A method for predicting the slow change of analog quantity of hydropower units based on characteristic value monitoring

Abstract

The invention relates to a method for predicting the slow change of analog quantity of a hydropower unit based on characteristic value monitoring. The method forms a time series sample set by regularly sampling the temperature and vibration frequency of the core components of a water turbine; after the samples are normalized, a weighted sum and standard deviation are calculated as the analog quantity; the current slope value is obtained by using local linear regression, and the analog quantity value of the next time series is predicted by using a random forest model; the current and predicted slope values ​​are compared to reflect the slow change trend of the analog quantity; the method of the invention can effectively predict the slow change of the analog quantity of the hydropower unit, help to discover potential faults in advance, and improve the operating safety and stability of the hydropower unit.

Description

Technical Field

[0001] The invention relates to the technical field of hydropower unit monitoring, and in particular to a method for predicting slow changes in analog quantities of hydropower units based on characteristic value monitoring. Background Art

[0002] With the continuous development and application of hydropower unit technology, the requirements for its operational stability and safety are also increasing. During the operation of a hydropower unit, the slow change trend of its analog quantity can often reflect the potential problems of the unit's operating status. Therefore, accurately predicting and monitoring the slow change of these analog quantities is of great significance for timely discovery and resolution of problems and ensuring the safe and stable operation of the hydropower unit.

[0003] At present, there are some traditional prediction methods for the slow-changing prediction of the simulated quantity of hydropower units, such as prediction models based on statistics and prediction methods based on physical models. However, these methods have some shortcomings in practical applications. For example, the prediction model based on statistics usually requires statistical analysis of a large amount of historical data, and has high requirements for the integrity and accuracy of the data. Once the data is missing or abnormal, the prediction performance of the model will be seriously affected; although the prediction method based on the physical model can more accurately reflect the physical process of the unit, the establishment and maintenance of the model is relatively complicated, and it has poor adaptability to the complex and changeable actual operating environment.

[0004] In addition, existing technologies often use a single parameter as a simulation quantity for slow-change prediction. However, the operating status of a hydropower unit is a comprehensive reflection. Although a single parameter (such as temperature or frequency) can reflect a certain aspect of the unit's status, it may not be able to fully reveal the overall operating status of the unit. Summary of the invention

[0005] The purpose of the present invention is to combine the core component temperature of the turbine and the vibration frequency of the turbine, and to find a method for predicting the slow change of the analog quantity of a hydropower unit based on characteristic value monitoring based on a random forest model and a regression function.

[0006] In order to achieve the above-mentioned purpose of the present invention, the present invention adopts the following technical solutions:

[0007] A method for predicting slow changes in simulated quantities of hydropower units based on characteristic value monitoring, the method comprising the following steps:

[0008] Step 1: based on the preset time interval and the preset time window length, the parameters in the hydropower unit: the core component temperature of the turbine and the vibration frequency of the turbine are sampled, and the sampled samples are divided according to the preset time length to form a time series sample set in units of time windows.

[0009] Furthermore, the preset time interval t is 2 to 5S, and continuous sampling is performed q times in a time window, then the preset time window length L=qt, where 4≤q≤12; dividing the sampled samples according to the preset time length means that the preset time length M=kL, where k represents the number of time windows, 15≤k≤20.

[0010] The time series sample set is represented as U = {A1, A2, ..., A i ,…,A n}, where A i represents the i-th time series sample, n represents the number of time series samples, each time series sample is composed of sampling data of k time windows, and the i-th time series sample A i Indicated as A i ={a i1 ,a i2 ,……,a ik}, a ik Represents the sampling data of k time windows in the i-th time series sample.

[0011] Step 2, performing minimum-maximum normalization processing on the sampled data of each time window, and then calculating the standard deviation of the weighted sum of each time series sample in the time series sample set as the simulation quantity of the hydropower unit.

