Early Warning Method and System for Thermal Turbine Units Based on Mann-Kendall Algorithm and LSTM Neural Network

Abstract

This invention provides a method and system for early warning of thermal turbine units based on the Mann-Kendall algorithm and LSTM neural network. The method includes: acquiring unit operation data from a power plant database; filtering data under normal operating conditions from historical power plant data to construct a prediction library and a standard library neural network; inputting real-time power plant operation data into the standard and prediction libraries for comparison; triggering an alarm if thermal parameters exceed set thresholds; and identifying the changing trends of various thermal parameters using the Mann-Kendall algorithm after the alarm. This invention uses an LSTM neural network to consider the influence of the thermal capacity of the thermal turbine unit, predicting thermal parameters to achieve more advanced power plant operation prediction results. Furthermore, it employs a sliding window residual analysis method to evaluate the prediction results, thereby achieving more accurate evaluation and reducing the influence of random signals.

Description

Technical Field

[0001] This invention relates to the field of thermal turbine units, specifically to an early warning method and system for thermal turbine units based on the Mann-Kendall algorithm and LSTM neural network. Background Technology

[0002] Existing parameter early warning systems mostly focus on predicting vibration signals. However, when vibration signals change significantly, the system has already experienced a major malfunction, leaving staff with little time to respond effectively. Furthermore, predictions based on BP neural networks fail to take time-series signals into account and neglect the gradient information of parameters.

[0003] The shortcomings of the existing scheme are: it only incorporates vibration parameters, making it difficult to effectively predict gradual faults; it only considers the state transition matrix of parameters at the current time and a certain time before, without considering the influence of parameter gradients on the results; and it is susceptible to disturbances from random signals.

[0004] Patent document CN112487910A (application number: CN202011329125.0) discloses a method and system for early warning of faults in a nuclear power turbine system. The method includes: acquiring real-time signal data of multiple monitoring parameters of the turbine, processing the signal data of each monitoring parameter, and generating multi-dimensional matrix data based on the processed signal data of the multiple monitoring parameters; feeding the multi-dimensional matrix data into a pre-established state prediction model, and obtaining parameter prediction information for the turbine at the next moment based on the output of the state prediction model; outputting the state prediction result in real time based on the parameter prediction information; determining whether the turbine has experienced an anomaly based on the state prediction result, and generating fault warning information when an anomaly occurs. However, this invention does not select data from historical power plant data under normal operating conditions to construct a prediction library and a standard library neural network; and it does not identify the changing trends of various thermodynamic parameters after an alarm is triggered. Summary of the Invention

[0005] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for early warning of thermal turbine units based on the Mann-Kendall algorithm and LSTM neural network.

[0006] A method for early warning of thermal turbine units based on the Mann-Kendall algorithm and LSTM neural network according to the present invention includes:

[0007] Step S1: Obtain unit operation data from the power plant database. The data includes cleaned real-time power plant operation data and historical power plant operation data.

[0008] Step S2: Filter out data under normal operating conditions from the power plant's historical data, and construct a prediction library and a standard library neural network;

[0009] Step S3: Input the real-time operation data of the power plant into the standard library and the prediction library for comparison. If the thermal parameters exceed the set threshold, an alarm will be triggered.

[0010] Step S4: After the alarm is triggered, identify the changing trends of each thermal parameter.

[0011] Preferably, in step S2:

[0012] Step S2.1: Set standards for normal unit operation data, judge historical data, and put data that meets the standards into the training set for training;

[0013] Step S2.2: Optimize the parameters of the initially trained neural network library;

[0014] Step S2.3: Test the library after optimization.

[0015] Preferably, in step S3:

[0016] Step S3.1: Extract the real-time regulating stage pressure, main steam valve pressure, first stage extraction pressure, second stage extraction pressure, high-pressure cylinder exhaust pressure and power data from the database as state parameters;

[0017] Step S3.2: Input the status parameters into the standard library and determine their deviation from the normal value. If the deviation from the normal standard exceeds the threshold, an alarm will be triggered.

[0018] Step S3.3: If there is no alarm in the standard library, input the status parameters into the prediction library, predict the pressure value after the regulation stage, and calculate the standard deviation and mean of the residuals within the window using the sliding window residual analysis method. If the value exceeds the set threshold, an alarm will be triggered.

[0019] Preferably, in step S4:

[0020] Step S4.1: After the alarm is triggered, the user inputs the parameters to be viewed and their corresponding time periods;

[0021] Step S4.2: Visualize the parameters according to the user's needs;

[0022] Step S4.3: Identify the trend of parameter changes; determine the degree of drastic change using the growth coefficient discrimination method. If the growth coefficient value is greater than the preset value, the change is considered drastic.

