deviation:

[0020]

[0021] Among them, Zi (Std_temp) is the standard deviation of temperature variation in the i-th data window;

[0022] Average rate of change of temperature in the data window:

[0023]

[0024] Among them, Z i (Ave_rate_temp) is the average rate of change of temperature data in the i-th data window, and t is the time interval between adjacent temperature rise changes.

[0025] Furthermore, the step S3 specifically includes:

[0026] (1) Data mapping: Map the temperature feature engineering indicators established based on the historical data of normal temperature rise of the thrust bearing into the high-dimensional feature space so that the feature engineering indicators obtained from the normal temperature rise data can be surrounded by a hyperplane; this hyperplane is the decision boundary of the normal temperature rise;

[0027] (2) Finding the optimal hyperplane: By maximizing the distance between the hyperplane and the normal data, find an optimal segmentation hyperplane so that the abnormal points are as far away from the hyperplane as possible;

[0028] The One-Class SVM anomaly detection model is trained through the above steps.

[0029] Furthermore, the step S4 specifically includes:

[0030] For the newly collected thrust bearing temperature rise data, the corresponding temperature characteristic index is calculated, and then the established One-Class SVM anomaly detection model is used to detect it; finally, the distance between it and the hyperplane is calculated through the decision function to determine whether the data point is abnormal.

[0031] The decision function of One-Class SVM is calculated as follows:

[0032] D(x) = sign(w·xb);

[0033] Where D(x) is the decision function value of the data point; w is the normal vector of the hyperplane; x is the eigenvector of the data point; b is the bias;

[0034] During the training process, the goal of One-Class SVM is to minimize the objective function:

[0035]

[0036] In the formula, ||w|| 2is the L2 norm of the weight vector; v represents the proportion of normal samples; n is the number of normal samples; D(x i ) is the training sample x i The decision function value of One-Class SVM; by minimizing the objective function, One-Class SVM finds a hyperplane that can wrap normal samples; in the decision function, the positive and negative signs of D(x) reflect the position of the data point x relative to this hyperplane, thereby judging whether it is abnormal; positive values ​​indicate normality, and negative values ​​indicate abnormality.

[0037] Beneficial effects of the present invention:

[0038] 1. Based on a large amount of known historical normal temperature rise data, the present invention can establish a temperature rise data feature engineering library and train a model for detecting temperature rise anomalies of thrust bearings.

[0039] 2. Based on the historical normal temperature rise data of the thrust bearing of the hydro-turbine generator set, the present invention realizes the detection of abnormal temperature rise of the thrust bearing by using data preprocessing and constructing a temperature feature engineering library method, providing a more important detection method for abnormal temperature rise detection of the thrust bearing of the hydro-turbine generator set.

[0040] 3. The present invention combines the algorithm model in the field of anomaly detection algorithms, innovatively proposes a temperature rise anomaly detection model for the thrust bearing of a hydro-turbine generator, and obtains a single-classification support vector machine anomaly detection model by calculating various temperature characteristic indicators for training. Compared with traditional detection methods, this model has the advantages of being easier to implement and more interpretable.

[0041] 4. The method of the present invention realizes abnormal detection of bearing temperature rise by establishing a feature engineering library of bearing temperature rise data and calculating the decision function value of each group of temperature rise data, thereby effectively solving the problems of poor data feature extraction effect and low detection accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a schematic flow diagram of the method of the present invention;

[0043] Figure 2 It is a temperature feature engineering indicator established by the normal temperature rise data based on the One-Class SVM anomaly detection model provided by the present invention;

[0044] Figure 3 It is a visualization diagram of the decision function value of normal temperature rise data based on the One-Class SVM anomaly detection model provided by the present invention;

[0045] Figure 4 It is the anomaly detection result of new data of temperature rise of thrust bearing of hydro-turbine generator set based on One-Class SVM anomaly detection model provided by the present invention;

[0046] Figure 5 It is a visualization diagram of the decision function value of the new data anomaly detection of the temperature rise of the thrust bearing of a hydro-turbine generator set based on the One-Class SVM anomaly detection model provided by the present invention. DETAILED DESCRIPTION

[0047] In order to make the purpose, scheme and advantages of the embodiments of the present invention more clear, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0048] Figure 1 The present invention is a schematic flow chart of a method for detecting abnormal temperature rise of a turbine thrust bearing based on a one-class support vector machine (One-Class SVM) according to an embodiment of the present invention, which comprises the following steps:

[0049] S1. Obtain normal temperature rise historical data and turbine start-stop schedule. The real-time temperature rise data of the thrust bearing is obtained through the temperature sensor installed on the turbine thrust bearing seat, and the temperature rise data is recorded and stored using the data acquisition system. After the data is acquired, the historical data is processed and cleaned to ensure that the acquired temperature rise data are all normal temperature rise change data of the thrust bearing. Data processing and cleaning include steps such as removing outliers and filling missing data. Finally, data archiving and backup are established to preserve the data for a long time.