[0012] Furthermore, the weighted sum S is calculated as: S = w T ·T+w f f, where T is the normalized temperature value of a single time window, f is the normalized vibration frequency value of a single time window, and w T and w f are the weights of temperature and vibration frequency respectively, w T +w f =1.

[0013] For the time series sample A, there are k time windows of normalized data, expressed as: (T1, f1), (T2, f2), …, (T k ,f k ), the weighted sum of the time window data is expressed as: S1,S2,…,S k .

[0014] Standard deviation of the weighted sum of time series samples: in, is the average of the weighted sums, S i It is the weighted sum of the i time window data in the time series sample.

[0015] Then for time series sample A, the standard deviation of the weighted sum of its time series samples is:

[0016]

[0017] Among them, T i and f i They respectively represent the normalized temperature value and vibration frequency value at i times.

[0018] Step 3: Perform local linear regression on the simulated values ​​in the time series sample set to obtain the slope value of the current regression function.

[0019] Step 4: Input the simulated quantity of the hydropower unit in the continuous time window into the trained random forest model, and predict the simulated quantity value of the hydropower unit in the next time series sample based on the random forest model.

[0020] Step 5, based on the predicted simulated value of the hydropower unit in the next time series sample, perform local linear regression calculation on the simulated value again to obtain the predicted value of the slope of the regression function; compare the size and change of the slope value of the current regression function and the predicted value of the slope of the regression function to reflect the slow-changing trend of the simulated value.

[0021] Furthermore, the minimum-maximum normalization calculation formula is: Where T is the normalized temperature value of a single time window, is the average temperature of the time window, max(T) and min(T) are the maximum and minimum values ​​of the temperature data in the time window respectively; f is the vibration frequency value of a single time window after normalization, is the average value of the vibration frequency in the time window, max(f) and min(f) are the maximum and minimum values ​​of the vibration frequency data in the time window, respectively.

[0022] Furthermore, the local linear regression operation in step 3 adopts the weighted least square method, and its formula is: β=(X T WX) -1 X T WY, where β is the regression coefficient vector, X is the design matrix, W is the diagonal weight matrix, and Y is the response variable vector.

[0023] Furthermore, in step 4, the random forest model uses characteristic data in the time series sample during training, wherein the characteristic data includes temperature values, vibration frequency values, and standard deviations of their weighted sums;

[0024] When predicting, the random forest model outputs the predicted simulated value based on the input time series sample, and its prediction formula is: in, is the predicted simulated value, m is the number of trees in the random forest, and h j(x) is the prediction result of the jth tree for input x.

[0025] Furthermore, the size and change of the slope value of the current regression function and the slope prediction value of the regression function reflect the slow change trend of the simulation quantity. The specific method is:

[0026] A slope threshold G is set, and the slope threshold G=0.08. When the absolute value of the slope value of the current regression function or the absolute value of the slope prediction value of the regression function is greater than the slope threshold G, the early warning mechanism is triggered, indicating that the simulated quantity of the hydropower unit is in a significantly slow-changing state.

[0027] A slope slow change threshold D is set, wherein the slow change threshold D=0.03. The absolute value of the difference between the absolute value of the slope prediction value of the regression function obtained in step 5 and the absolute value of the slope prediction value of the regression function is taken, and then the absolute value of the difference is taken and compared with the slow change threshold D. If the absolute value of the difference is greater than the slow change threshold D, the early warning mechanism is triggered, indicating that the simulated quantity of the hydropower unit is expected to undergo a significant slow change.

[0028] Furthermore, according to the slope prediction value of the regression function obtained in step 5, the change rate of the slope prediction value is calculated to quantify the slow change speed of the simulation quantity, and the change rate is the ratio of the difference of the slope prediction values ​​in adjacent time series samples to the time interval.

[0029] The formula for calculating the rate of change of the slope prediction value is: Where R is the rate of change of the slope prediction value, σ t+1 and σ t are the slope prediction values ​​of two adjacent time series samples, Δt is the time interval between two time series samples, Δt=M=kL.