[0023] The Mann-Kendall algorithm is used to determine whether data within a time period is increasing, decreasing, or remaining stable; the results are then correlated with the growth coefficient to determine whether the data is stable, experiencing a sudden increase, a sudden decrease, a gradual increase, or a gradual decrease.

[0024] Step S4.4: Determine whether the parameter changes are proportional by calculating the standard deviation of the ratio of the slopes of the parameters corresponding to the time step over the entire time period.

[0025] Preferably, the trend of parameter change is identified by combining the Mann-Kendall algorithm with the growth coefficient discrimination method; the degree of change is judged by the growth coefficient discrimination method, and if the growth coefficient value is greater than the preset value, the change is considered to be drastic; the Mann-Kendall algorithm is combined to determine whether the data increases, decreases or remains stable within the time period; the result is correlated with the result of the growth coefficient to find out whether the data remains stable, increases sharply, decreases sharply, increases slowly or decreases slowly.

[0026] According to the present invention, a thermal turbine unit early warning system based on the Mann-Kendall algorithm and LSTM neural network includes:

[0027] Module M1: Retrieves unit operation data from the power plant database, including cleaned real-time power plant operation data and historical power plant operation data;

[0028] Module M2: Filters data from historical power plant data under normal operating conditions to construct a prediction library and a standard library neural network;

[0029] Module M3: Inputs real-time operating data of the power plant into the standard library and prediction library for comparison. If the thermal parameters exceed the set threshold, an alarm will be triggered.

[0030] Module M4: After an alarm is triggered, it identifies the changing trends of various thermal parameters.

[0031] Preferably, in module M2:

[0032] Module M2.1: Sets the standards for normal unit operation data, judges historical data, and puts data that meets the standards into the training set for training;

[0033] Module M2.2: Optimizes the parameters of the initially trained neural network library;

[0034] Module M2.3: After optimization, the library is tested.

[0035] Preferably, in module M3:

[0036] Module M3.1: Extracts real-time data on the pressure after the regulating stage, the main steam valve pressure, the first stage extraction pressure, the second stage extraction pressure, the exhaust pressure of the high-pressure cylinder, and the power as status parameters from the database;

[0037] Module M3.2: Input the status parameters into the standard library, determine their deviation from the normal value, and trigger an alarm if the deviation from the normal standard exceeds the threshold.

[0038] Module M3.3: If there is no alarm in the standard library, the status parameters are input into the prediction library to predict the pressure value after the regulation stage. The standard deviation and mean of the residuals within the window are calculated by combining the sliding window residual analysis method. If the value exceeds the set threshold, an alarm is triggered.

[0039] Preferably, in module M4:

[0040] Module M4.1: After an alarm is triggered, the user inputs the parameters to be viewed and their corresponding time period;

[0041] Module M4.2: Visualizes parameters according to user needs;

[0042] Module M4.3: Identifies the trend of parameter changes; determines the degree of drastic change using the growth coefficient discrimination method; if the growth coefficient value is greater than the preset value, the change is considered drastic.

[0043] The Mann-Kendall algorithm is used to determine whether data within a time period is increasing, decreasing, or remaining stable; the results are then correlated with the growth coefficient to determine whether the data is stable, experiencing a sudden increase, a sudden decrease, a gradual increase, or a gradual decrease.

[0044] Module M4.4: Determines whether parameter changes are proportional by calculating the standard deviation of the ratio of the slopes of parameters at corresponding time steps over the entire time period.

[0045] Preferably, the trend of parameter change is identified by combining the Mann-Kendall algorithm with the growth coefficient discrimination method; the degree of change is judged by the growth coefficient discrimination method, and if the growth coefficient value is greater than the preset value, the change is considered to be drastic; the Mann-Kendall algorithm is combined to determine whether the data increases, decreases or remains stable within the time period; the result is correlated with the result of the growth coefficient to find out whether the data remains stable, increases sharply, decreases sharply, increases slowly or decreases slowly.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] 1. This invention uses an LSTM neural network to consider the influence of the turbine's heat capacity and predicts thermal parameters, thereby achieving more advanced power plant operation prediction results.

[0048] 2. This invention uses the sliding window residual analysis method to evaluate the prediction results, thereby achieving a more accurate evaluation effect and reducing the influence of random signals;

[0049] 3. This invention uses a symptom recognition algorithm that combines the growth coefficient discrimination method with the Mann-Kendall algorithm to quantify the changing trend, thereby achieving quantitative indicators of various parameters, gradients, and proportions after an alarm is triggered, providing a strong reference for subsequent fault diagnosis by staff. Attached Figure Description

[0050] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0051] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0052] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0053] Example 1:

[0054] This invention provides an algorithm based on big data neural networks to predict thermal parameters of power plants and evaluate the operating status of thermal turbine units, which can efficiently process power plant operating data.

