Anomaly Detection Method for Start-up and Shutdown Processes of Hydropower Units Based on Dynamic Time Warping

Abstract

The present invention discloses a method for detecting abnormalities in the start-up and shutdown processes of a hydroelectric generator set based on dynamic time regularization, and relates to the field of electronic technology. The present invention is based on characteristic indicators such as vibration swing, pressure, temperature, air gap, etc. during the start-up and shutdown processes of a hydroelectric generator set, and based on a dynamic time regularization method, combined with the operating characteristics of the unit, to achieve identification and detection of whether the unit status is abnormal during the start-up and shutdown processes, thereby achieving early warning of the health status of the unit.

Description

Technical Field

[0001] The present invention relates to the field of electronic technology, and in particular to a method for detecting abnormalities in the start-up and shutdown processes of a hydropower unit based on dynamic time regularization. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and may not constitute prior art.

[0003] The startup and shutdown processes of the hydro-turbine generator set are transition processes that must be experienced during the normal operation of the hydro-turbine generator set. During the above transition process, the unit needs to perform relatively complex actions. Taking the startup process as an example, the startup process needs to perform actions such as opening the guide vanes, increasing the speed, switching on the excitation, and synchronous grid connection. During the above process, the structural loads of the various components of the unit change dramatically, the operating conditions change complexly, and the probability of unit failure is relatively high.

[0004] During the normal startup process, the unit operates at a small opening, and the hydraulic factor has little effect on the unit vibration. The main factors affecting the unit's vibration, temperature, air gap and other characteristics are the unit's mechanical imbalance factors, electromagnetic tension imbalance factors, such as mass imbalance, axis bending, adjustment of the gap between the guide bearings of each part of the unit, non-roundness of the stator and rotor air gap, stator and rotor eccentricity and other factors. During the shutdown process, the unit's guide vanes are completely closed, and the main factors affecting the unit's operating status are the unit's mechanical imbalance factors and electromagnetic tension imbalance factors. Because the hydraulic force has a small impact, the discreteness and randomness of the vibration, swing, and air gap characteristic values ​​are reduced. For the same unit, when the unit's health state and control law have not changed, its vibration and air gap characteristics are also convergent and consistent. This lays the foundation for realizing abnormal detection during the startup and shutdown process.

[0005] In addition, since the no-load opening during the unit startup process is related to the unit's working water head, and the size of the no-load opening will affect the changes in the unit's swing and vibration, the change pattern of the vibration, swing and other states during the startup process is related to the working water head when the unit is started. Another factor is the influence of temperature. When the unit is started in cold and hot states, the differences in the tile temperature, oil temperature, and stator core temperature will directly affect the unit's swing, vibration measurement value, and the change pattern of the stator-rotor air gap. Therefore, the influence of temperature factors on the swing, vibration, and air gap characteristics needs to be considered in the abnormal detection of the startup process.

[0006] Therefore, it can be considered that for the same unit, the measured values ​​of unit vibration, swing, air gap, etc. during the start-up and shutdown process are directly related to the unit working water head, unit temperature (tile temperature, oil temperature, stator core temperature), etc. However, under the same working water head and unit temperature conditions, when the unit health status and control law have not changed, its vibration and air gap characteristics are also convergent and approaching consistency.

[0007] In summary, during the above transition process, the unit needs to perform relatively complex actions. During the process, the combined effects of mechanical force, electromagnetic force, fluid force, etc. on the unit change in a complex manner, and the probability of failure is relatively high. Therefore, it is particularly important to detect abnormalities in the unit's operating status during the start-up and shutdown process and identify whether the unit's health status has changed for the safe operation of the unit. Summary of the invention

[0008] The purpose of the present invention is to provide a method for detecting abnormalities in the start-up and shutdown processes of hydropower generator sets based on dynamic time regularization in response to the problems existing in the prior art. According to characteristic indicators such as vibration swing, pressure, temperature, air gap, etc. during the start-up and shutdown processes of the hydropower generator sets, based on the dynamic time regularization method and combined with the operating characteristics of the unit, it is possible to identify and detect whether the unit status is abnormal during the start-up and shutdown processes, thereby achieving early warning of the health status of the unit.

[0009] The technical solution of the present invention is as follows:

[0010] The abnormality detection method of the start-up and shutdown process of a hydropower unit based on dynamic time regularization includes:

[0011] Step S1: According to the historical health data samples of the startup process and the historical health data samples of the shutdown process, classification statistics and training are performed to obtain the benchmark time series of each characteristic indicator of the startup process and the shutdown process and the distance deviation threshold;

[0012] Step S2: Based on the benchmark time series of each characteristic index of the startup process and the shutdown process obtained by training in step S1 and the distance deviation threshold, abnormality detection is performed on the characteristic index of the startup process and the characteristic index of the shutdown process obtained by real-time measurement.