[0030] Furthermore, in step 4, the training process of the random forest model uses a cross-validation approach to optimize the model;

[0031] The formula for cross validation is expressed as: Among them, CV is the cross-validation error, K is the number of cross-validation folds, MSE p is the mean square error of the p-fold validation.

[0032] The optimal random forest model parameters are selected by minimizing the cross-validation error value.

[0033] Furthermore, when sampling the parameters in the hydropower unit, the unit active power, unit speed and oil cooler water flow rate should meet the following constraints:

[0034]

[0035] Only when the above constraints are fully satisfied, the parameters in the hydropower unit are sampled.

[0036] Furthermore, the number n of time series samples in the time series sample set satisfies 5≤n≤15.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] The present invention samples and monitors key parameters in the hydropower unit in real time, weights multiple parameters to form analog quantities, and can predict the stability of the unit based on the slow-changing trend of the analog quantities, discover faults in advance and take corresponding measures to deal with them; the present invention uses local linear regression operation to calculate the standard deviation of the weighted sum of time series samples, which can accurately reflect the dynamic changes of the analog quantities. At the same time, the random forest model is used to predict the analog quantities in the next time series sample, thereby improving the accuracy of the prediction; the method has relatively low requirements on the integrity and accuracy of the data, and can be adaptively adjusted according to the actual conditions of different units, and has strong adaptability and generalizability. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 The present invention is a flow chart of the method for predicting the slow change of analog quantity of a hydropower unit based on characteristic value monitoring. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention is described clearly and completely below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] like Figure 1 As shown, it is a flow chart of the method for predicting the slow change of the simulated quantity of a hydropower unit based on characteristic value monitoring of the present invention, and the method comprises the following steps:

[0042] Step 1: based on the preset time interval and the preset time window length, the parameters in the hydropower unit: the core component temperature of the turbine and the vibration frequency of the turbine are sampled, and the sampled samples are divided according to the preset time length to form a time series sample set in units of time windows.

[0043] It should be noted that the core component temperature generally refers to one or more of the stator winding temperature, the stator core temperature and the bearing bushing temperature. If it is the temperature of multiple components, the temperatures of the multiple components need to be weighted into one temperature value before performing the minimum-maximum normalization processing.

[0044] The preset time interval t is 2 to 5 seconds, and the continuous sampling is q times in a time window, then the preset time window length L = qt, where 4≤q≤12. The sample is divided according to the preset time length, which means that the preset time length M = kL, where k represents the number of time windows, 15≤k≤20. In this way, according to 15 to 20 time windows as a time series, it can be ensured that the data of the hydropower unit within 2 to 20 minutes forms a sequence data.

[0045] The time series sample set is represented as U = {A1, A2, ..., A i ,…,A n}, where A i represents the i-th time series sample, n represents the number of time series samples, each time series sample is composed of sampling data of k time windows, and the i-th time series sample A i Indicated as A i ={a i1 ,a i2 ,……,a ik}, a ik Represents the sampling data of k time windows in the i-th time series sample; the number n of time series samples in the time series sample set satisfies 5≤n≤15, ensuring that the data of 30 to 300 minutes of the unit can form a data set.

[0046] Step 2, performing minimum-maximum normalization processing on the sampled data of each time window, and then calculating the standard deviation of the weighted sum of each time series sample in the time series sample set as the simulation quantity of the hydropower unit.

[0047] The calculation formula of weighted sum S is: S = w T ·T+w f f, where T is the normalized temperature value of a single time window, f is the normalized vibration frequency value of a single time window, and w T and w f are the weights of temperature and vibration frequency respectively, w T +w f =1.

