JP2001174569A - Seabed sedimentary layer parameter estimation device using genetic algorithm - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a seabed sedimentary layer parameter estimation device stably estimating the parameter values expressing the characteristics of the seabed sedimentary layer from measurement data. SOLUTION: Observation characteristic value calculation parts 204, 306 find dispersion characteristics C0 of group speeds of elastic waves propagating the seabed sedimentary layer from the observation data of the seabed sedimentary layer, or an estimation object. An estimation characteristic value operation part 301 theoretically estimate the dispersion characteristic C of the group speeds of the elastic waves based on combination of candidates values of a plurality of parameters expressing the physical characteristics of the seabed sedimentary layer. An initial group generation part 305 and a gene operation part 303 generate and evolute the combination of the candidate values of the parameters as individual by genetic algorithm. An adaptability operation part 302 finds the adaptability expressing the extent of the accordance between the dispersion characteristics C0 and dispersion characteristics C and set a combination of the candidate values comprising the most adaptable individual to the solution.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus for estimating parameters indicating physical characteristics of a seabed sedimentary layer, and is particularly used for evaluating sound propagation characteristics in a shallow sea sedimentary layer and searching for mineral resources. The present invention relates to an apparatus for estimating parameters of a marine sediment layer.

[0002]

2. Description of the Related Art Inversion methods, multiple regression analysis methods, maximum likelihood estimation methods, and the like have been known as methods for estimating the values of parameters indicating the physical characteristics of sedimentary layers on the sea floor. About these, for example, RDStoll and 2 others: Es
timation of Shear Wave Velocity in Shallow Marine
Sediments: IEEE J.Oce.Eng., Vol.19, No.1, Jan., (199
4), Kimura, et al .: Velocity dispersion characteristics of Rayleigh waves on the ground surface, Proceedings of the Acoustical Society of Japan, Mar., 1163-1164
(1996) and A. Dziewonski and 2 others: A Technique for An
alysis of Transient Seismic Signal, Bulletin of S.
SA, Feb., 427-444 (1969). Each of these estimation methods estimates a physical characteristic value of the seabed sedimentary layer by performing a numerical analysis based on a measurement result actually measured for the seabed sedimentary layer.

[0003]

The measurement data obtained by actually measuring the seabed sedimentary layer is insufficient in quantity and contains much noise. For this reason, when estimating the characteristic parameter values of the marine sedimentary layer based on these measurement data by a conventional numerical analysis method, the solution sometimes falls to a local minimum value or diverges. In addition, in the conventional numerical analysis method, in order to use a higher mode as well as the fundamental mode of the dispersion characteristic in order to make the analysis model more precise and realistic, the number of variables increases due to an increase in the number of variables. There are also problems that the stability of the analysis is lacking and the amount of calculation is significantly increased. Further, in some cases, it is difficult to grasp the statistical properties of the errors included in the actual measured data of the marine sediment. In such a case, it is difficult for the above-described conventional numerical analysis method to properly construct a mathematical model for expressing a problem and finding a solution.

An object of the present invention is to provide a seabed sediment layer parameter estimating apparatus capable of stably estimating the value of a parameter indicating the characteristics of a seabed sedimentary layer from measured data.

[0005]

According to the present invention, there is provided the following apparatus for estimating parameters of a seabed sedimentary layer.

That is, an observation characteristic value calculation unit for obtaining characteristic values of an elastic wave propagating through the seabed sedimentary layer from data observed for the seabed sedimentary layer to be estimated, and a plurality of data indicating physical characteristics of the seabed sedimentary layer. A candidate value of the parameter as a gene, an individual obtained by combining the candidate values as an individual, an initial population generator for generating the individual by a predetermined number of individuals, and performing a genetic operation on the individual by a genetic algorithm. A genetic operation unit for applying the evolved to the next generation individual, an estimated characteristic value calculation unit for theoretically estimating the characteristic value of the elastic wave for the individual, and the estimated characteristic value calculation unit estimated for the individual. The fitness value indicating the degree of coincidence between the characteristic value and the characteristic value determined by the observation characteristic value calculation unit is determined, and the individual having the highest fitness among the individuals is configured. The candidate value has a fitness calculation unit that is a solution of the value of the parameter, the characteristic value is a dispersion characteristic at least about the frequency of the group velocity of the elastic wave, the parameter is the An apparatus for estimating a parameter of a seabed sediment, including a layer thickness, a density, a longitudinal wave velocity, a shear wave velocity, and a shear wave attenuation factor for each of one or more layers constituting the seabed sedimentary layer.

[0007]

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, an apparatus for estimating parameters of a seabed sedimentary layer according to an embodiment of the present invention will be described with reference to the drawings.

The marine sediment layer parameter estimating apparatus according to the present embodiment obtains a dispersion curve from actually observed data, and theoretically calculates a combination of parameter candidate values (parameter set: individual) generated by a genetic algorithm. Calculate the dispersion characteristics. Then, by comparing these two dispersion characteristics, an optimal parameter set is obtained, and this is set as an optimal solution or a quasi-optimal solution that can be practically satisfied. The configuration and operation of the marine sediment layer parameter estimation device of the present embodiment will be described with reference to FIGS.

