A turbine controller edge side data cleaning and anomaly detection method for an industrial internet

Abstract

This invention discloses a method for edge-side data cleaning and anomaly detection of turbine controllers for the Industrial Internet. The method includes: constructing a hierarchical matrix of variables such as speed, valve position, pressure, and temperature to uniformly organize the sampling positions of multi-variable monitored quantities in the same instance, thus synchronously expressing the positional and categorical relationships among the monitored quantities within the edge node; based on this, generating disturbance location indices and operating condition sequence indices to distinguish between operating condition switching responses and isolated sampling disturbances; then, eliminating isolated jumps in single variables and retaining continuous offsets in multi-variable quantities to reduce the accidental deletion of true response fragments; further, improving the accuracy of identifying continuous anomalies, transient anomalies, and linked anomalies through adjacency merging and cross-variable synchronous determination, thereby enhancing the rationality of edge-side data cleaning and the reliability of anomaly detection for turbine controllers.

Description

Technical Field

[0001] This invention relates to the technical field of edge data processing and anomaly detection in the Industrial Internet, and particularly to a method for edge data cleaning and anomaly detection of turbine controllers for the Industrial Internet. Background Technology

[0002] In recent years, the application of the Industrial Internet in the field of energy equipment operation monitoring and control has gradually deepened. Turbine unit control systems have also gradually expanded from simple field control to an operation mode that combines field control with edge node data processing. During the operation of turbine units, speed, valve position, pressure, and temperature are the basic monitoring quantities reflecting the unit's regulation and operating status. These monitoring quantities have a clear sequential correlation in the processes of valve opening changes, load increases and decreases, start-stop switching, and disturbance response. To meet the needs of remote monitoring, status analysis, and fault early warning, existing technologies typically upload the monitoring quantities output by the turbine controller to edge nodes or upper-level systems. Then, the monitored quantities are filtered, denoised, missing data filled, thresholded, or anomaly identified. For example, a fixed threshold or sliding window method can be used to remove outliers from a single monitored quantity, and the abnormal state can be classified according to statistical characteristics or model results. The above techniques have certain applicability in general industrial monitoring scenarios. However, the monitored quantities related to the turbine controller are not independent of each other. Especially in the process of speed response and pressure response caused by valve position changes, there is a strong process correlation between different monitored quantities at the same sampling position and adjacent sampling positions. Cleaning based solely on the local change amplitude of a single monitored quantity is difficult to accurately reflect the actual operating conditions of the turbine unit.

[0003] Existing technologies still have significant shortcomings in data processing at the edge of turbine controllers. First, most existing solutions handle abrupt changes, exceeding limits, or missing values ​​of single-channel monitoring separately, lacking a unified approach to organizing the sampling position relationships of multiple variables in the same instance. This makes it difficult to simultaneously combine changes in the correlation between speed, valve position, pressure, and temperature at the edge. Second, existing cleaning methods typically treat abrupt changes directly as noise points. However, during turbine controller operating condition switching, rapid changes in some monitoring quantities are themselves valid responses during unit regulation. Directly removing these during data cleaning can easily lead to the loss of genuine abnormal signs or response segments related to operating condition switching. Third, existing technologies struggle to handle different segments such as short-term spikes, continuous offsets, missing measurements, and drops. The existing anomaly detection methods often employ a decentralized processing approach, lacking unified location calibration and sequence constraints, making it difficult to form a continuous processing flow within the edge nodes that can both handle data cleaning and anomaly detection. Furthermore, current anomaly detection methods primarily focus on single-point amplitudes, local statistics, or trends in single-variable changes, neglecting the duration of segments, synchronous changes in multiple variables, and the continuity relationship between adjacent segments. Consequently, during turbine controller operation, occasional disturbances of single variables are easily confused with multivariate linkage anomalies, affecting the accuracy and stability of anomaly detection results. Therefore, how to uniformly organize multivariate monitoring quantities at the turbine controller edge and consider both the sequence of operating condition changes and the linkage relationship between multiple variables during the cleaning process has become a technical problem that needs further resolution in this field.

[0004] In summary, existing turbine controller edge-side data processing technologies suffer from several problems, including difficulty in distinguishing between effective change segments and isolated disturbance segments during operating condition switching, easy deletion of genuine abnormal signs during data cleaning, and incomplete criteria for anomaly determination. The present invention addresses the problem of cleaning and anomaly identification of multivariable monitoring quantities in turbine controller edge nodes. Summary of the Invention

[0005] The purpose of this section is to outline some aspects of the embodiments of the present invention and to briefly introduce some preferred embodiments. Some simplifications or omissions may be made in this section, as well as in the abstract and title of the present application, to avoid obscuring the purpose of this section, the abstract and title of the invention. Such simplifications or omissions shall not be used to limit the scope of the present invention.

[0006] In view of the aforementioned existing problems, the present invention is proposed.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] As a preferred embodiment of the turbine controller edge data cleaning and anomaly detection method for the industrial Internet described in this invention, the speed monitoring value, valve position monitoring value, pressure monitoring value and temperature monitoring value in the turbine controller edge node are arranged according to the same sampling position, and a variable hierarchical matrix is ​​formed according to the variable category.

[0009] In the variable hierarchical matrix, the positions of sudden jump segments, stable segments, missing segments and fall segments are calibrated to generate a disturbance position index, and the working condition sequence index is generated according to the order of valve position change start point, speed change start point and pressure response start point at the same sampling position.

[0010] Based on the disturbance location index and the working condition sequence index, abrupt jump segments that are not adjacent to the working condition switch adjacent positions and have no other variable segments changing in the same direction at the same sampling position are identified as isolated jump segments of a single variable and are removed. Offset segments located at the working condition switch adjacent positions or appearing consecutively in two or more variables at adjacent sampling positions are retained, and a cleaning result matrix is ​​generated.

[0011] Extract the retained segments from the cleaning result matrix according to the perturbation location index, merge the retained segments with continuous sampling locations and the same segment type, and generate anomaly judgment results based on the amplitude change, duration and cross-variable synchronization relationship of the merged segments.

