KR102124779B1 - Early warning device of plant of start and stop using multi-predictive model - Google Patents

Abstract

The present invention relates to an early warning device for starting and stopping a power plant using a multi-prediction model and a method thereof. The early warning device includes a multi-prediction model provided to receive information on the starting and stopping states of a power plant for each of a plurality of sections of the starting and stopping states and allow each of a plurality of prediction models to output a predicted value and a reliability of the predicted value in one section; a reliability analysis module for selecting one from the plurality of prediction models by comparing and analyzing the reliability output from each of the plurality of prediction models provided in the multi-prediction model in one of the plurality of sections in the starting state of the power plant or one of the plurality of sections in the stopping state and outputting a final predicted value of a corresponding prediction model for the one section; a predicted value regulation module for regulating discontinuities of the predicted values output from different prediction models for each of the plurality of sections and comparing the final predicted values output for each of the plurality of sections with actual measured values to output a residual; and a determination module for analyzing the residual to determine whether there are defects in the starting and stopping states of the power plant.

Description

Early warning device of plant and start and stop using multi-predictive model

The present invention relates to an early warning device and method for starting and stopping a power plant using multiple prediction models, and more specifically, to enable early warning of starting and stopping a power plant by changing and using a prediction model for each start and stop operation section. An early warning device and method for starting and stopping a power plant using multiple prediction models.

The early warning device is a system used to prevent equipment failure in advance and has been widely used in the military, aviation, and power generation fields.

The principle of the early warning device is that the predictive model uses the past normal operation data to make a predicted value for the driving condition, and compares the predicted value with the measured value to calculate the residual, thereby alerting if the residual is out of the normal operating range. Is a way to make it happen.

Conventionally, various mathematical methods are used to calculate the predicted value of the operating state of devices, and the representative methods include a kernel regression method using a statistical method, a Gaussian process regression method, a neural network method, and a Kalman filter.

The early warning device has a minimal effect on the occurrence of false alarms, so that the actual effect can be seen. Otherwise, while an operator analyzes a large number of alerts, important alerts may be missed, and the operator's concentration may be degraded, making it impossible to locate alerts that are caused by problems with the actual device.

In this situation, since the predictive model is trained using the data of the normal operation of the power plant, it is not easy to perform an early warning using the predictive model when the power plant is in the start and stop transient state. In the maneuvering and stopping states, there was a problem in that it was impossible to identify abnormal signs early.

Therefore, there is a need for an apparatus and a method capable of performing an early warning even in a transient state of starting and stopping of a power plant, not devices in a power plant.

Republic of Korea Registered Patent No. 10-1827108

The present invention is to solve the problems of the prior art as described above, it is an object to perform an early warning by detecting the abnormal signs at the start and stop of the power plant.

An early warning device for starting and stopping a power plant according to the present invention for achieving the above object receives information about a starting and stopping state of a power plant for each of a plurality of sections of the starting and stopping states, and predicts a plurality of sections in one section A multi-prediction model in which the model is provided to output a predicted value and a reliability for the predicted value, respectively; By comparing and analyzing the reliability outputted by the plurality of prediction models provided in the multiple prediction models in one of a plurality of sections of a power plant starting state or one of a plurality of sections when entering a stationary state of a power plant, A reliability analysis module that outputs a final prediction value of a corresponding prediction model for the one section by selecting one of the plurality of prediction models; A predicted value flattening module that flattens the discontinuity of the predicted values output from different prediction models for each of the plurality of sections, and outputs a residual by comparing the final predicted values output for each of the plurality of sections with actual values; And a determination module that analyzes the residuals to determine whether the power plant is in a start and stop state or not.

Here, the reliability analysis module may determine the switching of the section on the basis that the reliability is greater than or equal to a threshold when switching from one section to another section among the plurality of sections.

In addition, the multi-prediction model includes a maneuver prediction model, and the maneuver prediction model includes a plurality of first maneuver prediction models, a second maneuver prediction model, and an N maneuver prediction model. Prediction values and reliability can be generated for a plurality of sections classified as.

In addition, the multi-prediction model includes a static prediction model, and the static prediction model includes a plurality of first static prediction models, second static prediction models, and N static prediction models, so that the power plant is in a stationary state in a normal operation state. Prediction values and reliability can be generated for a plurality of sections classified as.

