Hydro-generator stator temperature monitoring and health assessment method based on multi-algorithm fusion

Abstract

The application discloses a kind of water turbine generator stator temperature monitoring and health evaluation method based on multi-algorithm fusion.The method constructs stator bar, iron core and tooth pressure plate, bus ring, air cooler and cooling system measuring point, collects and pre-processes temperature, flow and operating condition data;Respectively using Mahalanobis distance and chi-square distribution for multivariate anomaly detection, using isolated forest to calculate health degree, based on historical data to generate personalized dynamic threshold, using ARIMA and ARIMAX to predict temperature trend, and according to air cooler in and out air temperature difference, in and out water temperature difference and cooling water flow to evaluate heat transfer efficiency;Standardization is weighted to each analysis result, combined with abnormal measuring point position, temperature rise rate, cooling flow and heat transfer efficiency change, realize stator health classification, abnormal positioning and fault tracing.The application can improve the accuracy of early thermal anomaly identification and fault diagnosis under complex variable conditions.

Description

Technical Field

[0001] This invention relates to the field of stator temperature monitoring of hydro-generators, and more specifically to a method for stator temperature monitoring and health assessment of hydro-generators based on multi-algorithm fusion. Background Technology

[0002] Hydropower turbines are the core equipment in hydropower stations, converting mechanical energy into electrical energy. Their operating status directly affects the safety, stability, and efficiency of the hydropower unit. Components such as stator windings, stator cores, pressure plates, and busbars are affected by factors such as electromagnetic losses, insulation aging, increased contact resistance, and changes in cooling capacity during long-term operation. When localized overheating is not detected and addressed in a timely manner, it may further lead to decreased insulation performance, winding overheating, core damage, or unplanned unit shutdown. Therefore, online monitoring of the temperature of the hydropower turbine stator and cooling system, and identifying thermal anomalies based on monitoring data, is an important technical means for condition-based maintenance and fault early warning of hydropower units.

[0003] Existing hydro-generator temperature monitoring systems typically install temperature sensors at locations such as stator windings, stator core, bearings, air inlets and outlets, or air coolers. They determine overheating based on whether the temperature at these measuring points exceeds preset alarm values. While this method is simple in structure and provides a direct response, generator temperature is influenced by multiple factors, including active power, operating head, cooling medium flow rate, inlet water temperature, ambient temperature, and equipment aging. For units operating at low loads for extended periods, experiencing frequent peak shaving, or subject to fluctuating cooling conditions, relying solely on fixed temperature limits or single-point temperature changes is insufficient to adequately identify early anomalies before reaching critical temperatures.

[0004] Existing technology patent CN112855410B discloses a visualization monitoring system for the temperature field of a hydro-generator and its construction method. This technology installs temperature sensors on the stator, rotor, bearings, and air cooler of the hydro-generator, performs missing value identification, outlier processing, and spatial interpolation on the collected data, and displays the real-time temperature field of the hydro-generator through a 3D model, thereby improving the intuitiveness of temperature monitoring results. This technology can also use abnormal measuring points and historical fault data for operational analysis and maintenance reference.

[0005] Invention patent CN113588123B discloses a stator winding temperature early warning method. This method extracts continuous historical temperature values ​​from stator winding measuring points, calculates the predicted temperature value for the current moment based on the identified coefficient vector, and compares the deviation between the predicted and actual values ​​with a threshold value. When the number of anomalies within a certain judgment interval reaches a set number, an early warning signal is issued. Compared with fixed limit value alarms, this method can form a dynamically changing temperature judgment benchmark based on the recent operating characteristics of the measuring points themselves, and reduces the possibility of false alarms caused by instantaneous data fluctuations by accumulating the number of anomalies.

[0006] Furthermore, existing temperature monitoring technologies typically employ fixed thresholds, statistical prediction, spatial interpolation, or single anomaly detection models, with each method providing relatively independent outputs. A multi-level, joint judgment mechanism for the thermal state of hydro-generators is lacking. For example, while the temperature at a certain measuring point may not exceed the specified limit, it may still deviate significantly from the normal temperature under the same load and cooling conditions; or multiple measuring points may show temperature increases, but the cause could be a decrease in the cooling medium flow rate, rather than a simultaneous failure of multiple components. Using only a single threshold or a single prediction model may not be sufficient to balance anomaly detection sensitivity and false alarm control.

[0007] Meanwhile, there are clear spatial and thermal coupling relationships between the various temperature measuring points of the hydro-generator. Localized heating of the stator bars, abnormal heat dissipation of the iron core, increased contact resistance of the busbar rings, and blockage of a certain air cooler can lead to different temperature distributions and change patterns. Although existing technologies can display abnormal measuring points or determine temperature deviations, they lack unified fusion rules for situations where there are consistency or conflict between multi-source monitoring results, and they do not fully combine the spatial location of the measuring points, temperature change trends, and heat exchange status of the air cooler to attribute faults. Therefore, the monitoring results still rely heavily on the experience and judgment of the operators, making it difficult to form unified, quantitative, and traceable health assessment results.

[0008] In summary, the existing technology has at least the following areas that require improvement: The correlation between different types of monitored variables is not fully utilized, making it difficult to identify situations where a single variable is still within the normal range, but the combined relationship of multiple variables has become abnormal. Fixed thresholds or dynamic benchmarks based solely on historical temperatures at a single measurement point are insufficient to simultaneously adapt to changes in factors such as load, head, cooling flow, and equipment aging. Temperature field display, anomaly detection, trend prediction, and cooling performance analysis are usually independent of each other and lack a comprehensive evaluation mechanism that can verify each other. The lack of quantitative evaluation of the heat exchange capacity of each air cooler makes it difficult to distinguish between abnormal heat-generating components and a decline in the performance of the cooling system. The lack of health indicators that integrate the degree of anomalies, development trends, cooling efficiency, and spatial distribution characteristics makes it difficult to uniformly classify the status of the units. The correspondence between abnormal measuring points and the physical location of equipment and cooling areas is not fully utilized, and fault location and maintenance decisions still rely on human experience.

