A hydropower unit stator coil temperature early warning method and system
Through the GRU temperature prediction model of STL decomposition and operating state loss function, combined with confidence and error correction methods, the problems of insufficient prediction accuracy and false alarm in the temperature warning of stator coil of hydroelectric unit are solved, and higher prediction accuracy and accuracy are achieved.
Patent Information
- Application Number
- CN202211143698.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-09-20
- Publication Date
- 2025-08-19
- Estimated Expiration
- 2042-09-20
AI Technical Summary
The existing stator coil temperature warning method of hydroelectric unit has problems such as insufficient prediction accuracy and frequent false alarms caused by environmental temperature.
The periodized original temperature sequence is decomposed by STL decomposition method to obtain the overall temperature sequence and the periodic temperature sequence. Combined with the GRU temperature prediction model based on the operating state loss function, the standard deviation, mean and confidence of the historical error is calculated to determine whether to perform temperature warning.
The prediction accuracy of the model is improved, the impact of ambient temperature on temperature false alarms is reduced, the accuracy of early warning is improved, and the problem of false alarms and untimely alarms caused by model prediction errors is eliminated.
Smart Images
Figure CN115496278B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of hydropower unit status assessment and early warning, and more specifically, relates to a hydropower unit stator coil temperature early warning method and system. Background Art
[0002] The operating temperature of hydropower unit components is a critical non-electrical parameter for monitoring hydropower station operations. Operating temperature is closely related to the unit's operating status and component lifespan, providing a direct indicator of whether the unit is operating normally and stably. Researching prediction and early warning methods for hydropower unit operating temperature is of great practical significance for implementing predictive maintenance of equipment and reducing unnecessary losses. However, due to fluctuations in ambient temperature, unit operating temperature exhibits distinct characteristics throughout the year, and these differences in characteristics can affect the effectiveness of the prediction model. Furthermore, unit temperature is directly related to the unit's operating status. With frequent startup and shutdown cycles, the unit temperature also exhibits corresponding periodic changes. Only by enabling the model to deeply exploit the changes in the random operating state of temperature can the model's prediction accuracy be improved. Summary of the Invention
[0003] In view of the defects of the prior art, the purpose of the present invention is to provide a temperature warning method and system for the stator coil of a hydropower unit, aiming to solve the problems of insufficient prediction accuracy and being affected by ambient temperature in the existing temperature warning method for the stator coil of a hydropower unit, resulting in a large number of false alarms.
[0004] To achieve the above object, the present invention provides a method for early warning the temperature of a stator coil of a hydropower unit, comprising the following steps:
[0005] S1: According to the temperature variation characteristics of the stator winding of the hydropower unit, the original temperature sequence is periodized and decomposed using the STL decomposition method to obtain the overall temperature sequence and the periodic temperature sequence;
[0006] S2: Filter the entire temperature series, add the residual after filtering and the residual after decomposition by the STL decomposition method to the periodic temperature series to obtain periodic temperature trend data;
[0007] S3: Input several historical temperature actual values in the periodic temperature trend data of the unit's current operating stage into the GRU temperature prediction model based on the operating state loss function, and output the predicted temperature values of the unit at several future moments in the current operating stage;
[0008] S4: Divide the temperature change of the unit in the historical operation stage into a temperature rising stage and a temperature stabilization stage, determine the position of the temperature prediction point of the current operation stage in the temperature rising stage in the current operation stage and the historical operation stage, calculate the error between the historical temperature prediction value and the historical temperature actual value, and calculate the standard deviation, mean and confidence level of the historical error;
[0009] S5: When the standard deviation of the historical error is less than the preset standard deviation, the system error is corrected and the predicted temperature value of the unit at a future time during the current operation phase is combined to determine whether a temperature warning should be issued; when the standard deviation of the historical error is greater than the preset standard deviation, the confidence level and the predicted temperature value of the unit at a future time during the current operation phase are combined to determine whether a temperature warning should be issued.
[0010] Further preferably, the method for periodizing the original temperature sequence is:
[0011] Calculate the average running time of the unit based on the number of unit operations and the actual duration of each operation;
[0012] If the unit's operating time is longer than the unit's average operating time, the original temperature sequence corresponding to the time period in which the unit's operating time is longer than the average operating time will be removed; if the unit's operating time is less than the unit's average operating time, the temperature of the time period in which the unit's operating time is less than the average operating time will be padded, and the padded temperature value is the average temperature after the temperature of the unit reaches stability during the operating stage.