In addition, the multi-prediction model includes a maneuver prediction model and a stop prediction model, and the maneuver prediction model and the stop prediction model may use the same prediction model of the same type.

In addition, according to the classification criteria set by the user, the information on the starting and stopping states of the power plant is received to classify the plurality of sections on the normal operating state in the starting state of the power plant and the stopped state in the normal operating state of the power plant. It may include a section setting unit.

In addition, the prediction value flattening module may use either a moving average filter or a low-pass filter.

In addition, information regarding the start and stop states of the power plant may include reactor output or generator output.

On the other hand, the power plant start and stop early warning method for achieving the above object is a multi-prediction model receives information about the start and stop states of the power plant for each of the plurality of sections of the start and stop states, a plurality of predictions in one section The model outputting a predicted value and a reliability for the predicted value, respectively; The reliability analysis module compares the reliability output by each of the plurality of prediction models provided in the multiple prediction models in one of a plurality of sections of a power plant starting state or one of a plurality of sections when entering a stationary state, and Analyzing and selecting one of the plurality of prediction models to output a final prediction value of the prediction model for the one section; A predicted value flattening module flattening discontinuities of the predicted values output from different prediction models for each of the plurality of sections, and outputting a residual by comparing the final predicted values output for each of the plurality of sections with actual values; And determining, by the determination module, whether there is a defect in the start and stop states of the power plant by analyzing the residuals.

The early warning device and method for starting and stopping a power plant according to the present invention have the following effects.

First, early warning monitoring is possible in the start and stop sections of a power plant. Previously, a single prediction model was used in a power plant, and the existing single prediction model mainly used a prediction model using normal operation data. Accordingly, the existing single prediction model was applicable only in the normal operation section, and it was not practically applicable in the start and stop sections of the power plant, so that the start and stop sections could not be alarmed early. On the other hand, the early warning device and method according to the present invention is excellent in the scope of application because it is applied to the start and stop sections of a power plant using a plurality of predictive models to perform early warning.

Second, it is possible to overcome vulnerabilities occurring in a single prediction model. The early warning device according to the present invention uses multiple prediction models rather than a single prediction model. Accordingly, it is possible to overcome the vulnerability because it is possible to compensate for the case where the predicted value may differ from the measured value due to defects in each predicted model.

Third, it increases reliability. In the present invention, by applying a multi-prediction model, an early warning is performed on the operating state of devices of a power plant. As a result, the ability to detect defects can be strengthened and reliability for early warning can be increased rather than a single predictive model.

1 is a block diagram of an early warning device for starting and stopping a power plant using multiple prediction models.
FIG. 2 is a diagram schematically showing the performance of functions between components from a start state to a normal operation state in an early warning device for starting and stopping a power plant using multiple prediction models.
FIG. 3 is a diagram schematically showing functions performed between components from a normal operation state to a stop state in an early warning device for starting and stopping a power plant using a multi-prediction model.
4 is a flowchart of an early warning method for starting and stopping a power plant using multiple prediction models.
FIG. 5 is a diagram schematically showing that a multi-prediction model is applied for each section from an operation state of a power plant to a normal operation of an early warning device for starting and stopping a power plant using a multi-prediction model.
6 is a diagram schematically showing that a multi-prediction model is applied for each section from a normal operation state of a power plant to a stop when an early warning device for starting and stopping a power plant using the multi-prediction model is used.

Hereinafter, preferred embodiments of the present invention will be described with reference to the accompanying drawings. However, some components irrelevant to the subject matter of the invention will be omitted or compressed, but the omitted components are not necessarily necessary in the present invention, and may be used in combination by those skilled in the art to which the present invention pertains. Can be.

1 is a block diagram of an early warning device for starting and stopping a power plant using multiple prediction models.

As illustrated in FIG. 1, an early warning device for starting and stopping a power plant using a multi-prediction model according to an embodiment of the present invention includes a signal measurement unit 100, a multi-prediction model 200, and a reliability analysis module 300 ), the prediction value flattening module 400 and the determination module 500.

The signal measuring unit 100 is configured to receive a monitoring signal for output when the power plant is started and stopped, calculate the measured value, and transmit it. The monitoring signals related to output when the power plant starts and stops are related to pressure, temperature, and flow rate data. The signal measuring unit 100 is attached to and around the power plant system to measure and transmit a section of the power plant start and stop state. The transmission interval can be transmitted at a user-specified time interval. For example, it can be transmitted at 1 second intervals.