[0013] Furthermore, in the abnormality detection during the startup process, the step S1 includes:

[0014] Step S101: constructing a training set of characteristic indicators of the startup process under the same working conditions with working water head, guide bearing bush temperature, and stator core temperature as constraint conditions;

[0015] Step S102: iteratively calculate each characteristic indicator sample data in the original characteristic indicator set of the startup process under each working condition one by one to obtain the startup process benchmark time series and the maximum allowable distance deviation threshold of the corresponding characteristic indicator.

[0016] Furthermore, the step S101 includes:

[0017] set up is the working water head, is the upper average watt temperature, is the average watt temperature of the lower conductor, is the average watt temperature of the water conductor, is the average temperature of the core;

[0018] Extract the original feature data of each feature index of each boot process from the historical boot process health data samples, and construct the original eigenvalue matrix of each boot process ;

[0019] From the original eigenvalue matrix Calculate the operating characteristic value of the startup process:

[0020] Average working head : ;

[0021] Upper average watt temperature : ;

[0022] Average watt temperature of lower conductor : ;

[0023] Average water temperature : ;

[0024] Average core temperature : ;

[0025] Where:

[0026] For the The boot process;

[0027] is the total length of the time series;

[0028] Working water head at each moment;

[0029] is the measuring point;

[0030] The number of upper guide tile temperature measurement points;

[0031] is the temperature of each upper guide tile at each moment;

[0032] The number of temperature measurement points for the lower guide bearing;

[0033] is the temperature of each lower guide tile at each moment;

[0034] The number of water-conducting tile temperature measurement points;

[0035] is the temperature of each water guide tile at each moment;

[0036] is the number of core temperature measurement points;

[0037] is the core temperature at each moment;

[0038] by The allowable variation is within the range of ±1.0m. , , , The allowable variation is ±1.0℃. According to the same working condition , , , , Perform classification statistics to obtain the working condition The following is the original characteristic indicator set of the boot process: ;in, is the average water head under this condition, is the average watt temperature of the upper conductor under this condition, is the average watt temperature of the lower conductor under this condition, is the average temperature of the water conductor under this condition, is the average temperature of the core under this condition, For working conditions The number of boot processes under

[0039] After traversing all historical startup process health data samples, the original feature index set of the startup process under multiple working conditions is obtained:

[0040] , corresponding to the working condition ;

[0041] , corresponding to the working condition ;

[0042] …,

[0043] , corresponding to the working condition ;

[0044] Among them, each original feature index set For working conditions The original characteristic index set of multiple boot processes under For working conditions The number of boot processes under is the total number of working conditions obtained through traversal, ;

[0045] Original eigenvalue matrix ,include:

[0046]

[0047] in:

[0048] is the original discrete time series;

[0049] is the number of eigenvalues ​​in the characteristic indicator time series;

[0050] is the total number of original feature indicators.

[0051] Furthermore, the step S102 includes:

[0052] Step S1021: Setting ;

[0053] Step S1022: Traverse all characteristic indicators in the working condition The data under the condition are used to obtain the benchmark time series of all characteristic indicators and the maximum allowable distance deviation threshold;

[0054] Step S1023: construct working conditions The benchmark time series matrix under , and the maximum allowed distance deviation threshold vector ;

[0055] Step S1024: ,if , repeat step S1021-step S1024; otherwise, it means that the data under all working conditions have been traversed; output the benchmark time series matrix set under multiple working conditions , the maximum allowed distance deviation vector set , and their corresponding working condition quantity sets .

[0056] Furthermore, the step S2 comprises:

[0057] Step S201: Using the working water head, guide bearing bush temperature, and stator core temperature measured in real time during the startup process as working condition constraints, search for a benchmark time series matrix matching the working condition. and the maximum allowable distance deviation threshold vector ;

[0058] Step S202: performing anomaly detection on the time series of characteristic indicators of the startup process measured in real time.

[0059] Furthermore, the step S201 includes:

[0060] Step S2011: Calculate the operating characteristic value of the real-time startup process from the time series record of the startup process measured in real time: average working water head , upper average watt temperature , average watt temperature of lower conductor , average water conduction temperature , Average core temperature ;

[0061] Step S2012: According to , , , , Search for the benchmark time series matrix and the maximum distance deviation vector that meet the operating conditions.

[0062] Furthermore, the step S2012 includes:

[0063] Step S20121: Setting ;

[0064] Step S20122: Compare the working conditions and obtain the first Working conditions Next , , , and , if it satisfies:

[0065] ,

[0066] and ,

[0067] and ,

[0068] and ,

[0069] and , then the working condition search is successful, and step S202 is executed;

[0070] Otherwise, let , search again; if , that is, the data under all working conditions have been traversed, the working condition search fails, and the anomaly detection is terminated;

[0071] After the working condition search is successful, = , = , execute step S202.

[0072] Furthermore, the step S202 includes:

[0073] Step S2021: Setting ;

[0074] Step S2022: Obtain the The distance deviation of each characteristic indicator from the benchmark time series;

[0075] Step S2023: performing distance deviation detection;

[0076] Step S2024: ,if , that is, all feature indicator data have been traversed, the anomaly check has been completed, and the anomaly detection results are output.