[0013] Further preferably, the loss function of the GRU temperature prediction model includes a mean square error loss function and an operating state loss function; the operating state loss function includes an average temperature loss function, a temperature rise ratio loss function, a maximum temperature loss function and a median temperature loss function;
[0014] The average temperature loss function is used to calculate the error between the actual temperature average and the predicted temperature average during a single unit operation; the temperature rise ratio loss function is used to calculate the ratio of the time taken for the temperature to rise in the current operation stage of the computer unit to the total time; the maximum temperature loss function is used to calculate the error between the actual maximum temperature and the predicted maximum temperature in the current operation stage; the median temperature loss function is used to calculate the error between the median value of the actual temperature and the median value of the predicted temperature in the current operation stage.
[0015] Further preferably, the end condition of the temperature rising stage is that the change between the average temperatures of five consecutive adjacent temperature prediction points does not exceed 0.1°C.
[0016] Further preferably, the method for calculating the standard deviation, mean and confidence level of historical errors comprises the following steps:
[0017] Calculate the position of each temperature prediction point in the periodic temperature trend data in several historical operating stages and the current operating stage;
[0018] Find the temperature prediction point at the same position as the temperature prediction point in the current operation stage in the historical operation stage, and calculate the prediction error and confidence level based on the historical temperature prediction value and the historical temperature actual value at the temperature prediction point in the historical operation stage;
[0019] Calculate the error mean and standard deviation of the temperature prediction points during the historical operation phase.
[0020] The location of any temperature prediction point in the historical operation stage is:
[0021]
[0022] Among them, the temperature rising stage in the historical operation stage is t~t+T i r ; t′ represents the time of the temperature prediction point in the historical operation stage;
[0023] The location of the temperature prediction point in the current operation stage is:
[0024]
[0025] Where t″ represents the temperature prediction point time in the current operation stage; T r Indicates the temperature rise time in the current operating stage; are the temperature rise times in the five historical operation stages respectively.
[0026] Further preferably, the temperature warning rule is:
[0027]
[0028] in, is the standard deviation of the temperature prediction point in the historical operation stage; C is the confidence level of the temperature prediction point; Indicates the predicted temperature value at a future time in the current operation phase; TH indicates the threshold value; is the mean error of the temperature prediction points.
[0029] In another aspect, the present invention provides a hydropower unit stator coil temperature early warning system, comprising:
[0030] The temperature sequence periodic processing module is used to periodically process the original temperature sequence according to the temperature variation characteristics of the stator winding of the hydropower unit;
[0031] STL decomposition module, used to decompose the original temperature sequence after periodization using STL decomposition method to obtain the overall temperature sequence and periodic temperature sequence;
[0032] Filtering module, used for filtering the overall temperature series;
[0033] An addition module is used to add the residual after filtering and the residual after decomposition by the STL decomposition method to the periodic temperature series to obtain periodic temperature trend data;
[0034] The GRU temperature prediction module is equipped with a GRU temperature prediction model based on the operating state loss function, which is used to input several historical temperature actual values in the periodic temperature trend data of the unit's current operating stage and output the predicted temperature value of the unit at a future moment in the current operating stage;
[0035] An error parameter calculation module is used to divide the temperature change during the historical operation phase of the unit into a temperature rise phase and a temperature stabilization phase. During the temperature rise phase, the error between the historical temperature prediction value and the historical temperature actual value is calculated based on the position of the temperature prediction point in the historical operation phase, as well as the standard deviation, mean, and confidence level of the historical error.
[0036] The temperature warning module is used to correct the system error when the standard deviation of the historical error is less than the preset standard deviation, and determine whether to issue a temperature warning based on the predicted temperature value at a future time during the current operation phase of the unit. When the standard deviation of the historical error is greater than the preset standard deviation, the module determines whether to issue a temperature warning based on the confidence level and the predicted temperature value at a future time during the current operation phase.
[0037] Further preferably, the loss function of the GRU temperature prediction model includes a mean square error loss function and an operating state loss function; the operating state loss function includes an average temperature loss function, a temperature rise ratio loss function, a maximum temperature loss function and a median temperature loss function;
[0038] The average temperature loss function is used to calculate the error between the actual temperature average value and the predicted temperature average value during a single unit operation; the temperature rise ratio loss function is used to calculate the ratio of the time taken for the temperature to rise in the current operation stage of the computer unit to all the time; the maximum temperature loss function is used to calculate the error between the actual maximum temperature value and the predicted maximum temperature value in the current operation stage; the median temperature loss function is used to calculate the error between the median value of the actual temperature and the median value of the predicted temperature in the current operation stage.
