A method and system for pre-warning temperature rise of a water turbine generator set in starting operation mode

Abstract

This invention discloses a method and system for early warning of temperature rise during the start-up operation of a hydro-generator unit. The method includes: collecting and preprocessing multi-source operating data of the unit to form a structured dataset; identifying and verifying start-up conditions from the data and extracting key event information; grouping historical valid start-up data by initial temperature and using statistical methods to fit and generate a benchmark temperature rise function and normal fluctuation upper and lower limit functions for each group, constructing a benchmark model library for adaptive ambient temperature; when a new start-up condition is identified, matching the corresponding model according to its initial temperature, comparing and analyzing the real-time temperature rise data with the benchmark in multiple dimensions such as magnitude, rate, and peak time, and triggering corresponding early warnings based on set multi-level thresholds. This invention achieves refined and intelligent monitoring of the temperature rise status during the start-up process, significantly improving the accuracy and timeliness of early warnings.

Description

Technical Field

[0001] This invention relates to an early warning method, and more particularly to an early warning method and system for temperature rise during startup of a hydro-generator unit. Background Technology

[0002] During the start-up process of a hydro-generator unit, the temperature of key components such as the stator windings and bearings exhibits a rapid and non-linear dynamic rise. The magnitude, rate, and peak arrival time of this temperature rise are closely related to the health status of the equipment. Abnormal temperature rises often indicate potential faults such as poor lubrication, cooling failure, or insulation degradation. Therefore, effective monitoring and early warning of the temperature during the start-up process is an important means to achieve predictive maintenance of the unit and avoid major accidents.

[0003] Currently, the industry mainly relies on the following two methods to monitor the temperature of hydro-generator units:

[0004] 1. Fixed Threshold Alarm: This method sets fixed alarm and trip values ​​for the absolute temperature of the monitoring point. While simple and direct, it has significant drawbacks. First, fixed thresholds fail to account for seasonal variations in ambient temperature and the impact of diurnal temperature differences on the initial startup temperature and the temperature rise process, easily leading to false alarms in summer and missed alarms in winter. Second, it only focuses on the absolute value of the temperature "point," ignoring the dynamic trend of the temperature rise "process," making it insensitive to early-stage, slowly changing faults.

[0005] 2. Temperature prediction methods based on machine learning: Existing temperature prediction methods based on machine learning are mostly designed for steady-state operation and rely on long-term data training with smooth and low noise. They are difficult to adapt to the characteristics of rapid temperature rise, strong nonlinearity and sensitivity to initial conditions during the startup process.

[0006] Furthermore, existing monitoring systems typically analyze temperature signals in isolation, failing to perform deep fusion and correlation analysis with multi-source data characterizing unit operating status, such as guide vane opening, active power, and ambient temperature. This makes it difficult for the system to automatically and accurately identify the specific analysis object of "start-up condition" from the continuous operating data stream, and even more difficult to establish differentiated health benchmark models for the start-up process under different downtimes and ambient temperatures. Summary of the Invention

[0007] Objectives of the invention: The first objective of this invention is to provide a method for early warning of temperature rise during startup of hydro-generator units that can achieve early identification and warning, high sensitivity and accuracy of anomaly detection, and has the ability of model self-evolution; the second objective of this invention is to provide a system for implementing the above method.

[0008] Technical solution: The method for early warning of temperature rise during start-up of a hydro-generator unit as described in this invention includes the following steps:

[0009] (1) Data acquisition and preprocessing: Collect continuous multi-source operation data of the hydro-generator unit, preprocess the operation data to obtain a structured operation dataset; the preprocessing includes removing abnormal data, handling missing data and time alignment;

[0010] (2) Identify and verify the power-on condition: Identify and verify the power-on condition from the structured operation dataset and obtain power-on event information, which includes the power-on start time, the power-on process end time and the shutdown duration;

[0011] (3) Constructing a reference temperature rise function: Based on the startup event information, select the effective startup conditions of historical cold start, group the effective startup conditions according to the starting temperature, construct a reference temperature rise function and normal fluctuation upper and lower limit function for each group to characterize the relationship between temperature rise and time, and calculate the typical temperature rise rate and typical peak arrival time for each group.