[0077] Furthermore, the step S2022 includes:

[0078] Step a: Set It is represented as a time series of characteristic indicators measured in real time during the startup process; Extract The benchmark time series of characteristic indicators ,in, Characteristic index The number of eigenvalues ​​in the benchmark time series;

[0079] Step b: From The time when the generator output switch is closed is retrieved as the end point, and the time when the speed reaches 6r / min is retrieved as the starting point. The truncation process is performed to form a new time series. ,in, Characteristic index The number of eigenvalues ​​after truncation of the benchmark time series; The time when the generator output switch is closed is retrieved as the end point, and the time when the speed reaches 6r / min is retrieved as the starting point to form a new time series. in, Characteristic index The number of eigenvalues ​​after truncation of the time series;

[0080] Step c: Calculate using dynamic time warping algorithm and The cumulative distance matrix ,according to search , The shortest distance path between coordinates: in, The number of eigenvalues ​​of the new time series after being warped by the dynamic time warping algorithm is calculated; then according to Reconstruction of the value Get a new time series ,according to Reconstruction of the value Get a new time series ;

[0081] Step d: Calculation and Maximum distance deviation :

[0082]

[0083] in:

[0084] Indicates from Find the maximum value among the values, that is, Characteristic index The maximum distance deviation during the entire startup process;

[0085] The step S2023 includes:

[0086] According to judgment Feature indicators stored in Maximum allowable distance deviation , judging characteristic index Is it abnormal, that is:

[0087] if , then the characteristic index Abnormal, otherwise, characteristic indicators normal.

[0088] Furthermore, in the abnormality detection during the shutdown process, the same feature indicator set as that for the abnormality detection during the startup process is used, and the working water head, guide bearing temperature, and stator core temperature are also used as operating condition constraints; the specific calculation steps are the same as those for the abnormality detection during the startup process; the difference lies in the inconsistent interception methods for the feature indicator time series during the training and abnormality detection processes; the startup process uses the closing moment of the generator output switch as the end point, and the moment when the speed enters 6r / min as the starting point for truncation processing; while the shutdown process uses the disconnection of the generator output switch as the starting point, and the moment when the speed drops below 6r / min as the end point.

[0089] Compared with the prior art, the present invention has the following beneficial effects:

[0090] The present invention is based on the characteristic indicators of the start-up and shutdown process of the hydro-turbine generator set, such as vibration swing, pressure, temperature, air gap, etc., and on the basis of the dynamic time warping method, combined with the operating characteristics of the unit, to realize the identification and detection of whether the unit status is abnormal during the start-up and shutdown process, thereby realizing early warning of the health status of the unit. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] Figure 1 It is a flow chart of the abnormal detection method of the start-up and shutdown process of the hydropower unit based on dynamic time warping;

[0092] Figure 2 It is a characteristic sequence diagram of two original speech signals;

[0093] Figure 3 is the cumulative distance matrix diagram;

[0094] Figure 4 It is a characteristic sequence diagram of two speech signals after regularization;

[0095] Figure 5 This is the curve of the 1X amplitude change of the X-axis vibration of the upper frame of a certain unit during two startup processes;

[0096] Figure 6 This is the 1X amplitude change curve of the upper frame X-axis vibration during the two startup processes after DTW processing;

[0097] Figure 7 Flowchart for calculating the characteristic index benchmark time series and distance deviation threshold;

[0098] Figure 8a The upper part of the flowchart is the abnormal detection of characteristic indicators of the startup process measured in real time;

[0099] Figure 8b The lower part of the flowchart for abnormal detection of characteristic indicators of the startup process measured in real time. DETAILED DESCRIPTION

[0100] It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0101] The features and performance of the present invention are further described in detail below in conjunction with the embodiments.

[0102] First of all, it should be explained that the basic algorithm adopted by the present invention is the Dynamic Time Warping (DTW) algorithm.

[0103] The Dynamic Time Warping (DTW) algorithm is a machine learning algorithm that is often used to detect the similarity between two speech sounds. Since the length of each letter in each speech is different, the two speech sounds may not completely match. The core method of the dynamic time warping algorithm is to stretch or compress the speech sounds so that their shapes and change patterns are aligned as much as possible, so as to facilitate later comparison.

[0104] like Figure 2 As shown, A is a speech signal feature value sequence , B is another speech signal feature value sequence .

[0105] Intuitively, the two sequences have similar change rules, but the local rules are not consistent. After DTW calculation, the cumulative distance matrix is ​​obtained (the present invention uses Euclidean distance), and a path with the shortest distance is searched from the matrix, such as Figure 3 Medium and dark color coordinate part:

[0106] The path coordinates are as follows (A0, B0), (A1, B1), (A1, B2), (A1, B3), (A2, B4), ... (A9, B9), and the path is expanded along these corresponding coordinate points. and ,like Figure 4 As shown, a new regularized time series is formed and :

[0107] ,

[0108] Then the new time series formed has the following characteristics:

[0109] (1) The sequence length is consistent;

[0110] (2) The shape is closest;

[0111] (3) The point-by-point distance is minimal;

[0112] After regularization and Compared to the original and There has been a great improvement in the similarity of changing patterns and the closeness of point-by-point distances.