[0012] The beneficial effects of this invention are as follows: By constructing a hierarchical matrix of variables such as rotational speed, valve position, pressure, and temperature, this invention achieves unified organization of the sampling positions of multivariate monitoring quantities in the same instance, enabling the synchronous expression of the positional and categorical relationships among the monitoring quantities within the edge nodes. Based on this, by generating disturbance location indices and operating condition sequence indices, it distinguishes between operating condition switching responses and isolated sampling disturbances. Furthermore, it eliminates isolated jumps in single variables and retains continuous offsets in multivariate variables, reducing the accidental deletion of true response fragments. Moreover, through adjacency merging and cross-variable synchronous determination, it improves the accuracy of identifying continuous anomalies, transient anomalies, and linked anomalies, thereby enhancing the rationality of data cleaning on the edge side of the turbine controller and the reliability of anomaly detection. Attached Figure Description

[0013] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. Wherein:

[0014] Figure 1 This is a flowchart illustrating the edge-side data cleaning and anomaly detection method for turbine controllers oriented towards the Industrial Internet, as shown in this invention. Detailed Implementation

[0015] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0016] Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without inventive effort should fall within the scope of protection of this invention.

[0017] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0018] In this embodiment, the turbine controller edge-side data cleaning and anomaly detection method for the Industrial Internet is deployed at the turbine controller edge node. The turbine controller edge node is electrically connected to the speed sensor, actuator valve position feedback unit, pressure sensor, and temperature sensor. The speed sensor outputs the turbine spindle speed sampling value and the speed regulating shaft speed sampling value. The actuator valve position feedback unit outputs the main steam valve position sampling value, the regulating valve position sampling value, and the bypass valve position sampling value. The pressure sensor outputs the main steam pressure sampling value, the exhaust steam pressure sampling value, the lubricating oil pressure sampling value, and the control oil pressure sampling value. The temperature sensor outputs the main steam temperature sampling value, the exhaust steam temperature sampling value, the bearing temperature sampling value, and the lubricating oil temperature sampling value. Each sampling value is accompanied by a sampling time stamp generated by the same clock source. In this embodiment, the sampling interval is 100 ms, and the edge node continuously collects 300 sets of sampling values ​​as one analysis batch.

[0019] According to an embodiment of the present invention, in combination Figure 1 The flowchart shown illustrates a method for edge-side data cleaning and anomaly detection in turbine controllers for the Industrial Internet, specifically including the following steps:

[0020] S1. Arrange the speed monitoring values, valve position monitoring values, pressure monitoring values, and temperature monitoring values ​​in the edge nodes of the turbine controller according to the same sampling position, and form a variable hierarchical matrix according to variable category. Note that the following should be noted in this step:

[0021] The turbine controller edge node receives speed sampling values ​​from the speed sensor, valve position sampling values ​​from the actuator valve position feedback device, pressure sampling values ​​from the pressure sensor, and temperature sampling values ​​from the temperature sensor; it then merges these speed sampling values, valve position sampling values, pressure sampling values, and temperature sampling values ​​according to the sampling time marker to generate speed monitoring values, valve position monitoring values, pressure monitoring values, and temperature monitoring values.

[0022] The speed monitoring values ​​include turbine spindle speed sampling values ​​and speed regulating shaft speed sampling values; valve position monitoring values ​​include main steam valve position sampling values, regulating valve position sampling values ​​and bypass valve position sampling values; pressure monitoring values ​​include main steam pressure sampling values, exhaust steam pressure sampling values, lubricating oil pressure sampling values ​​and control oil pressure sampling values; temperature monitoring values ​​include main steam temperature sampling values, exhaust steam temperature sampling values, bearing temperature sampling values ​​and lubricating oil temperature sampling values.

[0023] S1.1. Group the speed monitoring value, valve position monitoring value, pressure monitoring value and temperature monitoring value into the same sampling monitoring group according to the sampling time mark, and form a sampling position sequence according to the order of the sampling time mark.

[0024] In a preferred embodiment, when each sampling time marker arrives, the turbine controller edge node receives the turbine main shaft speed sampling value, speed regulating shaft speed sampling value, main steam valve position sampling value, regulating valve position sampling value, bypass valve position sampling value, main steam pressure sampling value, exhaust steam pressure sampling value, lubricating oil pressure sampling value, control oil pressure sampling value, main steam temperature sampling value, exhaust steam temperature sampling value, bearing temperature sampling value, and lubricating oil temperature sampling value corresponding to that sampling time marker. The above sampling values ​​with the same sampling time marker are grouped into one same sampling monitoring group. If some sampling values ​​are missing under a certain sampling time marker, the missing position is left empty in the same sampling monitoring group and is not replaced by an adjacent sampling value. Then, all the same sampling monitoring groups are arranged in ascending order of sampling time marker. The first same sampling monitoring group corresponds to sampling position 1, the second same sampling monitoring group corresponds to sampling position 2, and so on, forming a sampling position sequence.

[0025] As an example, in a certain analysis batch, the sampling time markers are 0 ms, 100 ms, 200 ms, 300 ms, and 400 ms respectively. Then the sampling position sequence is sampling position 1, sampling position 2, sampling position 3, sampling position 4, and sampling position 5 respectively. If there is a lack of exhaust temperature sampling value in the same sampling monitoring group with a sampling time marker of 200 ms, the exhaust temperature position corresponding to that sampling position will remain vacant.

[0026] It should be noted that for the same sampling monitoring group with repeated sampling time markers, only the same sampling monitoring group that arrives at the edge node first and has a higher data item completeness is retained; for the same sampling monitoring group that arrives in reverse order of sampling time markers, it is still reordered according to the sampling time marker value, not according to the order of arrival; after adopting this processing method, the sampling position sequence only reflects the time sequence relationship and is not affected by network transmission jitter.

[0027] S1.2. Using each sampling position in the sampling position sequence as the column position of the matrix, and using the speed variable, valve position variable, pressure variable, and temperature variable as the row position of the matrix, generate a variable hierarchical arrangement matrix.

