Intelligent scheduling management method for detection tasks of water conservancy and hydropower engineering

Abstract

The invention relates to the technical field of task scheduling management, in particular to an intelligent scheduling management method for water conservancy and hydropower engineering detection tasks, which comprises the following steps of: acquiring active power data of a unit, smoothing, differentially generating a rate and an acceleration sequence to construct a load characteristic matrix, comparing the load characteristic matrix with a steady-state threshold to generate a quasi-steady-state interval, and calculating a quasi-steady-state interval; calculating a load instruction deviation to generate a steady-state confirmation identifier, calculating a predicted steady-state window based on the identifier, if the window is longer than the minimum sampling duration, generating a trigger acquisition signal, starting vibration acquisition to generate an original waveform, and uploading the original waveform. According to the method, the load characteristic matrix is constructed through power data differential operation, load instruction deviation dual verification is combined to recognize the steady-state time period, the steady-state window length is pre-judged, and collection is triggered when the requirement is met, so that noise interference introduced by working condition fluctuation is effectively avoided, and it is ensured that original vibration waveform data originates from a stable working condition; and the data sample purity and the fault analysis value are obviously improved.

Description

Technical Field

[0001] This invention relates to the field of task scheduling and management technology, and in particular to an intelligent scheduling and management method for water conservancy and hydropower engineering inspection tasks. Background Technology

[0002] The field of task scheduling and management technology involves the orderly allocation and execution monitoring of work tasks through the overall planning of business processes, time nodes, and execution resources, using data processing technology. Traditional intelligent scheduling and management methods for water conservancy and hydropower engineering inspection tasks involve managers manually entering the locations to be inspected and the scheduled completion dates of water conservancy facilities such as dams and generator units into spreadsheets or basic progress recording software on computer terminals. They then manually check paper schedules and equipment inventory lists to verify the availability of inspection personnel and testing instruments. Based on the verification results, they create task allocation sheets in paper or electronic document form, and administrative personnel assign specific inspection work to the corresponding engineering and technical teams via telephone calls or on-site distribution of documents.

[0003] Current technology relies on manual review of paper schedules and equipment lists to formulate inspection plans. This offline and lagging management model cannot perceive the real-time operating conditions of water conservancy facilities and units, resulting in a disconnect between the issuance of inspection tasks and the actual operating status of the units. Due to the lack of real-time monitoring of the active power and load changes of the units, it is difficult to accurately capture the steady-state window during the operation of the units. This easily leads to the collection of vibration data during periods of severe fluctuation or instability in the unit's operating conditions. As a result, the collected data contains a large amount of non-stationary interference noise, which seriously reduces the accuracy and reference value of the vibration waveform data and cannot provide reliable data support for subsequent fault diagnosis. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing an intelligent scheduling and management method for water conservancy and hydropower engineering inspection tasks.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: an intelligent scheduling and management method for water conservancy and hydropower engineering inspection tasks, comprising the following steps: S1: Obtain the active power data of the generator set, smooth the active power data of the generator set using Savitzky-Golay to generate smoothed power data, generate a power rate sequence by first-order difference of the smoothed power data, generate a power acceleration sequence by second-order difference of the power rate sequence, and construct a load feature matrix based on the power rate sequence and the power acceleration sequence. S2: Compare the load feature matrix with the steady-state threshold group, filter the time periods in the load feature matrix that are within the steady-state threshold group to generate a quasi-steady-state interval, calculate the deviation between the load target command and the active power data of the unit, and generate a steady-state confirmation mark if the deviation is less than the dead zone range. S3: Extract the minimum sampling duration based on the steady-state confirmation flag, calculate the time difference between the next load instruction and the current time to generate an expected steady-state window, and generate a high-priority trigger acquisition signal if the expected steady-state window is greater than the minimum sampling duration; S4: Receive the high-priority trigger acquisition signal, start vibration acquisition based on the high-priority trigger acquisition signal, generate the original vibration waveform within the minimum sampling time, add a label to the original vibration waveform and upload steady-state data.

[0006] As a further aspect of the present invention, step S1 specifically comprises: S11: The local control unit of the turbine reads the current operating power value of the unit in real time according to the preset sampling frequency. The Savitzky-Golay digital filter is used to perform polynomial least squares fitting on the obtained discrete power points to remove high-frequency random noise interference and retain the low-frequency trend characteristics of power change, thereby generating the smooth power data. S12: Perform a forward first-order difference operation on the smoothed power data in the time dimension to calculate the power change amplitude between adjacent sampling points to reflect the speed of power change, generate the power rate sequence, and perform a second difference operation on the power rate sequence to calculate the trend of power change rate to reflect the acceleration and deceleration state of unit regulation, and generate the power acceleration sequence. S13: Using timestamps as index bases, align and stitch the smoothed power data, the power rate sequence, and the power acceleration sequence on the same time dimension to construct a multi-dimensional time series set including power amplitude dimension, change rate dimension, and change acceleration dimension, and generate the load feature matrix.