[0039] Further preferably, the temperature warning rule is:
[0040]
[0041] in, is the standard deviation of the temperature prediction point in the historical operation stage; C is the confidence level of the temperature prediction point; Indicates the predicted temperature value at a future time in the current operation phase; TH indicates the threshold value; is the mean error of the temperature prediction points.
[0042] Further preferably, the method for calculating the standard deviation, mean and confidence level of historical errors comprises the following steps:
[0043] Calculate the position of each temperature prediction point in the periodic temperature trend data in several historical operating stages and the current operating stage;
[0044] Find the temperature prediction point at the same position as the temperature prediction point in the current operation stage in the historical operation stage, and calculate the prediction error and confidence level based on the historical temperature prediction value and the historical temperature actual value at the temperature prediction point in the historical operation stage;
[0045] Calculate the error mean and standard deviation of the temperature prediction points during the historical operation phase.
[0046] Further preferably, the location of any temperature prediction point in the historical operation stage is:
[0047]
[0048] Among them, the temperature rising stage in the historical operation stage is t~t+T i r ; t′ represents the time of the temperature prediction point in the historical operation stage;
[0049] The location of the temperature prediction point in the current operation stage is:
[0050]
[0051] Where t″ represents the temperature prediction point time in the current operation stage; T r Indicates the temperature rise time in the current operating stage; are the temperature rise times in the five historical operation stages respectively.
[0052] In general, the above technical solutions conceived by the present invention have the following beneficial effects compared with the prior art:
[0053] The present invention provides a method and system for early warning the stator coil temperature of a hydropower unit. Taking into account the influence of climate temperature on the temperature of unit components, the STL decomposition method is used to decompose the periodized original temperature sequence to obtain an overall temperature sequence and a periodic temperature sequence. The overall temperature sequence represents the climate temperature. The overall temperature sequence is removed from the original temperature sequence, and the periodic temperature trend data affected by the unit operation is extracted for prediction and early warning, thereby improving the effect of model training and reducing the influence of ambient temperature on false temperature alarms.
[0054] The present invention provides a temperature warning method and system for the stator coil of a hydropower unit, in which a GRU temperature prediction model based on an operating state loss function is constructed. The operating state loss function reflects the overall temperature characteristics by calculating the temperature rise ratio, temperature change mean, temperature maximum value, etc. in each operating stage to capture the key characteristics of the temperature trend in the unit operation stage, thereby improving the prediction effect of the model.
[0055] The present invention provides a temperature warning method and system for the stator coil of a hydropower unit. When the standard deviation of the historical error is less than the preset standard deviation, the system error is corrected, and the predicted temperature value of the unit at a future time during the current operation stage is combined to determine whether a temperature warning should be issued; when the standard deviation of the historical error is greater than the preset standard deviation, the confidence level and the predicted temperature value at a future time during the current operation stage are combined to determine whether a temperature warning should be issued; the temperature warning method fully considers the impact of the prediction error of the GRU temperature prediction model at different temperature change stages on the warning effect, and proposes a confidence level and error correction method to eliminate the problems of false alarms and untimely alarms caused by model prediction errors, thereby improving the accuracy of warnings. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] Figure 1 This is a flow chart of the stator coil temperature warning of a hydropower unit provided by an embodiment of the present invention;
[0057] Figure 2 This is a trend diagram of the original temperature sequence of the stator coil of the hydropower unit provided by the embodiment of the present invention;
[0058] Figure 3 This is an overall temperature change trend diagram obtained by STL decomposition provided by an embodiment of the present invention;
[0059] Figure 4 This is a periodic temperature change trend diagram obtained by STL decomposition provided by an embodiment of the present invention;
[0060] Figure 5 This is a prediction effect diagram of the operating state loss function under different weights α provided by an embodiment of the present invention;
[0061] Figure 6 This is a comparison of the effects of the loss function and MSE provided by the embodiment of the present invention under different prediction step sizes;
[0062] Figure 7 This is a diagram showing the temperature prediction and early warning effects of the stator coil of a hydropower unit provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0063] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0064] The present invention provides a temperature warning method and system for the stator coil of a hydropower unit. The overall technical solution is as follows: according to the characteristics of the unit temperature change, a non-periodic initial temperature sequence is periodized, and the periodized temperature sequence is decomposed using the STL decomposition method to extract the overall temperature sequence and the local periodic temperature sequence related to the unit operation; a GRU temperature prediction model based on the operating state loss function is used to predict the periodic temperature trend data, deeply explore the intrinsic connection between the temperature and the unit operating state, and propose the model prediction accuracy; a temperature warning method is adopted, combined with a confidence calculation method and an error correction method to eliminate false alarms caused by prediction errors in different operating stages; wherein, the overall temperature sequence represents the influence of the ambient temperature on the stator coil temperature; the periodic temperature sequence represents the influence of the unit operation stage on the stator coil.