[0028] Specifically, a matrix row and position hierarchy is first established according to variable categories. The speed variable layer includes two matrix rows corresponding to the turbine main shaft speed monitoring value and the governor shaft speed monitoring value; the valve position variable layer includes three matrix rows corresponding to the main steam valve position monitoring value, the regulating valve position monitoring value, and the bypass valve position monitoring value; the pressure variable layer includes four matrix rows corresponding to the main steam pressure monitoring value, the exhaust steam pressure monitoring value, the lubricating oil pressure monitoring value, and the control oil pressure monitoring value; and the temperature variable layer includes four matrix rows corresponding to the main steam temperature monitoring value, the exhaust steam temperature monitoring value, the bearing temperature monitoring value, and the lubricating oil temperature monitoring value. Then, sampling positions 1 to 300 are sequentially used as matrix columns to obtain a 13-row, 300-column variable hierarchical arrangement matrix. In this variable hierarchical arrangement matrix, matrix rows of the same variable category are arranged continuously, while different variable categories are arranged in a fixed order: speed variable layer - valve position variable layer - pressure variable layer - temperature variable layer. This allows valve position changes, speed changes, pressure responses, and temperature responses to form a horizontal comparison relationship under the same column position.

[0029] S1.3 Fill the speed monitoring value, valve position monitoring value, pressure monitoring value and temperature monitoring value from each sampling monitoring group into the corresponding matrix row position and corresponding matrix column position in the variable hierarchical arrangement matrix to generate the variable hierarchical matrix.

[0030] Specifically, using the sampling position corresponding to each sampling monitoring group as the matrix column, the turbine spindle speed monitoring value in that sampling monitoring group is filled into the turbine spindle speed matrix row, the speed regulating shaft speed monitoring value is filled into the speed regulating shaft speed matrix row, the main steam valve position monitoring value is filled into the main steam valve position matrix row, the regulating valve position monitoring value is filled into the regulating valve position matrix row, the bypass valve position monitoring value is filled into the bypass valve position matrix row, the main steam pressure monitoring value, exhaust steam pressure monitoring value, lubricating oil pressure monitoring value, and control oil pressure monitoring value are respectively filled into the corresponding pressure matrix row, and the main steam temperature monitoring value, exhaust steam temperature monitoring value, bearing temperature monitoring value, and lubricating oil temperature monitoring value are respectively filled into the corresponding temperature matrix row; if there is a missing state in a certain sampling monitoring group, a missing mark is retained in the corresponding matrix row and column.

[0031] For example, at sampling position 25, the turbine spindle speed monitoring value is 3000 r / min, the speed regulating shaft speed monitoring value is 1498 r / min, the main steam valve position monitoring value is 42%, the regulating valve position monitoring value is 38%, the bypass valve position monitoring value is 0%, the main steam pressure monitoring value is 9.8 MPa, the exhaust steam pressure monitoring value is 0.12 MPa, the lubricating oil pressure monitoring value is 0.28 MPa, the control oil pressure monitoring value is 1.45 MPa, the main steam temperature monitoring value is 538 ℃, the exhaust steam temperature monitoring value is 86 ℃, the bearing temperature monitoring value is 72 ℃, and the lubricating oil temperature monitoring value is 49 ℃. These values ​​are then filled into the corresponding row positions of the variable layer matrix and the column positions of the matrix corresponding to sampling position 25.

[0032] It should be noted that in this embodiment, the monitored values ​​of different variable categories are first merged according to the same sampling position, and then a variable hierarchical matrix is ​​formed according to the variable category. This is because anomalies on the edge side of the turbine controller may manifest as isolated jumps of a single variable, or as linked offsets with a causal relationship between valve position, speed, pressure and temperature. If each type of monitored value is processed separately according to time sequence, the sequential relationship between valve position action and speed response and pressure response is not easy to identify in a unified manner. If all variables are directly mixed into a single column data stream, the segment boundaries and variable category boundaries will overlap, which is not conducive to the subsequent formation of a clear disturbance location index and operating condition sequence index.

[0033] In this embodiment, the variable hierarchical matrix retains the differences in variable categories vertically and the consistency of the same sampling position horizontally. Therefore, it can identify abrupt segments, stable segments, missing segments, offset segments, and falling segments in a single matrix row position, and can also identify the synchronous change relationship between multiple variables in the same matrix column position. Compared with the existing technology of cleaning according to a single variable or screening only by thresholding the time series point by point, this embodiment incorporates the sampling position and variable category into a unified matrix structure. Subsequently, it can directly determine the adjacent position of the working condition switch and the isolated disturbance position based on the matrix position relationship, thereby reducing the situation of misjudging normal working condition switches as abnormal and improving the consistency of local cleaning of edge nodes and anomaly judgment.

[0034] S2. In the variable hierarchical matrix, calibrate the positions of sudden jump segments, stable segments, missing segments, and fallback segments to generate a disturbance position index. Then, generate a working condition sequence index based on the order of the valve position change start point, speed change start point, and pressure response start point at the same sampling position. Note that the following should be noted in this step:

[0035] S2.1 In the variable hierarchical matrix, the monitoring values ​​of adjacent sampling positions are compared sequentially according to the row position of each matrix. Segments that show a sudden increase or decrease at a single sampling position and are not continuous before and after sampling positions are marked as a sudden jump segment. Segments with consistent change amplitude in multiple consecutive sampling positions are marked as stable segments. Segments with missing monitoring values ​​are marked as missing measurement segments. Segments that continue in the same direction of change in multiple consecutive sampling positions are marked as offset segments. Segments that gradually decrease in multiple consecutive sampling positions are marked as falling segments.

[0036] In a preferred embodiment, taking the same physical quantity corresponding to each row of the matrix as the unit, the monitoring values ​​of two adjacent sampling positions are compared one by one in the order of sampling positions 1 to 300; for example, for the row of the turbine spindle speed matrix, the changes in monitoring values ​​between sampling positions 1 and 2, sampling positions 2 and 3, and sampling positions 3 and 4 are compared sequentially; the row of the main steam pressure matrix is ​​also compared sequentially in the same way.

[0037] Furthermore, in this embodiment, the monitoring value of each matrix row position for adjacent sampling positions specifically refers to the monitoring value of the same physical quantity corresponding to two adjacent matrix column positions in the same matrix row position, without involving cross-comparison between different matrix rows positions.

