JPS5810717B2 - Nuclear reactor core abnormality diagnosis method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to a method for diagnosing abnormalities within the core of an operating nuclear reactor.

An object of the present invention is to provide a nuclear reactor core abnormality diagnosis method that can accurately determine the cause of an abnormality.

A feature of the present invention is that a time-series signal consisting of a plurality of m measured values indicating the core state is converted into a power spectral density based on a multivariate autoregressive model, and m elements of the power spectral density are sequentially referenced. m compared to the spectral density
Determine whether or not elements exist within a predetermined dispersion range, and if there is an element that is outside the predetermined dispersion range, find the frequency region of that element that is outside the predetermined dispersion range, and The objective is to determine the presence or absence of other elements that fall outside a predetermined dispersion range in the frequency domain.

A comprehensive explanation will be given to achieve the above objectives.

Actual data measured from an operating nuclear reactor is considered to be the result of a sample process of irregular dynamic physical phenomena within the reactor.

If the measured data record follows a stationary normal stochastic process, then the stochastic process can be completely described by means and correlation functions (or spectral densities).

In this case, any stochastic process (time series) can be described as the output of a constant coefficient linear filter that receives normal white noise as input.

Therefore, by performing time series analysis and determining the characteristics of the linear filter, it becomes possible to estimate the dynamic characteristics of the nuclear reactor.

An m-dimensional (corresponding to m measured parameters describing the reactor's dynamic characteristics) stochastic process (X( The mathematical model for t)) is generally expressed as follows.

Here, ξ(1) is a normal white noise sequence, meaning an m-dimensional vector, and each moment has the following properties.

Here, if we use the delay operator z-1, from equation (1) we get A multivariate rational model is obtained.

Here, as p=q Then, equation (2) also becomes It can be written as a transfer function.

Now, the m-dimensional transfer function A(Z)-1 shown by equation (3)
Assuming that B(Z) and the variance Σ2 of the noise input are obtained, the spectral density of X(t) is given by.

Now, if m white noise sequences ξ(1) are mutually independent, Σ2 becomes a diagonal matrix, and the power spectral degree of an arbitrary parameter (i) is expressed as.

Here, σii2 is a diagonal term of Σ2, and is a relationship expressed by a linear combination of .

According to equation (5), the power spectrum of any measured parameter of a nuclear reactor is represented by a linear combination of the variances of the white noise generating the time series of the respective parameter.

Viewed from another perspective, according to equation (5), the PSD of any parameter of the reactor will be separated into proportions that depend on each of the m parameters.

Therefore, when the PSD of an arbitrary parameter is determined and the pattern of a frequency band of a certain characteristic differs from a normal pattern, it becomes possible to identify which parameter this difference in pattern depends on.

The present invention will be explained in detail below using examples.

FIG. 1 is an example of a nuclear reactor abnormality diagnosis device according to the present invention.

Measured quantities of the nuclear reactor core are input to the present abnormality diagnosis apparatus using the signal acquisition device 1.

Here, m measurement parameters are selected, and at the same time, specific measurement noise is filtered out, and the analog measurement signal is converted into a time series.

The stationarity test device 2 calculates a moving average of the time series signal 9 at appropriate time intervals, tests the stationarity of the first-order moment, and further averages it with a preset number of samples N. Find the time series by removing this average value from.

This time-series signal 10 is led to a statistical processing device 3, where a covariance matrix is calculated.

The obtained covariance matrix signal 11 is returned to the stationarity test device 2, and the stationarity of the second-order moment is tested.

In the stationary case, the tested covariance matrix signal 12 is input to the rational model estimator 4.

Taking as an example a case in which the time series of measurement signals from a nuclear reactor is expressed by an autoregressive model, Equation (1) is expressed as follows.

At this time, the parameter A(i) is the generalized Yule
- It is known that it can be obtained by solving the Walker equation.

C(j-k) is a covariance matrix with a delay number of j-, and is input as signal 12.

FIG. 2 shows an example of the configuration of the rational model estimation device 4 in the case of an autoregressive model, which includes a calculation device 18 for solving the canonical equation expressed by equation (7), a white noise time series variance calculation device 19, and a white noise time series variance calculation device 19. It is composed of an order determining device 20.

The canonical equation calculation device 18 uses an iterative solution method (P, Whittle) for degree p.

Biometrica, 50, 1963] is available.

Further, based on the parameters Ap1, .

Here, C'(i) is the transposed matrix of the dispersion matrix C(i).