[0055] To address the problems existing in the prior art, this invention proposes an early warning system for thermal turbine units based on the Mann-Kendall algorithm and LSTM neural network. By learning from historical data of normal power plant operation, it can predict subsequent normal data based on power plant operating parameters over a period of time. By comparing the predicted data with the actual data and combining the sliding window residual analysis method, it can issue an alarm when the deviation exceeds a set threshold and identify the trend of parameter changes after the alarm.

[0056] The technical solution adopted in this invention is as follows: Figure 1 As shown:

[0057] Step S1: Obtain unit operation data from the power plant database, including cleaned real-time power plant operation data and historical power plant operation data;

[0058] Step S2: Filter out data under normal operating conditions from the power plant's historical data, and construct a prediction library and a standard library neural network;

[0059] Step S3: Input the real-time operation data of the power plant into the standard library and the prediction library for comparison. If the thermal parameters exceed the set threshold, an alarm will be triggered.

[0060] Step S4: After the alarm is triggered, the changing trends of each thermal parameter are identified for reference in subsequent specific diagnosis.

[0061] Step S2 is as follows:

[0062] Step S2.1: Set standards for normal unit operation data, judge historical data, and put data that meets the standards into the training set for training;

[0063] Step S2.2: Optimize the parameters of the initially trained neural network library;

[0064] Step S2.3: After optimization, test the effectiveness of the neural network library.

[0065] Step S3 includes:

[0066] Step S3.1: Extract the real-time regulating stage pressure, main steam valve pressure, first stage extraction pressure, second stage extraction pressure, high pressure cylinder exhaust pressure and power data from the database as state parameters.

[0067] Step S3.2: Input the status parameters into the standard library and determine their deviation from the normal value. If the deviation from the normal standard exceeds the threshold, an alarm will be triggered.

[0068] Step S3.3: If there is no alarm in the standard library, input the status parameters into the prediction library to predict the pressure value after the regulation stage. Combine the sliding window residual analysis method to calculate the standard deviation and mean within the window. If the set threshold is exceeded, an alarm will be triggered.

[0069] Step S4 includes:

[0070] Step S4.1: After the alarm is triggered, the user inputs the parameters to be viewed and their corresponding time periods;

[0071] Step S4.2: Visualize the parameters according to the user's needs;

[0072] Step S4.3: Identify the trend of parameter change using the Mann-Kendall algorithm combined with the growth coefficient discriminant method; determine the degree of drastic change using the growth coefficient discriminant method, and consider the change to be drastic if the growth coefficient value is greater than 15%; combine the Mann-Kendall algorithm to determine whether the data is increasing, decreasing or remaining stable within the time period; correlate the result with the growth coefficient result to determine whether the data is stable, has increased sharply, decreased sharply, increased slowly or decreased slowly.

[0073] Step S4.4: Determine whether the parameter changes are proportional by calculating the standard deviation of the ratio of the slopes of the parameters corresponding to the time step over the entire time period.

[0074] Those skilled in the art will understand that, in addition to implementing the system, apparatus, and their modules provided by this invention in purely computer-readable program code, the same program can be implemented in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers by logically programming the method steps. Therefore, the system, apparatus, and their modules provided by this invention can be considered as a hardware component, and the modules included therein for implementing various programs can also be considered as structures within the hardware component; alternatively, the modules for implementing various functions can be considered as both software programs implementing the method and structures within the hardware component.

[0075] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A heat turbine unit early warning method based on Mann-Kendall algorithm and LSTM neural network, characterized in that, Comprising: Step S1: obtaining unit operation data in the power plant database, the data including cleaned real-time operation data and historical operation data of the power plant; Step S2: screening data in a normal operation state from the historical data of the power plant to construct a standard library and a prediction library of the thermal turbine unit early warning system neural network; Step S3: inputting real-time operation parameter data of the power plant into the standard library and the prediction library for comparison, and alarming if the thermal parameters exceed the set threshold; Step S4: after the alarm, identifying the change trend of each thermal parameter by combining the Mann-Kendall algorithm; In the step S3: Step S3.1: extracting real-time adjusted stage after pressure, main steam valve pressure, first-stage bleeding pressure, second-stage bleeding pressure, exhaust pressure of the high-pressure cylinder and power data from the edge database as state parameters; Step S3.2: inputting the state parameters into the standard library to judge the deviation from the normal value, and alarming if the deviation exceeds the threshold; Step S3.3: if the standard library does not alarm, inputting the state parameters into the prediction library to predict the value of the adjusted stage after pressure, and calculating the standard deviation and mean value of the residual error in the window by combining the sliding window residual error analysis method, and alarming if the value exceeds the set threshold; In the step S4: Step S4.1: after the alarm, inputting the parameters and the corresponding time period to be viewed by the user; Step S4.2: visualizing the parameters according to the needs of the user; Step S4.3: identifying the change trend of the parameters; determining the degree of change by the growth coefficient discrimination method, and considering the change to be violent if the growth coefficient value is greater than the preset value; determining whether the data is increasing, decreasing or remaining stable in the time period by the Mann-Kendall algorithm; correlating the results with the results of the growth coefficient to know that the data is remaining stable, sharply increasing, sharply decreasing, slowly increasing or slowly decreasing; Step S4.4: determining whether the change of the parameters is proportional by calculating the standard deviation of the slope ratio of the parameters in the corresponding time steps in the whole time period.