[0012] (4) Real-time monitoring and multi-level early warning: Calculate the temperature rise and temperature rise rate of the new start-up condition and record the peak temperature rise rate. Determine the corresponding group according to the starting temperature of the new start-up condition, call the corresponding reference temperature rise function, normal fluctuation upper and lower limit function, typical temperature rise rate and typical peak arrival time, and compare and analyze the real-time monitored and calculated data with the called data. The comparison and analysis includes at least comparing the real-time temperature rise with the normal fluctuation upper and lower limit function, comparing the real-time average temperature rise rate with the typical temperature rise rate, and comparing the real-time temperature rise rate peak arrival time with the typical peak arrival time. Execute multi-level early warning based on the comparison and analysis results.

[0013] Preferably, the abnormal data removal in step (1) refers to removing data points that exceed the physical reasonable range in temperature, active power, guide vane opening, or unit speed.

[0014] Preferably, the missing data processing in step (1) includes using linear interpolation to complete the data in consecutive missing time periods not exceeding a first threshold, and marking consecutive missing time periods exceeding the first threshold as invalid.

[0015] Preferably, the first threshold is two sampling periods.

[0016] Preferably, the identification and verification of the power-on condition in step (2) includes the following steps:

[0017] (21) Calculate the first difference of the guide vane opening sequence. When the increment of two consecutive sampling points is greater than the preset second threshold, record it as the potential start-up time.

[0018] (22) Starting from the potential start-up time, search backwards. When the active power fluctuation and the guide vane opening change rate meet the preset conditions, record it as the end time of the start-up process.

[0019] (23) Determine the startup condition by the power drop ratio, guide vane closure status or shutdown signal status from the potential startup start time to the startup process end time, and eliminate invalid startup conditions in the startup conditions;

[0020] (24) Verify that the unit was in a shutdown state before the potential start-up time, and calculate the shutdown duration from the last shutdown time to the current start-up time.

[0021] Preferably, the second threshold in step (21) is 10%.

[0022] Preferably, the preset condition in step (22) is that the standard deviation of active power fluctuation is less than 2% and the guide vane opening change rate is less than 1% / min, and the duration of this state is not less than 2 minutes.

[0023] Preferably, step (3) involves constructing a baseline temperature rise function and normal fluctuation upper and lower limit functions for each group, including:

[0024] (31) For all valid start-up conditions within the group, calculate the median, 5th percentile and 95th percentile of the temperature rise at each relative time point;

[0025] (32) Using relative time as the independent variable and the median temperature rise as the dependent variable, the least squares method is used to fit a polynomial to generate the reference temperature rise function for this group.

[0026] (33) Based on the 5th percentile and 95th percentile, construct the lower limit function and upper limit function of normal fluctuation.

[0027] Preferably, the multi-level warning in step (4) is a three-level warning, and the triggering condition of the warning is based on a combination of one or more of the following: the magnitude of the real-time temperature rise exceeding the upper limit function of the normal fluctuation, the proportion of the real-time average temperature rise rate exceeding the typical temperature rise rate, and the difference between the arrival time of the real-time temperature rise rate peak and the arrival time of the typical peak.

[0028] The hydro-generator unit start-up temperature rise early warning system of the present invention includes:

[0029] The data acquisition and preprocessing module is used to acquire and preprocess continuous multi-source operating data of the hydro-generator unit to form a structured operating dataset;

[0030] The power-on condition identification module is used to identify and verify the power-on condition from the structured operation dataset and extract power-on event information;

[0031] The benchmark model library construction module is used to construct benchmark temperature rise functions and normal fluctuation upper and lower limit functions based on the selected effective start-up conditions of historical cold start, grouped by starting temperature, and store them as a benchmark model library.

[0032] The online monitoring and early warning module is used to call the corresponding model from the benchmark model library after a new startup condition is identified, compare and analyze the real-time temperature rise data with the benchmark, and execute multi-level early warnings.

