FR2910689A1 - Noisy temporal neutron flow representing signal processing method for nuclear power plant, involves filtering Gaussian transformed signal by kalman filter, and performing inverse transformation for filtered signal - Google Patents

Abstract

The method involves capturing a Gaussian signal of a neutron flow by a sensor in impulse mode, and performing a transformation of a stabilization and gaussianisation of signal of a neutron flow by using an anscombe transform. The gaussian-transformed signal is filtered using a kalman filter. The inverse transformation of stabilization and gaussianisation of a filtered signal is performed, where the average and variance of the signal are greater than or equal to four. An independent claim is also included for a device for processing a signal representing a neutron flow comprising a sensor for capturing a signal representing a noisy temporal neutron flow.

Description

1 METHOD OF FILTERING A NEUTRONIC TIME NEUTRON FLOW SIGNAL

  TECHNICAL FIELD The present invention relates to the nuclear field and is applicable with respect to reactor neutronics in measuring the neutron fluence rate. The measurement of the neutron fluence rate is particularly important in nuclear power plants where it must be continuously monitored. The invention relates in particular to a method of filtering a noisy temporal neutron flux signal of fish distribution and to a device implementing said method.

  STATE OF THE ART Neutronics in nuclear reactors is concerned with the chain reactions taking place in the fuel (core of the reactors).

  These reactions are fissions which follow one another according to the scheme: neutrons give fissions, fissions give neutrons, neutrons give fissions, etc. The production of energy by induced fissions is the objective. It is the number of neutrons created in these reactions that conditions the kinetics of the reaction. The measurement of the neutron flux is therefore an indicator representative of the energy delivered in real time by a nuclear reactor. For a nuclear reactor, it is important to be able to accurately measure the energy supplied by the fissions in the nominal operating mode, but also during the change of the reactor regime: increase or decrease in speed, start-up (divergence of the core) , stop, etc. 2910689 2 For the safe conduct of nuclear reactors, it is necessary to measure the energy produced. This measurement is carried out thanks to the monitoring of the neutron flux. The neutron flux also called Neutronic Fluence Flow (DFN) 5 (expressed in number of neutrons per cm2 and per second) is continuously monitored in power plants, it measures the number of fissions taking place per unit of time. Thus, there is a dedicated neutron instrumentation (internal and external) whose function is to measure the neutron flux. The measurement of the neutron flux is carried out by means of a measuring chain, schematized in FIG. 1, consisting of: a neutron detector 1 (fission chamber or proportional counters), a coaxial cable 2, 15 a signal conditioning assembly 3 (amplification, filtering, digitization), a treatment unit 4, display means and alarms 5. Depending on the operating phases of the reactor, the neutron flux dynamics is very wide. . It typically varies from 102 to 1014 neutrons.cm "2.s 1, which corresponds in decibels to a variation of 12 decades, and thus represents a considerable measurement dynamic.In addition to the fact that the neutron flux signal is noisy, it appears That measuring this signal over such a range is not without posing real technical problems.To cover the operating range, each detector is associated with conditioning electronics according to the nature of the signal delivered: circuits for the pulses, circuits for current measurements, circuits for the measurement of fluctuations Figure 2 shows the ranges of variations of the different types of the delivered signal 2910689 3 The detector 1 counts the neutrons arriving on it. The pulsed operating mode follows a time-varying Poisson distribution of parameter X (t), which is called a count rate measurement. In this text we will refer only to the signal acquired in the pulse mode. It is known to those skilled in the art that the mean and the variance of a random variable according to a Poisson distribution of parameter X (t) are respectively equal to X (t). Thus the neutron flux signal sees its variance fluctuate as its average changes over time, which is a real problem in filtering this signal. The data filtering methods having a Poisson distribution ignore this property and on the one hand make the assumption of a Gaussian distribution and on the other hand consider the variance of this Gaussian constant over time. These assumptions made, many statistical signal processing tools (stationary linear filters) exist. To overcome the difficulty imposed by a non-constant variance, over the entire measurement dynamics it is common to use linear filters whose time constant varies as a function of the reactor power (these constants are previously calibrated ). Neutron flux filtering methods are available in the literature.