[0048] The selection of weight values ​​takes a large hydropower station as an example. During the operation of the hydropower unit of the hydropower station, engineers found that the temperature and frequency parameters have a significant impact on the operating status of the unit. In order to more accurately predict the potential problems of the unit and improve the operating efficiency and safety of the unit, the temperature and frequency parameters are combined into a simulated quantity for monitoring and analysis in a weighted manner. When the unit is operating normally, the change of temperature parameters is relatively stable, but once a fault or abnormal situation occurs, the change of temperature can often be quickly reflected. The frequency parameter is directly related to the load and speed of the unit, and its change can reflect the operating status and performance of the unit.

[0049] Under normal circumstances, the temperature parameters and frequency parameters have roughly the same impact on the unit's operating status, so the initial weights of the two were set to 0.5 each. However, in subsequent actual operation, it was found that when the unit load was high, the change in frequency parameters had a more significant impact on the unit, so it was decided to increase the weight of the frequency parameter to 0.6 and reduce the weight of the temperature parameter to 0.4. After a period of operation and data analysis, it was found that the new weight setting can more accurately reflect the actual operating status of the unit, especially when operating at high load, and can promptly detect potential faults and abnormal conditions.

[0050] However, in some special cases (such as low load operation or large changes in ambient temperature), the change in temperature parameters has a more significant impact on the unit. Therefore, the weight values ​​of temperature and frequency parameters can also be dynamically adjusted according to the actual operation and environmental conditions of the unit.

[0051] In addition, the preset time interval t and the preset time window length L can be adaptively adjusted according to the actual operation of the hydropower unit and the data sampling conditions; the basis for the adaptive adjustment can be factors such as the load change and operation stability of the hydropower unit; the adaptive adjustment formula of the preset time interval t and the preset time window length L can be expressed as: t′=f 1(P,S) ; L′=f 2(t′,q) ; Wherein, t′ is the adjusted preset time interval, L′ is the adjusted preset time window length, f1 and f2 are adaptive adjustment functions, P represents the load condition of the hydropower unit, S represents the operating stability of the hydropower unit, and q is the number of sampling times in the time window.

[0052] There is no restriction on the dynamic adjustment method of the weight value. For example, an intelligent monitoring system has been developed that can monitor the operating status and environmental conditions of the unit in real time, and dynamically calculate and adjust the weight values ​​of the temperature and frequency parameters according to preset rules and algorithms. When the unit load is high and the ambient temperature is stable, the system will automatically increase the weight value of the frequency parameter; when the unit load is low or the ambient temperature changes greatly, the system will increase the weight value of the temperature parameter accordingly.

[0053] For the time series sample A, there are k time windows of normalized data, expressed as: (T1, f1), (T2, f2), …, (T k ,f k ), the weighted sum of the time window data is expressed as: S1,S2,…,S k .

[0054] Standard deviation of the weighted sum of time series samples: in, is the average of the weighted sums, S i It is the weighted sum of the i time window data in the time series sample.

[0055] Then for time series sample A, the standard deviation of the weighted sum of its time series samples is:

[0056]

[0057] Among them, T i and f i They respectively represent the normalized temperature value and vibration frequency value at i times.

[0058] Step 3: Perform local linear regression on the simulated values ​​in the time series sample set to obtain the slope value of the current regression function.

[0059] Step 4: Input the simulated quantity of the hydropower unit in the continuous time window into the trained random forest model, and predict the simulated quantity value of the hydropower unit in the next time series sample based on the random forest model.

[0060] Step 5: According to the predicted value of the hydropower unit simulation in the next time series sample, the simulation is again subjected to local linear regression calculation to obtain the predicted value of the slope of the regression function. Comparing the size and change of the slope value of the current regression function and the predicted value of the slope of the regression function reflects the slow change trend of the simulation.

[0061] The minimum-maximum normalization calculation formula is: Where T is the normalized temperature value of a single time window, is the average temperature of the time window, max(T) and min(T) are the maximum and minimum values ​​of the temperature data in the time window, respectively. f is the vibration frequency value of a single time window after normalization, is the average value of the vibration frequency in the time window, max(f) and min(f) are the maximum and minimum values ​​of the vibration frequency data in the time window, respectively.