[0007] As a further aspect of the present invention, step S2 specifically comprises: S21: Obtain the pre-set steady-state threshold group, extract the feature vector of each time segment in the load feature matrix, determine row by row whether the power rate value in the feature vector is less than the rate threshold in the steady-state threshold group, and whether the power acceleration value is less than the acceleration threshold in the steady-state threshold group, and filter the continuous time segments that meet the dual constraint conditions. S22: Calculate the duration of the continuous time segment. If the duration exceeds the preset minimum stable period, determine that the unit operating condition during the time period is in a stable operating state during the non-adjustment process. Mark the start and end times of the time period as the effective interval boundary and generate the quasi-steady state interval. S23: Obtain the current load target command issued by the automatic power generation control system, calculate the absolute difference between the mean of the smoothed power data in the quasi-steady-state interval and the load target command, compare the absolute difference with the preset dead zone range, and when the absolute difference falls within the dead zone range, confirm that the unit has accurately responded to the load command and achieved output balance, and generate the steady-state confirmation flag.

[0008] As a further aspect of the present invention, step S3 specifically comprises: S31: In response to the generation of the steady-state confirmation flag, read the vibration signal acquisition standard required for the current detection item from the preset detection task configuration table, extract the minimum continuous data length required to meet the frequency domain resolution requirements, and generate the minimum sampling duration; S32: Obtain the expected issuance time of the next pre-planned load adjustment instruction through the data interface of the power plant monitoring system, calculate the time difference between the expected issuance time and the current system time, assess the remaining time length that the current steady-state condition can be maintained, and generate the expected steady-state window; S33: Compare the expected steady-state window with the minimum sampling duration to determine whether the remaining steady-state time is sufficient to complete a complete vibration data acquisition task. If and only if the value of the expected steady-state window is strictly greater than the minimum sampling duration, activate the acquisition task triggering logic and generate the high-priority trigger acquisition signal.

[0009] As a further aspect of the present invention, step S4 specifically comprises: S41: The vibration monitoring subsystem listens to the high-priority trigger acquisition signal in real time. When a signal level change is detected, it sends a synchronization trigger command to the high-speed data acquisition card at the front end to drive the vibration acceleration sensors of multiple channels to synchronously start the analog signal to digital signal conversion process. S42: Within the time period defined by the minimum sampling duration, continuously read the digital sequence converted by the sensor through the buffer, perform DC component removal and windowing processing on the data stream, extract a complete and continuous time-domain waveform data that can reflect the mechanical vibration characteristics of the unit under the current steady-state load, and generate the original vibration waveform. S43: Obtain the current unit head data, guide vane opening data, and active power data of the unit; encapsulate the unit head data, guide vane opening data, and active power data of the unit as operating condition description metadata into the data packet header; associate and bind them with the original vibration waveform; construct a standardized data packet with an operating condition index; and send the data packet to a remote server via industrial Ethernet to upload steady-state data.

[0010] As a further aspect of the present invention, the process of constructing the load characteristic matrix includes: Obtain the smoothed power data, the power rate sequence, and the power acceleration sequence, and calculate the maximum absolute value of each sequence within the current statistical period as a normalization benchmark. The normalization benchmark is used to perform dimensionless processing on each element in multiple sequences to eliminate the influence of numerical differences between different physical dimensions, and the normalized power vector, velocity vector and acceleration vector are obtained. The three normalized vectors are used as column vectors and stacked horizontally in time index order to construct a three-column multi-row numerical matrix, ensuring that each row in the matrix fully represents the instantaneous load dynamic characteristics of the unit at the target sampling time, thus generating the load feature matrix.

[0011] As a further aspect of the present invention, the process of selecting continuous time segments that satisfy the dual constraint conditions in S21 includes: Traverse all row vectors of the load feature matrix, perform a logical AND operation on each row vector, and determine whether the absolute value of the velocity element in the row vector is less than the preset velocity fluctuation limit, and at the same time determine whether the absolute value of the acceleration element in the row vector is less than the preset acceleration fluctuation limit. Establish a binary logic sequence corresponding to the time axis, set the time that satisfies the above logical AND operation conditions to logical true value, and set the time that does not satisfy the conditions to logical false value; Morphological filtering is performed on the binary logic sequence to remove isolated truth points with a length less than a preset glitch threshold, and discontinuous truth segments with adjacent spacing less than a preset tolerance are merged to obtain a smooth and continuous time index set, which serves as the selected continuous time segment.

[0012] As a further aspect of the present invention, the calculation process of the expected steady-state window includes: By using the dispatch interface of the power plant's automatic generation control system, query the planned execution timestamp of the first unexecuted instruction in the dispatch team list; If the query result is empty or the planned execution timestamp is earlier than the current time, the default setting is to use a future preset long period of time as the stable period for the unit to maintain its current output, and the long period of time is directly assigned as the window length. If a valid future plan execution timestamp is found, the difference in seconds between the timestamp and the current system atomic clock time is calculated, and a preset safety redundancy coefficient is introduced to reduce the difference to reserve the mechanical delay time before the unit responds to the command. The remaining available sampling time is then used to generate the expected steady-state window.

[0013] As a further aspect of the present invention, the process of adding a label to the original vibration waveform in S43 includes: Read the real-time register address of the unit's local control unit, and collect the current upstream water level, downstream water level, guide vane servo travel value, and generator stator current value. The gross head is calculated using the difference between upstream and downstream water levels. The stroke value of the guide vane servo is combined with the pre-stored comprehensive characteristic curve of the turbine operation to determine the relative position zone of the current operating point on the characteristic curve. The capillary head, guide vane relay stroke value, relative position partition number, and acquisition start timestamp down to milliseconds are encoded into metadata text in JSON format; Write the metadata text into the header extension field of the original vibration waveform file to establish a strong mapping relationship between the waveform data and the operating conditions, and complete the tag addition.