[0113] In fact, speech signals are time series signals. In other words, the DTW algorithm is actually used to compare the similarity of two time series. Conversely, each characteristic indicator of the start-up and shutdown process of the water machine, such as vibration, swing, speed, etc., is also a time series, and has similar and consistent change patterns under the same head and temperature. However, due to slight differences in the startup time, synchronization time, and grid-connected time of different startup processes, directly using the original eigenvalue sequence for proximity comparison will lead to non-convergence of errors or high discreteness, resulting in invalid comparison. Therefore, the DTW algorithm can be used to regularize and align the eigenvalue sequences of different startup processes, compare the similarity of the change process of different startup and shutdown processes, and detect whether there is an abnormality in the process state.

[0114] For example Figure 5 The above two curves are all data from the time when the generator output switch is closed from 6r / min to the time when the generator output switch is closed. Since the start-up time of the two start-up processes is different, the excitation time is different, and the synchronization time is also different, although the two curves have similar change rules, they are obviously misaligned in time. Directly using the above curves for point-by-point proximity comparison will result in a huge error and invalid judgment.

[0115] Figure 6 The two curves are stretched and aligned after the dynamic time warping process. Figure 6 It can be seen that the fit of the two curves after stretching and alignment has been greatly improved. The processed curves can be evaluated for their closeness by calculating the distance deviation point by point, and then the consistency of the characteristic changes in the two startup processes can be determined.

[0116] Embodiment 1

[0117] In the present invention, based on the collected vibration signals of the frame, top cover, stator base and other parts of the hydro-generator set, the swing signals of each guide bearing part, the air gap signal, the pressure signal, the temperature signal, the speed signal, the generator outlet switch and other signals, the start-up process and the shutdown process, the working water head, and the unit temperature are distinguished, and the dynamic time warping method (DTW) is used to perform path regularization on the characteristic indicators such as vibration, swing, and air gap according to the generator outlet switch and the unit speed to form a consistent change time series; and the start-up and shutdown process sample data collected historically are used to obtain the start-up process and shutdown process under different water heads and different temperatures. The benchmark characteristic indicator change sequence under normal conditions is obtained. Based on this, the maximum distance deviation between the time series and the benchmark sequence under each start-up and shutdown process is calculated, and the allowable distance deviation threshold is constructed based on the 3σ distribution of the maximum distance deviation. Finally, based on the benchmark change sequence and the allowable distance deviation threshold, the dynamic time warping method (DTW) is used to detect the vibration, swing, air gap, speed and other signals of each start-up and shutdown process to determine whether the unit operation state is abnormal. Compared with traditional methods, this method is completely data-driven, relies on historical measurement data for identification, and uses all characteristic indicators of vibration, swing, air gap, and pressure signals. It does not rely on any fault mechanism and has strong adaptability.

[0118] This method performs abnormality detection based on various characteristic indicators obtained by feature extraction calculation based on the vibration, swing, pressure, air gap, temperature, speed, working head, generator outlet switch and other signals collected by the online monitoring system or offline measurement system installed on the hydropower generator set. These characteristic indicators specifically include:

[0119] Table 1 Anomaly detection feature index table

[0120]

[0121] Suppose the original discrete time series of a characteristic index of a startup process is ( , , ),in Represents a certain boot process, in which The total number of boot processes participating in statistical training; Represents the A characteristic indicator involved in anomaly detection, is the total number of feature indicators involved in the detection, such as Represents the peak-to-peak value of the upper X-direction swing, Represents the upper guide X-direction swing clearance value, Represents the 1X amplitude of the upper guide X-direction swing, etc. Represents the characteristic parameter in the first The characteristic value at the moment, this characteristic index has a total of values, , … It contains the time series of characteristic changes corresponding to each moment of the startup process from the unit's 0r / min speed to the unit's grid connection. Construct the process characteristic indicator matrix as follows:

[0122] (Formula 1)

[0123] The above matrix Contains the During the boot process The characteristic index is the entire time series data from the unit's 0r / min speed to the start-up process after the unit is connected to the grid.

[0124] In this method, we first extract characteristic index data of multiple boot processes from the stored health history characteristic value data samples, and construct The startup processes with similar or same working water head, stator core temperature, guide bearing temperature, etc. are constructed into a matrix set under the same working conditions. :

[0125] (Formula 2)

[0126] Then, yes The generator output switch and speed are used as constraints to perform DTW calculations on each characteristic index sequence of each startup process, align the characteristic index time series of different startup processes, and then iterate to obtain the process benchmark time series of each characteristic index to obtain the benchmark time series matrix ;by Based on the calculation of each startup process The maximum distance deviation between the next time series and the benchmark series is constructed by using the 3σ distribution of the maximum distance deviations to construct the allowable distance deviation threshold.