[0062] The local linear regression operation in step 3 uses the weighted least squares method, and its formula is: β = (X TWX) -1 X T WY, where β is the regression coefficient vector, X is the design matrix, W is the diagonal weight matrix, and Y is the response variable vector.

[0063] It should be noted that in the given weighted least squares formula, β is the regression coefficient vector, which contains the slope and intercept of the linear regression model. The model is y = β0 + β1x, where β0 is the intercept and β1 is the slope. In multiple linear regression, β is a vector containing the coefficients of all independent variables. If only one independent variable is concerned (i.e., univariate linear regression), then β will be a two-dimensional vector, where β0 is the first element (intercept) and β1 is the second element (slope).

[0064] To get the slope β1, we need to do the following: Construct a design matrix X with all 1s in the first column and the values ​​of the independent variable x in the second column. Select or calculate a weight matrix W where the elements on the diagonal are weights based on the uncertainty or importance of the observations. Compute β using the weighted least squares formula. Extract the slope β1 from the β vector, which is the second element of the β vector (usually the element with index 1 in Python).

[0065] In step 4, the random forest model uses characteristic data in the time series samples during training, and the characteristic data includes temperature values, vibration frequency values, and the standard deviation of their weighted sum.

[0066] When predicting, the random forest model outputs the predicted simulated value based on the input time series sample, and its prediction formula is: in, is the predicted simulated value, m is the number of trees in the random forest, and h j (x) is the prediction result of the jth tree for input x.

[0067] It should be noted that in step 4, the training process of the random forest model uses cross-validation to optimize the model.

[0068] The formula for cross validation is expressed as: Among them, CV is the cross-validation error, K is the number of cross-validation folds, MSE p is the mean square error of the P-fold validation.

[0069] The random forest model involved in step 4 consists of multiple decision trees, which are independently constructed during the training process. Each tree is trained based on samples and feature subsets randomly selected from the original dataset. The structure of the random forest model ensures that it can capture complex patterns and relationships in the data while reducing the risk of overfitting.

[0070] The training process of the random forest model includes: data preparation, building decision trees, and integrating decision trees.

[0071] Collect historical data containing relevant features of the simulated quantities of hydropower units, and perform necessary preprocessing, such as missing value filling, outlier processing, and data standardization. For each tree, a portion of samples is randomly selected from the original data set as the training set, and a portion of features is randomly selected for training. Each tree is independently built based on this information, and the structure of the tree is created by recursively dividing the training set into smaller subsets. All constructed decision trees are combined into a random forest model. In the prediction stage, new input data is input into the random forest model, and each tree will give a predicted value based on its training results. Finally, the final prediction result is obtained by summarizing the predicted values ​​of all trees (taking the average or mode).

[0072] This example provides a code for training a random forest model:

[0073] from sklearn.ensemble import RandomForestRegressor

[0074] from sklearn.model_selection import train_test_split

[0075] from sklearn.metrics import mean_squared_error

[0076] #Assume that df is a DataFrame containing temperature and vibration frequency, and necessary preprocessing has been performed

[0077] #df['weighted_sum'] is the weighted sum of temperature and vibration frequency

[0078] #df['other_relevant_features'] is other relevant features

[0079] #df['trend_indicator'] is the simulated quantity we want to predict

[0080] # Split features and target variables

[0081] X = df[['weighted_sum','other_relevant_features']] #Assume 'other_relevant_features' is a column name or a list of column names of DataFrame

[0082] y = df['trend_indicator']

[0083] # Divide into training set and test set

[0084] X_train,X_test,y_train,y_test=train_test_split(X,y,test_size=0.2,random_state=42)

[0085] # Initialize the random forest model

[0086] model=RandomForestRegressor(n_estimators=100,random_state=42)

[0087] #Train model

[0088] model.fit(X_train,y_train)

[0089] #Predict the test set

[0090] y_pred=model.predict(X_test)

[0091] #Evaluate the model

[0092] mse=mean_squared_error(y_test,y_pred)

[0093] print(f'Mean Squared Error:{mse}')

[0094] The optimal random forest model parameters are selected by minimizing the cross-validation error value.