The marine sediment layer parameter estimating apparatus according to the present embodiment includes a vibrator 104, a geophone 103, and an analysis processor 102. The analysis processor 102 includes the observation ship 101
It is installed in. The exciter 104 and the receiver 103 are
As shown in FIG.
05 are installed at mutually distant positions. Geophone 103
Includes two geophones 103a and 103b,
The receiver 103b is installed at a position farther from the exciter 104 than the receiver 03a. The exciter 104 has a sea bottom 105
Has a transmitter that sends out a signal by giving a vibration to. The receiver 103 has a receiver that receives an elastic wave that has propagated through the seabed sedimentary layer 106 among the signals sent from the exciter 104.

As shown in FIG. 2, the analysis processor 102 includes a distance calculator 201, a waveform accumulator 202, and a frequency analyzer 2.
03, an observation dispersion characteristic calculation unit 204, a deposited layer parameter estimation unit 205, a display unit 206, and a set value reception unit 207 that receives settings from a user. In FIG. 2, each unit of the analysis processor 102 is shown as an independent device, but the entire analysis processor 102 is
Arithmetic devices, storage devices, reception units such as keyboards and mice,
It is configured by a computer having a display device and the like. In the program storage area of the storage device, a distance calculation program, a waveform addition program, a frequency analysis program, an observation dispersion characteristic calculation program, and a sedimentary layer parameter estimation program are stored in advance, and the calculation device stores these programs respectively. By reading and executing it, the distance calculation unit 201 and the waveform accumulation unit 20
2, frequency analysis unit 203, observation dispersion characteristic calculation unit 204,
And the operation of the deposition layer parameter estimation unit 205 is realized. Data obtained by executing these programs is stored in a data storage area of a storage device. The setting value receiving unit 207 is configured by a receiving unit such as a mouse or a keyboard of the computer, and the display unit 206 is configured by a display device such as a CRT of the computer.

The set value receiving unit 207 includes a distance calculating unit 201
The setting of the water depth, water temperature, salinity, cumulative number, frequency analysis resolution, dispersion characteristic calculation range, number of layers, generation number, group size, etc. necessary for processing in each part of the deposition layer parameter estimation unit 205 is received from the operator. is there.

The display unit 206 is a device for displaying the set values of the respective processing units input by the set value receiving unit 207 and the calculation results of the respective processing units.

The distance calculation unit 201 receives the transmission data of the oscillator 104 and the reception data of the oscillators 103a and 103b, and uses these data to calculate the waveform transmitted by the transmitter of the oscillator 104. The operation of measuring the time until is received by the receivers of the receivers 103a and 103b is performed.
Further, the separately measured water temperature, salinity, and water depth are received from the set value receiving unit 207, and the sound speed is obtained by substituting them into a predetermined mathematical expression. From the sensor to the geophone 103a,
Further, the distance between the receiver 103a and the receiver 103b is calculated.

The waveform accumulating section 202 performs an accumulating process for accumulating, in the amplitude direction, a received waveform received by the oscillator 103a on the vibration repeatedly applied to the sea bottom by the oscillator 104, and taking an average. The number of waveforms to be added is the number of waveforms set by the user in the set value receiving unit 207. Similarly, the geophone 103
The accumulation processing is also performed on the waveform received by b. Thereby, the influence of noise included in the received waveform can be reduced.

The frequency analyzer 203 includes a waveform accumulator 202
A process is performed for analyzing the level of the received waveform accumulated in the above for each frequency. The frequency resolution, such as the number of hertz in the frequency to be analyzed, is specified by the resolution set by the user in the set value receiving unit 207.

The observation dispersion characteristic calculation section 204 propagates the seabed sedimentary layer 106 based on the distance from the exciter 104 to the receiver 103 obtained by the distance calculation section 201 and the frequency analysis result obtained by the frequency analysis section 203. The dispersion characteristic of the group velocity, which indicates the relationship between the group velocity and the frequency of the elastic wave, is calculated. When the dispersion characteristics of the group velocity thus obtained are shown on the graph of group velocity-frequency, a closed curve 70 concentrically overlapping as shown in FIG.
It will be 1 mag. As a method of calculating the dispersion characteristic indicating the relationship between the group velocity and the frequency, various generally well-known methods can be used. Here, the multiple filter method often used in geophysics is used here. Is used. The multiplex filter method is a method of obtaining a relationship between a frequency and a group velocity, that is, a dispersion characteristic, by cutting out a measured signal for each minute band which is a frequency domain and returning it to a time domain. See "A Techni" for more information on the multiple filter method.
que for the Analysis of Transient Seismic Signals "
byA. Dziewonski, S. Bloch and M. Landisman Bullet
in of the Seismological Society of America.Vol.5
9, No. 1 p. 427-444. February, 1969.