[0127] Finally, based on the benchmark change sequence and the allowable distance deviation threshold, the dynamic time warping (DTW) method is used to detect the vibration, swing, air gap, speed and other signals of each startup process to determine whether the unit operating status is abnormal.

[0128] See also Figure 1 , the abnormality detection method of the start-up and shutdown process of the hydropower unit based on dynamic time warping includes:

[0129] Step S1: According to the historical health data samples of the startup process and the historical health data samples of the shutdown process, classification statistics and training are performed to obtain the benchmark time series of each characteristic indicator of the startup process and the shutdown process and the distance deviation threshold;

[0130] Step S2: Based on the benchmark time series of each characteristic index of the startup process and the shutdown process obtained by training in step S1 and the distance deviation threshold, abnormality detection is performed on the characteristic index of the startup process and the characteristic index of the shutdown process obtained by real-time measurement.

[0131] In this embodiment, specifically, the method and process of abnormality detection during startup and abnormality detection during shutdown are substantially the same, so abnormality detection during startup is described in detail below;

[0132] For details, please refer to the abnormal detection during the boot process. Figure 7 , the step S1 comprises:

[0133] Step S101: constructing a training set of characteristic indicators of the startup process under the same working conditions with working water head, guide bearing bush temperature, and stator core temperature as constraint conditions;

[0134] Step S102: iteratively calculate each characteristic indicator sample data in the original characteristic indicator set of the startup process under each working condition one by one to obtain the startup process benchmark time series and the maximum allowable distance deviation threshold of the corresponding characteristic indicator.

[0135] In this embodiment, specifically, step S101 includes:

[0136] set up is the working water head, is the upper average watt temperature, is the average watt temperature of the lower conductor, is the average watt temperature of the water conductor, is the average temperature of the core;

[0137] Extract the original feature data of each feature index of each boot process from the historical boot process health data sample, and construct the original eigenvalue matrix of each boot process according to (Equation 1): ;

[0138] From the original eigenvalue matrix Calculate the operating characteristic value of the startup process:

[0139] Average working head : ;

[0140] Average temperature of upper guide : ;

[0141] Average watt temperature of lower conductor : ;

[0142] Average water temperature : ;

[0143] Average core temperature : ;

[0144] Where:

[0145] For the The boot process;

[0146] is the total length of the time series;

[0147] Working water head at each moment;

[0148] is the measuring point;

[0149] The number of upper guide tile temperature measurement points;

[0150] is the temperature of each upper guide tile at each moment;

[0151] The number of temperature measurement points for the lower guide bearing;

[0152] is the temperature of each lower guide tile at each moment;

[0153] The number of water-conducting tile temperature measurement points;

[0154] is the temperature of each water guide tile at each moment;

[0155] is the number of core temperature measurement points;

[0156] is the core temperature at each moment;

[0157] by The allowable variation is within the range of ±1.0m. , , , The allowable variation is ±1.0℃. According to the same working condition , , , , Perform classification statistics to obtain the working condition The following is the original characteristic indicator set of the boot process: ;in, is the average water head under this condition, is the average watt temperature of the upper conductor under this condition, is the average watt temperature of the lower conductor under this condition, is the average temperature of the water conductor under this condition, is the average temperature of the core under this condition, For working conditions The number of boot processes under

[0158] After traversing all historical startup process health data samples, the original feature index set of the startup process under multiple working conditions is obtained:

[0159] , corresponding to the working condition ;

[0160] , corresponding to the working condition ;

[0161] …,

[0162] , corresponding to the working condition ;

[0163] Among them, each original feature index set For working conditions The original characteristic index set of multiple boot processes under For working conditions The number of boot processes under is the total number of working conditions obtained through traversal, .

[0164] In this embodiment, specifically, step S102 includes:

[0165] Step S1021: Setting ;

[0166] Step S1022: Traverse all characteristic indicators in the working condition The data under the condition are used to obtain the benchmark time series of all characteristic indicators and the maximum allowable distance deviation threshold;

[0167] Step S1023: construct working conditions The benchmark time series matrix under , and the maximum allowed distance deviation threshold vector ;

[0168] (Formula 3)

[0169] In the above formula It is working condition The startup process benchmark time series matrix of each characteristic index is as follows: It is a vector composed of the maximum allowable distance deviation vectors of each characteristic index under this working condition; synchronously record each , And the working head of the corresponding working condition , upper average watt temperature , average watt temperature of lower conductor , average water conduction temperature , the average temperature of the core , for use in abnormality detection in step S2;

[0170] Step S1024: ,if , repeat step S1021-step S1024; otherwise, it means that the data under all working conditions have been traversed; output the benchmark time series matrix set under multiple working conditions , the maximum allowed distance deviation vector set , and their corresponding working condition quantity sets ;