[0009] Therefore, it is necessary to provide a method for temperature monitoring and health assessment of the stator and cooling system of hydro-generators, which can uniformly process multi-location and multi-type monitoring data, combine multivariate correlation anomaly detection, unsupervised anomaly identification, unit-specific thresholds, temperature trend prediction related to operating conditions, and air cooler heat exchange efficiency evaluation, and fuse different analysis results to improve the reliability of early thermal anomaly identification, status classification, and fault cause location. Summary of the Invention

[0010] To address the above problems, this invention proposes a method for stator temperature monitoring and health assessment of hydro-generators based on multi-algorithm fusion, the method comprising the following steps: S1. For the stator of the hydro-generator, four categories of measuring points are constructed, including stator bar temperature measuring points, stator core and tooth pressure plate temperature measuring points, busbar temperature measuring points, and air cooler and cooling system temperature and flow measuring points. S2. Based on the measurement points in step S1, collect raw data from multiple sources and perform standardized preprocessing; S3. Construct a multivariate detection method for stator temperature based on Mahalanobis distance and chi-square distribution to obtain Mahalanobis distance anomaly detection results; S4. Construct a quantitative assessment method for the operational health of the stator system based on the isolated forest algorithm to obtain the health assessment level; S5. Construct a personalized dynamic threshold adaptive matching and limit-over monitoring method based on historical unit data to obtain personalized threshold over-limit situations; S6. Construct a stator temperature time series trend method based on ARIMA and ARIMAX algorithms to obtain the temperature time series prediction trend; S7. Construct a method for monitoring the temperature trend and cooling efficiency of an air cooler to obtain the real-time heat exchange efficiency of the air cooler; S8. Integrate the Mahalanobis distance anomaly detection results, health assessment level, personalized threshold exceedance situation, temperature time series prediction trend, and real-time heat exchange efficiency of air cooler from steps S3 to S7, and score them according to weight to obtain the stator temperature monitoring and health assessment results of hydro-generator based on multi-algorithm fusion.

[0011] Furthermore, the specific implementation method of step S1 includes the following steps: S1.1. Arrange temperature measuring points for stator bars. On average, multiple temperature measuring points are arranged in each of the upper, middle and lower layers of the winding. S1.2. Arrange temperature measuring points on the stator core and tooth pressure plate: Arrange multiple measuring points evenly on the core, the upper tooth pressure plate of the core, and the lower tooth pressure plate of the core. S1.3. Arrange temperature measuring points on the bus ring: Arrange multiple temperature measuring points evenly on the bus ring; S1.4. Arrange measuring points for air coolers and cooling system: Set up inlet and outlet air temperature measuring points for each of the multiple air coolers, and set up measuring points and flow sensors for the total water supply / total drainage temperature.

[0012] Furthermore, the specific implementation method of step S2 includes the following steps: S2.1. Based on the measurement points in step S1, collect multi-source raw data, standardize and uniformly preprocess the multi-source raw data, and convert the collected raw packaged format data into a standard structured data format. S2.2. Calculate the peak-to-peak value of the temperature and cooling water flow signals one by one according to the fixed sampling period and the number of sampling points, and then calculate the average value. S2.3. Use time-domain signal denoising algorithm to filter out invalid noise data caused by on-site electromagnetic interference and environmental disturbances, mark abnormal data with constant value labeling, and remove invalid abnormal rows of data in batches to complete data cleaning and optimization; S2.4. Perform time-series resampling and alignment on all time-series data at fixed time intervals to ensure that all time-series data are evenly distributed at equal time intervals.

[0013] Furthermore, the specific implementation method of step S3 includes the following steps: S3.1. Select the pre-processed stator bar temperature, core temperature, air cooler inlet and outlet air temperature, and cooling medium flow characteristics to construct a multivariate sample vector; S3.2. Calculate the Mahalanobis distance for a single sample. The formula is as follows: ; in, The Mahalanobis distance for a single sample. This is a sample vector of multidimensional temperature and flow characteristics of the stator. This is the mean vector of all normally functioning samples. The sample covariance matrix, It is the inverse of the covariance matrix; S3.3. Squaring the Mahalanobis distance to make it conform to the chi-square distribution of the corresponding degrees of freedom, and combining the pre-set abnormal sample ratio in the field operation and maintenance, the stator temperature abnormality-specific judgment threshold is adaptively calculated through the statistical characteristics of the chi-square distribution; the Mahalanobis distance squared value calculated in real time is compared with the adaptive threshold point by point to obtain the Mahalanobis distance abnormality detection result.

[0014] Furthermore, the specific implementation method of step S4 includes the following steps: S4.1. Select the pre-processed stator bar temperature and cooling medium flow characteristics to construct the training dataset for the isolated forest algorithm model; S4.2. The isolated forest unsupervised machine learning algorithm is adopted. Multiple isolated decision trees are randomly constructed based on the training dataset. By random feature cutting and random sample sampling, the normal and abnormal samples of the stator are differentiated and isolated. Because the data distribution of abnormal samples deviates from the normal, the isolation cutting path length is significantly shorter than that of normal samples.