[0038] Preferably, the calibration method for various segments in this embodiment is as follows: when a matrix row position shows a significant increase or decrease compared to the previous sampling position at a certain sampling position, and the subsequent sampling position returns to the adjacent level before the aforementioned change, or no longer continues along the direction of change, then the single sampling position is calibrated as a sudden jump segment; in this embodiment, taking the turbine spindle speed matrix row position as an example, if the turbine spindle speed monitoring values ​​corresponding to sampling positions 120, 121, and 122 are 3001 r / min, 3065 r / min, and 3003 r / min respectively, then sampling position 121 constitutes a sudden jump segment; segments with consistent change amplitudes within multiple consecutive sampling positions are calibrated as stable segments; in this embodiment, if the main steam pressure matrix row position at sampling positions 40 to 48 is 9.80 MPa, 9.80 MPa, 9.81 MPa, ... If the measured values ​​are 35%, 37%, 39%, 41%, and 43% respectively, then the measured values ​​are 9.60 MPa, 9.45 MPa, 9.31 MPa, 9.18 MPa, and 9.05 MPa respectively, then the measured values ​​are 9.60 MPa, 9.45 MPa, 9.31 MPa, 9.18 MPa, and 9.05 MPa respectively. In this embodiment, if the measured values ​​of the exhaust steam temperature matrix rows at sampling positions 66 to 68 are all empty, then the measured values ​​are 66 to 68 as missing segments. In this embodiment, if the measured values ​​of the regulating valve position matrix rows at sampling positions 150 to 154 are 35%, 37%, 39%, 41%, and 43% respectively, then the measured values ​​are 150 to 154 as offset segments. In this embodiment, if the measured values ​​of the main steam pressure matrix rows at sampling positions 210 to 214 are 9.60 MPa, 9.45 MPa, 9.31 MPa, 9.18 MPa, and 9.05 MPa respectively, then the measured values ​​are 210 to 214 as falling segments.

[0039] S2.2. Arrange the matrix row positions, starting matrix column positions, and ending matrix column positions of the sudden jump segment, stable segment, missing segment, offset segment, and fallback segment in sequence to form a disturbance position index.

[0040] Specifically, each calibrated segment is registered one by one, using the matrix row position as the first-level arrangement reference, the starting matrix column position as the second-level arrangement reference, and the ending matrix column position as the third-level arrangement reference. Each registration includes the segment type, matrix row position, starting matrix column position, and ending matrix column position. For example, if the offset segment on the row position of the control valve position matrix is ​​located between sampling positions 150 and 154, it is recorded in the disturbance position index as "Control Valve Position Matrix Row Position - Offset Segment - Starting Matrix Column 150 - Ending Matrix Column 154". If the missing segment on the row position of the exhaust temperature matrix is ​​located between sampling positions 66 and 68, it is recorded in the disturbance position index as "Exhaust Temperature Matrix Row Position - Missing Segment - Starting Matrix Column 66 - Ending Matrix Column 68".

[0041] S2.3 In the variable hierarchical matrix, the first column position of the matrix row where the valve position monitoring value transitions from a stationary segment to an offset segment is determined as the starting point of the valve position change;

[0042] S2.4. The first column of the matrix where the speed monitoring value is located is located after the valve position change start point and is transitioned from a stable segment to an offset segment, and is determined as the speed change start point.

[0043] S2.5. The first column of the matrix containing the pressure monitoring value, which is located after the starting point of the speed change and transitions from a stable segment to an offset segment, is determined as the starting point of the pressure response.

[0044] S2.6. The operating condition sequence index is formed by arranging the valve position change start point, speed change start point, and pressure response start point in the variable hierarchical matrix in that order.

[0045] In a preferred embodiment, the valve position change start point, speed change start point, and pressure response start point are arranged in ascending order of matrix column position, and the column position interval between two adjacent start points is marked as the adjacent position of the operating condition switching. For example, if the valve position change start point is 150, the speed change start point is 153, and the pressure response start point is 156, then the operating condition sequence index is valve position change start point 150 - speed change start point 153 - pressure response start point 156. At the same time, sampling positions 149 to 151, sampling positions 152 to 154, and sampling positions 155 to 157 are marked as adjacent positions of the operating condition switching. The reason for using the method of extending one sampling position before and after the start point to form the adjacent positions of the operating condition switching is that there is a delay in the action transmission and a time difference in sensor reporting within the sampling interval on the edge side of the turbine controller, and the adjacent sampling positions before and after the start point usually still belong to the continuous performance of the same operating condition switching process.

[0046] It should be noted that this embodiment first calibrates the disturbance segments in the variable hierarchical matrix and then generates the operating condition sequence index because the valve position action, speed change, and pressure response in the turbine control process have a clear sequential relationship. If anomalies are screened only based on the magnitude of the deviation of a single variable, the speed increase and pressure change caused by the valve opening adjustment may be misjudged as abnormal deviations. If noise is judged only based on statistical fluctuations, the linkage changes that truly reflect the switching of operating conditions may be mistakenly rejected. This embodiment calibrates the positions of sudden jump segments, stable segments, missing segments, deviation segments, and fallback segments, and then constructs the operating condition sequence index based on the valve position change start point, speed change start point, and pressure response start point. This enables subsequent steps to distinguish between ordered deviations caused by operating condition switching and isolated disturbances that are not dependent on operating condition switching. Compared with the existing technology that divides each variable into anomaly segments separately according to a fixed threshold, this embodiment combines segment type identification with operating condition sequence identification, which can directly establish the temporal constraint relationship of the turbine control process at the edge, thereby providing a more consistent positional basis for data cleaning.

[0047] S3. Based on the disturbance location index and operating condition sequence index, abrupt jump segments that are not adjacent to the operating condition switch adjacent positions and have no other variable segments changing in the same direction at the same sampling position are identified as isolated jumps of a single variable and are removed. Offset segments located at the operating condition switch adjacent positions or appearing consecutively in two or more variables at adjacent sampling positions are retained, generating a cleaning result matrix. Note that the following should be noted in this step:

[0048] S3.1. Based on the disturbance location index, filter out the matrix row position, starting matrix column position, and ending matrix column position of the jump segment in the variable hierarchical matrix.

[0049] Specifically, all entries of the segment type "jump segment" are retrieved in the perturbation location index, and the corresponding matrix row position, starting matrix column position and ending matrix column position are extracted in the order of matrix row position and starting matrix column position. Since the jump segment can appear in one sampling position or in two consecutive sampling positions in this embodiment, its starting matrix column position and ending matrix column position are retained at the same time during screening.

[0050] For example, if the jump segment of the turbine spindle speed matrix row is located at sampling positions 121 to 121, and the jump segment of the bearing temperature matrix row is located at sampling positions 208 to 209, then the turbine spindle speed matrix row -121-121 and the bearing temperature matrix row -208-209 are respectively retrieved from the disturbance position index.