[0171] It should be noted that, in step S1022, the The characteristic index is in the working condition The following are the benchmark time series and allowable distance deviations:

[0172] (1) Settings ;

[0173] (2) Assume ( Characteristic index The number of eigenvalues ​​of the time series), from Next During the boot process Copy indicator Original time series;

[0174] (3) Initial settings ;

[0175] (4) Assume For the The characteristic index is in the working condition The benchmark time series under the initial setting , that is, initial Indicator for the first startup process time series;

[0176] (5) From The time when the generator output switch is closed is retrieved as the end point, and the time when the speed reaches 6r / min is retrieved as the starting point. The truncation process is performed to form a new time series. ( For indicators The number of eigenvalues ​​after truncation of the benchmark time series); The time when the generator output switch is closed is the end point, and the time when the speed reaches 6r / min is the start point, forming a new time series. ( For indicators The number of eigenvalues ​​after truncation of the original time series);

[0177] (6) Calculation using dynamic time warping algorithm and The cumulative distance matrix ,according to search , The shortest distance path between coordinates:

[0178] ( is the number of eigenvalues ​​in the new time series after DTW calculation and regularization); then according to Reconstruction of the value Get a new time series ,

[0179] according to Reconstruction of the value Get a new time series . and Compared with the following features:

[0180] 1. The lengths of the two time series are the same, both are ;

[0181] 2. The shapes of the two time series are closest;

[0182] 3. The point-by-point distance between two time series is the smallest;

[0183] (7) Calculation and Maximum distance deviation :

[0184] (Formula 5)

[0185] In the above formula, Indicates from Find the maximum value among the values, that is, Characteristic index In the entire The maximum distance deviation during the startup process;

[0186] (8) According to and Calculate the new , as follows:

[0187] (Formula 6)

[0188] (9) Order ,like , which means that the traversal has been completed. If all the boot processes below are completed, proceed to the next step, otherwise repeat step (4);

[0189] (10) Obtaining characteristic index In working condition Permissible distance deviation under:

[0190] (Formula 7)

[0191] In the above formula, Indicates from indivual Find it in the data (99.7% probability) width range, that is, the distance deviation data 1.25 times the distribution as the working condition The maximum permissible distance deviation under

[0192] (11) Records Characteristic index In working condition The benchmark time series under Characteristic index In working condition The allowable distance deviation under the condition is as follows; proceed to the next step;

[0193] (12) Order ,if , that is, all characteristic indicators have been traversed in the working condition The data under is obtained to obtain the benchmark time series and the maximum allowable distance deviation of all characteristic indicators, and then execute the next step; otherwise, repeat step S1022.

[0194] In this embodiment, specifically, Figure 8a and Figure 8b As shown, the step S2 includes:

[0195] Step S201: Using the working water head, guide bearing bush temperature, and stator core temperature measured in real time during the startup process as working condition constraints, search for a benchmark time series matrix matching the working condition. and the maximum allowable distance deviation threshold vector ;

[0196] Step S202: performing anomaly detection on the time series of characteristic indicators of the startup process measured in real time.

[0197] In this embodiment, specifically, step S201 includes:

[0198] Step S2011: Calculate the operating characteristic value of the real-time startup process from the time series record of the startup process measured in real time: average working water head , upper average watt temperature , average watt temperature of lower conductor , average water conduction temperature , Average core temperature ;

[0199] Average working head: ;

[0200] Average upper watt temperature: ;

[0201] Average watt temperature of lower conductor: ;

[0202] Average water conduction tile temperature: ;

[0203] Average core temperature: ;

[0204] In the above formula, , is the total length of the time series of the boot process, It is the working water head measured at each moment during the startup process; is the temperature of each upper guide tile at each moment of the process, is the number of upper conduction temperature measurement points; is the temperature of each lower guide watt at each moment of the process, is the number of downward temperature measurement points; is the temperature of each water conductor at each time of the process, is the number of water conduction temperature measurement points; is the core temperature at each time of the process, The number of core temperature measurement points.

[0205] Step S2012: According to , , , , Search for the benchmark time series matrix and the maximum distance deviation vector that meet the operating conditions.

[0206] In this embodiment, specifically, step S2012 includes:

[0207] Step S20121: Setting ;

[0208] Step S20122: Compare the working conditions and obtain the first Working conditions Next , , , and , if it satisfies:

[0209] ,

[0210] and ,

[0211] and ,

[0212] and ,

[0213] and , then the working condition search is successful, and step S202 is executed;

[0214] Otherwise, let , search again; if , that is, the data under all working conditions have been traversed, the working condition search fails, and the anomaly detection is terminated;

[0215] After the working condition search is successful, = , = , execute step S202.