[0033] Beneficial effects: Compared with the prior art, the present invention has the following significant advantages: (1) It realizes early identification, hierarchical alarm and adaptive early warning of abnormal temperature rise during startup; (2) It combines the three criteria of temperature rise amplitude, rate and peak arrival time to improve the sensitivity and accuracy of abnormal detection; (3) It supports sensor performance verification and online model evolution to realize continuous optimization of early warning strategy; (4) It improves the autonomous operation safety and intelligent operation and maintenance level of small hydropower stations under centralized control mode. Attached Figure Description

[0034] Figure 1 This is a flowchart of the present invention;

[0035] Figure 2 It is the temperature curve of the power-on event under effective power-on conditions. Detailed Implementation

[0036] The technical solution of the present invention will be further described below with reference to the accompanying drawings.

[0037] This invention provides a method for early warning of temperature rise during startup of a hydro-generator unit. The method constructs a personalized temperature rise benchmark model based on historical operating data and achieves graded early warning of temperature status during startup through multi-dimensional comparison of real-time data with the benchmark model. The overall process is attached. Figure 1 As shown, the process mainly includes four steps: data acquisition and preprocessing, identification and verification of startup conditions, construction of a baseline temperature rise function, and real-time monitoring and multi-level early warning.

[0038] Step 1: Data Acquisition and Preprocessing

[0039] This step aims to obtain high-quality, time-synchronized multi-source running datasets to lay the foundation for subsequent analysis.

[0040] 1. Data Collection

[0041] Continuous collection of multi-source operational data related to the start-up temperature rise of the hydro-generator unit. The data is determined based on the turbine generator unit model and measuring points, and includes:

[0042] (1) Operating parameters: active power, guide vane opening, unit speed.

[0043] (2) Process parameters: stator winding temperature and bearing temperature.

[0044] (3) Environmental parameters: ambient temperature of the plant.

[0045] All data were collected at a fixed sampling period and initially organized into a dataset arranged in time series.

[0046] 2. Data Preprocessing

[0047] The original dataset is preprocessed to form a structured runtime dataset that can be used for analysis.

[0048] (1) Abnormal data removal: Remove data points that exceed the physical reasonable range in temperature, active power, guide vane opening or unit speed.

[0049] (2) Missing data processing: Missing values ​​in the data are processed. A threshold for consecutive missing values ​​is set to two sampling periods. For periods with consecutive missing values ​​not exceeding this threshold, linear interpolation of the preceding and following valid data is used to complete the data; for periods with consecutive missing values ​​exceeding this threshold, the entire period is marked as "invalid data" and is not included in subsequent modeling and analysis.

[0050] (3) Time alignment: Since the data from different sensors may have slight delays, the timestamps of all parameters are uniformly calibrated to ensure that the correlation analysis is carried out under the same time reference, so as to achieve accurate time alignment of multi-source data.

[0051] Step 2: Identify and verify the power-on status

[0052] This step automatically and accurately extracts the valid boot process time periods from continuously running data.

[0053] 1. Identify potential boot start times

[0054] Extract the preprocessed guide vane opening time series, calculate its first-order difference, and set a guide vane opening mutation threshold. Based on expert experience and data analysis, the optimal value for this threshold is 6%-10%. When the guide vane opening increments at two consecutive sampling points are both greater than this threshold, the time of the first increment point is recorded as the potential start-up time t. start .

[0055] 2. Identify the end time of the boot process

[0056] From t start Begin the backward search, calculating the rolling standard deviation of active power and the rate of change of guide vane opening. When both of the following conditions are simultaneously met and maintained for at least 2 minutes, the unit is deemed to have entered a stable operating state, and this moment is recorded as the end time t of the start-up process. end .

[0057] Condition (1): The standard deviation of active power fluctuation is less than 2%;

[0058] Condition (2): The rate of change of guide vane opening is less than 1% / min.

[0059] 3. Eliminate invalid startup conditions

[0060] Check from t start To t end Does any of the following events occur within the time interval:

[0061] (1) Active power decreases by more than 20% of its current value;

[0062] (2) Guide vanes closed;

[0063] (3) A shutdown signal appears.