[0051] S3.2. Compare the positions of the column before the starting matrix column and the column after the ending matrix column with the adjacent positions of the working condition switching in the working condition sequence index, and compare the matrix row positions of the same sampling positions from the starting matrix column to the ending matrix column.

[0052] In a preferred embodiment, if the column preceding the starting matrix column or the column following the ending matrix column falls into the adjacent position of the operating condition switching, then the jump segment is adjacent to the operating condition switching process; if neither falls into the adjacent position of the operating condition switching, then the jump segment is not adjacent to the operating condition switching process. For example, if the starting matrix column of a certain jump segment is 121 and the ending matrix column is 121, then the preceding column is 120 and the following column is 122; if the adjacent positions of the operating condition switching include 149 to 151, 152 to 154, and 155 to 157, then 120 and 122 are not adjacent positions of the operating condition switching, and the jump segment is not adjacent to the operating condition switching process.

[0053] In another preferred embodiment, within all matrix columns covered by the jump segment, the remaining matrix rows are horizontally checked to see if there are any segments with the same direction of change as the jump segment; if the jump segment is a sudden increase, the remaining matrix rows at the same sampling position are horizontally checked to see if there are any offset segments, jump segments, or upward surge segments before falling back in the upward direction; if the jump segment is a sudden decrease, the remaining matrix rows at the same sampling position are horizontally checked to see if there are any offset segments, jump segments, or falling back segments in the downward direction.

[0054] S3.3 When neither the current column nor the next column belongs to the adjacent position of the working condition switch, and there are no segments of change in the same direction in the other matrix rows of the same sampling position, the jump segment is judged as a single variable isolated jump segment.

[0055] In this embodiment, the adjacent positions for switching operating conditions are uniformly taken as the 3-column interval formed by the sampling position one before the starting point and the sampling position one after the starting point; the same-direction change segment specifically refers to the segment with the same change direction appearing in different matrix rows at the same sampling position; if the target jump segment is manifested as a numerical increase, then the other matrix rows also show offset segments or jump segments in the same matrix column range, which are considered to be the existence of same-direction change segments; if the target jump segment is manifested as a numerical decrease, then the other matrix rows also show offset segments or fall segments in the same matrix column range, which are considered to be the existence of same-direction change segments.

[0056] Preferably, in this embodiment, the method for determining a single-variable isolated jump segment is as follows: when neither the preceding nor following column of a certain jump segment belongs to the adjacent position of the operating condition switch, and within the range from the starting column to the ending column of the matrix corresponding to the jump segment, there are no segments with the same direction of change in the other matrix rows, the jump segment is determined to be a single-variable isolated jump segment; for example, if the turbine spindle speed matrix row shows a jump segment at sampling position 121 where the speed increases suddenly from 3001 r / min to 3065 r / min and then drops back to 3003 r / min, while the regulating valve position matrix row at the same sampling position is still 38%, the main steam pressure matrix row is still 9.80 MPa, and the bearing temperature matrix row is still 72 ℃, and sampling positions 120 and 122 do not belong to the adjacent position of the operating condition switch, then the turbine spindle speed jump segment is determined to be a single-variable isolated jump segment.

[0057] S3.4 Remove the corresponding row and column positions of isolated jump segments of a single variable from the variable hierarchical matrix.

[0058] It should be noted that for matrix positions that have been identified as isolated jump segments of a single variable, their abnormal monitoring values ​​are deleted, and the corresponding matrix row and column positions are marked as empty, while the original monitoring values ​​of other matrix rows and columns of that matrix column are retained.

[0059] S3.5. Based on the disturbance position index, filter out the matrix row position, starting matrix column position, and ending matrix column position of the offset segment in the variable hierarchical matrix.

[0060] Specifically, all entries of the segment type "offset segment" are retrieved in the disturbance position index, and the corresponding matrix row position, starting matrix column position, and ending matrix column position are extracted in the order of matrix row position and starting matrix column position. For example, if the offset segment on the control valve position matrix row position is located between 150 and 154, the offset segment on the turbine main shaft speed matrix row position is located between 153 and 158, and the offset segment on the main steam pressure matrix row position is located between 156 and 160, then the above three offset segment entries are screened out respectively.

[0061] S3.6. Determine the offset segment located at the adjacent position of the working condition switch as the retained segment.

[0062] S3.7. The offset segments that are continuously expanded along adjacent sampling positions and appear synchronously in two or more matrix rows are determined as retained segments.

[0063] Specifically, the starting matrix column of the offset segment is checked column by column from the ending matrix column according to adjacent sampling positions. If there is no interruption column between each adjacent matrix column, it is considered to be continuously expanded along the adjacent sampling positions.

[0064] In this embodiment, the offset segments that occur synchronously in two or more matrix rows specifically refer to: within the same matrix column range, at least two different matrix rows simultaneously have offset segments with the same direction of change and partially overlapping column ranges; for example, if the regulating valve position matrix row has an upward offset segment from 150 to 154, the turbine main shaft speed matrix row has an upward offset segment from 153 to 158, and the main steam pressure matrix row has a downward offset segment from 156 to 160, then the regulating valve position matrix row and the turbine main shaft speed matrix row form a synchronous upward offset from 153 to 154.

[0065] S3.8. Retain the original matrix position of the retained fragment in the variable hierarchical matrix, and combine it with the remaining matrix content after removing isolated jump fragments of a single variable to generate the cleaning result matrix.

[0066] Specifically, for offset segments that have been determined to be retained, their matrix row and column positions are not changed; for isolated jump segments of single variables that have been removed, the corresponding matrix positions are left empty; for other segments that have not been removed, the original values ​​are maintained, thus forming a cleaned result matrix; this cleaned result matrix has the same matrix dimensions and sampling position order as the original variable hierarchical matrix, and only the content changes at the positions of segments determined to be isolated jump segments of single variables.