[0014] As a further aspect of the present invention, the process of determining whether the deviation is less than the dead zone range in S23 involves calculating a steady-state determination index, the formula for which the steady-state determination index is calculated is as follows: ; in, Represents the steady-state determination index, This represents the arithmetic mean of the smoothed power data within the quasi-steady-state interval. The value representing the load target instruction. This represents the rated power value of the unit. This represents the standard deviation of the smoothed power data within the quasi-steady-state interval. The weighting coefficients representing the deviation term. The weighting coefficients represent the fluctuation term.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: In this invention, a load feature matrix is ​​constructed by smoothing and performing multi-order differential operations on the active power data of the unit. This matrix is ​​then compared with a steady-state threshold and double-verified in conjunction with load command deviation. This allows for the accurate identification of the actual steady-state period during unit operation. The steady-state window length is predicted based on the time difference of subsequent load commands, and data acquisition is triggered only when the expected window meets the minimum sampling duration requirement. This effectively avoids non-stationary noise interference introduced by operating condition fluctuations, ensuring that the acquired original vibration waveform data originates from stable operating conditions, and significantly improving the purity of the data sample and the value of fault analysis. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the intelligent scheduling and management method for water conservancy and hydropower engineering inspection tasks according to the present invention. Figure 2 This is a flowchart of the load feature matrix construction process of the present invention; Figure 3 This is a flowchart illustrating the steady-state confirmation identifier generation process of the present invention; Figure 4 This is a flowchart of the high-priority trigger acquisition signal generation process of the present invention; Figure 5 This is a flowchart of the steady-state data acquisition and uploading process of the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the software-based technical solution is described in detail below with reference to system architecture diagrams and embodiments. It should be understood that the specific embodiments described herein are only for explaining the technical solutions of this invention and do not constitute a limitation on the scope of protection.

[0018] In the description of this invention, the system architecture relationships or data processing flows indicated by terms such as "layer," "module," "interface," "data flow," "client," and "server" are all defined based on the architecture diagram or flowchart corresponding to the embodiments. This way of describing is only used to clearly illustrate the logical relationships between the elements in the technical solution, and not to limit the physical deployment form. The term "multiple" includes two or more technical units, including but not limited to multiple data nodes, processing threads, service instances, or functional components and other scalable elements. The specific number is determined according to the actual business scenario and needs to be specifically specified.

[0019] Please see Figure 1 and Figure 2 This invention provides a technical solution: an intelligent scheduling and management method for water conservancy and hydropower engineering inspection tasks, comprising the following steps: S1: Obtain active power data of the unit, generate smoothed power data by Savitzky-Golay smoothing of active power data of the unit, generate power rate sequence by first-order differential smoothing of power data, generate power acceleration sequence by second-order differential power rate sequence, and construct load feature matrix based on power rate sequence and power acceleration sequence. The specific steps of S1 are as follows: S11: The local control unit of the turbine reads the current operating power value of the unit in real time according to the preset sampling frequency. The Savitzky-Golay digital filter is used to perform polynomial least squares fitting on the obtained discrete power points to remove high-frequency random noise interference and retain the low-frequency trend characteristics of power change, thereby generating smooth power data. S12: Perform forward first-order difference operation on the smoothed power data in the time dimension to calculate the power change amplitude between adjacent sampling points to reflect the speed of power change, generate a power rate sequence, and perform a second difference operation on the power rate sequence to calculate the trend of power change rate to reflect the acceleration and deceleration state of unit regulation, and generate a power acceleration sequence. S13: Using timestamps as indexes, smooth power data, power rate sequences, and power acceleration sequences are aligned and stitched together on the same time dimension to construct a multi-dimensional time series set including power amplitude dimension, change rate dimension, and change acceleration dimension, and generate a load feature matrix. The process of constructing the load characteristic matrix includes: Obtain smoothed power data, power rate sequence, and power acceleration sequence, and calculate the maximum absolute value of each sequence within the current statistical period as the normalization benchmark. By using a normalized benchmark, each element in multiple sequences is made dimensionless to eliminate the influence of numerical differences between different physical dimensions, resulting in normalized power vector, velocity vector, and acceleration vector. The three normalized vectors are used as column vectors and stacked horizontally in time index order to construct a three-column multi-row numerical matrix. This ensures that each row in the matrix fully represents the instantaneous load dynamic characteristics of the unit at the target sampling time, thus generating a load feature matrix.

[0020] This process, executed between the local control unit and data processing server of a hydro-generator unit, is illustrated using a case study of a large mixed-flow hydro-generator unit (rated power 600MW). First, a power transmitter installed at the generator outlet continuously collects the unit's active power analog signal at a sampling frequency of 50Hz. This analog signal is then quantized into a digital signal via an analog-to-digital converter, forming a discrete power point sequence. The current moment is selected. A 60-second time window was used to obtain the original power sequence containing 3000 sampling points. For this original power sequence, the Savitzky-Golay digital filter algorithm was applied, with a filter window length of 51 (25 points on each side for fitting) and a polynomial fitting order of 3. The algorithm slid the window point by point, performing least-squares fitting on the 51 data points within the window. The center point value of the fitted curve was used to replace the original center point value, thereby eliminating high-frequency random noise caused by electrical interference and generating smooth power data. .