The section setting unit 150 receives the classification criteria set by the user or the expert and receives information on the start and stop states of the power plant from the signal measuring unit 100 according to the criteria classified by the user or the expert, thereby starting the power station. In the classification of the plurality of sections regarding the normal operation state and the stationary state in the normal operation state of the power plant.

In the present invention, the classification of the section on the starting and operating state of the power plant is classified by the user or the expert by looking at the data on the monitoring signal, and the section setting unit 150 acquires the section and starts the operation from the starting state of the power plant to the normal operating state. It is to classify the section from the normal operation state to the stationary state of the power plant.

For example, for some monitoring signals related to Component Cooling Water (CCW), the speed of the Reactor Coolant Pump (RCP) is 1100 RPM or higher, and the Reactor Coolant System (RCS) cold tube temperature is 294 If it is ℃, it can be operated in the same way as normal operation.

On the other hand, some monitoring signals of the CVCS (chemical and volume control system) system can start normal operation from the point when the reactor output exceeds 0.1%. These monitoring signals can be used by making two groups of predictive models, starting and stopping, and operating normally, and switching them by section.

In addition, in the case of some monitoring signals of the main steam system or the condenser system, operation data may be continuously changed depending on the output of the power plant. These monitoring signals can be selectively used by providing a number of prediction models for each output section.

Therefore, the section is not classified uniformly by applying the same criteria, and the user or expert classifies the section by considering the characteristics of the operation data for each monitoring signal. The criteria for dividing the intervals are the values of the representative monitoring signals, and the approximate criteria are as follows.

For example, a criterion of RCP speed of 1100 RPM or higher, a criterion of RCS low temperature pipe of 294°C or higher, a reactor power of 0.1%, 20% or 80%, a turbine speed of 20 RPM or higher, and a generator output of 20%, 80%, etc. Can be divided into:

The multi-prediction model 200 is a component that receives a monitoring signal received from the signal measurement unit 100 and generates a predicted value for the output of the power plant for each section of the start and stop states of the power plant. The multi-prediction model 200 may use a plurality of prediction models from 1 to N for each section of a power plant start and stop state, and one prediction model may be selected, and the maneuver prediction model 210 and the stop prediction Model 220 may be included.

The maneuver prediction model 210 is configured to generate a predicted value for the output of the power plant for each of a plurality of sections of the power plant's start state when the power plant enters the start state. The starting state of such a power plant may be classified into a plurality of sections from the start of starting to the state of normal operation.

The stationary prediction model 220 is configured to generate a predicted value for the output of the power plant for each of a plurality of sections of the stationary state of the power plant when the power station enters the stationary state. The stationary state of such a power plant may be classified into a plurality of sections from the time when normal operation starts to the state of stoppage.

The maneuver prediction model 210 and the stop prediction model 220 are said to be able to use a plurality of prediction models for each section. Accordingly, in the maneuver prediction model 210, a plurality of prediction models include a first maneuver prediction model 211, a second maneuver prediction model 212,. , Referred to as the Nth starting prediction model, and in the static prediction model 220, the first static prediction model 221, the second static prediction model 220,. , It can be referred to as an Nth static prediction model. Different algorithms may be applied to a plurality of prediction models of the maneuver prediction model 210 and the stop prediction model 220, and each prediction model may output prediction values of the current monitoring signal and reliability of the prediction values, respectively.

The reliability analysis module 300 compares and analyzes the reliability output by each of the plurality of prediction models provided in the multiple prediction models 200 in one section of the starting state of the power plant or one section when entering the stationary state, and analyzes the plurality of prediction models. By selecting one of the prediction models, the final prediction value of the prediction model for a section is output. Here, a prediction model for each section may be selected through a representative signal for determining each section switch in a plurality of sections.

The predicted value flattening module 400 is configured to flatten discontinuities of predicted values output from different prediction models for a plurality of sections, and output a residual by comparing final predicted values output for each of the plurality of sections with actual values.

The determination module 500 is a component that analyzes the residual output from the comparison module 400 to determine the presence or absence of a defect in the operation state of power plant equipment.

In addition, hereinafter, a functionally connected relationship between components of an early warning device for starting and stopping a power plant using the multi-prediction model 200 will be described.