  Reference is made, for example, to the document M. KONRAD, Weighting functions for neutron flux measurement and its relative speed of change, Elektrotechnika, vol. 19, num. 2-3, 1976. In this paper, it is proposed to use weighting functions (rectangular or triangular) as a function of the counting rate (that is, as a function of X (t)) and the time width of the measurement window. Although these methods correctly follow the slow evolutions of the signal and make it possible to reduce the measurement noise, they do not make it possible to follow the fast discontinuities as closely as possible and therefore introduce errors in these during the filtering. These processing methods are based on the gaussianity assumption of flux, which is true only for sufficiently large fluxes (above 500 neutrons.cm "2.s'). On the other hand these methods are based on a hypothesis stationarity (at least local) that is to say of average and variance fixed in a certain time window These assumptions are not, in practice, always verified, especially when the neutron flux signal has strong discontinuities As a result, these methods are not optimal because the non-stationary fish nature of the neutron flux is not taken into account because, due to the difficult application environment, the signal may be subject to abrupt changes. which are non-linear: increase in power, decrease in power, emergency stop, etc. It is on this basis that the Applicant finds it useful to integrate into the neutron measurement chain a filt algorithm. optimal rage vis-à-vis the nature of the fish signal - in particular close to the discontinuities - and low complexity.

BASE OF THE INVENTION Applying the Haar-Fisz or Anscombe transforms to the field of neutronics for the purpose of filtering a noisy temporal neutron flux the Applicant has found an improvement in the measurement of the signal regardless of its variations. . Thus according to a first aspect, the invention proposes a method of processing a signal representative of a noisy temporal neutron flux of a nuclear power plant reactor, after acquisition by a sensor, said neutron flux signal being of average and variance varying during the acquisition characterized in that it comprises the following successive steps a) transformation (31) of stabilization and gaussianization of the neutron flux signal, b) filtering (32) of the transformed-Gaussian signal to stabilized variance by means of a Kalman filter, c) inverse transformation (33) of stabilization and gaussianization of the filtered signal obtained in step b) of said method so as to obtain a filtered neutron flux signal. The characteristics of the method are as follows - the stabilization and Gaussianization transform is an Anscombe or Haar-Fisz transform, - the neutron flux signal is acquired in the pulse mode.

The invention also relates to a device comprising means for implementing the steps of the method of processing the noisy temporal neutron flux signal. PRESENTATION OF THE FIGURES Other features and advantages of the invention will become apparent from the description which follows, which is purely illustrative and nonlimiting, and should be read with reference to the appended figures in which, in addition to FIGS. 1 and 2 already discussed: FIG. 3 represents the various steps of the method of the present invention, FIG. 4 illustrates the results of a comparison of the filtering method according to the invention.

DETAILED DESCRIPTION OF THE INVENTION The method of the invention provides a method of filtering a noisy temporal neutron flux. The invention relates to the signal acquired in the pulse mode.

As already mentioned, the measurement of the neutron fluence rate is subject to strong real-time measurement constraints.

It will be recalled that the problematic inherent in the filtering of the noisy temporal neutron flux comes from the fact that the signal has a nonstationary Poissonian statistic. Transforms applied to imaging signals contaminated with fish noise exist. In the context of the invention, the Applicant has investigated in order to apply these algorithms to its problematic: the signal to be filtered is a temporal signal and not an imaging signal. Anscombe and Haar-Fisz transforms are statistical signal processing. After transformation, the transformed signal can be considered as a noisy gaussian temporal signal which subsequently greatly simplifies the filtering steps: conventional and uncomplicated filtering techniques apply.