[0021] Next, smooth the power data. Perform discrete-time first-order difference operations. Specifically, take the time step... smooth power value Subtract time smooth power value ,Right now This value directly reflects the power increment within a sampling interval of 0.02 seconds. Performing this operation on the data throughout the entire time window constructs a power rate sequence of length 2999. Subsequently, the power rate sequence was analyzed. Perform the difference operation again to calculate the time. rate With time rate The difference, that is This quantifies the acceleration of power change, generating a power acceleration sequence of length 2998. To ensure consistent matrix dimensions, the beginnings of the power rate sequence and power acceleration sequence are padded with zeros to restore their length to 3000 points.

[0022] In the stage of constructing the load characteristic matrix, the three sequences mentioned above are first normalized to eliminate dimensional differences. Then, the power data is smoothed. Identify the maximum absolute value within the time window. MW; Traversing the power rate sequence Identify the largest absolute value MW / sample; Traverse the power acceleration sequence Identify the largest absolute value MW / sample². Using these three benchmark values, perform division operations on the corresponding sequences respectively: Divide each element in the matrix by 605.2, and... Divide each element by 2.5, and... Divide each element by 0.8 to obtain the normalized power vector. velocity vector and acceleration vector Finally, using the timestamp as the row index, the normalized power vector is... velocity vector acceleration vector By concatenating the column vectors horizontally, a numerical matrix of 3000 rows and 3 columns is constructed.

[0023] Table 1 illustrates the processing steps from some of the original data to the elements of the feature matrix: Table 1 Comparison Table of Unit Load Characteristic Data Processing: ; As shown in Table 1, through the above calculation process, the unit load status at each moment is transformed into dimensionless characteristic values, eliminating the influence of physical units, and finally generating a load characteristic matrix.

[0024] The Savitzky-Golay digital filter algorithm described above is a filtering method based on local polynomial least squares fitting. Its core lies in using low-order polynomial fitting within a moving window to replace mean filtering, thereby better preserving the signal's peak value, width, and other higher-order moment characteristics while smoothing out noise.

[0025] The above normalization process refers to mapping data to a specific interval of 0 to 1 or -1 to 1. In this embodiment, the maximum value normalization method is adopted, that is, by dividing by the maximum absolute value in the sequence, the absolute value of all data is less than or equal to 1. The purpose is to eliminate the weight influence of different physical dimensions on the subsequent threshold discrimination.

[0026] Please see Figure 1 and Figure 3 S2: Compare the load feature matrix with the steady-state threshold group, filter the time periods in the load feature matrix that are within the steady-state threshold group to generate a quasi-steady-state interval, calculate the deviation between the load target command and the unit active power data, and generate a steady-state confirmation mark if the deviation is less than the dead zone range. The specific steps of S2 are as follows: S21: Obtain a pre-set steady-state threshold group, extract the feature vector of each time segment in the load feature matrix, and determine row by row whether the power rate value in the feature vector is less than the rate threshold in the steady-state threshold group and whether the power acceleration value is less than the acceleration threshold in the steady-state threshold group, and filter out continuous time segments that meet the dual constraint conditions. The process of selecting continuous time segments that satisfy the dual constraints in S21 includes: Traverse all row vectors of the load feature matrix, perform a logical AND operation on each row vector, and determine whether the absolute value of the velocity element in the row vector is less than the preset velocity fluctuation limit. At the same time, determine whether the absolute value of the acceleration element in the row vector is less than the preset acceleration fluctuation limit. Establish a binary logic sequence corresponding to the time axis, set the time that satisfies the above logical AND operation conditions to logical true value, and set the time that does not satisfy the conditions to logical false value; Morphological filtering is performed on the binary logic sequence to remove isolated truth points with a length less than a preset spurt threshold, and discontinuous truth segments with adjacent spacing less than a preset tolerance are merged to obtain a smooth and continuous time index set, which is used as the selected continuous time segment. S22: Statistically determine the duration of a continuous time segment. If the duration exceeds the preset minimum stable period, the unit operating condition during the time segment is determined to be in a stable operating state during the non-regulation process. The start and end times of the time segment are marked as the effective interval boundaries to generate a quasi-steady-state interval. S23: Obtain the current load target command issued by the automatic generation control system, calculate the absolute difference between the mean of the smoothed power data in the quasi-steady-state interval and the load target command, compare the absolute difference with the preset dead zone range, and when the absolute difference falls within the dead zone range, confirm that the unit has accurately responded to the load command and achieved output balance, and generate a steady-state confirmation mark. The process of determining whether the deviation is less than the dead zone in S23 involves calculating the steady-state determination index. The formula for calculating the steady-state determination index is as follows: ; in, Represents the steady-state determination index. This represents the arithmetic mean of the smoothed power data within the quasi-steady-state region. The value representing the load target command. This represents the rated power value of the unit. This represents the standard deviation of smoothed power data within the quasi-steady-state region. The weighting coefficients representing the deviation term. The weighting coefficients represent the fluctuation term.

[0027] Steady-state logic discrimination is performed based on the 3000-row, 3-column load feature matrix generated by S1. First, a preset steady-state threshold set is read from the system configuration database. For the 600MW unit in this embodiment, the normalized rate threshold is set to 0.005 (corresponding to approximately 0.0125MW / 20ms in actual physical quantity), and the normalized acceleration threshold is set to 0.002 (corresponding to approximately 0.0016MW / 20ms² in actual physical quantity). The program scans the load feature matrix row by row, extracting the second column element (rate feature) and the third column element (acceleration feature) of each row. For the... Okay, perform Boolean operations: if and If the condition is met, the logical state at that moment is marked as "1" (true); otherwise, it is marked as "0" (false). After traversing all 3000 rows of data, a binary logical sequence of length 3000 is generated.