2. The method of Claim 1, wherein the method is based on the Mann-Kendall algorithm and the LSTM neural network. In the step S2: Step S2.1: setting the standard for normal unit operation data, judging the historical data, and putting the data meeting the standard into the training set for training; Step S2.2: optimizing the parameters of the initially trained neural network library; Step S2.3: testing the effectiveness of the neural network library after optimization.

3. The thermal turbine unit early warning method based on the Mann-Kendall algorithm and the LSTM neural network according to claim 1, characterized in that: the change trend of the parameters is identified by the Mann-Kendall algorithm combined with the growth coefficient discrimination method; the degree of change is determined by the growth coefficient discrimination method, and the change is considered to be violent if the growth coefficient value is greater than the preset value; whether the data is increasing, decreasing or remaining stable in the time period is determined by the Mann-Kendall algorithm; the results are correlated with the results of the growth coefficient to know that the data is remaining stable, sharply increasing, sharply decreasing, slowly increasing or slowly decreasing.

4. A heat turbine unit early warning system based on the Mann-Kendall algorithm and the LSTM neural network, characterized in that, Comprising: Module M1: Obtain unit operation data in power plant database, including cleaned real-time operation data and historical operation data of power plant; Module M2: Screen data in normal operation state from power plant historical data, construct prediction library and standard library neural network of thermal turbine unit early warning system; Module M3: Input real-time operation parameter data of power plant into standard library and prediction library for comparison, and alarm if thermal parameters exceed set threshold; Module M4: After alarm, identify change trend of each thermal parameter by Mann-Kendall algorithm; In the module M3: Module M3.1: Extract real-time regulated stage post-pressure, main steam valve pressure, first-stage extraction pressure, second-stage extraction pressure, exhaust pressure of high-pressure cylinder and power data as state parameters from database; Module M3.2: Input state parameters into standard library to judge deviation from normal value, and alarm if deviation from normal value exceeds threshold; Module M3.3: If standard library does not alarm, input state parameters into prediction library to predict value of regulated stage post-pressure, and calculate standard deviation and mean value of residual error in window by sliding window residual error analysis method, and alarm if it exceeds set threshold; In the module M4: Module M4.1: After alarm, input parameters and corresponding time period to be viewed by user; Module M4.2: Visualize parameters according to user's needs; Module M4.3: Identify change trend of parameters; determine change intensity by growth coefficient discrimination method, and consider change as intense if growth coefficient value is greater than preset value; Determine whether data is growing, decreasing or remaining stable in time period by Mann-Kendall algorithm; correlate results with growth coefficient results to know that data is remaining stable, sharply increasing, sharply decreasing, slowly increasing or slowly decreasing; Module M4.4: Determine whether parameter change is proportional by calculating standard deviation of parameter slope ratio in corresponding time step in entire time period.

5. The heat turbine unit early warning system based on the Mann-Kendall algorithm and the LSTM neural network according to claim 4, characterized in that, In the module M2: Module M2.1: Set standard for normal unit operation data, judge historical data, and put data meeting standard into training set for training; Module M2.2: Optimize parameters of initially trained neural network library; Module M2.3: Test effectiveness of optimized neural network library.

6. The thermal turbine unit early warning system based on Mann-Kendall algorithm and LSTM neural network according to claim 4, characterized in that: Identify change trend of parameters by Mann-Kendall algorithm combined with growth coefficient discrimination method; determine change intensity by growth coefficient discrimination method, and consider change as intense if growth coefficient value is greater than preset value; determine whether data is growing, decreasing or remaining stable in time period by Mann-Kendall algorithm; correlate results with growth coefficient results to know that data is remaining stable, sharply increasing, sharply decreasing, slowly increasing or slowly decreasing.

Images