[0065] More specifically, if Figure 1 As shown, on the one hand, the present invention provides a hydropower unit stator coil temperature early warning method, comprising the following steps:
[0066] S1: Based on the temperature variation characteristics of the hydropower unit stator winding, the original temperature sequence is periodized and decomposed using the STL decomposition method to obtain the overall temperature sequence and the periodic temperature sequence. The overall temperature sequence is further filtered to obtain smoother overall temperature trend data. The filtered residual is then added to the periodic temperature sequence to obtain the periodic temperature trend data.
[0067] More specifically, the periodic processing and STL decomposition of the original temperature series in S1 specifically include the following steps:
[0068] The average running time of the computer group is calculated as follows:
[0069]
[0070] Among them, T i Indicates the actual duration of each run; Indicates the average running time of the unit; N indicates the number of times the unit has run;
[0071] Each operation stage is processed, and the original temperature sequence corresponding to the time period with operation time longer than the average operation time is removed, and the temperature of the operation time period that is insufficient is supplemented. The supplemented value is the average temperature after the unit temperature reaches stability in this operation stage, and the unit operation time becomes
[0072] The original periodic temperature series is subjected to STL decomposition to obtain the overall temperature series, the periodic temperature series and the residual value; the overall temperature series is filtered to obtain smoother overall temperature trend data, and the periodic temperature series and the residual value obtained in the above two steps of decomposition and filtering are added together to obtain the periodic temperature trend data;
[0073] S2: Establish a GRU temperature prediction model based on the operating state loss function. This model uses the operating state loss function to capture the key characteristics of temperature during each startup and shutdown process, improving the model's prediction effect.
[0074] More specifically, the GRU temperature prediction model based on the running state loss function in S2 has the following main structure:
[0075] The loss functions of the GRU temperature prediction model include the mean square error loss function (mse) and the operating state loss function. The operating state loss function includes: average temperature loss, temperature rise ratio loss, maximum temperature loss, and median temperature loss.
[0076] The average temperature loss is used to calculate the error between the actual temperature average and the predicted temperature average during a single unit operation. The calculation formula is as follows:
[0077]
[0078] Among them, t represents the time when the unit starts running, y represents the actual temperature value, Indicates the predicted temperature value; T i Indicates the actual duration of each run;
[0079] The temperature rise ratio loss function calculates the ratio of the time taken for temperature rise to the total time during the operation. The calculation formula is as follows:
[0080]
[0081] Among them, t up Indicates the actual time taken for the temperature to rise. Indicates the time taken to predict the temperature rise;
[0082] The maximum temperature loss function calculates the error between the actual maximum temperature and the predicted maximum temperature during the run. The calculation formula is as follows:
[0083]
[0084] The median temperature loss function calculates the error between the median value of the actual temperature during the run and the median value of the predicted temperature. The calculation formula is as follows:
[0085]
[0086] The calculation formula of the mean square error loss function mse is as follows:
[0087]
[0088] From the above, we can see that the total loss function of the GRU temperature prediction model is calculated as follows:
[0089] loss=mse+α·(loss1+loss2+loss3+loss4)
[0090] Among them, α represents the weight of the running state loss function;
[0091] S3: Divide the temperature changes during the unit's historical operating phase into two phases: a temperature rise phase and a temperature stabilization phase. Calculate the error between the historical predicted values and the actual values of the corresponding points in the past five operating phases based on the location of the temperature prediction point in the current operating phase. Calculate the standard deviation and mean of the five historical errors. Calculate the confidence level of the predicted value based on the error. Process the predicted value based on the standard deviation of the error and compare it with the alarm threshold to implement unit temperature warning.