[0050] The initial data source is a dataset of measured temperature rises of thrust bearings of hydropower units in a domestic hydropower station. This dataset uses 279 sets of measured normal temperature rise data of three thrust bearings of the unit as training samples. The test set is processed as follows: some abnormal samples are randomly selected from the provided dataset of measured temperature rises of thrust bearings of hydropower units and mixed into the training data to form new input sample data, and the group to which the abnormal samples belong is recorded.

[0051] S2. Process the missing values ​​and abnormal values ​​of the thrust bearing temperature rise data to obtain the normal temperature rise history data of the thrust bearing, and establish a temperature feature engineering library (such as Figure 2 The engineering library consists of the thrust bearing temperature average value, standard deviation, maximum and minimum values, and average rate of change.

[0052] (1) Taking each startup time of the hydro-turbine generator as the starting time of the thrust bearing temperature rise, and the working time required for the thrust bearing temperature rise to be stable as the total segmentation length, the temperature rise data of the thrust bearing of the hydro-turbine generator is segmented according to a preset window length and a preset moving length to obtain a plurality of temperature rise data windows;

[0053] (2) Establish a temperature feature engineering library for each data window to obtain the temperature mean, standard deviation, maximum and minimum values, and average change rate of each data window.

[0054] The temperature feature engineering library is expressed as:

[0055] Average value of temperature rise data in data window:

[0056]

[0057] Among them, Z i (Ave_temp) is the average temperature change of the i-th data window, n is the number of data windows, (x1, x2, ..., x m ) is the temperature data in each data window, Z i =(x1,x2,...,x m ) i .

[0058] Standard deviation of temperature rise data in data window:

[0059]

[0060] Among them, Z i (Std_temp) is the standard deviation of temperature variation in the i-th data window.

[0061] Average change rate of temperature rise data in the data window:

[0062]

[0063] Among them, Z i (Ave_rate_temp) is the average rate of change of temperature data in the i-th data window, and t is the time interval between adjacent temperature rise changes.

[0064] S3, input the temperature characteristic index established in S2 into the One-Class SVM model to train the anomaly detection model;

[0065] (1) Data mapping: The temperature feature engineering index established based on the historical data of normal temperature rise of the thrust bearing is mapped to the high-dimensional feature space, so that the feature engineering index obtained from the normal temperature rise data can be surrounded by a hyperplane. This hyperplane is the decision boundary of the normal temperature rise (e.g. Figure 3 shown). Figure 3 The red solid line in the middle is the contour line determined by the decision function value of each set of temperature rise data in the temperature feature engineering library, so as to intuitively obtain the range of the decision function value; the green dotted line is the two-dimensional threshold line corresponding to the optimal hyperplane. When the decision function value is lower than the threshold line, it is detected as an abnormality.

[0066] (2) Finding the optimal hyperplane: By maximizing the distance between the hyperplane and the normal data, an optimal segmentation hyperplane is found so that the abnormal points are as far away from the hyperplane as possible.

[0067] The one-class support vector machine (One-Class SVM) anomaly detection model is trained through the above steps.

[0068] S4. Use the above-trained anomaly detection model to detect the anomaly of the new data of thrust bearing temperature rise and determine whether the new data of thrust bearing temperature rise is abnormal based on the decision function value of the One-Class SVM model.

[0069] For the newly collected thrust bearing temperature rise data, the corresponding temperature characteristic index is calculated, and then the established One-Class SVM anomaly detection model is used to detect it (such as Figure 4 As shown); finally, the distance between the data point and the hyperplane is calculated through the decision function to determine whether the data point is abnormal.

[0070] The decision function of One-Class SVM is calculated as follows:

[0071] D(x)=sign(w·xb)

[0072] Where D(x) is the decision function value of the data point; w is the normal vector of the hyperplane; x is the eigenvector of the data point; and b is the bias.

[0073] During the training process, the goal of One-Class SVM is to minimize the objective function:

[0074]

[0075] In the formula, ||w|| 2 is the L2 norm of the weight vector; v represents the proportion of normal samples; n is the number of normal samples; D(x i ) is the training sample x i By minimizing the objective function, One-Class SVM finds a hyperplane that can wrap normal samples. In the decision function, the sign of D(x) reflects the position of the data point x relative to the hyperplane, thereby determining whether it is abnormal (e.g. Figure 5 As shown in the figure). The red solid line in the figure is the contour line of the decision function value of each group of temperature rise data in the temperature feature engineering library; the green dotted line is the two-dimensional threshold line corresponding to the optimal hyperplane. The decision function values ​​greater than the threshold line are all normal temperature rise data, otherwise they are all abnormal. The serial number in the figure is the group containing abnormal temperature rise sample data.

[0076] The One-Class SVM anomaly detection model is obtained by training the initial sample data, and the trained model is used to detect the new input sample data. Since the abnormal sample groups mixed into the training initial sample data have been marked, the detection effect of the model can be directly judged by observing the results of the One-Class SVM anomaly detection model. In the embodiment of the present invention, the groups of abnormal temperature rise data mixed into the new input sample data after the One-Class SVM anomaly detection are all detected, which is sufficient to verify the effectiveness of the proposed anomaly detection model in the abnormal detection of temperature rise of the thrust bearing of the hydropower unit.