First, let's take a brief look at how early warnings are performed when the power plant is running.

FIG. 2 is a diagram schematically showing functions performed between components from an initial state to a normal operation state in an early warning device for starting and stopping a power plant using a multi-prediction model 200.

As shown in FIG. 2, the early warning device for starting and stopping a power plant according to the present invention outputs the signal measuring unit 100 to start the power of the power plant in order to perform an early warning when there is a problem with the power plant in the starting state. A monitoring signal is received for a plurality of sections classified from a start state to a normal operation state and a plurality of sections classified from a normal operation state to a stop state.

The received monitoring signal is received by the maneuver prediction model 210 provided in the multi-prediction model 200 and outputs the predicted values and reliability for each section of the power plant's maneuvering state and transmits the predicted value to the predicted value flattening module 400.

The prediction value flattening module 400 flattens the prediction values output from different prediction models, selects the final prediction values, compares and analyzes the measured values, and outputs a residual.

Thereafter, the residual is transmitted to the determination module 500, and the determination module 500 can analyze the residual and perform an early warning.

Let's take a brief look at the early warnings that are performed from the start-up state of the plant to the stationary state of the plant in the opposite process.

3 is a diagram schematically showing functions performed between components from a normal operation state to a stop state in an early warning device for starting and stopping a power plant using a multi-prediction model 200.

As shown in FIG. 3, the early warning device for starting and stopping a power plant according to the present invention is a signal measuring unit for performing an early warning when there is a problem in a power plant from a normal operation state to a time when it enters a stop state ( 100) receives a monitoring signal for the output at the time of entering the stationary state of the power plant for each of the plurality of sections classified from the normal operation state to the stationary state and the plurality of sections classified from the normal operation state to the stationary state.

The received monitoring signal is received by the stationary prediction model 220 provided in the multi-prediction model 200 and outputs the predicted values and reliability every several sections from the normal operation state to the stationary state of the power plant to the predicted value flattening module 400 send.

The predicted value flattening module 400 flattens the predicted values output from different predictive models, selects the final predicted values, compares and analyzes the measured values, and outputs a residual.

Thereafter, the residual is transmitted to the determination module 500, and the determination module 500 can analyze the residual and perform an early warning.

Hereinafter, a process of an early warning regarding starting and stopping of a power plant will be described with reference to the drawings.

4 is a flowchart of an early warning method for starting and stopping a power plant using multiple prediction models 200.

As illustrated in FIG. 4, initially, the multiple prediction models 200 receive information about the start and stop states of the power plant for each of the plurality of sections in the start and stop states, so that the plurality of prediction models in one section are predicted values and The reliability of the predicted value is output. <S40>

By sensing the output of the power plant in relation to the start and stop states of the power plant, the signal measuring unit 100 may receive a monitoring signal. Since the received monitoring signals are analog signals, the signal measuring unit 100 processes data and undergoes an analog-to-digital conversion step, whereby it can be converted into a digital signal. The signal measurement unit 100 may transmit the measured values of the monitoring signals converted to digital signals to the prediction value flattening module 400.

Here, the output of the power plant can be largely divided into a reactor output (heat output) and a generator output. Such reactor output and generator output are in megawatts (MW), but for convenience,% units can be used as the maximum output power. .

In the present invention, since section classification according to a plurality of prediction models is largely associated with the power plant output, it will be described as an example of the power plant output.

In addition, the multi-prediction model 200 including the maneuver prediction model 210 and the stop prediction model 220 each composed of a plurality of prediction models, the power of the power plant in the corresponding section from the output of the power plant measured by the signal measuring unit 100 Output the predicted value for the output.

Such a plurality of prediction models may be used from the first to the Nth for a plurality of sections from the starting state of the power plant to the normal operating state and the normal operating state to the stationary state of the power plant.

Specifically, the maneuver prediction model 210 is composed of a plurality of prediction models, the first maneuver prediction model 211, the second maneuver prediction model 212,. In addition, the N-th maneuver prediction model may be provided. Accordingly, each prediction model includes a first maneuver prediction model 211, a second maneuver prediction model 212,... , It can be used up to the N-th maneuver prediction model to output predicted values and reliability of predicted values for a plurality of sections.

The maneuver prediction model 210 may output a prediction value of a device by applying different algorithms according to use.