[0015] S4.3. Traverse all isolated decision trees to calculate the average isolated path length of a single running sample. After normalization, calculate the sample anomaly score, and then convert it into a stator system health score using a linear conversion formula. The calculation formula is as follows: ; in, The final health score for the stator unit. The normalized outlier score is the sample output by the Isolation Forest algorithm, with a fixed score range of [0,1]. S4.4. Based on the hydropower plant unit operation and maintenance classification standard, the system classifies the operating status into four levels: normal, attention, abnormal, and fault according to the health status threshold, and outputs the quantitative assessment results of the overall health status of the stator in real time.

[0016] Furthermore, step S5 performs normal distribution probability fitting on the preprocessed stator bar temperature, core temperature, and busbar temperature data; based on the proportion of custom abnormal data in power plant operation and maintenance, the upper and lower bounds of the exclusive confidence interval for temperature monitoring are calculated based on the principle of normal distribution statistics, and personalized dynamic early warning thresholds adapted to the operating years, equipment losses, and operating conditions of a single unit are generated.

[0017] Furthermore, the specific implementation method of step S6 includes the following steps: S6.1. For core and critical temperature measurement points such as stator bars, iron core, and busbars, the ARIMA univariate time series prediction model is adopted. The optimal parameters of the model are determined by ADF stationarity test, autocorrelation and partial autocorrelation plot order determination, and after fitting the model coefficients, the short-term and medium-term time series change trends of stator temperature are predicted. S6.2. Combining the active power, operating head, and cooling water flow of the hydro-generator unit, the ARIMAX multivariate coupled time-series prediction model is activated to take into account both the time-series variation of temperature itself and the influence of external operating condition disturbances, and the temperature prediction results are corrected to improve the prediction accuracy under complex variable operating conditions; the effectiveness of the prediction model fitting is verified by white noise test, and the prediction curves of the continuous future temperature changes of each measuring point and the prediction extreme value data are output.

[0018] Furthermore, the specific implementation method of step S7 includes the following steps: S7.1. Batch extract real-time time-series data of inlet and outlet air temperatures for each air cooler within a continuous cycle, calculate temperature characteristic values, and plot temperature trend curves. The formula for calculating the real-time heat exchange temperature difference of the air cooler is: ; in, For the real-time heat exchange temperature difference of the air cooler, This refers to the outlet air temperature of the air cooler. This refers to the air inlet temperature of the air cooler. S7.2. Combining the total inlet water temperature, total outlet water temperature, and real-time cooling water flow rate parameters of the air cooler, calculate the real-time heat exchange efficiency of the air cooler using the following formula: ; in, For the cooling and heat exchange efficiency of the air cooler, The specific heat capacity of cooling water, This refers to the real-time mass flow rate of the cooling water. This refers to the outlet water temperature of the air cooler. This refers to the inlet water temperature of the air cooler. The rated heat exchange capacity of the air cooler; S7.3. Compare the real-time heat exchange efficiency with the rated standard performance threshold, and combine the temperature trend curve fluctuation characteristics from step S7.1 to troubleshoot air cooler faults, and simultaneously mark the abnormal air cooler number and fault degree.

[0019] Compared with the prior art, the beneficial effects of the present invention include: (1) This invention performs joint analysis on multi-source data such as stator bars, iron core, tooth pressure plate, busbar, air cooler and cooling medium flow rate, which can identify the state where a single parameter has not exceeded the limit but the multivariate correlation has become abnormal, thereby improving the early detection capability of local overheating and cooling abnormalities.

[0020] (2) This invention combines historical data of the unit with operating parameters such as active power, operating head, and cooling water flow rate to establish a dynamic judgment benchmark and temperature prediction model that adapts to the characteristics of a single unit, which can reduce false alarms caused by load changes, environmental changes and cooling condition fluctuations.

[0021] (3) This invention converts abnormal detection results into health indicators and classifies the unit status; at the same time, it predicts the development trend of stator temperature and can detect the risk of continuous temperature rise before the temperature reaches the fixed alarm limit, providing a basis for operation adjustment and maintenance arrangement.

[0022] (4) The present invention calculates the heat exchange efficiency based on the air inlet and outlet air temperature, the inlet and outlet water temperature and the cooling water flow rate, and analyzes it together with the stator temperature anomaly and prediction deviation, which is helpful to distinguish the abnormal causes such as stator local heating, poor heat dissipation of iron core, contact heating of bus ring and air cooler blockage, scaling or insufficient flow.

[0023] (5) This invention integrates multivariate statistical detection, unsupervised anomaly identification, personalized threshold, temperature prediction and heat exchange efficiency evaluation results, and combines the physical location and spatial distribution of abnormal measurement points to determine the abnormal area and fault type, thereby reducing the reliance on a single algorithm and human experience and improving the accuracy of condition assessment and maintenance decision-making. Attached Figure Description

[0024] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0025] Figure 1 This is a flowchart of the method for monitoring and assessing the stator temperature of a hydro-generator based on multi-algorithm fusion as described in this invention; Figure 2 This is a schematic diagram of the stator measuring points of the hydro-generator in an embodiment of the method of the present invention; Figure 3 This is a diagram illustrating the temperature time-series prediction effect in an embodiment of the method of the present invention. Figure 4 This is a diagram illustrating the personalized dynamic threshold effect in an embodiment of the method of the present invention. Figure 5 This is a diagram illustrating the three-dimensional fault location of the stator in an embodiment of the method of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention; that is, the described specific embodiments are merely a part of the embodiments of the invention, and not all of them. The components of the specific embodiments of the invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations, and the invention may also have other embodiments.