[0067] It should be noted that this embodiment does not uniformly delete all jump segments, nor does it uniformly retain all offset segments. Instead, it distinguishes and processes them based on both the disturbance location index and the operating condition sequence index. This is because the sources of anomalies on the edge side of the turbine controller have two distinct mechanisms: one is isolated jumps in a single variable caused by sensor jitter, sampling glitches, or communication interference, which are not adjacent to the operating condition switch and are not accompanied by changes in other variables in the same direction at the same sampling position; the other is the chain response of speed and pressure caused by valve position action, which occurs near the adjacent position of the operating condition switch or occurs continuously and synchronously in multiple variables. If both types of disturbances are deleted together, the key segments that truly reflect the changes in operating conditions will be mistakenly deleted; if both are retained together, obvious glitches will be introduced into the subsequent anomaly judgment. This embodiment removes isolated jumps from the variable hierarchical matrix through dual screening of position adjacency relationship and cross-variable synchronization relationship, while retaining offset segments with operating condition correlation or multi-variable synchronization characteristics. Therefore, the cleaned result matrix reduces the impact of noise while retaining the key changes in the turbine control process.

[0068] S4. Extract the retained segments from the cleaning result matrix according to the perturbation location index. Merge the retained segments with continuous sampling locations and the same segment type, and generate anomaly judgment results based on the amplitude changes, durations, and intervariate synchronization relationships of the merged segments. Note that the following should be noted in this step:

[0069] S4.1 Based on the perturbation location index, filter out the matrix row position, starting matrix column position, and ending matrix column position of the retained fragment in the cleaning result matrix.

[0070] In a preferred embodiment, all offset segment entries that have been determined to be retained segments are first retrieved from the disturbance location index. Then, the cleaning result matrix is ​​compared to confirm that the matrix positions corresponding to these offset segments still retain the original monitoring values. Thus, the matrix row position, starting matrix column position, and ending matrix column position of each retained segment are obtained. For example, if the control valve position matrix rows 150 to 154, the turbine main shaft speed matrix rows 153 to 158, and the main steam pressure matrix rows 156 to 160 have all been determined to be retained segments in step S3, then they are respectively screened as subsequent merging and determination objects in step S4.1.

[0071] S4.2 Merge the retained segments with the same matrix row position, the same segment type, and whose terminating matrix column position is consecutively connected to the starting matrix column position of the next retained segment to generate merged segments.

[0072] Specifically, in the same matrix row, if the ending matrix column of the previous retained segment differs from the starting matrix column of the next retained segment by 1, then the two are connected end to end into a new merged segment; for example, if there are two offset segments 153 to 155 and 156 to 158 in the turbine spindle speed matrix row, then the two are merged into a merged segment 153 to 158; if there is a gap of 2 or more matrix columns between the two, then no merging is performed; after adopting this method, the segmented changes of the same variable at continuous sampling positions can be reorganized into continuous change segments, which is convenient for subsequent anomaly judgment based on the overall change trend.

[0073] S4.3 Determine the amplitude change based on the starting and ending monitoring values ​​of the merged segment, determine the duration based on the number of columns between the starting and ending matrix columns, and perform a horizontal comparison of the remaining matrix rows at the same sampling position to determine the cross-variable synchronization relationship.

[0074] As an example, the monitoring value corresponding to the column position of the starting matrix of the merged segment is taken as the starting monitoring value, and the monitoring value corresponding to the column position of the ending matrix of the merged segment is taken as the ending monitoring value. The direction and magnitude of the difference between the ending monitoring value and the starting monitoring value are used as the amplitude change. For example, if the merged segment of the control valve position matrix is ​​located between rows 150 and 154, and the monitoring value of the control valve position at sampling position 150 is 35%, and the monitoring value of the control valve position at sampling position 154 is 43%, then the amplitude change of this merged segment is an increase of 8%. If the merged segment of the main steam pressure matrix is ​​located between rows 156 and 160, and the monitoring value of the main steam pressure at sampling position 156 is 9.80 MPa, and the monitoring value of the main steam pressure at sampling position 160 is 9.10 MPa, then the amplitude change of this merged segment is a decrease of 0.70 MPa.

[0075] As an example, the duration is calculated by subtracting the column position of the starting matrix from the column position of the ending matrix and then adding 1, corresponding to the number of consecutive sampling positions covered by the merged segment. For example, if the merged segment is located between 150 and 154, the duration is 5 sampling positions; if the merged segment is located between 208 and 209, the duration is 2 sampling positions. In this embodiment, the sampling interval is 100 ms, so a duration of 5 sampling positions corresponds to 500 ms, and a duration of 2 sampling positions corresponds to 200 ms.

[0076] As an example, for the entire matrix column range covered by a certain merged segment, check column by column whether there are merged or retained segments with the same direction and overlapping column intervals in the remaining matrix rows. If at least one other matrix row satisfies the above condition, it is determined that there is a cross-variable synchronization relationship between the merged segment and the other matrix row. If two or more other matrix rows satisfy the above condition, it is determined that the merged segment has a multivariable cross-variable synchronization relationship. For example, if the row position of the regulating valve matrix continuously rises from 150 to 154, and the row position of the turbine spindle speed matrix continuously rises from 153 to 158, then the two have a cross-variable synchronization relationship from 153 to 154. If the row position of the main steam temperature matrix also rises synchronously from 154 to 158, then the merged segment has a cross-variable synchronization relationship with both the row position of the turbine spindle speed matrix and the row position of the main steam temperature matrix.

[0077] S4.4. Merged segments with continuously increasing amplitude changes (defined in this embodiment as continuous changes in the same direction within 3 or more consecutive sampling positions) and whose duration covers multiple consecutive sampling positions (defined in this embodiment as no less than 3 sampling positions) are determined as continuous abnormal segments. Merged segments with a single sudden increase in amplitude changes and whose duration is less than that of continuous abnormal segments are determined as transient abnormal segments. Merged segments with cross-variable synchronization relationships are determined as linkage abnormal segments, and anomaly determination results are generated.

[0078] Specifically, the retained segments are offset segments located at adjacent positions during operating condition switching or continuously distributed across two or more matrix rows along adjacent sampling positions; the merged segments are the result of connecting retained segments with consistent matrix rows, segment types, and consecutive beginning and end of matrix columns; the cross-variable synchronization relationship is the correspondence between segments that change in the same direction at the same sampling position in two or more matrix rows within the matrix column range of the merged segments; the anomaly determination results include continuous abnormal segments, transient abnormal segments, and linked abnormal segments.