For example, the first maneuvering prediction model 211 may use a data model using an Auto Associative Kernel Regression (AAKR) technique, and the second maneuvering prediction model 212 may use a physical model using a Kalman Filter. Can be used. Accordingly, the first maneuvering prediction model 211 may output a prediction value using an AAKR technique, and the second maneuvering prediction model 212 may output a prediction value using a Kalman filter.

The AAKR technique generates a pattern model by learning past data, compares the current measured value with the pattern model to determine similarity, and weights the pattern model based on the similarity to generate a predicted value. Therefore, the AAKR technique is an interpolation-type prediction model that can be predicted in a trained pattern model, and can be predicted only when the actual value exists in the trained pattern model. In the similarity determination, it is possible to check whether the actual value exists in the learned pattern region, and the reliability of the predicted value can be calculated by synthesizing these similarities.

When a prediction model to which a plurality of AAKR techniques are applied is applied, each AAKR model can output prediction values and reliability. Here, the reliability means the similarity between the measured value and the pattern model.

The Kalman filter is a statistical model based on a physical model, and prediction values and covariance can be output. Such a Kalman filter can be predicted even in an unlearned region, and it is possible to determine the uncertainty (reliability) of the predicted value through covariance.

After the predicted value is output, the first maneuvering prediction model 211, the second maneuvering prediction model 212,. , Reliability for each predicted value may be output from the N-th maneuver prediction model. At this time, the size, meaning, and calculation method of the reliability may be different for each algorithm.

The reliability is the first maneuver prediction model 211, the second maneuver prediction model 212,. , It is an index roughly showing how accurate the predicted value outputted to the device in the Nth prediction model is.

Further, the first maneuvering prediction model 211, the second maneuvering prediction model 212, ... , N-th maneuver prediction model, the first maneuver prediction model 211, the second maneuver prediction model 212, such as all using the AAKR technique or Kalman filter. Also, all the N-th maneuver prediction models may use the same algorithm.

In the same way, the stationary prediction model 220 may be used for a plurality of sections from the normal operation state of the power plant to the stationary state.

Specifically, the static prediction model 220 is composed of a plurality of predictive models, the first static prediction model 221, the second static prediction model 222,. , N-th stationary prediction model. Accordingly, each prediction model includes a first stop prediction model 221, a second stop prediction model 222,... , It can be used up to the Nth stop prediction model to output predicted values and reliability of predicted values for a plurality of sections.

Since the algorithm applied to the static prediction model 220 can also be applied in the same manner as the predictive prediction model 210, the description of the same content will be omitted.

Thereafter, the reliability analysis module 300 outputs a plurality of prediction models provided in the multi-prediction model 200 in one of a plurality of sections of a power plant start-up state or one of a plurality of sections when entering a stationary state, respectively. By comparing and analyzing one reliability and selecting one of a plurality of prediction models, the final prediction value of the prediction model for one interval is output. <S41>

This will be continued with reference to the drawings.

FIG. 5 is a diagram schematically showing that the multi-prediction model 200 is applied for each section from the start state of the power plant to the normal operation of the early warning device for starting and stopping the power plant using the multi-prediction model 200.

As shown in Fig. 5, a plurality of sections are formed between the starting state and the normal driving state. Accordingly, the first maneuvering prediction model 211, the second maneuvering prediction model 212, which are configured in the maneuvering prediction model 210,. , The N-th start prediction model includes a first start prediction model 211, a second start prediction model 212 for each of a plurality of sections from section #1 to section #n,. , The N-th maneuver prediction model is used to output the predicted value and reliability for one section.

Thereafter, in the reliability analysis module 300, the first maneuvering prediction model 211, the second maneuvering prediction model 212,... , By comparing and analyzing the reliability output from the N-th maneuvering prediction model, a prediction model having the highest reliability can be selected in a corresponding section.

In the same way, the first static prediction model 221, the second static prediction model 222 provided in the static prediction model 220,. , By comparing and analyzing the reliability output from the Nth static prediction model, the most reliable prediction model can be selected. This will be described with reference to FIG. 6.

6 is a diagram schematically showing that the multi-prediction model 200 is applied for each section from a normal operation state of a power plant to a stop when an early warning device for starting and stopping a power plant using the multi-prediction model 200 is stopped.