[0027] Therefore, the following detailed description of specific embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected specific embodiments of the invention. All other specific embodiments obtained by those skilled in the art based on these specific embodiments without inventive effort are within the scope of protection of this invention.

[0028] To further understand the invention's content, features, and effects, the following specific embodiments are provided, along with accompanying drawings. Figures 1 to 5 Detailed explanation is as follows: like Figure 1 As shown, a method for stator temperature monitoring and health assessment of hydro-generators based on multi-algorithm fusion includes the following steps: S1. For the stator of the hydro-generator, four categories of measuring points are constructed, including stator bar temperature measuring points, stator core and tooth pressure plate temperature measuring points, busbar temperature measuring points, and air cooler and cooling system temperature and flow measuring points. Furthermore, the specific implementation method of step S1 includes the following steps: S1.1. Arrange temperature measuring points for stator bars. On average, 24 temperature measuring points are arranged in each of the upper, middle and lower layers of the winding, for a total of 72 temperature measuring points. S1.2. Arrange temperature measuring points on the stator core and tooth pressure plate: 18 measuring points are arranged on the core, the upper tooth pressure plate of the core, and the lower tooth pressure plate of the core, for a total of 54 temperature measuring points. S1.3. Arrange temperature measuring points on the bus ring: Arrange 12 temperature measuring points on the bus ring on an average basis; S1.4. Arrangement of measuring points for air coolers and cooling system: Set up inlet and outlet air temperature measuring points for each of the 20 air coolers, and set up 1 measuring point and 1 flow sensor for the total water supply / total drainage temperature, for a total of 42 measuring points and 2 cooling water flow channels.

[0029] S2. Based on the measurement points in step S1, collect raw data from multiple sources and perform standardized preprocessing; Furthermore, the specific implementation method of step S2 includes the following steps: S2.1. Based on the measurement points in step S1, collect multi-source raw data, standardize and uniformly preprocess the multi-source raw data, and convert the collected raw packaged format data into a standard structured data format. S2.2. Calculate the peak-to-peak value of the temperature and cooling water flow signals one by one according to the fixed sampling period and the number of sampling points, and then calculate the average value. S2.3. Use time-domain signal denoising algorithm to filter out invalid noise data caused by on-site electromagnetic interference and environmental disturbances, mark abnormal data with constant value labeling, and remove invalid abnormal rows of data in batches to complete data cleaning and optimization; S2.4. Perform time-series resampling and alignment on all time-series data at fixed time intervals to ensure that all time-series data are evenly distributed at equal time intervals.

[0030] S3. Construct a multivariate detection method for stator temperature based on Mahalanobis distance and chi-square distribution to obtain Mahalanobis distance anomaly detection results; Furthermore, the specific implementation method of step S3 includes the following steps: S3.1. Select the pre-processed stator bar temperature, core temperature, air cooler inlet and outlet air temperature, and cooling medium flow characteristics to construct a multivariate sample vector; S3.2. Calculate the Mahalanobis distance for a single sample. The formula is as follows: ; in, The Mahalanobis distance for a single sample. This is a sample vector of multidimensional temperature and flow characteristics of the stator. This is the mean vector of all normally functioning samples. The sample covariance matrix, It is the inverse of the covariance matrix; S3.3. Squaring the Mahalanobis distance to make it conform to the chi-square distribution of the corresponding degrees of freedom, and combining the pre-set abnormal sample ratio in the field operation and maintenance, the stator temperature abnormality-specific judgment threshold is adaptively calculated through the statistical characteristics of the chi-square distribution; the Mahalanobis distance squared value calculated in real time is compared with the adaptive threshold point by point to obtain the Mahalanobis distance abnormality detection result. Furthermore, the calculated Mahalanobis distance is squared to obtain the squared Mahalanobis distance value, which strictly follows a chi-square distribution of the number of monitoring variables of the stator of the turbine generator unit with matched degrees of freedom. A fixed proportion of abnormal samples is set based on the actual operation and maintenance needs of the hydropower plant units. Based on the quantile characteristics of the chi-square distribution, the specific threshold for stator temperature anomalies is adaptively calculated. Core calculation formula: ; In the formula: This is the squared value of the Mahalanobis distance. For degrees of freedom The chi-square distribution, The values ​​are matched to the total number of multidimensional feature variables involved in the calculation of the stator. The Mahalanobis distance squared and chi-square distribution adaptive thresholds are calculated in real time by comparing samples one by one, and the output is the accurate identification result of abnormal measurement points, the quantitative value of the degree of stator temperature anomaly, and the special judgment conclusion of parameter exceeding the limit.

[0031] S4. Construct a quantitative assessment method for the operational health of the stator system based on the isolated forest algorithm to obtain the health assessment level; Furthermore, the specific implementation method of step S4 includes the following steps: S4.1. Select the pre-processed stator bar temperature and cooling medium flow characteristics to construct the training dataset for the isolated forest algorithm model; S4.2. The isolated forest unsupervised machine learning algorithm is adopted. Multiple isolated decision trees are randomly constructed based on the training dataset. By random feature cutting and random sample sampling, the normal and abnormal samples of the stator are differentiated and isolated. Because the data distribution of abnormal samples deviates from the normal, the isolation cutting path length is significantly shorter than that of normal samples.

[0032] S4.3. Traverse all isolated decision trees to calculate the average isolated path length of a single running sample. After normalization, calculate the sample anomaly score, and then convert it into a stator system health score using a linear conversion formula. The calculation formula is as follows: ; in, The final health score for the stator unit. The normalized outlier score is the sample output by the Isolation Forest algorithm, with a fixed score range of [0,1]. S4.4. Based on the hydropower plant unit operation and maintenance classification standard, the system classifies the operating status into four levels: normal, attention, abnormal, and fault according to the health status threshold, and outputs the quantitative assessment results of the overall health status of the stator in real time.