[0064] If the above event occurs, the power-on condition will be deemed invalid and discarded, and will not be used for subsequent modeling.

[0065] 4. Verify and calculate downtime.

[0066] For power-on events that pass the above checks, further verification of t is needed. start The unit was in a shutdown state before time t. start The temperature at the monitoring point deviates from the ambient temperature within a reasonable range. Simultaneously, the exact last shutdown time t is determined from historical data. last_shutdown Calculate the downtime Δt before this startup. off = t start – t last_shutdown Complete boot event information includes: t start t end and Δt off .

[0067] Step 3: Construct the reference temperature rise function

[0068] This step utilizes historical valid start-up data to establish a temperature rise baseline model for the unit under different ambient temperatures.

[0069] 1. Selecting and modeling samples

[0070] Cold start samples for building the baseline model were selected from all historical boot events. The selection criteria were:

[0071] (1) Stop duration Δt off The time exceeds the preset cooling time threshold. This threshold is determined through analysis of historical data.

[0072] (2) The steady-state running time after startup shall not be less than the preset duration to ensure that the complete startup temperature rise process can be captured. The preset duration shall be determined based on the experience of unit commissioning and testing.

[0073] Samples that meet all of the above conditions are marked as valid power-on conditions.

[0074] 2. Grouped by ambient temperature

[0075] The temperature T at the start of power-on start Based on this, and following the principle of similar temperature rise trends, all valid start-up conditions are grouped. For example: see attached... Figure 2 As shown, the temperature rise trend at the start of power-on is similar between 31 and 34 degrees Celsius, and similar between 36 and 37 degrees Celsius. Therefore, the power-on conditions with a temperature of 31 to 34 degrees Celsius at the start of power-on are divided into one group, and the power-on conditions with a temperature of 36 to 37 degrees Celsius at the start of power-on are divided into another group.

[0076] 3. Calculate group statistics

[0077] For all valid operating conditions within a specific temperature group, at each relative time point t, calculate the median, 5th percentile, and 95th percentile of the temperature rise for all operating conditions within that specific temperature group.

[0078] 4. Constructing the benchmark and fluctuation boundary functions

[0079] Using relative time t as the independent variable and median temperature rise as the dependent variable, a polynomial curve is fitted using the least squares method, which serves as the baseline temperature rise function T for this group. base (t).

[0080] Construct the upper limit function T for the normal temperature rise fluctuation of this group using the 5% and 95% quantiles respectively. upper (t) and lower bound function T lower (t) serves as the boundary for judging the temperature rise range.

[0081] 5. Extract key feature parameters for grouping

[0082] (1) Typical temperature rise rate R med Calculate the time t for each operating condition within the group, starting from the start-up time. start At the end time t end Average temperature rise rate R i =[T i (t start )- T i (t end )] / ( t end - t start Take all R. iThe median is used as the typical temperature rise rate R for this group. med .

[0083] (2) Typical peak arrival time t peak For each operating condition, calculate its temperature rise rate r. i (t)=dT i / dt, and find r i (t) The time t when the maximum value is reached peak,i Calculate the relative time t for the peak temperature rise rate to be reached. peak,i =t peak,i - t start,i Take all t peak,i The median is used as the typical peak arrival time t for this group. peak .

[0084] 6. Storage baseline model

[0085] The reference temperature rise function T for each temperature group base (t), upper and lower bound functions T lower (t), T upper (t) Typical temperature rise rate R med Typical peak arrival time t peak These are stored to form the start-up temperature rise baseline model library for the unit.

[0086] Step 4: Real-time monitoring and multi-level early warning

[0087] When a new unit starts up, the system calls the baseline model for real-time comparison and early warning.

[0088] 1. Model matching

[0089] The system identifies the start time and start temperature of a new startup condition, determines the corresponding ambient temperature group based on the start temperature, and calls the corresponding baseline function and feature parameters from the model library.