[0028] Next, mathematical morphological filtering is applied to the binary logic sequence. First, an "erosion" operation is performed, setting the structuring element length to 100 sampling points (i.e., 2 seconds). This sets all isolated pulses with consecutive "1"s less than 100 points in length to "0", eliminating pseudo-steady-state discrimination caused by transient interference. Then, an "expansion" operation is performed, setting the tolerance interval to 50 sampling points (i.e., 1 second). If the interval between two consecutive "1" sequence segments is less than 50 "0"s, the intermediate "0"s are filled with "1", thus merging discontinuous segments caused by minor jitter. After processing, the longest consecutive "1" segment in the binary logic sequence is scanned, and its start timestamp is recorded. and end timestamp Calculate the duration of this segment. Set the minimum stable period to 30 seconds. If... If the interval is 1 second, then the time period is determined to be the quasi-steady-state interval.

[0029] After determining the quasi-steady-state range, the system obtains the current load target command issued by the automatic generation control system. Assuming the current load target is 450MW and the unit's rated power... The power is 600MW. Extract the corresponding smoothed power data segment within the quasi-steady-state interval and calculate its arithmetic mean. and standard deviation Subsequently, the steady-state determination index formula was used to calculate... .

[0030] The steady-state determination index calculation formula used in this embodiment is as follows: ; in, This represents the steady-state determination index; the smaller the value, the higher the degree of steady-state stability. This represents the arithmetic mean of smoothed power data within the quasi-steady-state range, used to reflect the center position of the unit's actual output. The numerical value representing the load target command serves as the ideal control target; The rated power value of the unit is used to normalize deviations and fluctuations. The standard deviation represents the smoothed power data within the quasi-steady-state interval and is used to characterize the degree of dispersion of power around the mean. The weighting coefficient representing the deviation term is set to 0.6, which is used to adjust the importance of static deviation in steady-state determination; The weighting coefficient representing the fluctuation term is set to 0.4, which is used to adjust the importance of dynamic fluctuations in steady-state determination.

[0031] The following example illustrates the parameter settings and calculation process: After processing with S1 and initial filtering with S2, the data obtained within the quasi-steady-state range is as follows: 1. Rated power of the unit MW.

[0032] 2. Current load target instruction MW.

[0033] 3. Calculated mean power over the interval MW.

[0034] 4. Calculated standard deviation of interval power MW.

[0035] 5. The weighting coefficient is set to , .

[0036] Substitute the above values ​​into the formula to calculate: First item (deviation item): ; Second item (fluctuation item): ; Weighted calculation: ; ; The steady-state determination index was calculated. The value is 0.0011. The system's preset steady-state verification threshold is 0.005. Comparing the calculated result with the steady-state verification threshold, 0.0011 < 0.005. Simultaneously, the absolute deviation is checked separately. MW, the set dead zone range is Rated power is MW. Since 0.8MW < 6MW, and If the steady-state confirmation threshold requirement is met, the system determines that the unit has accurately responded to the command and is in a highly stable state, and generates a steady-state confirmation flag.

[0037] The results indicate that the unit's current operating conditions meet the triggering conditions for precision testing in terms of both numerical accuracy and stability.

[0038] The aforementioned mathematical morphology filtering process refers to a method of analyzing and processing the signal shape using structuring elements. The erosion operation is used to shrink the signal boundary to eliminate small noise points, while the dilation operation is used to expand the signal boundary to fill the voids or cracks inside the signal.

[0039] Please see Figure 1 and Figure 4 S3: Extract the minimum sampling duration based on the steady-state confirmation flag, calculate the time difference between the next load instruction and the current time to generate the expected steady-state window, and generate a high-priority trigger acquisition signal if the expected steady-state window is greater than the minimum sampling duration; The specific steps for S3 are as follows: S31: In response to the generation of the steady-state confirmation flag, read the vibration signal acquisition standard required for the current detection item from the preset detection task configuration table, extract the minimum continuous data length required to meet the frequency domain resolution requirements, and generate the minimum sampling duration; S32: Obtain the expected issuance time of the next pre-planned load adjustment instruction through the data interface of the power plant monitoring system, calculate the time difference between the expected issuance time and the current system time, assess the remaining time length that the current steady-state condition can be maintained, and generate the expected steady-state window; The calculation process for the steady-state window is expected to include: By using the dispatch interface of the power plant's automatic generation control system, query the planned execution timestamp of the first unexecuted instruction in the dispatch team list; If the query result is empty or the planned execution timestamp is earlier than the current time, the default setting is to use a future preset long period of time as the stable period for the unit to maintain its current output, and the long period of time is directly assigned as the window length. If a valid future plan execution timestamp is found, the difference between the timestamp and the current system atomic clock time is calculated in seconds. A preset safety redundancy coefficient is introduced to reduce the difference to reserve the mechanical delay time before the unit responds to the command. The remaining available sampling time is then used to generate the expected steady-state window. S33: Compare the expected steady-state window with the minimum sampling duration to determine whether the remaining steady-state time is sufficient to complete a complete vibration data acquisition task. If and only if the value of the expected steady-state window is strictly greater than the minimum sampling duration, activate the acquisition task triggering logic and generate a high-priority trigger acquisition signal.