As shown in Fig. 6, a plurality of sections are formed between the normal operation state and the stop state. Accordingly, the first static prediction model 221, the second static prediction model 222 configured in the static prediction model 220,. , The Nth stop prediction model includes a first stop prediction model 221, a second stop prediction model 222 for each of a plurality of sections from interval #1 to interval #n,. , The Nth static prediction model is used to output the predicted value and reliability for one section.

Thereafter, in the reliability analysis module 300, the first static prediction model 221, the second static prediction model 222,… , By comparing and analyzing the reliability output from the Nth static prediction model, a prediction model having the highest reliability in a corresponding section may be selected.

This means that the reliability analysis module 300 selects a predictive model and outputs a final predicted value for each section among a plurality of sections of a plurality of sections of a starting state of a power plant or a plurality of sections when entering a stationary state.

At this time, switching from one section to another section is determined by the reliability analysis module 300 that the reliability analysis module 300 determines that a normal signal for classifying the corresponding section is received when the reliability is higher than a preset threshold based on the reliability. Can be.

In addition, as another embodiment of determining the section switching, the section switching may be determined when the reliability of the next section is high by comparing the reliability of the current section with the next section in real time.

Also, for each section, the reliability analysis module 300 may select a prediction model having the highest reliability and output the final prediction value.

In addition, it is also possible to determine the section switching through the representative signal.

Specifically, the representative signal is the main signal of the power plant, such as the speed of the RCP, the temperature of the RCS low-temperature tube, the reactor power, the turbine speed, and the generator output, and can be used as an auxiliary by adding it to the result of the reliability analysis module.

For example, even if the reliability of the third prediction model is highest according to the output of the reliability analysis module 300, when it is determined that it is appropriate to use the fourth prediction model based on the state of the representative signal, the reliability analysis module 300 You can ignore the results and use the predicted values from the fourth model.

Here, the start prediction model 210 and the stop prediction model 220 may use a plurality of prediction models, but the types of the prediction models may use the same prediction model.

Next, the predicted value flattening module 400 flattens the discontinuity of the predicted values output from different prediction models for each of the plurality of sections, and compares the final predicted values output for each of the plurality of sections with actual values to determine a residual. Output it. <S42>

Since the final prediction value output from the maneuver prediction model 210 or the static prediction model 220 may be output by different algorithms for a plurality of sections, it is necessary to convert to a single criterion in order to compare the residuals. Accordingly, the final predicted value may be converted through a process of flattening the discontinuity.

A motion average filter or a low pass filter may be used for the prediction value flattening operation.

The moving average filter is a filter that outputs continuously input values as an average of an arbitrary number, and the low-pass filter is a filter that blocks the high-frequency portion and passes the low-frequency portion of the frequency components of the input values. Both of these filters smoothly connect sudden changes occurring in the input value.

For example, the moving average filter,

Can be represented as Here, N may represent the number of samples.

In addition, in the case of the first-order digital low-pass filter of the low-pass filter,

는 샘플링 타임, r은 시정수를 나타낼 수 있다.Can be represented as here,

Is a sampling time and r is a time constant.

In addition, the residual is calculated using the difference between the measured value and the predicted value of the corresponding device. In other words, the residual can be calculated and output through the |measured value-predicted value|.

Finally, the determination module 500 analyzes the residuals to determine whether the power plant is in a start or stop state or not. <S43>

The determination module 500 may determine whether a monitoring signal for the corresponding device is included in a preset normal operation range based on the residual output from the prediction value flattening module 400.

Here, the normal driving range may be determined differently according to a user setting. Therefore, if the residual value is outside the preset normal operating range, the monitoring signal for the device determines that the relevant device is in an abnormal operating state and generates an early warning, and if the residual value does not exceed the preset normal operating range, an alarm Does not cause

At this time, the normal operating range for generating an alarm can be adjusted by the user.

For example, the normal operating range of the water supply pump is 45 (minimum) to 55 (maximum), and the deviation between the minimum and maximum values is 10.

Therefore, when the residual is programmed to generate an alarm when 30% (ㅁ3) of the deviation of the maximum [min.-min. value] of the normal operating range (W3), the residual when the input measured value is 60 and the calculated predicted value is 52 Becomes 8, and an alarm is generated in excess of 30% of the deviation.

Conversely, if the measured value is 43 and the predicted value is 45, the residual becomes 2, so that no alarm is generated.