[0090] 2. Real-time computation and feature extraction

[0091] (1) Real-time temperature rise curve: Calculate the temperature rise ΔT real (t)=T(t) – T(t start );

[0092] (2) Real-time average temperature rise rate: based on the start-up time t start The average temperature rise rate R up to the current time t is calculated in real time, with zero as the starting point. real (t)=[T(t) – T(t start )] / t; until the end time t of the boot process is detected. end Calculate the final average temperature rise rate R. real and with Rmed Compare.

[0093] (3) Real-time peak arrival time: based on the power-on start time t start Starting from zero, calculate the real-time temperature rise rate r(t) = dT / dt. If a peak temperature rise rate is detected, record the corresponding relative arrival time t. real-peak =t real-peak – t start and t peak Compare.

[0094] 3. Multi-level early warning judgment

[0095] A three-level early warning mechanism is set up, and its triggering conditions are based on a comprehensive judgment of the following three dimensions. The early warning threshold parameters (a, b, c, d, e, f, g) need to be determined based on the historical operating data of the corresponding units.

[0096] (1) Level 1 warning (moderate abnormality): triggered when any of the following conditions are met.

[0097] Condition 1: There exists a time t (t ∈ [0, t)). end ]), making ΔT real (t)>T upper (t)+a.

[0098] Condition 2: R real >b×R med .

[0099] Condition 3: T real_peak <T peak – c.

[0100] (2) Level II warning (low-level anomaly): triggered when any of the following conditions are met, but the Level I warning is not reached.

[0101] Condition 1: T upper (t)<ΔT real (t)≤T upper (t)+a.

[0102] Condition 2: d×R med <R real ≤b×R med .

[0103] Condition 3: T peak – e≤T real_peak < T peak – c.

[0104] (3) Level 3 alert (early attention): It is triggered when any of the following conditions are met, and the first or second level warning is not reached.

[0105] Condition 1: ΔTreal (t)> f×T upper (t).

[0106] Condition 2: R real (t)>g×R med .

[0107] When the warning conditions are met, the system automatically generates a warning message containing the warning level, type, and trigger time, and pushes it to the front end of the monitoring system. If the warning conditions are not met, only the comparison results of the real-time temperature rise curve and the baseline curve are pushed; then the warning process ends.

[0108] System Implementation

[0109] Corresponding to the above method, the present invention also provides a temperature rise monitoring and early warning system for the start-up process of a hydro-generator unit. This system can be deployed on a power plant monitoring server and includes:

[0110] Data acquisition and preprocessing module: responsible for executing step one.

[0111] Power-on status identification module: responsible for executing step two.

[0112] The benchmark model library construction module is responsible for executing step three and managing the benchmark model library.

[0113] Online monitoring and early warning module: responsible for executing step four, performing real-time comparison, judgment and alarm output.

[0114] The modules work collaboratively to achieve the early warning function described in this invention. Furthermore, the model also has an online evolution function; when the cumulative number of newly added valid events reaches a preset number, or when a preset number of days have passed since the last model update, the system automatically triggers the baseline function and R... med The model is self-evolved through refitting and updating parameters such as the preset number of times and the preset number of days, which are set according to the actual operating conditions of the corresponding units.

Claims

1. A method for early warning of temperature rise during start-up of a hydro-generator unit, characterized in that, Includes the following steps: (1) Data acquisition and preprocessing: Collect continuous multi-source operation data of the hydro-generator unit, preprocess the operation data to obtain a structured operation dataset; the preprocessing includes removing abnormal data, handling missing data and time alignment; (2) Identify and verify the power-on condition: Identify and verify the power-on condition from the structured operation dataset and obtain power-on event information, which includes the power-on start time, the power-on process end time and the shutdown duration; (3) Constructing a reference temperature rise function: Based on the startup event information, select the effective startup conditions of historical cold start, group the effective startup conditions according to the starting temperature, construct a reference temperature rise function and normal fluctuation upper and lower limit function for each group to characterize the relationship between temperature rise and time, and calculate the typical temperature rise rate and typical peak arrival time for each group. (4) Real-time monitoring and multi-level early warning: Calculate the temperature rise and temperature rise rate of the new start-up condition and record the peak temperature rise rate. Determine the corresponding group according to the starting temperature of the new start-up condition, call the corresponding reference temperature rise function, normal fluctuation upper and lower limit function, typical temperature rise rate and typical peak arrival time, and compare and analyze the real-time monitored and calculated data with the called data. The comparison and analysis includes at least comparing the real-time temperature rise with the normal fluctuation upper and lower limit function, comparing the real-time average temperature rise rate with the typical temperature rise rate, and comparing the real-time temperature rise rate peak arrival time with the typical peak arrival time. Execute multi-level early warning based on the comparison and analysis results.