[0077] The above embodiments are only preferred technical solutions of the present invention and should not be regarded as limiting the present invention. The protection scope of the present invention shall be the technical solutions recorded in the claims, including equivalent replacement solutions of the technical features in the technical solutions recorded in the claims. That is, equivalent replacement improvements within this scope are also within the protection scope of the present invention.

Claims

1. A method for detecting abnormal temperature rise of thrust bearing of hydro-turbine generator based on One-Class SVM. Features: The following steps are involved: S1. Obtain the historical data of normal temperature rise of the thrust bearing of the turbine generator and the start and stop schedule of the turbine unit; S2. Perform missing value processing and abnormal value processing operations on the thrust bearing temperature rise data to obtain the normal temperature rise history data of the thrust bearing, and establish a temperature feature engineering library for the processed thrust bearing temperature data; S3, input the temperature characteristic index established in S2 into the One-Class SVM model to train the anomaly detection model; S4. Use the above-trained anomaly detection model to detect the anomaly of the new data of thrust bearing temperature rise, and judge whether the new data of thrust bearing temperature rise is abnormal based on the decision function value of the One-Class SVM model.

2. According to the method for detecting abnormal temperature rise of thrust bearing of a hydro-turbine generator based on One-Class SVM in claim 1, Features: The step S1 specifically includes: The real-time temperature rise data of the thrust bearing is obtained through the temperature sensor installed on the thrust bearing seat of the turbine, and the temperature rise data is recorded and stored using the data acquisition system. After the data is acquired, the historical data is processed and cleaned to ensure that the acquired temperature rise data are all normal temperature rise change data of the thrust bearing. The data processing and cleaning include the steps of removing outliers and filling in missing data. Finally, data archiving and backup are established to preserve the data for a long time.

3. According to the method for detecting abnormal temperature rise of thrust bearing of a hydro-turbine generator based on One-Class SVM in claim 1, Features: The step S2 specifically includes: (1) Taking each startup time of the hydro-turbine generator as the starting time of the thrust bearing temperature rise, and the working time required for the thrust bearing temperature rise to be stable as the total segmentation length, the temperature rise data of the thrust bearing of the hydro-turbine generator is segmented according to a preset window length and a preset moving length to obtain a plurality of temperature rise data windows; (2) Establish a temperature feature engineering library for each data window to obtain the temperature mean, standard deviation, maximum and minimum values, and average change rate of each data window.

4. According to claim 3, a method for detecting abnormal temperature rise of a thrust bearing of a hydro-turbine generator based on One-Class SVM, Features: The temperature characteristic index is expressed as: Average temperature of data window: Among them, Z i (Ave_temp) is the average temperature change of the i-th data window, n is the number of data windows, (x 1 ,x 2 ,...,x m ) is the temperature data in each data window, Z i =(x 1 ,x 2 ,...,x m ) i ; Data window temperature standard deviation: Among them, Z i (Std_temp) is the standard deviation of temperature variation in the i-th data window; Average rate of change of temperature in the data window: Among them, Z i (Ave_rate_temp) is the average rate of change of temperature data in the i-th data window, and t is the time interval between adjacent temperature rise changes.

5. According to the method for detecting abnormal temperature rise of thrust bearing of a hydro-turbine generator based on One-Class SVM in claim 1, Features: The step S3 specifically includes: (1) Data mapping: Map the temperature feature engineering indicators established based on the historical data of normal temperature rise of the thrust bearing into the high-dimensional feature space so that the feature engineering indicators obtained from the normal temperature rise data can be surrounded by a hyperplane; this hyperplane is the decision boundary of the normal temperature rise; (2) Finding the optimal hyperplane: By maximizing the distance between the hyperplane and the normal data, find an optimal segmentation hyperplane so that the abnormal points are as far away from the hyperplane as possible; The One-Class SVM anomaly detection model is trained through the above steps.

6. According to the method for detecting abnormal temperature rise of thrust bearing of a hydro-turbine generator based on One-Class SVM in claim 1, Features: The step S4 specifically includes: For the newly collected thrust bearing temperature rise data, the corresponding temperature characteristic index is calculated, and then the established One-Class SVM anomaly detection model is used to detect it; finally, the distance between it and the hyperplane is calculated through the decision function to determine whether the data point is abnormal. The decision function of One-Class SVM is calculated as follows: D(x) = sign(w·xb); Where D(x) is the decision function value of the data point; w is the normal vector of the hyperplane; x is the eigenvector of the data point; b is the bias; During the training process, the goal of One-Class SVM is to minimize the objective function: In the formula, ||w|| 2 is the L2 norm of the weight vector; v represents the proportion of normal samples; n is the number of normal samples; D(x i ) is the training sample x i The decision function value of One-Class SVM; by minimizing the objective function, One-Class SVM finds a hyperplane that can wrap normal samples; in the decision function, the positive and negative signs of D(x) reflect the position of the data point x relative to this hyperplane, thereby judging whether it is abnormal; positive values ​​indicate normality, and negative values ​​indicate abnormality.

Images