The document F.J. ANSCOMBE, The Transformation of Binomial and Negative Fish Binomial Data, Biometrika vol. 35, p. 246-254, 1948, proposes a transformed, so-called Anscombe Transform, making it possible to make Gaussian and to stabilize the variance of a fish distribution signal. One of the main characteristics of the Anscombe transform is that the variance 20 is stabilized as long as the parameter X (t) is greater than or equal to four. More recently, P. Fry lewicz and GP Nason, A Haar-Fisz algorithm for Fish Intensity Estimation, Technical Report, Statistics Group, Department of Mathematics, University of Bristol, UK, proposes to combine so-called Haar wavelets with the transform of Fisz (see document, The Limiting Distribution of a Function of Two Independent Random Variables and its Statistical Applications, Colloquium Mathematicum, vol 3, pp. 138-146, 1955) which makes it possible to gaussianize a Poisson signal. One of the main features of the Haar-Fisz transform is that the variance is stabilized as long as the parameter Δ (t) of the Poisson distribution is greater than or equal to one. The Haar-Fisz transform is a wavelet transform. As pointed out above, in the field of neutron flux measurement for the control of nuclear reactors, the use of mainly linear (gaussian) signal processing methods, which do not take into account the corpuscular nature of the phenomena, is mainly used. observed. The Kalman filter alone has been widely used and tested, but without taking into account the statistics of phenomena.

The Kalman filter is described, for example, in AH JAZWINSKI, Filtering for stochastic processes, Interscience Publishers, 1968. However, the direct application of a Kalman filter to a standard neutron flux signal is not optimal because the mean and the variance of the signals fluctuate.

In addition, since the neutron flux signal is not Gaussian, the direct application of a Kalman filter can give biased results or even diverge, which is not acceptable for the control of a nuclear reactor. Figure 3 shows the different steps of the method of the invention. The method aims to filter the noisy temporal neutron flux noted v (t). The signal v (t) has a Poisson distribution of parameter X (t). This signal undergoes a stabilization transform of variance and Gaussianization denoted T which may be, or an Anscombe transform, or a Haar-Fisz transform. The resulting signal is denoted v * (t) and is Gaussian with stabilized variance (constant). The signal v * (t) is subsequently filtered by means of a Kalman filter. Since the signal v * (t) is Gaussian, the results obtained by means of the Kalman filter are optimal since it is known that this filter is optimal in the case of a nonstationary Gaussian signal. The signal filtered by means of the Kalman filter is denoted x (t). The filtered signal z (t) is then subjected to the inverse stabilization and gaussianization transform 33 denoted T-. The desired filtered neutron flux, denoted 1 (t), is then obtained. . As previously discussed, the method according to the present invention has two modes of operation. The first mode is that implementing the Anscombe transform and the second is the one implementing the Haar-Fisz transform.

Below, we will describe the implementation of the transforms according to the two possible embodiments. Anscombe Transform Let a signal noted v whose samples denoted v (t) are realizations of the Poisson's law P (λ (t)), the Anscombe transform of the signal v (t) is given by T. [v (t)} = v * (t) = aJv (t) +3/8, where a is a constant to adjust the variance. If a = l, the variance of v (t) is equal to 1/4, to obtain a variance equal to 1 i 1 suffices to fix a = 2. The inverse Anscombe transform of the signal v * (t) allowing to find the signal v (t) is such that T - '[v * (t)] = v (t) = (v * (t) / a) 2 -3 / 8. As already mentioned, the Anscombe transform stabilizes the variance of the signal v (t) at a chosen value (via a) when X (t) 4. Haar-Fisz Transform The Haar-Fisz transform is a somewhat discrete wavelet decomposition.