To determine the parameters of the autoregressive model, for example, Final Prediction Error (
FPE) [H.

Akaike, Ann, of the In5t, o
f5tatist.

Math, vol, 22. No. 2.1970] may be minimized.

Here, 1Σ21 is the determinant of the white noise matrix given from the calculation device 19, and the order determination device 20 uses equation (9)
FPE(p) is calculated for successively increasing p, and the value of p that minimizes FPE(p) is automatically selected.

Parameters A1 .

The spectral density in the case of the autoregressive model is determined from the following equation according to equation (4).

Assuming that m white noise time series are independent, the power spectral density of any measurement time series i is given by:

Here, (A(e-J2πf)-1)iK is the matrix A(eJ
2πf)-1, and σ2KK indicates an element of the matrix Σ2.

Next, the spectrum calculation device 5 calculates each power spectrum according to the formula 0υ.

The value on the left side and each term on the right side of equation (11) are input to the abnormality determination device 6 as the reactor spectrum signal 14.

FIG. 3 shows an example of the configuration of the abnormality determination device 6.

The spectrum signal is input to the abnormality determination control device 21.

The control device 21 periodically or at an appropriate time,
The reference spectral density is stored in the reference spectrum storage device 2.
Output to 3.

Furthermore, the density of the spectrum is outputted to the diagnostic spectrum storage device 22 periodically or at a time required for diagnosis.

The spectrum comparison device 24 outputs a synchronization signal 15 for comparing the spectra stored in the area storage devices 22, 23,
First, it has a function of sequentially comparing each (m) power spectrum densities defined in (11).

When these m power spectral densities are within a preset dispersion range, the time series of the core at the time of diagnosis coincides with the standard steady state, and is determined to be normal.

Among m psds, a certain i-th (measurement variable 1) ps
If d becomes biased by more than the preset variance value,
First, an alarm signal 16 is issued to indicate on the alarm display device 7 that there is an abnormality in the psd of the measured variable i.

At the same time, a spectrum pattern output signal 17 is output to a display device 8 such as a CRT in order to indicate in which frequency band of the i-th PSD there is a bias.

In addition, a synchronization signal is sent to the surface storage device again, and the i-th psd is
In order to compare in more detail, m spectra on the right side of 2(11) are sequentially retrieved from the area storage device and compared.

As a result, for example, if it is determined that patterns of l measurement variables out of m are abnormal in the same frequency band, the results are displayed together with both spectrum patterns.

Next, in the i-th psd, 1(A(e-
A periodic signal is output for comparing J2πf)-t)1112σ112.

It is determined whether this spectrum is biased in the target frequency band.

If it is determined that there is a bias, the abnormality in a certain frequency band of measurement channel i is caused by the relationship with measurement channel l, and it is also an abnormal phenomenon of measurement channel l itself (which has no correlation with other channels). It is shown that it is based on

Furthermore, an abnormality in measurement channel i that is determined to be unbiased can be determined to be an abnormality occurring in the information transmission path from measurement channel l to measurement channel 1.

According to the present invention, since the determination is made based on the frequency range outside the predetermined dispersion range, the cause of the abnormality in the reactor core can be clearly determined regardless of whether there is one cause of abnormality or multiple related causes. can.

Furthermore, as described above, since the frequency domain is taken into consideration for determination, it is possible to distinguish between causes of abnormality caused by the reactor itself and causes of abnormality caused by the instrumentation system itself, such as the measurement channel.

This is because the frequency range in which an abnormality occurs is different between the abnormality cause due to the nuclear reactor itself and the abnormality cause due to the instrumentation system itself.

[Brief explanation of drawings]

FIG. 1 is a block diagram showing the configuration of an embodiment of the nuclear reactor abnormality diagnosis device according to the present invention, FIG. 2 is a block diagram showing the configuration of an autoregressive rational model estimation device, and FIG. FIG. 2 is a block diagram showing the configuration.

Claims (1)

[Claims] 1 A time-series signal consisting of a plurality of m measurement values indicating the core state measured simultaneously from an operating nuclear reactor is converted into a power spectral density based on a multivariable autoregressive model, and m of the power spectral density is It is determined whether or not the m elements are within a predetermined dispersion range by sequentially comparing the m elements with a reference spectral density, and if there is an element that is outside the predetermined dispersion range, A method for diagnosing abnormality in a nuclear reactor core, the method comprising: determining a frequency region outside the predetermined dispersion range of the element, and determining whether or not there is another element outside the predetermined dispersion range in this frequency region.