[0216] In this embodiment, specifically, step S202 includes:

[0217] Step S2021: Setting ;

[0218] Step S2022: Obtain the The distance deviation of each characteristic indicator from the benchmark time series;

[0219] Step S2023: performing distance deviation detection;

[0220] Step S2024: ,if , that is, all feature indicator data have been traversed, the anomaly check has been completed, and the anomaly detection results are output.

[0221] Furthermore, the step S2022 includes:

[0222] Step a: Set It is represented as a time series of characteristic indicators measured in real time during the startup process; Extract The benchmark time series of characteristic indicators ,in, Characteristic index The number of eigenvalues ​​in the benchmark time series;

[0223] Step b: From The time when the generator output switch is closed is retrieved as the end point, and the time when the speed reaches 6r / min is retrieved as the starting point. The truncation process is performed to form a new time series. ,in, Characteristic index The number of eigenvalues ​​after truncation of the base time series; The time when the generator output switch is closed is retrieved as the end point, and the time when the speed reaches 6r / min is retrieved as the starting point to form a new time series. in, Characteristic index The number of eigenvalues ​​after truncation of the time series;

[0224] Step c: Calculate using dynamic time warping algorithm and The cumulative distance matrix ,according to search , The shortest distance path between coordinates: in, The number of eigenvalues ​​of the new time series after being warped by the dynamic time warping algorithm is calculated; then according to Reconstruction of the value Get a new time series ,according to Reconstruction of the value Get a new time series ;

[0225] Step d: Calculation and Maximum distance deviation :

[0226] (Formula 8)

[0227] in:

[0228] Indicates from Find the maximum value among the values, that is, Characteristic index The maximum distance deviation during the entire startup process;

[0229] The step S2023 includes:

[0230] According to judgment Feature indicators stored in Maximum allowable distance deviation , judging characteristic index Is it abnormal, that is:

[0231] if , then the characteristic index Abnormal, otherwise, characteristic indicators normal.

[0232] The above-mentioned embodiments only express the specific implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the protection scope of the present application. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the technical solution concept of the present application, and these all belong to the protection scope of the present application.

[0233] This background section is provided to generally present the context of the invention, and the work of the presently named inventors, the work to the extent described in this background section, and aspects of this section that did not constitute prior art at the time of application are neither explicitly nor implicitly admitted to be prior art to the present invention.

Claims

1. A method for detecting abnormalities in the start-up and shutdown process of a hydropower unit based on dynamic time regularization, characterized in that: include: Step S1: According to the historical health data samples of the startup process and the historical health data samples of the shutdown process, classification statistics and training are performed to obtain the benchmark time series of each characteristic indicator of the startup process and the shutdown process and the distance deviation threshold; Step S2: Based on the benchmark time series of each characteristic index of the startup process and the shutdown process obtained by training in step S1 and the distance deviation threshold, abnormal detection is performed on the characteristic index of the startup process and the characteristic index of the shutdown process obtained by real-time measurement; In the abnormality detection during the startup process, the step S1 includes: Step S101: constructing a training set of characteristic indicators of the startup process under the same working conditions with working water head, guide bearing bush temperature, and stator core temperature as constraint conditions; Step S102: iteratively calculating each characteristic indicator sample data in the original characteristic indicator set of the startup process under each working condition, and obtaining a startup process reference time series and a maximum allowable distance deviation threshold of the corresponding characteristic indicator; The step S101 includes: set up is the working water head, is the upper average watt temperature, is the average watt temperature of the lower conductor, is the average watt temperature of the water conductor, is the average temperature of the core; Extract the original feature data of each feature index of each boot process from the historical boot process health data samples, and construct the original eigenvalue matrix of each boot process ; From the original eigenvalue matrix Calculate the operating characteristic value of the startup process: Average working head : ; Upper average watt temperature : ; Average watt temperature of lower conductor : ; Average water temperature : ; Average core temperature : ; Where: For the The boot process; is the total length of the time series; Working water head at each moment; is the measuring point; The number of upper guide bearing temperature measurement points; is the temperature of each upper guide tile at each moment; The number of temperature measurement points for the lower guide bearing; is the temperature of each lower guide tile at each moment; The number of water-conducting tile temperature measurement points; is the temperature of each water guide tile at each moment; is the number of core temperature measurement points; is the core temperature at each moment; by The allowable variation is within the range of ±1.0m. , , , The allowable variation is ±1.0℃. According to the same working condition , , , , Perform classification statistics to obtain the working condition The following is the original characteristic indicator set of the boot process: ;in, is the average water head under this condition, is the average watt temperature of the upper conductor under this condition, is the average watt temperature of the lower conductor under this condition, is the average temperature of the water conductor under this condition, is the average temperature of the core under this condition, For working conditions The number of boot processes under After traversing all historical startup process health data samples, the original feature index set of the startup process under multiple working conditions is obtained: , corresponding to the working condition ; , corresponding to the working condition ; …, , corresponding to the working condition ; Among them, each original feature index set For working conditions The original characteristic index set of multiple boot processes under For working conditions The number of boot processes under is the total number of working conditions obtained through traversal, ; Original eigenvalue matrix ,include: in: is the original discrete time series; is the number of eigenvalues ​​in the characteristic indicator time series; is the total number of original feature indicators; The step S102 includes: Step S1021: Setting ; Step S1022: Traverse all characteristic indicators in the working condition The data under the condition are used to obtain the benchmark time series of all characteristic indicators and the maximum allowable distance deviation threshold; Step S1023: construct working conditions The benchmark time series matrix under , and the maximum allowed distance deviation threshold vector ; Step S1024: ,if , repeat step S1021-step S1024; otherwise, it means that the data under all working conditions have been traversed; output the benchmark time series matrix set under multiple working conditions , the maximum allowed distance deviation vector set , and their corresponding working condition quantity sets .