The Haar-Fisz transform combines the Haar wavelets with the Fisz transform and is constructed in the following manner. Let the signal v (t) admitting the discrete writing v = (vo, v ,, ..., vN_t) of size N = 2J, where J is the number of scales. The signal v (t) is decomposed in two dimensions (in scale (index j), and in time (index i)). Let s = (so, s;, ..., sZ_,) and s' The Haar-Fisz transform is obtained in the following successive steps. DEBUT s "= v 30 For j = J1, J2, ..., 0 {STEP 1 2910689 For i = 0.1, ..., 21- '{si = s2i + S2i + ii = S2i + s21 + lt 2 / s} 5 END STEP 1} For j = 0,1, ..., Jù1 {STEP 2 if -1 = s; + s st, ', j J = if ù END STEP 2} 10 V = SJ END The vector v contains the stabilized signal in variance and gaussianized The inverse Haar-Fisz transform is obtained by applying step 2 to v then the modified step 1 in the following way: s 2 '= s ; + d; Js', j-1 = j û ~ J j S21 + 1 if Ji Si. As already mentioned, the Haar-Fisz transform stabilizes the variance of the signal v (t) at a chosen value (via (a) since (t)> 1. In the case of signals sampled at the sampling period Te the Anscombe and Haar-Fisz transforms are of course valid as long as the product A (k) Te is greater than or equal to 4Te for the Anscombe transform and Te for the Haar-Fisz transform, regardless of Te, after Gaussianization and stab Variance is applied by applying Kalman filtering. The process terminates by applying to the filtered signal the Anscombe or Haar-Fisz inverse transform according to the chosen embodiment. According to the method of the invention, a neutron flux signal is acquired for a certain period of time.

FIG. 4 illustrates a result obtained by applying the method of the invention applied to a real noisy time neutron flux signal. In practice, it should be noted that the mean and variance of the signal 40 are not exactly identical but vary over the course of the acquisition time. The signal 40 (see FIG. 4) acquired via the neutron measurement chain exhibits a strong discontinuity at around 200 seconds and passes abruptly by approximately λ. 2x102 to A. z-2x104. It should be noted that the fact of previously applying an Anscombe or Haar-Fisz transform makes it possible to advantageously follow the variations of the signal even when it has a very large discontinuity while optimally filtering it. Thus the signal is filtered even when it has strong discontinuities.

The signal 41 is obtained according to an earlier conventional linear filtering method. There is a significant shift between this filtered signal 41 and the raw signal 40, especially after the discontinuity. The signal 42 is obtained according to the method of the invention using an Anscombe transform and a Kalman filter.

Furthermore, in the example of FIG. 4, the Anscombe and Haar-Fisz transforms (not shown in the example of FIG. 4) substantially lead to the same result since the minimum value of X (t) is greater than 4, and that in this range both transforms apply. The method of the invention can integrate with the measuring device as previously described and as known, this has the advantage of not having to modify the measurement chain. The invention also relates to a device implementing the steps of the filtering method described. Consequently, the method and the device as described make it possible to filter a time signal of noisy neutron flux of fish distribution. One of the advantages of the method and of the associated device is that they make it possible to follow the variations of the signal itself. it has very strong discontinuities, such as in the case of a Poisson distribution signal. In 2910689 it effect as already mentioned the averages and variances of such a signal vary in time. 12

Claims (7)

  A method of processing a signal representative of a noisy temporal neutron flux of a nuclear power plant reactor, after acquisition by a sensor, said neutron flux signal being of varying average and variance during the acquisition characterized by it comprises the following successive steps a) transformation (31) of stabilization and Gaussianization of the neutron flux signal, b) filtering (32) of the transformed Gaussian signal with stabilized variance by means of a Kalman filter, c) transformation inverse (33) stabilization and Gaussianization of the filtered signal obtained in step b) of said method so as to obtain a filtered neutron flux signal.   2. Method according to claim 1, characterized in that the stabilization and Gaussianization transform (31) is an Anscombe transform.   3. Method according to claim 2, characterized in that the average and the variance of the neutron flux signal are greater than or equal to four.   4. Process according to claim 1, characterized in that the stabilization and Gaussianization transform (31) is a Haar-Fisz transform.   5. Method according to the preceding claim, characterized in that the average and the variance of the neutron flux signal are greater than or equal to one.   6. Method according to any one of the preceding claims, characterized in that the neutron flux signal is acquired in the pulse mode. 13 2910689   7. Apparatus for processing a signal representative of a neutron flux comprising a sensor for acquiring the neutron flux, said signal being noisy after acquisition, said neutron flux signal being of varying average and variance during the acquisition, characterized in that said device 5 comprises means able to implement the steps of the method according to one of the preceding claims.