[0092] More specifically, the specific steps of temperature warning in S3 are as follows:
[0093] The temperature changes in the past five historical operation stages are divided, and process i can be divided into the temperature rising stage t~t+T i r and temperature stabilization stage t+T i r ~t+T i r +T i s , where i = 1, 2, ..., 5; the end condition of the temperature rise phase is that the change between the average temperatures of five consecutive adjacent points does not exceed 0.1°C. The formula is as follows:
[0094]
[0095] According to the above division rules, the position p of any temperature point in history can be calculated. The calculation process of the position p in the historical operation stage is as follows:
[0096]
[0097] Where exp represents the exponential function with e as the base, and t′ represents the time corresponding to the prediction point;
[0098] The specific calculation method of p in the current running stage is as follows:
[0099]
[0100] Where t″ represents the current prediction point time, T r Indicates the temperature rise time in the current operation stage. Due to the temperature rise in the current operation stage, it is not possible to obtain T r , so the corresponding historical value is selected for replacement, and the above-mentioned stage division rules need to be continuously used to divide the new prediction points into stages;
[0101] According to the above rules, find the points in the past five historical operation stages that have the same p-value as the corresponding stage in the current operation stage, and calculate their confidence formula as follows:
[0102]
[0103] C=σ(-2ln(|e p |))
[0104] Among them, |e p | is the prediction error, represents the historical temperature actual value and historical temperature forecast value with the same p value as the forecast point in the nth historical operation stage, σ represents the Sigmoid function, and C is the confidence level of the forecast point. According to the formula, when the mean forecast error is 0.1, the confidence level reaches 0.99; when the mean forecast error is 0.5, the confidence level reaches 0.8; when the mean forecast error is 1, the confidence level drops to 0.5.
[0105] Prediction error for historical p-value points Calculate the mean when n=1,2,...,5 The standard deviation is When the standard deviation is small, it indicates that the prediction error is mostly systematic error and error correction is required. For the temperature series, the main cause of systematic error is the change of the temperature series, which usually manifests as an abnormal increase trend. Therefore, when a systematic error occurs, the early warning intensity should be appropriately increased. When the standard deviation is large, it indicates that the prediction error is mostly random error. The confidence level can be used to reflect the reliability of the prediction. The early warning rule is:
[0106]
[0107] in, Indicates the current predicted temperature value, and TH indicates the threshold value.
[0108] In another aspect, the present invention provides a hydropower unit stator coil temperature early warning system, comprising:
[0109] The temperature sequence periodic processing module is used to periodically process the original temperature sequence according to the temperature variation characteristics of the stator winding of the hydropower unit;
[0110] STL decomposition module, used to decompose the original temperature sequence after periodization using STL decomposition method to obtain the overall temperature sequence and periodic temperature sequence;
[0111] Filtering module, used for filtering the overall temperature series;
[0112] An addition module is used to add the residual after filtering and the residual after decomposition by the STL decomposition method to the periodic temperature series to obtain periodic temperature trend data;
[0113] The GRU temperature prediction module is equipped with a GRU temperature prediction model based on the operating state loss function, which is used to input several historical temperature actual values in the periodic temperature trend data of the unit's current operating stage and output the predicted temperature value of the unit at a future moment in the current operating stage;
[0114] An error parameter calculation module is used to divide the temperature change in the historical operation stage of the unit into a temperature rise stage and a temperature stabilization stage. In the temperature rise stage, the error between the historical temperature prediction value and the historical temperature actual value is calculated based on the position of the temperature prediction point in the historical operation stage, and the standard deviation, mean and confidence level of the historical error are calculated;
[0115] The temperature warning module is used to correct the system error when the standard deviation of the historical error is less than the preset standard deviation, and determine whether to issue a temperature warning based on the predicted temperature value at a future time during the current operation phase of the unit. When the standard deviation of the historical error is greater than the preset standard deviation, the module determines whether to issue a temperature warning based on the confidence level and the predicted temperature value at a future time during the current operation phase.