[0095] The comparison of the slope value of the current regression function and the predicted slope value of the regression function reflects the slow change trend of the simulation value. The specific method is:

[0096] A slope threshold G is set, and the slope threshold G=0.08. When the absolute value of the slope value of the current regression function or the absolute value of the slope prediction value of the regression function is greater than the slope threshold G, the early warning mechanism is triggered, indicating that the simulated quantity of the hydropower unit is in a significantly slow-changing state.

[0097] A slope slow change threshold D is set, wherein the slow change threshold D=0.03. The absolute value of the difference between the absolute value of the slope prediction value of the regression function obtained in step 5 and the absolute value of the slope prediction value of the regression function is taken, and then the absolute value of the difference is taken and compared with the slow change threshold D. If the absolute value of the difference is greater than the slow change threshold D, the early warning mechanism is triggered, indicating that the simulated quantity of the hydropower unit is expected to undergo a significant slow change.

[0098] According to the slope prediction value of the regression function obtained in step 5, the change rate of the slope prediction value is calculated to quantify the slow change speed of the simulation quantity. The change rate is the ratio of the difference of the slope prediction values ​​in adjacent time series samples to the time interval.

[0099] The formula for calculating the rate of change of the slope prediction value is: Where R is the rate of change of the slope prediction value, σ t+1 and σ t are the slope prediction values ​​of two adjacent time series samples, Δt is the time interval between two time series samples, Δt=M=kL.

[0100] In addition to comparing the slope value of the current regression function and the size and change of the slope prediction value of the regression function, the confidence interval of the slope prediction value can also be calculated to further evaluate the reliability of the prediction result. The calculation formula of the confidence interval can be derived based on the output variance or standard error of the random forest model.

[0101] When sampling the parameters in the hydropower unit, the unit active power, unit speed and oil cooler water flow should meet the following constraints:

[0102]

[0103] Only when the above constraints are fully satisfied, the parameters in the hydropower unit are sampled.