[0040] Once the steady-state confirmation flag is generated in step S2, the system immediately enters the time window evaluation process. First, it consults the "Detection Task Configuration Table" stored on the server. For the current detection project, "Vibration Spectrum Analysis of the Turbine Top Cover," the detection task configuration table requires a frequency resolution of at least 0.5Hz and 16 averages to eliminate noise. According to signal processing principles, the single sampling length is the reciprocal of the frequency resolution, i.e. seconds. The minimum continuous data length required to complete 16 averages is calculated as follows: Seconds. Considering the edge effects caused by data truncation and windowing, a redundancy coefficient of 1.25 is introduced to finally determine the minimum sampling duration. Second.

[0041] Subsequently, the system accesses the dispatch queue interface of the power plant's automatic generation control system via the IEC60870-5-104 communication protocol. The query result shows that the current time is 10:15:00, and the expected timestamp for the next planned load adjustment command in the queue is 10:16:30. The system calculates the difference between the two: 90 seconds. This represents the theoretical maximum steady-state holding time. To prevent data acquisition interruption due to network latency or premature adjustments by the automatic power generation control system, a safety redundancy logic is introduced: a fixed mechanical response delay of 10 seconds is subtracted from the 90 seconds to calculate the expected steady-state window. Second.

[0042] Finally, the numerical comparison logic is executed: the expected steady-state window is... (80 seconds) and minimum sampling duration (40 seconds) Compare the results. The judgment logic is as follows: The conditions are met. This means that there is sufficient time to complete a high-quality vibration acquisition that meets the standards before the next instruction changes the unit's operating conditions. Based on this comparison result, the system's internal task scheduler immediately raises the priority of the current acquisition task to the highest level and generates a high-priority trigger acquisition signal, which includes the task ID and the maximum allowed acquisition duration.

[0043] Table 2 illustrates the time parameter evaluation process in this step: Table 2 Sampling Window Evaluation Parameters: ; As shown in Table 2, through rigorous calculation and comparison of the steady-state window, it is ensured that the data acquisition task is executed only under conditions with sufficient time margin, thus avoiding the risk of data acquisition being interrupted by operating condition adjustments.

[0044] The aforementioned IEC60870-5-104 communication protocol is a network transmission protocol widely used in the field of power system automation, used to realize remote monitoring and data transmission between control centers and substations or power plants.

[0045] Please see Figure 1 and Figure 5 S4: Receive high-priority trigger acquisition signal, start vibration acquisition based on high-priority trigger acquisition signal, generate original vibration waveform within minimum sampling time, add tag to original vibration waveform and upload steady-state data; The specific steps for S4 are as follows: S41: The vibration monitoring subsystem listens to high-priority trigger acquisition signals in real time. When a signal level change is detected, it sends a synchronization trigger command to the high-speed data acquisition card at the front end to drive the vibration acceleration sensors of multiple channels to synchronously start the analog-to-digital signal conversion process. S42: Within the time period defined by the minimum sampling duration, continuously read the digital sequence converted by the sensor through the buffer, perform DC component removal and windowing processing on the data stream, extract a complete and continuous time-domain waveform data that can reflect the mechanical vibration characteristics of the unit under the current steady-state load, and generate the original vibration waveform. S43: Obtain the current unit head data, guide vane opening data, and unit active power data. Encapsulate the unit head data, guide vane opening data, and unit active power data as operating condition description metadata into the data packet header, associate and bind them with the original vibration waveform, construct a standardized data packet with operating condition index, and send the data packet to the remote server via industrial Ethernet to upload steady-state data. The process of adding labels to the original vibration waveform in S43 includes: Read the real-time register address of the unit's local control unit, and collect the current upstream water level, downstream water level, guide vane servo travel value, and generator stator current value. The gross head is calculated using the difference between upstream and downstream water levels. The stroke value of the guide vane servo is combined with the pre-stored comprehensive characteristic curve of the turbine operation to determine the relative position zone of the current operating point on the characteristic curve. Encode the capillary head, guide vane relay stroke value, relative position partition number, and acquisition start timestamp down to milliseconds into metadata text in JSON format; Write the metadata text into the header extension field of the original vibration waveform file to establish a strong mapping relationship between the waveform data and the operating conditions, and complete the tag addition.

[0046] The vibration monitoring subsystem continuously polls the status bit of high-priority trigger acquisition signals. When the high-priority trigger acquisition signal changes from "0" to "1", the data acquisition card is immediately activated. In this embodiment, the system controls an 8-channel high-speed synchronous data acquisition card, connected to piezoelectric accelerometers located on the turbine top cover, lower guide bearing, and tailrace pipe. The sampling rate is set to 2560Hz to ensure coverage of the effective analysis frequency band from 0-1000Hz. Within 40 seconds of receiving the trigger signal (i.e., the minimum sampling time)... The data acquisition card continuously converts analog voltage signals and writes the data stream to a high-speed ring buffer. After acquisition, the system reads a data stream of length from the high-speed ring buffer. A digital sequence of data points. Preprocessing is performed on this digital sequence: first, the sequence mean is calculated and subtracted to remove the DC component; then, a Hanning window function is applied to both ends of the time-domain waveform to suppress spectral leakage in subsequent spectral analysis, generating a standardized raw vibration waveform data block.