[0116] Further preferably, the loss function of the GRU temperature prediction model includes a mean square error loss function and an operating state loss function; the operating state loss function includes an average temperature loss function, a temperature rise ratio loss function, a maximum temperature loss function and a median temperature loss function;
[0117] The average temperature loss function is used to calculate the error between the actual temperature average value and the predicted temperature average value during a single unit operation; the temperature rise ratio loss function is used to calculate the ratio of the time taken for the temperature to rise in the current operation stage of the computer unit to all the time; the maximum temperature loss function is used to calculate the error between the actual maximum temperature value and the predicted maximum temperature value in the current operation stage; the median temperature loss function is used to calculate the error between the median value of the actual temperature and the median value of the predicted temperature in the current operation stage.
[0118] Further preferably, the temperature warning rule is:
[0119]
[0120] in, is the standard deviation of the temperature prediction point in the historical operation stage; C is the confidence level of the temperature prediction point; Indicates the predicted temperature value at a future time in the current operation phase; TH indicates the threshold value; is the mean error of the temperature prediction points.
[0121] Further preferably, the method for calculating the standard deviation, mean and confidence level of historical errors comprises the following steps:
[0122] Calculate the position of each temperature prediction point in the periodic temperature trend data in several historical operating stages and the current operating stage;
[0123] Find the temperature prediction point at the same position as the temperature prediction point in the current operation stage in the historical operation stage, and calculate the prediction error and confidence level based on the historical temperature prediction value and the historical temperature actual value at the temperature prediction point in the historical operation stage;
[0124] Calculate the error mean and standard deviation of the temperature prediction points during the historical operation phase.
[0125] Further preferably, the location of any temperature prediction point in the historical operation stage is:
[0126]
[0127] Among them, the temperature rising stage in the historical operation stage is t~t+T i r ; t′ represents the time of the temperature prediction point in the historical operation stage;
[0128] The location of the temperature prediction point in the current operation stage is:
[0129]
[0130] Where t″ represents the temperature prediction point time in the current operation stage; T r Indicates the temperature rise time in the current operating stage; are the temperature rise times in the five historical operation stages respectively.
[0131] Example
[0132] Step 1: The temperature dataset covers the period from January 2, 2018 to December 29, 2018. The original sequence of stator coil temperature changes during unit operation is as follows: Figure 2As shown in the figure, it can be seen that in a year, the stator core temperature has an obvious changing trend with the change of ambient temperature. The overall temperature is low in winter and high in summer. The original temperature sequence is periodized and then STL decomposed. The decomposed overall temperature sequence is filtered to obtain the overall temperature trend data, as shown in Figure 3 As shown in the figure, the overall temperature trend extracted is consistent with the trend in the original temperature series; the residual obtained by STL decomposition and filtering is added to the periodic temperature series, and the periodic temperature trend is obtained as shown in the figure. Figure 4 As shown, the temperature change at this time is only related to the unit operation process, and the overall distribution of the periodic temperature trend data is stable;
[0133] Step 2: Use the GRU temperature prediction model based on the running state loss function to predict the temperature cycle change trend. The GRU model input sequence length is 9 and the output length is 1, that is, 9 points are used to predict 1 point; the learning rate of the model is set to 0.005, and the ratio of the training set to the test set is 3:1; the weight α of the running state loss function affects the single-step prediction result as follows Figure 5 As shown in the figure, the RMSE indicator is the root mean square error, which reflects the error between the temperature prediction value and the actual temperature value. The smaller the RMSE, the higher the prediction accuracy. Figure 5 It can be seen that when the weight value is 0.1, each loss function can achieve the best prediction effect and is better than the prediction model without using the state loss function; further, the prediction step of the model is increased to realize the multi-step prediction of the model; single-step prediction means using historical data to predict the latest next data point, which in this embodiment means predicting the point one minute later. The relationship between the prediction step and the prediction time in multi-step prediction is similar; Figure 6 It can be seen that when the step size ranges from 1 to 10, the total loss function has advantages over using a single MSE loss function and can improve the prediction accuracy of the model;
[0134] Step 3: To realize the temperature warning function based on the accuracy of multi-step prediction, in order to ensure the accuracy of warning, the prediction step of the present invention is set to 3, that is, the temperature of the unit is predicted 3 minutes later; Figure 7 The figure shows the multi-step temperature prediction results and warning diagram; after reaching the temperature warning value, Figure 7 It can be seen that about five minutes before the actual alarm point, the model uses historical data to predict the temperature three minutes later. The predicted value meets the requirements of error correction. After error correction, it reaches the alarm threshold and realizes early alarm. It can be seen that the error warning method of the present invention can avoid the problem of untimely alarm due to changes in temperature data.