[0104] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. 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 method for predicting slow changes in simulated quantities of hydropower units based on characteristic value monitoring, characterized in that: The method comprises the following steps: Step 1, based on a preset time interval and a preset time window length, the parameters in the hydropower unit: the core component temperature of the turbine and the vibration frequency of the turbine are sampled, and the sampled samples are divided according to the preset time length to form a time series sample set with the time window as the unit; The core component temperature refers to one or more of the stator winding temperature, the stator core temperature and the bearing bush temperature; The preset time interval t is 2 to 5 seconds, and the continuous sampling is q times in a time window, then the preset time window length L=qt, where 4≤q≤12; dividing the sample according to the preset time length means that the preset time length M=kL, where k represents the number of time windows, 15≤k≤20; The time series sample set is represented as U = {A1, A2, ..., A i ,…,A n }, where A i represents the i-th time series sample, n represents the number of time series samples, each time series sample is composed of sampling data of k time windows, and the i-th time series sample A i Indicated as A i ={a i1 ,a i2 ,……,a ik }, a ik Represents the sampling data of k time windows in the i-th time series sample; When sampling the parameters in the hydropower unit, the unit active power, unit speed and oil cooler water flow should meet the following constraints: Only when the above constraints are fully satisfied, the parameters in the hydropower unit are sampled; Step 2, performing minimum-maximum normalization processing on the sampled data of each time window, and then calculating the standard deviation of the weighted sum of each time series sample in the time series sample set as the simulation quantity of the hydropower unit; The calculation formula for the minimum-maximum normalization is: Where T is the normalized temperature value of a single time window, is the average temperature of the time window, max(T) and min(T) are the maximum and minimum values ​​of the temperature data in the time window respectively; f is the vibration frequency value of a single time window after normalization, is the average value of the vibration frequency in the time window, max(f) and min(f) are the maximum and minimum values ​​of the vibration frequency data in the time window respectively; The calculation formula of weighted sum S is: S = w T ·T+w f f, where T is the normalized temperature value of a single time window, f is the normalized vibration frequency value of a single time window, and w T and w f are the weights of temperature and vibration frequency respectively, w T +w f =1; For the time series sample A, there are k time windows of normalized data, expressed as: (T1, f1), (T2, f2), …, (T k ,f k ), the weighted sum of the time window data is expressed as: S1,S2,…,S k ; Standard deviation of the weighted sum of time series samples: in, is the average of the weighted sums, S i It is the weighted sum of the i time window data in the time series sample; Then for time series sample A, the standard deviation of the weighted sum of its time series samples is: Among them, T i and f i Respectively represent the normalized temperature value and vibration frequency value of i times; Step 3, perform local linear regression operation on the simulated quantity in the time series sample set to obtain the slope value of the current regression function; The local linear regression operation uses the weighted least squares method, and its formula is: β = (X T WX) -1 X T WY, where β is the regression coefficient vector, X is the design matrix, W is the diagonal weight matrix, and Y is the response variable vector; Step 4, inputting the simulated quantity of the hydropower unit in the continuous time window into the trained random forest model, and predicting the simulated quantity value of the hydropower unit in the next time series sample based on the random forest model; When training the random forest model, feature data in the time series samples are used, and the feature data include temperature values, vibration frequency values, and standard deviations of their weighted sums; When predicting, the random forest model outputs the predicted simulated value based on the input time series sample, and its prediction formula is: in, is the predicted simulated value, m is the number of trees in the random forest, and h j (x) is the prediction result of the jth tree for input x; The training process of the random forest model uses cross-validation to optimize the model; The formula for cross validation is expressed as: Among them, CV is the cross-validation error, MSE p is the mean square error of the p-fold validation; By minimizing the cross-validation error value, the optimal random forest model parameters are selected; Step 5: perform local linear regression calculation on the simulated quantity again according to the simulated quantity value of the hydropower unit in the predicted next time series sample to obtain the slope prediction value of the regression function; compare the size and change of the slope value of the current regression function and the slope prediction value of the regression function to reflect the slow change trend of the simulated quantity; The comparison of the current regression function slope value and the predicted slope value of the regression function reflects the slow change trend of the simulation value. The specific method is: A slope threshold G is set, wherein the slope threshold G=0.

08. When the absolute value of the slope value of the current regression function or the absolute value of the slope prediction value of the regression function is greater than the slope threshold G, an early warning mechanism is triggered, indicating that the simulated quantity of the hydropower unit is in a significantly slow-changing state; A slope slow change threshold D is set, wherein the slow change threshold D=0.03, and the absolute value of the difference between the absolute value of the slope prediction value of the regression function obtained in step 5 and the absolute value of the slope prediction value of the regression function is taken, and then the absolute value of the difference is taken and compared with the slow change threshold D. If the absolute value of the difference is greater than the slow change threshold D, the early warning mechanism is triggered, indicating that the simulated quantity of the hydropower unit is expected to undergo a significant slow change; According to the slope prediction value of the regression function obtained in step 5, the change rate of the slope prediction value is calculated to quantify the slow change speed of the simulation quantity, and the change rate is the ratio of the difference of the slope prediction values ​​in adjacent time series samples to the time interval; The formula for calculating the rate of change of the slope prediction value is: Where R is the rate of change of the slope prediction value, σ t+1 and σ t are the slope prediction values ​​of two adjacent time series samples, Δt is the time interval between two time series samples, Δt=M=kL.

2. The method for predicting slow changes in simulated quantities of hydropower units based on characteristic value monitoring according to claim 1 is characterized in that: The number n of the time series samples in the time series sample set satisfies 5≤n≤15.

Images