[0033] S5. Construct a personalized dynamic threshold adaptive matching and limit-over monitoring method based on historical unit data to obtain personalized threshold over-limit situations; Furthermore, step S5 performs normal distribution probability fitting on the preprocessed stator bar temperature, core temperature, and busbar temperature data; based on the proportion of custom abnormal data in power plant operation and maintenance, the upper and lower bounds of the exclusive confidence interval for temperature monitoring are calculated based on the principle of normal distribution statistics, and personalized dynamic early warning thresholds adapted to the operating years, equipment losses, and operating conditions of a single unit are generated.

[0034] Furthermore, based on the statistical properties of normal distribution and combined with the proportion of abnormal data customized by power plant operation and maintenance personnel, the upper and lower limits of the corresponding confidence interval are calculated. This abandons the industry-standard fixed threshold approach and adapts to the individual operating losses and equipment characteristics of a single unit. Core calculation formula: ; in, For personalized temperature warning upper limit threshold, For personalized temperature warning lower limit threshold, This represents the average of historical normal temperatures. The standard deviation of historical normal temperature samples. To match the normal distribution quantile coefficients of the anomaly proportion, the upper and lower bounds of the confidence interval obtained from the fitting solution are used as the unit-specific dynamic personalized thresholds, replacing the traditional industry-wide fixed alarm thresholds, and adapting to the differentiated monitoring needs of hydro-generator units with different installed capacities and operating years. The system compares the monitored temperature at each measuring point with the personalized upper and lower limit thresholds in real time, and outputs the personalized early warning threshold values ​​for each measuring point and the real-time alarm trigger signal for temperature exceeding the limit.

[0035] S6. Construct a stator temperature time series trend method based on ARIMA and ARIMAX algorithms to obtain the temperature time series prediction trend; Furthermore, the specific implementation method of step S6 includes the following steps: S6.1. For core and critical temperature measurement points such as stator bars, iron core, and busbars, the ARIMA univariate time series prediction model is adopted. After ADF stationarity test, autocorrelation and partial autocorrelation plot order determination, the optimal p, d, and q parameters of the model are determined. After fitting the model coefficients, the short-term and medium-term time series change trends of stator temperature are predicted. Furthermore, temperature time-series data from key stator measurement points are selected, and time-series resampling preprocessing is performed to ensure equal time intervals in the data distribution, forming a stationary univariate time-series sequence to be predicted, which is adapted to the input requirements of the time-series prediction model. The stationarity of the time-series sequence is tested by ADF unit root, and the tailing and trailing characteristics of autocorrelation plots and partial autocorrelation plots are combined to accurately determine the three core parameters of the ARIMA model: autoregression order p, difference order d, and moving average order q. Differential processing is used to eliminate trend fluctuations in the time-series data.

[0036] Furthermore, an ARIMA(p,d,q) univariate temperature time-series prediction model is constructed. Based on historical temperature time-series data, the model coefficients are fitted, and the predicted short- to medium-term temperature changes at the stator measuring points are iteratively calculated. The calculation formula is as follows: ; in, The temperature time series value at time t after d-th order difference is given. These are p-order autoregressive coefficients. The coefficients are the q-th order moving average. Let be the random white noise interference term at time t.

[0037] Furthermore, adopt Ljung-box The white noise test method is used to verify the effectiveness of the model prediction residuals. If the residuals meet the white noise standard, the model fit is valid, and the final output is the real-time predicted value of the stator temperature and the continuous trend curve. First, the prediction residuals are calculated: ; Build Ljung Box The test statistic is calculated using the following model: ; Statistic Approximately follows the degrees of freedom Chi-square distribution .

[0038] The verification process is as follows: Set the null hypothesis Since the residual is white noise, calculate Statistics and Value; if Then accept The model is deemed to be fit effectively; otherwise, it needs to be re-optimized. If the residuals meet the white noise standard, the model is considered to be fit effectively, and the final output is the real-time predicted value of the stator temperature and the continuous trend curve.

[0039] S6.2. Combining the active power, operating head, and cooling water flow of the hydro-generator unit, the ARIMAX multivariate coupled time series prediction model is activated to take into account both the time series variation law of temperature itself and the influence of external operating condition disturbances, and to correct the temperature prediction results to improve the prediction accuracy under complex variable operating conditions; the effectiveness of the prediction model fitting is verified by white noise test, and the prediction curves of the future continuous temperature change of each measuring point and the prediction extreme value data are output. Furthermore, with the temperature at key stator measuring points as the core dependent variable for prediction, operating parameters that directly affect stator temperature rise, such as unit active power, operating head, guide vane opening, and air cooler cooling flow rate, are selected as exogenous explanatory variables to complete synchronous resampling and standardized preprocessing of multivariate time series data.

[0040] The basic orders of the ARIMAX model p, d, and q were determined using the ARIMAX model parameter order determination method. Based on the univariate time series prediction model, exogenous variable regression coefficients were introduced to construct a multivariate coupled time series prediction model, taking into account both the time series regularity of temperature itself and the influence of external operating conditions.