[0135] The present invention provides a method and system for early warning the stator coil temperature of a hydropower unit. Taking into account the influence of climate temperature on the temperature of unit components, the STL decomposition method is used to decompose the periodized original temperature sequence to obtain an overall temperature sequence and a periodic temperature sequence. The overall temperature sequence represents the climate temperature. The overall temperature sequence is removed from the original temperature sequence, and the periodic temperature trend data affected by the unit operation is extracted for prediction and early warning, thereby improving the effect of model training and reducing the influence of ambient temperature on false temperature alarms.
[0136] The present invention provides a temperature warning method and system for the stator coil of a hydropower unit, in which a GRU temperature prediction model based on an operating state loss function is constructed. The operating state loss function reflects the overall temperature characteristics by calculating the temperature rise ratio, temperature change mean, temperature maximum value, etc. in each operating stage to capture the key characteristics of the temperature trend in the unit operation stage, thereby improving the prediction effect of the model.
[0137] The present invention provides a temperature warning method and system for the stator coil of a hydropower unit. When the standard deviation of the historical error is less than the preset standard deviation, the system error is corrected, and the predicted temperature value of the unit at a future time during the current operation stage is combined to determine whether a temperature warning should be issued; when the standard deviation of the historical error is greater than the preset standard deviation, the confidence level and the predicted temperature value at a future time during the current operation stage are combined to determine whether a temperature warning should be issued; the temperature warning method fully considers the impact of the prediction error of the GRU temperature prediction model at different temperature change stages on the warning effect, and proposes a confidence level and error correction method to eliminate the problems of false alarms and untimely alarms caused by model prediction errors, thereby improving the accuracy of warnings.
[0138] It will be easily understood by those skilled in the art that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.
Claims
1. A hydropower unit stator coil temperature early warning method, characterized in that: The following steps are involved: S1: According to the temperature variation characteristics of the stator winding of the hydropower unit, the original temperature sequence is periodized and decomposed using the STL decomposition method to obtain the overall temperature sequence and the periodic temperature sequence; S2: Filter the entire temperature series, add the residual after filtering and the residual after decomposition by the STL decomposition method to the periodic temperature series to obtain periodic temperature trend data; S3: Input several historical temperature actual values in the periodic temperature trend data of the unit's current operating stage into the GRU temperature prediction model based on the operating state loss function, and output the predicted temperature value of the unit at a future moment in the current operating stage; S4: Divide the temperature change of the unit's historical operation stage into a temperature rising stage and a temperature stabilization stage, determine the position of the temperature prediction point of the current operation stage in the temperature rising stage in the historical operation stage, calculate the error between the historical temperature prediction value and the historical temperature actual value, and calculate the standard deviation, mean and confidence level of the historical error; S5: When the standard deviation of the historical error is less than the preset standard deviation, the system error is corrected and the predicted temperature value at the future moment of the unit's current operation stage is combined to determine whether to issue a temperature warning; When the standard deviation of the historical error is greater than the preset standard deviation, the confidence level and the predicted temperature value at a future moment in the current operation phase are combined to determine whether to issue a temperature warning.
2. The hydropower unit stator coil temperature early warning method according to claim 1, characterized in that: The method for periodizing the original temperature series is: Calculate the average running time of the unit based on the number of unit operations and the actual duration of each operation; If the unit's operating time is longer than the unit's average operating time, the original temperature sequence corresponding to the time period in which the unit's operating time is longer than the average operating time will be removed; if the unit's operating time is less than the unit's average operating time, the temperature of the time period in which the unit's operating time is less than the average operating time will be padded, and the padded temperature value is the average temperature after the temperature of the unit reaches stability during the operating stage.
3. The hydropower unit stator coil temperature early warning method according to claim 2, characterized in that: The loss function of the GRU temperature prediction model includes a mean square error loss function and an operating state loss function; the operating state loss function includes an average temperature loss function, a temperature rise ratio loss function, a maximum temperature loss function, and a median temperature loss function; The average temperature loss function is used to calculate the error between the actual temperature average value and the predicted temperature average value during a single unit operation; the temperature rise ratio loss function is used to calculate the ratio of the time taken for the temperature to rise in the current operation stage of the computer unit to all the time; the maximum temperature loss function is used to calculate the error between the actual maximum temperature value and the predicted maximum temperature value in the current operation stage; the median temperature loss function is used to calculate the error between the median value of the actual temperature and the median value of the predicted temperature in the current operation stage.