[0079] Furthermore, the specific methods for determining each abnormal segment include: when a merged segment continuously changes in the same direction within three or more consecutive sampling positions, and the difference between the terminated monitoring value and the initial monitoring value reaches more than twice the upper limit of the normal fluctuation of the corresponding variable, it is determined to be a continuous abnormal segment; in this embodiment, the upper limit of the normal fluctuation of the turbine spindle speed monitoring value under steady-state conditions is taken as 15 r / min, the upper limit of the normal fluctuation of the regulating valve position monitoring value under steady-state conditions is taken as 1.5%, the upper limit of the normal fluctuation of the main steam pressure monitoring value under steady-state conditions is taken as 0.08 MPa, and the upper limit of the normal fluctuation of the bearing temperature monitoring value under steady-state conditions is taken as 1.0 ℃; accordingly, if the continuous length of the merged segment on the turbine spindle speed matrix row is greater than or equal to three sampling positions, and the amplitude change is greater than or equal to 30 r / min, it is determined to be a continuous abnormal segment; if the continuous length of the merged segment on the main steam pressure matrix row is greater than or equal to three sampling positions, and the amplitude change is greater than or equal to 0.16 If the value is less than 1 MPa, it is determined to be a continuous abnormal segment; if the merged segment only shows a single sudden increase or decrease in 1 or 2 sampling positions, and the adjacent sampling positions after the termination matrix column do not continue in that direction, it is determined to be a transient abnormal segment; if the merged segment has a cross-variable synchronization relationship with 2 or more other matrix rows within its matrix column range, it is determined to be a linked abnormal segment.

[0080] In a preferred example, a turbine unit is operating at 60% load, and the sampling interval is 100 ms. At sampling positions 150 to 154, the regulating valve position monitoring value increased from 35% to 43%; at sampling positions 153 to 158, the turbine spindle speed monitoring value increased from 3002 r / min to 3045 r / min; at sampling positions 156 to 160, the main steam pressure monitoring value decreased from 9.80 MPa to 9.12 MPa; at sampling positions 154 to 159, the main steam temperature monitoring value increased from 538 ℃ to 545 ℃. After cleaning in step S3, all of the above segments were retained. Subsequently, in step S4, the duration of the regulating valve position merged segment was 5 sampling positions, and the amplitude change was an increase of 8%, which is more than twice 1.5%. Furthermore, it had a cross-variable synchronous relationship with the turbine spindle speed merged segment and the main steam temperature merged segment. Therefore, this regulating valve position merged segment was determined to be a linkage abnormal segment. The duration of the turbine spindle speed merged segment was 6 sampling positions, and the amplitude change was an increase of 43%. The turbine spindle speed merging segment is determined to be both a continuous abnormal segment and a linkage abnormal segment because the main steam pressure merging segment has a cross-variable synchronous relationship with the regulating valve position merging segment and the main steam temperature merging segment. The main steam pressure merging segment has a duration of 5 sampling positions, an amplitude change of 0.68 MPa, which is more than twice 0.08 MPa, and its starting point lags behind the turbine spindle speed merging segment, which is consistent with the pressure response characteristics during the turbine operating condition change process. Therefore, the main steam pressure merging segment is determined to be a continuous abnormal segment. If the bearing temperature matrix row position shows a sudden increase from 72 ℃ to 76 ℃ at sampling position 208 and then drops back to 72.5 ℃ at sampling position 209, and there are no segments with the same direction of change in the other matrix rows at the same sampling position, then this segment has been removed as an isolated jump segment with a single variable in step S3 and will not proceed to step S4.

[0081] It should also be noted that, based on the cleaning result matrix, this embodiment does not directly draw conclusions on a single sampling point. Instead, it first extracts and retains segments, then merges them according to adjacency relationships, and finally forms anomaly judgment results based on amplitude changes, duration, and cross-variable synchronization relationships. This is because effective anomalies on the edge side of the turbine controller are often not isolated single points, but rather form continuously changing segments at continuous sampling positions, or synchronous changing segments between multiple variables. If anomaly judgment is made only based on a single-point threshold, it is easy to confuse short-term spikes with real anomalies, and it is also difficult to distinguish between single-variable deviations and multi-variable linked deviations.

[0082] This embodiment merges segments before making a judgment, which can combine multiple sub-segments of the same abnormal process that unfold continuously in time into a complete change segment. Then, it identifies the correlation changes between valve position, speed, pressure and temperature through cross-variable synchronization relationships. Compared with the existing technology that only alarms based on single variable thresholds, this embodiment can not only determine whether an abnormality exists, but also determine whether the abnormality belongs to a continuous abnormal segment, a transient abnormal segment or a linked abnormal segment, so that the abnormality judgment results output by the edge node are more consistent with the actual operating state of the turbine control process.