2. The method for detecting abnormalities in the start-up and shutdown process of a hydropower unit based on dynamic time warping according to claim 1 is characterized in that: The step S2 comprises: Step S201: Using the working water head, guide bearing bush temperature, and stator core temperature measured in real time during the startup process as working condition constraints, search for a benchmark time series matrix matching the working condition. and the maximum allowable distance deviation threshold vector ; Step S202: performing anomaly detection on the time series of characteristic indicators of the startup process measured in real time.

3. The method for detecting abnormalities during the start-up and shutdown process of a hydropower unit based on dynamic time warping according to claim 2 is characterized in that: The step S201 includes: Step S2011: Calculate the operating characteristic value of the real-time startup process from the time series record of the startup process measured in real time: average working water head , upper average watt temperature , average watt temperature of lower conductor , average water conduction temperature , Average core temperature ; Step S2012: According to , , , , Search for the benchmark time series matrix and the maximum distance deviation vector that meet the operating conditions.

4. The method for detecting abnormalities in the start-up and shutdown process of a hydropower unit based on dynamic time warping according to claim 3 is characterized in that: The step S2012 includes: Step S20121: Setting ; Step S20122: Compare the working conditions and obtain the first Working conditions Next , , , and , if it satisfies: ， and , and , and , and , then the working condition search is successful, and step S202 is executed; Otherwise, let , search again; if , that is, the data under all working conditions have been traversed, the working condition search fails, and the anomaly detection is terminated; After the working condition search is successful, = , = , execute step S202.

5. The method for detecting abnormalities during the start-up and shutdown process of a hydropower unit based on dynamic time warping according to claim 4 is characterized in that: The step S202 includes: Step S2021: Setting ; Step S2022: Obtain the The distance deviation of each characteristic indicator from the benchmark time series; Step S2023: performing distance deviation detection; Step S2024: ,if , that is, all feature indicator data have been traversed, the anomaly check has been completed, and the anomaly detection results are output.

6. The method for detecting abnormalities during the start-up and shutdown process of a hydropower unit based on dynamic time warping according to claim 5 is characterized in that: The step S2022 includes: Step a: Set It is represented as a time series of characteristic indicators measured in real time during the startup process; Extract The benchmark time series of characteristic indicators ,in, Characteristic index The number of eigenvalues ​​in the benchmark time series; Step b: From The time when the generator output switch is closed is retrieved as the end point, and the time when the speed reaches 6r / min is retrieved as the starting point. The truncation process is performed to form a new time series. ,in, Characteristic index The number of eigenvalues ​​after truncation of the benchmark time series; The time when the generator output switch is closed is retrieved as the end point, and the time when the speed reaches 6r / min is retrieved as the starting point to form a new time series. in, Characteristic index The number of eigenvalues ​​after the time series is truncated; Step c: Calculate using dynamic time warping algorithm and The cumulative distance matrix ,according to search , The shortest distance path between coordinates: in, The number of eigenvalues ​​of the new time series after being warped by the dynamic time warping algorithm is calculated; then according to Reconstruction of the value Get a new time series ,according to Reconstruction of the value Get a new time series ; Step d: Calculation and Maximum distance deviation : in: Indicates from Find the maximum value among the values, that is, Characteristic index The maximum distance deviation during the entire startup process; The step S2023 includes: According to judgment Feature indicators stored in Maximum allowable distance deviation , judging characteristic index Is it abnormal, that is: if , then the characteristic index Abnormal, otherwise, characteristic indicators normal.

7. The method for detecting abnormalities during the start-up and shutdown process of a hydropower unit based on dynamic time warping according to claim 6 is characterized in that: In the abnormality detection of the shutdown process, the same feature index set as that of the abnormality detection of the startup process is used, and the working head, guide bearing bush temperature, and stator core temperature are also used as working condition constraints; the specific calculation steps are the same as those of the abnormality detection of the startup process; the difference is that the interception method of the feature index time series is inconsistent during the training and abnormality detection processes; the startup process uses the closing moment of the generator output switch as the end point and the moment when the speed reaches 6r / min as the starting point for truncation processing; The shutdown process starts when the generator output switch is disconnected, and ends when the speed drops below 6r / min.

Images