[0047] Simultaneously, the system reads the real-time registers of the unit's local control unit via the Modbus TCP protocol. At the center of the data acquisition period, the upstream water level is read. meters, downstream water level Meters, guide vane relay stroke Generator stator current A. The system uses formulas The calculated gross head is 105.3 meters. Next, the built-in turbine comprehensive characteristic curve database is invoked, using the gross head of 105.3 meters and the guide vane servo travel of 65.4% as coordinate inputs, to perform spatial interpolation on the characteristic curve. The positioning result shows that the current operating point is located in the "high-efficiency and stable operation zone," defined as "Zone-3."

[0048] Finally, the system encapsulates the aforementioned physical quantities and location information into metadata text in JSON format. For example: {"timestamp":"2025-12-18T10:15:20.000","head":105.3,"guide_vane":65.4,"power":450.8,"zone_id":3}. This metadata text is used as an extended header file and merged with the original vibration waveform data block in binary format to construct a standardized data package with a complete operating condition index. The system establishes an FTP connection with the remote data center and uploads this standardized data package to the historical database for storage, completing a full steady-state data upload task.

[0049] The Hanning window function mentioned above is a raised cosine window function commonly used in signal processing. Its function is to smooth the truncation boundary of the signal's time-domain waveform, thereby reducing sidelobe interference caused by energy leakage during discrete Fourier transform.

[0050] The above embodiments illustrate preferred embodiments of the present invention. Any equivalent adjustments to the technical solution based on software engineering methods are within the scope of protection, including but not limited to: implementing algorithm logic using different programming languages, refactoring functional modules into services, adjusting data interaction protocols, and optimizing resource scheduling strategies. Any implementation scheme derived from reasonable modifications to the data processing flow, service call chain, or system architecture layer without departing from the core technology of the present invention should be considered within the protection scope defined by the technical solution of the present invention.

Claims

1. A method for intelligent scheduling and management of water conservancy and hydropower engineering inspection tasks, characterized in that, Includes the following steps: S1: Obtain the active power data of the generator set, smooth the active power data of the generator set using Savitzky-Golay to generate smoothed power data, generate a power rate sequence by first-order difference of the smoothed power data, generate a power acceleration sequence by second-order difference of the power rate sequence, and construct a load feature matrix based on the power rate sequence and the power acceleration sequence. S2: Compare the load feature matrix with the steady-state threshold group, filter the time periods in the load feature matrix that are within the steady-state threshold group to generate a quasi-steady-state interval, calculate the deviation between the load target command and the active power data of the unit, and generate a steady-state confirmation mark if the deviation is less than the dead zone range. S3: Extract the minimum sampling duration based on the steady-state confirmation flag, calculate the time difference between the next load instruction and the current time to generate an expected steady-state window, and generate a high-priority trigger acquisition signal if the expected steady-state window is greater than the minimum sampling duration; S4: Receive the high-priority trigger acquisition signal, start vibration acquisition based on the high-priority trigger acquisition signal, generate the original vibration waveform within the minimum sampling time, add a label to the original vibration waveform and upload steady-state data.

2. The intelligent scheduling and management method for water conservancy and hydropower engineering testing tasks according to claim 1, characterized in that, The specific steps of S1 are as follows: S11: The local control unit of the turbine reads the current operating power value of the unit in real time according to the preset sampling frequency. The Savitzky-Golay digital filter is used to perform polynomial least squares fitting on the obtained discrete power points to remove high-frequency random noise interference and retain the low-frequency trend characteristics of power change, thereby generating the smooth power data. S12: Perform a forward first-order difference operation on the smoothed power data in the time dimension to calculate the power change amplitude between adjacent sampling points to reflect the speed of power change, generate the power rate sequence, and perform a second difference operation on the power rate sequence to calculate the trend of power change rate to reflect the acceleration and deceleration state of unit regulation, and generate the power acceleration sequence. S13: Using timestamps as index bases, align and stitch the smoothed power data, the power rate sequence, and the power acceleration sequence on the same time dimension to construct a multi-dimensional time series set including power amplitude dimension, change rate dimension, and change acceleration dimension, and generate the load feature matrix.

3. The intelligent scheduling and management method for water conservancy and hydropower engineering testing tasks according to claim 1, characterized in that, The specific steps of S2 are as follows: S21: Obtain the pre-set steady-state threshold group, extract the feature vector of each time segment in the load feature matrix, determine row by row whether the power rate value in the feature vector is less than the rate threshold in the steady-state threshold group, and whether the power acceleration value is less than the acceleration threshold in the steady-state threshold group, and filter the continuous time segments that meet the dual constraint conditions. S22: Calculate the duration of the continuous time segment. If the duration exceeds the preset minimum stable period, determine that the unit operating condition during the time period is in a stable operating state during the non-adjustment process. Mark the start and end times of the time period as the effective interval boundary and generate the quasi-steady state interval. S23: Obtain the current load target command issued by the automatic power generation control system, calculate the absolute difference between the mean of the smoothed power data in the quasi-steady-state interval and the load target command, compare the absolute difference with the preset dead zone range, and when the absolute difference falls within the dead zone range, confirm that the unit has accurately responded to the load command and achieved output balance, and generate the steady-state confirmation flag.

4. The intelligent scheduling and management method for water conservancy and hydropower engineering inspection tasks according to claim 1, characterized in that, The specific steps of S3 are as follows: S31: In response to the generation of the steady-state confirmation flag, read the vibration signal acquisition standard required for the current detection item from the preset detection task configuration table, extract the minimum continuous data length required to meet the frequency domain resolution requirements, and generate the minimum sampling duration; S32: Obtain the expected issuance time of the next pre-planned load adjustment instruction through the data interface of the power plant monitoring system, calculate the time difference between the expected issuance time and the current system time, assess the remaining time length that the current steady-state condition can be maintained, and generate the expected steady-state window; S33: Compare the expected steady-state window with the minimum sampling duration to determine whether the remaining steady-state time is sufficient to complete a complete vibration data acquisition task. If and only if the value of the expected steady-state window is strictly greater than the minimum sampling duration, activate the acquisition task triggering logic and generate the high-priority trigger acquisition signal.