[0083] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for data cleaning and anomaly detection on the edge side of a turbine controller for the industrial internet, characterized by include: Arrange the speed monitoring values, valve position monitoring values, pressure monitoring values, and temperature monitoring values ​​in the edge nodes of the turbine controller according to the same sampling position, and form a variable hierarchical matrix according to the variable category; In the variable hierarchical matrix, the positions of sudden jump segments, stable segments, missing segments and fall segments are calibrated to generate a disturbance position index, and the working condition sequence index is generated according to the order of valve position change start point, speed change start point and pressure response start point at the same sampling position. Based on the disturbance location index and the working condition sequence index, abrupt jump segments that are not adjacent to the working condition switch adjacent positions and have no other variable segments changing in the same direction at the same sampling position are identified as isolated jump segments of a single variable and are removed. Offset segments located at the working condition switch adjacent positions or appearing consecutively in two or more variables at adjacent sampling positions are retained, and a cleaning result matrix is ​​generated. Extract the retained segments from the cleaning result matrix according to the perturbation location index, merge the retained segments with continuous sampling locations and the same segment type, and generate anomaly judgment results based on the amplitude change, duration and cross-variable synchronization relationship of the merged segments. 2.The industrial internet of things oriented turbine controller edge side data cleaning and anomaly detection method according to claim 1, characterized in that, Forming the variable hierarchical matrix includes: The speed monitoring value, valve position monitoring value, pressure monitoring value, and temperature monitoring value are grouped into the same sampling monitoring group according to the sampling time mark, and a sampling position sequence is formed according to the order of the sampling time marks; Using each sampling position in the sampling position sequence as a matrix column and the speed variable, valve position variable, pressure variable, and temperature variable as matrix row positions, a variable hierarchical arrangement matrix is ​​generated; The speed monitoring value, valve position monitoring value, pressure monitoring value, and temperature monitoring value from each sampling monitoring group are filled into the corresponding matrix row and column positions in the variable hierarchical arrangement matrix to generate the variable hierarchical matrix. 3.The industrial internet of things oriented turbine controller edge side data cleaning and anomaly detection method according to claim 1 or 2, characterized in that, The turbine controller edge node receives speed sampling values ​​from the speed sensor, valve position sampling values ​​from the actuator valve position feedback unit, pressure sampling values ​​from the pressure sensor, and temperature sampling values ​​from the temperature sensor; the speed sampling values, valve position sampling values, pressure sampling values, and temperature sampling values ​​are merged according to the sampling time marker to generate the speed monitoring value, valve position monitoring value, pressure monitoring value, and temperature monitoring value; wherein: The speed monitoring values ​​include turbine spindle speed sampling values ​​and speed regulating shaft speed sampling values; The valve position monitoring values ​​include the main steam valve position sampling value, the regulating valve position sampling value, and the bypass valve position sampling value; The pressure monitoring values ​​include main steam pressure sampling values, exhaust steam pressure sampling values, lubricating oil pressure sampling values, and control oil pressure sampling values; The temperature monitoring values ​​include main steam temperature sampling values, exhaust steam temperature sampling values, bearing temperature sampling values, and lubricating oil temperature sampling values. 4.The industrial internet of things oriented turbine controller edge side data cleaning and anomaly detection method according to claim 1, characterized in that, The generation of the disturbance location index includes: In the variable hierarchical matrix, the monitoring values ​​of adjacent sampling positions are compared sequentially according to the row position of each matrix. Segments that show a sudden increase or decrease at a single sampling position and are not continuous before and after sampling positions are marked as a sudden jump segment. Segments with consistent change amplitude in multiple consecutive sampling positions are marked as stable segments. Segments with missing monitoring values ​​are marked as missing measurement segments. Segments that continue in the same direction of change in multiple consecutive sampling positions are marked as offset segments. Segments that gradually decrease in multiple consecutive sampling positions are marked as falling segments. The matrix row positions, starting matrix column positions, and ending matrix column positions of the sudden jump segment, the stable segment, the missing measurement segment, the offset segment, and the fallback segment are arranged sequentially to form the disturbance position index.

5. The industrial internet of things oriented turbine controller edge side data cleaning and anomaly detection method according to claim 4, characterized in that, The generation of the operating condition sequence index includes: In the variable hierarchical matrix, the first column position of the matrix row where the valve position monitoring value transitions from a stable segment to an offset segment is determined as the starting point of the valve position change; The first column position of the matrix containing the speed monitoring value, which is located after the valve position change starting point and transitions from a stable segment to an offset segment, is determined as the speed change starting point. The first column of the matrix containing the pressure monitoring value, which is located after the starting point of the speed change and transitions from a stable segment to an offset segment, is determined as the starting point of the pressure response. The operating condition sequence index is formed by arranging the valve position change start point, the speed change start point, and the pressure response start point in the variable hierarchical matrix in that order. 6.The industrial internet of things oriented turbine controller edge side data cleaning and anomaly detection method according to claim 1, characterized in that, The determination and removal of isolated jump segments of a single variable includes: Based on the perturbation location index, the matrix row position, starting matrix column position, and ending matrix column position of the sudden jump segment in the variable hierarchical matrix are selected; The positions of the column before the starting matrix column and the column after the ending matrix column are compared with the adjacent positions of the working condition switching in the working condition sequence index, and the matrix row positions of the same sampling positions from the starting matrix column to the ending matrix column are compared. When neither the preceding column nor the following column belongs to the adjacent position of the working condition switch, and there are no segments of change in the same direction in the other matrix rows of the same sampling position, the jump segment is determined to be a single variable isolated jump segment. The corresponding row and column positions of the isolated jump fragment of the single variable in the variable hierarchical matrix are removed.

7. The industrial internet of things oriented turbine controller edge side data cleaning and anomaly detection method according to claim 6, characterized in that, The generation of the cleaning result matrix includes: Based on the perturbation location index, the matrix row position, starting matrix column position, and ending matrix column position of the offset segment in the variable hierarchical matrix are selected; The offset segment located adjacent to the working condition switching position is determined to be a retained segment; The offset segment that is continuously expanded along adjacent sampling positions and appears synchronously in two or more matrix rows is determined to be a retained segment; The original matrix position of the retained fragment in the variable hierarchical matrix is ​​preserved, and the remaining matrix content after removing the isolated jump fragment of the single variable is combined to generate the cleaning result matrix.

8. The method for edge-side data cleaning and anomaly detection of turbine controllers for the Industrial Internet according to claim 1, characterized in that, The generation of the anomaly determination result includes: Based on the disturbance location index, the matrix row position, starting matrix column position, and ending matrix column position of the retained segment are selected from the cleaning result matrix; The segments with the same matrix row position, the same segment type, and whose terminating matrix column position is consecutively connected to the starting matrix column position of the next segment are merged to generate a merged segment; The amplitude change is determined by the starting and ending monitoring values ​​of the merged segment, the duration is determined by the number of columns between the starting and ending matrix columns, and the remaining matrix rows at the same sampling position are compared laterally to determine the cross-variable synchronization relationship. Merged segments with continuously increasing amplitude changes and a duration covering multiple consecutive sampling positions are identified as continuous abnormal segments. Merged segments with a single sudden increase in amplitude changes and a duration shorter than the continuous abnormal segments are identified as transient abnormal segments. Merged segments with cross-variable synchronization relationships are identified as linked abnormal segments, and the abnormal determination results are generated.

9. The industrial internet of things oriented turbine controller edge side data cleaning and anomaly detection method according to claim 8, characterized in that, The retained segment is an offset segment located at the adjacent position of the working condition switching or continuously distributed in two or more matrix rows along adjacent sampling positions; the merged segment is the result of connecting retained segments with consistent matrix rows, consistent segment types, and consecutive beginning and end of matrix columns; the cross-variable synchronization relationship is the correspondence between two or more matrix rows with segments changing in the same direction at the same sampling position within the matrix column range of the merged segment.

10. The industrial internet of things oriented turbine controller edge side data cleaning and anomaly detection method according to claim 8, characterized in that, The anomaly determination results include continuous anomaly segments, transient anomaly segments, and linked anomaly segments.

Images