[0041] An ARIMAX(p,d,q) multivariate coupled prediction model is constructed to fit the correlation between exogenous variables and stator temperature, accurately calculating the future trend of stator temperature under complex variable operating conditions, and achieving advanced temperature prediction under multi-factor coupling. The calculation formula is as follows: ; in, Let be the value of the k-th exogenous operating condition variable at time t. For the regression coefficients of the corresponding exogenous variables, This represents the total number of exogenous variables; the remaining parameters are consistent with the ARIMA model definition. The model's prediction accuracy was verified using white noise testing. Based on the prediction results, a temperature advance warning threshold was set to achieve early prediction and tiered advance warning of stator temperature anomalies under varying operating conditions. First, the residual sequence of the temperature prediction model was extracted: ; Calculate the autocorrelation coefficients of each order of residuals : According to the formula Calculate the test statistic Its approximate degree of freedom Chi-square distribution ; With a significance level of 0.05, the null hypothesis... H 0 represents white noise in the residual; if the calculated p ≥ 0.05, then it is accepted. H A value of 0 indicates that the residuals have no significant correlation, the model has fully extracted the temperature time-series information, and the prediction is effective; otherwise, the prediction is rejected. H If the result is 0, the model order needs to be readjusted and the model retrained. Only after passing the verification can the stator temperature prediction results and advance predictions be output.

[0042] S7. Construct a method for monitoring the temperature trend and cooling efficiency of an air cooler to obtain the real-time heat exchange efficiency of the air cooler; Furthermore, the specific implementation method of step S7 includes the following steps: S7.1. Batch extract real-time time-series data of inlet and outlet air temperatures for each air cooler within a continuous cycle, calculate temperature characteristic values, and plot temperature trend curves. The formula for calculating the real-time heat exchange temperature difference of the air cooler is: ; in, For the real-time heat exchange temperature difference of the air cooler, This refers to the outlet air temperature of the air cooler. This refers to the air inlet temperature of the air cooler. S7.2. Combining the total inlet water temperature, total outlet water temperature, and real-time cooling water flow rate parameters of the air cooler, calculate the real-time heat exchange efficiency of the air cooler using the following formula: ; in, For the cooling and heat exchange efficiency of the air cooler, The specific heat capacity of cooling water, This refers to the real-time mass flow rate of the cooling water. This refers to the outlet water temperature of the air cooler. This refers to the inlet water temperature of the air cooler. The rated heat exchange capacity of the air cooler; S7.3. Compare the real-time heat exchange efficiency with the rated standard performance threshold, and combine the temperature trend curve fluctuation characteristics from step S7.1 to troubleshoot air cooler faults, and simultaneously mark the abnormal air cooler number and fault degree.

[0043] S8. Integrate the Mahalanobis distance anomaly detection results, health assessment level, personalized threshold exceedance situation, temperature time series prediction trend, and real-time heat exchange efficiency of air cooler from steps S3 to S7, and score them according to weight to obtain the stator temperature monitoring and health assessment results of hydro-generator based on multi-algorithm fusion.

[0044] Furthermore, the output results of each step S3-S7 are standardized and weighted to obtain a comprehensive anomaly index; multi-indicator linkage logic distinguishes between real faults and single-point interference; and stator typical fault modes are matched according to anomaly characteristics to achieve automatic fault type identification. Combining the actual physical installation distribution of 181 measuring points, and relying on a dynamic display interface with 3D solid modeling, the specific area of ​​stator overheating anomaly and the corresponding anomaly measuring point number are accurately located; based on correlation characteristics such as temperature rise rate, cooling flow rate change, air cooler heat exchange efficiency decay, and health decline, the fault causes are intelligently identified, and different fault types such as local winding overheating, poor core heat dissipation, busbar contact heating, air cooler scaling and blockage, insufficient cooling flow, and insulation degradation are accurately distinguished, thus completing intelligent fault tracing.

[0045] Furthermore, Figure 2 This diagram illustrates the overall distribution of measuring points on the stator of a hydro-generator, showing the specific installation locations of four main categories of measuring points within the stator structure: temperature measuring points in the upper, middle, and lower layers of stator bars (24 points per layer, totaling 72); temperature measuring points in the stator core and upper and lower tooth pressure plates (54 points in total); temperature measuring points in the busbar area (12 points in total); and the locations of inlet and outlet air temperatures, inlet and outlet water temperatures, and flow sensors for the air cooler and cooling system (42 measuring points + 2 flow channels in total). This diagram visually reflects the physical distribution of measuring points throughout the entire space, providing a spatial reference basis for subsequent multi-source data fusion, thermal anomaly area location, and 3D fault display. Simultaneously, it demonstrates that this invention achieves comprehensive, multi-layered monitoring coverage of the stator's thermal state.

[0046] Furthermore, Figure 3 The figure displays the measured temperature curves of several key measuring points (such as conductor bars and iron core) over a continuous time period, along with the temperature trend curves predicted by the ARIMA / ARIMAX models. The figure includes historical measured values ​​(solid blue line), predicted future temperature values ​​(red curve), prediction confidence intervals (pink area), and warning thresholds (dashed orange line). This figure verifies the predictive ability of the ARIMA and ARIMAX models for stator temperature changes under varying operating conditions, demonstrating that the models can detect the risk of continuous temperature rise in advance through time-series prediction before the temperature reaches the alarm limit, achieving the core capability of early warning.

[0047] Furthermore, Figure 4This figure displays the temperature change curve of a specific measuring point over a long period of operation, overlaid with personalized dynamic upper and lower thresholds (rather than fixed thresholds) generated by fitting a normal distribution. The graph shows the historical temperature sample distribution (blue bars), the normal fitting curve (blue curve), the personalized upper threshold (red dashed line), the lower threshold (orange dashed line), and the industry-fixed threshold (gray dashed line). This figure illustrates that the invention can generate adaptive thresholds based on historical data of a single unit, avoiding false alarms or missed alarms caused by fixed limits, and improving adaptability to units with different operating conditions, years of operation, load characteristics, and loss states.