When an alarm occurs, the determination module 500 may transmit the alarm status information to an integrated center device (not shown) that monitors all power plants so that the user can confirm the alarm.

As described above, in the present invention, multiple prediction models 200 are used to output a plurality of prediction values and reliability based on a monitoring signal for starting and stopping of a power plant, and outputting residuals by comparing predicted values having a high ranking with actual values based on reliability. By doing so, it is determined whether the output residual is included in the normal operation range and an early warning is generated.

The above-described preferred embodiments of the present invention are disclosed for the purpose of illustration, and those skilled in the art having ordinary knowledge of the present invention will be able to make various modifications, changes, and additions within the spirit and scope of the present invention. And additions should be considered to fall within the scope of the claims of the present invention.

100: signal measuring unit
150: section setting unit
200: multiple prediction model
210: maneuver prediction model
211: first maneuver prediction model
212: second maneuver prediction model
220: static prediction model
221: first static prediction model
222: second stop prediction model
300: reliability analysis module
400: prediction value flattening module
500: decision module

Claims (9)

In the early warning device for starting and stopping the power plant,
A multiple prediction model provided to receive information on a start and stop state of a power plant for a plurality of sections of the start and stop states, so that a plurality of prediction models in one section output prediction values and reliability of the prediction values, respectively;
By comparing and analyzing the reliability outputted by the plurality of prediction models provided in the multiple prediction models in one of a plurality of sections of a power plant starting state or one of a plurality of sections when entering a stationary state of a power plant, A reliability analysis module that outputs a final prediction value of a corresponding prediction model for the one section by selecting one of the plurality of prediction models;
A predicted value flattening module that flattens the discontinuity of the predicted values output from different prediction models for each of the plurality of sections, and outputs a residual by comparing the final predicted values output for each of the plurality of sections with actual values; And
An early warning device for starting and stopping a power plant, including; a determination module for analyzing the residuals to determine whether the power plant is in a start and stop state or not.
According to claim 1,
The reliability analysis module is an early warning device for starting and stopping a power plant that determines the switching of the section based on the reliability being greater than or equal to a threshold when switching from one section to another section among the plurality of sections.
According to claim 1,
The multiple prediction model includes a maneuver prediction model,
The maneuver predictive model includes a plurality of first maneuver predictive models, a second maneuver predictive model, and an N maneuver predictive model to generate predicted values and reliability for a plurality of sections categorized from a start state to a normal operation state of the plant Early warning device for starting and stopping.
According to claim 1,
The multiple prediction model includes a static prediction model,
The stationary predictive model includes a plurality of first stationary predictive models, a second stationary predictive model, and an N stationary predictive model, to generate predicted values and reliability for a plurality of sections classified from a normal operation state to a stationary state of the power plant. Early warning device for starting and stopping.
According to claim 1,
The multi-prediction model includes a maneuver prediction model and a stop prediction model,
The start prediction model and the stop prediction model are early warning devices for starting and stopping a power plant using the same kind of prediction model.
According to claim 1,
Sections for receiving the information on the start and stop states of the power plant according to the classification criteria set by the user and classifying the plurality of sections on the normal operation state on the power plant start state and the stop state on the power plant normal operation state An early warning device for starting and stopping a power plant including a setting unit.
According to claim 1,
The predicted value flattening module is an early warning device for starting and stopping a power plant using either a moving average filter or a low-pass filter.
According to claim 1,
The information regarding the start and stop state of the power plant includes a reactor output or a generator output, an early warning device for starting and stopping the power plant.
In the early warning method for starting and stopping the power plant,
A multiple prediction model receiving information regarding a start and stop state of a power plant for each of a plurality of sections of the start and stop states, and outputting prediction values and reliability of the prediction values in a plurality of prediction models in one section, respectively;
The reliability analysis module compares the reliability output by each of the plurality of prediction models provided in the multiple prediction models in one of a plurality of sections of a power plant starting state or one of a plurality of sections when entering a stationary state, and Analyzing and selecting one of the plurality of prediction models to output a final prediction value of the prediction model for the one section;
Flattening a discontinuity of the predicted value output from a predictive model different for each of the plurality of sections, and outputting a residual by comparing the final predicted value output for each of the plurality of sections with an actual value; And
Determining module by analyzing the residual to determine whether the fault of the start and stop state of the power plant; early warning method for power plant startup and shutdown comprising a.

Images