2. The early warning method according to claim 1, characterized in that, The abnormal data removal mentioned in step (1) refers to removing data points that exceed the physical reasonable range in temperature, active power, guide vane opening, or unit speed.

3. The early warning method according to claim 1, characterized in that, The missing data processing in step (1) includes using linear interpolation to complete the data in consecutive missing time periods not exceeding the first threshold, and marking consecutive missing time periods exceeding the first threshold as invalid.

4. The early warning method according to claim 3, characterized in that, The first threshold is two sampling periods.

5. The early warning method according to claim 1, characterized in that, Step (2) involves identifying and verifying the power-on status, which includes the following steps: (21) Calculate the first difference of the guide vane opening sequence. When the increment of two consecutive sampling points is greater than the preset second threshold, record it as the potential start-up time. (22) Starting from the potential start-up time, search backwards. When the active power fluctuation and the guide vane opening change rate meet the preset conditions, record it as the end time of the start-up process. (23) Determine the startup condition by the power drop ratio, guide vane closure status or shutdown signal status from the potential startup start time to the startup process end time, and eliminate invalid startup conditions in the startup conditions; (24) Verify that the unit was in a shutdown state before the potential start-up time, and calculate the shutdown duration from the last shutdown time to the current start-up time.

6. The early warning method according to claim 5, characterized in that, In step (21), the second threshold is 10%.

7. The early warning method according to claim 5, characterized in that, The preset conditions in step (22) are that the standard deviation of active power fluctuation is less than 2% and the guide vane opening change rate is less than 1% / min, and the duration of this state is not less than 2 minutes.

8. The early warning method according to claim 1, characterized in that, The steps in step (3) described above for constructing the baseline temperature rise function and the upper and lower limits of normal fluctuation functions for each group include: (31) For all valid start-up conditions within the group, calculate the median, 5th percentile and 95th percentile of the temperature rise at each relative time point; (32) Using relative time as the independent variable and the median temperature rise as the dependent variable, the least squares method is used to fit a polynomial to generate the reference temperature rise function for this group. (33) Based on the 5th percentile and 95th percentile, construct the lower limit function and upper limit function of normal fluctuation.

9. The early warning method according to claim 1, characterized in that, The multi-level warning mentioned in step (4) is a three-level warning. The triggering condition of the warning is based on one or more of the following: the magnitude of the real-time temperature rise exceeding the upper limit function of the normal fluctuation, the proportion of the real-time average temperature rise rate exceeding the typical temperature rise rate, and the difference between the arrival time of the real-time temperature rise rate peak and the arrival time of the typical peak.

10. A temperature rise early warning system for hydro-generator unit start-up operation, characterized in that, include: The data acquisition and preprocessing module is used to acquire and preprocess continuous multi-source operating data of the hydro-generator unit to form a structured operating dataset; The power-on condition identification module is used to identify and verify the power-on condition from the structured operation dataset and extract power-on event information; The benchmark model library construction module is used to construct benchmark temperature rise functions and normal fluctuation upper and lower limit functions based on the selected effective start-up conditions of historical cold start, grouped by starting temperature, and store them as a benchmark model library. The online monitoring and early warning module is used to call the corresponding model from the benchmark model library after a new startup condition is identified, compare and analyze the real-time temperature rise data with the benchmark, and execute multi-level early warnings.

Images