[0048] Furthermore, Figure 5 This is a dynamic display interface built based on a 3D solid model of a hydro-generator. The image includes normal operating measurement points (blue dots), overheating fault points (red asterisks), and fault areas (red dashed boxes). This image provides an intuitive visualization of the spatial location of abnormal measurement points and the causes of faults. It marks abnormal temperature areas (red indicates overheating) with color and highlighting. It can also combine multi-dimensional information such as the spatial location of measurement points, temperature rise rate, and cooling efficiency to help maintenance personnel quickly identify problematic components and areas. Combined with the results of the fusion algorithm, it can complete fault tracing and improve the efficiency of maintenance decision-making.

[0049] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0050] Although this application has been described above with reference to specific embodiments, various modifications can be made and components can be replaced with equivalents without departing from the scope of this application. In particular, as long as there is no structural conflict, the features in the specific embodiments disclosed in this application can be combined with each other in any way. The lack of an exhaustive description of these combinations in this specification is merely for the sake of brevity and resource conservation. Therefore, this application is not limited to the specific embodiments disclosed herein, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for stator temperature monitoring and health assessment of hydro-generators based on multi-algorithm fusion, characterized in that, Includes the following steps: S1. For the stator of the hydro-generator, four categories of measuring points are constructed, including stator bar temperature measuring points, stator core and tooth pressure plate temperature measuring points, busbar temperature measuring points, and air cooler and cooling system temperature and flow measuring points. S2. Based on the measurement points in step S1, collect raw data from multiple sources and perform standardized preprocessing; S3. Construct a multivariate detection method for stator temperature based on Mahalanobis distance and chi-square distribution to obtain Mahalanobis distance anomaly detection results; S4. Construct a quantitative assessment method for the operational health of the stator system based on the isolated forest algorithm to obtain the health assessment level; S5. Construct a personalized dynamic threshold adaptive matching and limit-over monitoring method based on historical unit data to obtain personalized threshold over-limit situations; S6. Construct a stator temperature time series trend method based on ARIMA and ARIMAX algorithms to obtain the temperature time series prediction trend; S7. Construct a method for monitoring the temperature trend and cooling efficiency of an air cooler to obtain the real-time heat exchange efficiency of the air cooler; S8. Integrate the Mahalanobis distance anomaly detection results, health assessment level, personalized threshold exceedance situation, temperature time series prediction trend, and real-time heat exchange efficiency of air cooler from steps S3 to S7, and score them according to weight to obtain the stator temperature monitoring and health assessment results of hydro-generator based on multi-algorithm fusion.

2. The method for stator temperature monitoring and health assessment of hydro-generators based on multi-algorithm fusion according to claim 1, characterized in that, The specific implementation method of step S1 includes the following steps: S1.

1. Arrange temperature measuring points for the stator bars, and arrange temperature measuring points evenly in the upper, middle and lower layers of the winding; S1.

2. Arrange temperature measuring points on the stator core and tooth pressure plate: Arrange temperature measuring points evenly on the core, the upper tooth pressure plate of the core, and the lower tooth pressure plate of the core. S1.

3. Arrange temperature measuring points on the bus ring: Arrange temperature measuring points evenly on the bus ring; S1.

4. Arrange measuring points for air coolers and cooling systems: Set up measuring points for inlet and outlet air temperatures on each air cooler, and set up measuring points and flow sensors for the total water supply / total drainage temperatures.

3. The method for stator temperature monitoring and health assessment of hydro-generators based on multi-algorithm fusion according to claim 1, characterized in that, The specific implementation method of step S2 includes the following steps: S2.

1. Based on the measurement points in step S1, collect multi-source raw data, standardize and uniformly preprocess the multi-source raw data, and convert the collected raw packaged format data into a standard structured data format. S2.

2. Calculate the peak-to-peak value of the temperature and cooling water flow signals one by one according to the fixed sampling period and the number of sampling points, and then calculate the average value. S2.

3. Use time-domain signal denoising algorithm to filter out invalid noise data caused by on-site electromagnetic interference and environmental disturbances, mark abnormal data with constant value labeling, and remove invalid abnormal rows of data in batches to complete data cleaning and optimization; S2.

4. Perform time-series resampling and alignment on all time-series data at fixed time intervals to ensure that all time-series data are evenly distributed at equal time intervals.

4. The method for stator temperature monitoring and health assessment of hydro-generators based on multi-algorithm fusion according to claim 1, characterized in that, The specific implementation method of step S3 includes the following steps: S3.

1. Select the pre-processed stator bar temperature, core temperature, air cooler inlet and outlet air temperature, and cooling medium flow characteristics to construct a multivariate sample vector; S3.

2. Calculate the Mahalanobis distance for a single sample. The formula is as follows: ； in, The Mahalanobis distance for a single sample. This is a sample vector of multidimensional temperature and flow characteristics of the stator. This is the mean vector of all normally functioning samples. The sample covariance matrix, It is the inverse of the covariance matrix; S3.

3. Squaring the Mahalanobis distance to make it conform to the chi-square distribution of the corresponding degrees of freedom, and combining the pre-set abnormal sample ratio in the field operation and maintenance, the stator temperature abnormality-specific judgment threshold is adaptively calculated through the statistical characteristics of the chi-square distribution; the Mahalanobis distance squared value calculated in real time is compared with the adaptive threshold point by point to obtain the Mahalanobis distance abnormality detection result.