4. The hydropower unit stator coil temperature early warning method according to any one of claims 1 to 3, characterized in that: The end condition of the temperature rise phase is that the change between the average temperatures of five consecutive adjacent temperature prediction points does not exceed 0.1℃.
5. The hydropower unit stator coil temperature early warning method according to claim 3, characterized in that: The method for calculating the standard deviation, mean, and confidence level of historical errors includes the following steps: Calculate the position of each temperature prediction point in the periodic temperature trend data in several historical operating stages and the current operating stage; Find the temperature prediction point at the same position as the temperature prediction point in the current operation stage in the historical operation stage, and calculate the prediction error and confidence level based on the historical temperature prediction value and the historical temperature actual value at the temperature prediction point in the historical operation stage; Calculate the error mean and standard deviation of the temperature prediction points during the historical operation phase.
6. The hydropower unit stator coil temperature early warning method according to claim 5, characterized in that: The location of any temperature prediction point in the historical operation stage is: Among them, the temperature rising stage in the historical operation stage is t~t+T i r ; t′ represents the time of the temperature prediction point in the historical operation stage; The location of the temperature prediction point in the current operation stage is: Where t″ represents the temperature prediction point time in the current operation stage; T r Indicates the temperature rise time in the current operating stage; T1 r ,T2 r ,...,T5 r are the temperature rise times in the five historical operation stages respectively.
7. The hydropower unit stator coil temperature early warning method according to claim 5, characterized in that: The temperature warning rules are: in, is the standard deviation of the temperature prediction point in the historical operation stage; C is the confidence level of the temperature prediction point; Indicates the predicted temperature value at a future time in the current operation phase; TH indicates the threshold value; is the mean error of the temperature prediction points.
8. A hydropower unit stator coil temperature warning system, characterized in that: include: The temperature sequence periodic processing module is used to periodically process the original temperature sequence according to the temperature variation characteristics of the stator winding of the hydropower unit; STL decomposition module, used to decompose the original temperature sequence after periodization using STL decomposition method to obtain the overall temperature sequence and periodic temperature sequence; Filtering module, used for filtering the overall temperature series; An addition module is used to add the residual after filtering and the residual after decomposition by the STL decomposition method to the periodic temperature series to obtain periodic temperature trend data; The GRU temperature prediction module is equipped with a GRU temperature prediction model based on the operating state loss function, which is used to input several historical temperature actual values in the periodic temperature trend data of the unit's current operating stage and output the predicted temperature value of the unit at a future moment in the current operating stage; An error parameter calculation module is used to divide the temperature change during the historical operation phase of the unit into a temperature rise phase and a temperature stabilization phase. During the temperature rise phase, the error between the historical temperature prediction value and the historical temperature actual value is calculated based on the position of the temperature prediction point in the historical operation phase, as well as the standard deviation, mean, and confidence level of the historical error. The temperature warning module is used to correct the system error when the standard deviation of the historical error is less than the preset standard deviation, and determine whether to issue a temperature warning based on the predicted temperature value at a future time during the current operation phase of the unit; When the standard deviation of the historical error is greater than the preset standard deviation, the confidence level and the predicted temperature value at a future moment in the current operation phase are combined to determine whether to issue a temperature warning.
9. The hydropower unit stator coil temperature early warning system according to claim 8, characterized in that: The loss function of the GRU temperature prediction model includes a mean square error loss function and an operating state loss function; the operating state loss function includes an average temperature loss function, a temperature rise ratio loss function, a maximum temperature loss function, and a median temperature loss function; The average temperature loss function is used to calculate the error between the actual temperature average value and the predicted temperature average value during a single unit operation; the temperature rise ratio loss function is used to calculate the ratio of the time taken for the temperature to rise in the current operation stage of the computer unit to all the time; the maximum temperature loss function is used to calculate the error between the actual maximum temperature value and the predicted maximum temperature value in the current operation stage; the median temperature loss function is used to calculate the error between the median value of the actual temperature and the median value of the predicted temperature in the current operation stage.
10. The hydropower unit stator coil temperature early warning system according to claim 8 or 9, characterized in that: The temperature warning rules are: in, is the standard deviation of the temperature prediction point in the historical operation stage; C is the confidence level of the temperature prediction point; Indicates the predicted temperature value at a future time in the current operation phase; TH indicates the threshold value; is the mean error of the temperature prediction points.