5. The intelligent scheduling and management method for water conservancy and hydropower engineering testing tasks according to claim 1, characterized in that, The specific steps of S4 are as follows: S41: The vibration monitoring subsystem listens to the high-priority trigger acquisition signal in real time. When a signal level change is detected, it sends a synchronization trigger command to the high-speed data acquisition card at the front end to drive the vibration acceleration sensors of multiple channels to synchronously start the analog signal to digital signal conversion process. S42: Within the time period defined by the minimum sampling duration, continuously read the digital sequence converted by the sensor through the buffer, perform DC component removal and windowing processing on the data stream, extract a complete and continuous time-domain waveform data that can reflect the mechanical vibration characteristics of the unit under the current steady-state load, and generate the original vibration waveform. S43: Obtain the current unit head data, guide vane opening data, and active power data of the unit; encapsulate the unit head data, guide vane opening data, and active power data of the unit as operating condition description metadata into the data packet header; associate and bind them with the original vibration waveform; construct a standardized data packet with an operating condition index; and send the data packet to a remote server via industrial Ethernet to upload steady-state data.

6. The intelligent scheduling and management method for water conservancy and hydropower engineering testing tasks according to claim 2, characterized in that, The process of constructing the load characteristic matrix includes: Obtain the smoothed power data, the power rate sequence, and the power acceleration sequence, and calculate the maximum absolute value of each sequence within the current statistical period as a normalization benchmark. The normalization benchmark is used to perform dimensionless processing on each element in multiple sequences to eliminate the influence of numerical differences between different physical dimensions, and the normalized power vector, velocity vector and acceleration vector are obtained. The three normalized vectors are used as column vectors and stacked horizontally in time index order to construct a three-column multi-row numerical matrix, ensuring that each row in the matrix fully represents the instantaneous load dynamic characteristics of the unit at the target sampling time, thus generating the load feature matrix.

7. The intelligent scheduling and management method for water conservancy and hydropower engineering inspection tasks according to claim 3, characterized in that, The process of selecting continuous time segments that satisfy the dual constraints in S21 includes: Traverse all row vectors of the load feature matrix, perform a logical AND operation on each row vector, and determine whether the absolute value of the velocity element in the row vector is less than the preset velocity fluctuation limit, and at the same time determine whether the absolute value of the acceleration element in the row vector is less than the preset acceleration fluctuation limit. Establish a binary logic sequence corresponding to the time axis, set the time that satisfies the above logical AND operation conditions to logical true value, and set the time that does not satisfy the conditions to logical false value; Morphological filtering is performed on the binary logic sequence to remove isolated truth points with a length less than a preset glitch threshold, and discontinuous truth segments with adjacent spacing less than a preset tolerance are merged to obtain a smooth and continuous time index set, which serves as the selected continuous time segment.

8. The intelligent scheduling and management method for water conservancy and hydropower engineering inspection tasks according to claim 4, characterized in that, The calculation process for the expected steady-state window includes: By using the dispatch interface of the power plant's automatic generation control system, query the planned execution timestamp of the first unexecuted instruction in the dispatch team list; If the query result is empty or the planned execution timestamp is earlier than the current time, the default setting is to use a future preset long period of time as the stable period for the unit to maintain its current output, and the long period of time is directly assigned as the window length. If a valid future plan execution timestamp is found, the difference in seconds between the timestamp and the current system atomic clock time is calculated, and a preset safety redundancy coefficient is introduced to reduce the difference to reserve the mechanical delay time before the unit responds to the command. The remaining available sampling time is then used to generate the expected steady-state window.

9. The intelligent scheduling and management method for water conservancy and hydropower engineering testing tasks according to claim 5, characterized in that, The process of adding a label to the original vibration waveform in S43 includes: Read the real-time register address of the unit's local control unit, and collect the current upstream water level, downstream water level, guide vane servo travel value, and generator stator current value. The gross head is calculated using the difference between upstream and downstream water levels. The stroke value of the guide vane servo is combined with the pre-stored comprehensive characteristic curve of the turbine operation to determine the relative position zone of the current operating point on the characteristic curve. The capillary head, guide vane relay stroke value, relative position partition number, and acquisition start timestamp down to milliseconds are encoded into metadata text in JSON format; Write the metadata text into the header extension field of the original vibration waveform file to establish a strong mapping relationship between the waveform data and the operating conditions, and complete the tag addition.

10. The intelligent scheduling and management method for water conservancy and hydropower engineering testing tasks according to claim 3, characterized in that, The process of determining whether the deviation is less than the dead zone range in S23 involves calculating the steady-state determination index, and the formula for calculating the steady-state determination index is as follows: ； in, Represents the steady-state determination index, This represents the arithmetic mean of the smoothed power data within the quasi-steady-state interval. The value representing the load target instruction. This represents the rated power value of the unit. This represents the standard deviation of the smoothed power data within the quasi-steady-state interval. The weighting coefficients representing the deviation term. The weighting coefficients represent the fluctuation term.

Images