5. The method for stator temperature monitoring and health assessment of hydro-generators based on multi-algorithm fusion according to claim 1, characterized in that, The specific implementation method of step S4 includes the following steps: S4.

1. Select the pre-processed stator bar temperature and cooling medium flow characteristics to construct the training dataset for the isolated forest algorithm model; S4.

2. The isolated forest unsupervised machine learning algorithm is adopted. Multiple isolated decision trees are randomly constructed based on the training dataset. Through random feature cutting and random sample sampling, the normal and abnormal samples of the stator are differentiated and isolated. Because the data distribution of abnormal samples deviates from the normal, the isolation cutting path length of abnormal samples is significantly shorter than that of normal samples. S4.

3. Traverse all isolated decision trees to calculate the average isolated path length of a single running sample. After normalization, calculate the sample anomaly score, and then convert it into a stator system health score using a linear conversion formula. The calculation formula is as follows: ； in, The final health score for the stator unit. The normalized outlier score is the sample output by the Isolation Forest algorithm, with a fixed score range of [0,1]. S4.

4. Based on the hydropower plant unit operation and maintenance classification standard, the system classifies the operating status into four levels: normal, attention, abnormal, and fault according to the health status threshold, and outputs the quantitative assessment results of the overall health status of the stator in real time.

6. The method for stator temperature monitoring and health assessment of hydro-generators based on multi-algorithm fusion according to claim 1, characterized in that, Step S5 performs normal distribution probability fitting on the preprocessed stator bar temperature, core temperature, and busbar temperature data; based on the proportion of custom abnormal data in power plant operation and maintenance, calculates the upper and lower bounds of the exclusive confidence interval for temperature monitoring based on the principle of normal distribution statistics, and generates personalized dynamic early warning thresholds that are adapted to the operating years, equipment losses, and operating conditions of a single unit.

7. The method for stator temperature monitoring and health assessment of hydro-generators based on multi-algorithm fusion according to claim 1, characterized in that, The specific implementation method of step S6 includes the following steps: S6.

1. For core and critical temperature measurement points such as stator bars, iron core, and busbars, the ARIMA univariate time series prediction model is adopted. The optimal parameters of the model are determined by ADF stationarity test, autocorrelation and partial autocorrelation plot order determination, and after fitting the model coefficients, the short-term and medium-term time series change trends of stator temperature are predicted. S6.

2. Combining the active power, operating head, and cooling water flow of the hydro-generator unit, the ARIMAX multivariate coupled time-series prediction model is activated to take into account both the time-series variation of temperature itself and the influence of external operating condition disturbances, and the temperature prediction results are corrected to improve the prediction accuracy under complex variable operating conditions; the effectiveness of the prediction model fitting is verified by white noise test, and the prediction curves of the continuous future temperature changes of each measuring point and the prediction extreme value data are output.

8. The method for stator temperature monitoring and health assessment of hydro-generators based on multi-algorithm fusion according to claim 1, characterized in that, The specific implementation method of step S7 includes the following steps: S7.

1. Batch extract real-time time-series data of inlet and outlet air temperatures for each air cooler within a continuous cycle, calculate temperature characteristic values, and plot temperature trend curves. The formula for calculating the real-time heat exchange temperature difference of the air cooler is: ； in, For the real-time heat exchange temperature difference of the air cooler, This refers to the outlet air temperature of the air cooler. This refers to the air inlet temperature of the air cooler. S7.

2. Combining the total inlet water temperature, total outlet water temperature, and real-time cooling water flow rate parameters of the air cooler, calculate the real-time heat exchange efficiency of the air cooler using the following formula: ； in, For the cooling and heat exchange efficiency of the air cooler, The specific heat capacity of cooling water, This is the real-time mass flow rate of the cooling water. This refers to the outlet water temperature of the air cooler. This refers to the inlet water temperature of the air cooler. The rated heat exchange capacity of the air cooler; S7.

3. Compare the real-time heat exchange efficiency with the rated standard performance threshold, and combine the temperature trend curve fluctuation characteristics from step S7.1 to troubleshoot air cooler faults, and simultaneously mark the abnormal air cooler number and fault degree.

9. The method for stator temperature monitoring and health assessment of hydro-generators based on multi-algorithm fusion according to claim 1, characterized in that, The specific implementation method of step S8 includes: The results of Mahalanobis distance anomaly detection, health assessment level, personalized threshold exceedance, temperature time series prediction trend, and real-time heat exchange efficiency of air cooler output from steps S3 to S7 are standardized and weighted to obtain a comprehensive anomaly index. Distinguish between real faults and single-point interference through multi-indicator linkage logic; Based on the abnormal characteristics represented by the Mahalanobis distance anomaly detection results, health assessment level, personalized threshold exceedance, temperature time series prediction trend, and real-time heat exchange efficiency of the air cooler, the stator typical fault modes are matched to obtain the fault type identification results.

10. The method for stator temperature monitoring and health assessment of a hydro-generator based on multi-algorithm fusion according to claim 9, characterized in that, Step S8 also includes: Based on the actual physical installation and distribution of each measuring point, and relying on the dynamic display interface of the three-dimensional solid model of the hydro-generator, the stator overheating abnormal area and the corresponding abnormal measuring point number are determined. Based on the rate of temperature rise, changes in cooling water flow, decrease in air cooler heat exchange efficiency, and decline in health, the causes of the fault are identified to distinguish the fault types among local winding overheating, poor core heat dissipation, contact heating of the bus ring, scaling and blockage of the air cooler, insufficient cooling water flow, and insulation deterioration, thereby obtaining the results of fault location and fault tracing.

Images