The observation dispersion characteristic calculation unit 204 also calculates the dispersion characteristic indicating the relationship between the phase velocity and the frequency of the elastic wave when the received data is obtained by both the geophones 103a and 103b. calculate. The dispersion characteristic of the phase velocity is obtained from the distance between the geophone 103a and the geophone 103b and the frequency analysis result obtained by the frequency analysis unit 203. Therefore, it cannot be obtained when only one of the geophone 103a and the geophone 103b can receive a wave. As a method of calculating the dispersion characteristic of the phase velocity, a generally well-known method can be used.
Use the filter method. For details of the time-variant filter method, see the document “Pilant, W. and L. Knopoff (1964). Ob
servations of multiple seismic events, Bull.Seis
m. Soc. Am. 54, 19-39 ".

Next, the deposition layer parameter estimation unit 205 will be described in detail. Deposition layer parameter estimation unit 205
Calculates the observation dispersion curve from the observation dispersion characteristics calculated by the observation dispersion characteristics calculation unit 204, and uses a genetic algorithm technique to combine candidate values of various parameters indicating the physical characteristics of the marine sedimentary layer (parameter set). : Individual), and a dispersion curve (estimated dispersion curve) is calculated from this parameter set. Then, by comparing these two dispersion characteristics, an optimal parameter set is obtained, and this is set as a solution. When generating a parameter set, the generation set is changed by performing genetic operations such as selection 401, crossover 402, mutation 403, etc. on the parameter set of the initial generation, so that the fitness of the observed dispersion curve is Use a technique to generate high individuals. In addition,
General genetic algorithms are described in detail in the following documents. Holland, John (1975) .Adapta
tionin Natural and Artificial Systems, MIT press 1
992 ", Goldberg, David (1987). Genetic Algorithm in
Search, Optimization, and Machine Learning.Readin
g, Mass .: Addison-Wesley, Davis, Lawrence (1990).
Handbook of Genetic Algorithms, Van Nostrand (Japanese translation: Genetic Algorithm Handbook, Kazuzu et al., Morikita Publishing).

The structure of the deposition layer parameter estimating unit 205 will be specifically described. As shown in FIG. 3, the sedimentary layer parameter estimation unit 205 includes an initial
4. Gene manipulation unit 303, estimated dispersion curve calculation unit 301,
A fitness calculator 302 and an observation dispersion curve calculator 306 are provided.

The observation dispersion curve calculation unit 306 obtains a dispersion curve C ₀ from the observation dispersion characteristic of the group velocity obtained by the observation dispersion characteristic calculation unit 204. As a method of obtaining the dispersion curve C ₀ , first, the dispersion characteristic is plotted on a group velocity-frequency graph as shown in FIG. 5, thereby obtaining a contour-shaped closed curve 701 indicating the observed dispersion characteristic, and the like. select the coordinate indicating the peaks of equal, obtaining the dispersion curve C ₀ in the form of the set of coordinates. The obtained dispersion curve C ₀ is stored in the data storage area. The observation dispersion curve calculation unit 306 generates an image showing the dispersion curve C ₀ of the obtained group velocity together with the dispersion characteristics on a graph, outputs the image to the display 206, and displays it as shown in FIG. When the dispersion characteristic of the phase velocity has been obtained by the observation dispersion characteristic calculation unit 204, the observation dispersion curve calculation unit 306 generates a dispersion curve D which is a curve indicating the peak.
Also ask for ₀ . The dispersion characteristic and dispersion curve D _{0 of the} obtained phase velocity are also displayed on the display unit 206.

Further, the observation dispersion curve calculation unit 306 includes a correction unit 316 that receives correction of the dispersion curves C ₀ and D ₀ . The correction unit 316 adjusts the coordinates on the dispersion curve C ₀ (or dispersion curve D ₀ ) manually specified by the operator using the mouse or the like of the set value reception unit 207 by the operator viewing the graph displayed on the display unit 206. Performs the operation of replacing with another coordinate manually specified. Thereby, the coordinates (FIG. 5) of the curve 702 due to the noise included in the dispersion curve C ₀ can be removed by the judgment of the operator as shown in FIG. The coordinates of the corrected dispersion curve C ₀ (or dispersion curve D ₀ ) are stored in the data storage area of the storage device.

In the present embodiment, the observation dispersion curve calculation unit 306 obtains the dispersion curve C ₀ of group velocity described above for at least the Rayleigh wave which is the 0th mode of the elastic wave. Since dispersion characteristics of higher-order modes can be obtained depending on observation conditions, a dispersion curve of group velocities for higher-order modes is also obtained and stored in the data storage area. If the dispersion characteristic of the phase velocity of the higher-order mode is obtained according to the observation conditions, the dispersion curve of the phase velocity of the higher-order mode is also obtained and stored in the data storage area.

On the other hand, the initial group generation unit 305 generates individuals (parameter sets) of the initial generation and generates
04. The number of individuals to be generated is the number of individuals set by the operator in the set value receiving unit 207.
The group pool unit 304 stores the individual in a predetermined storage area of the storage device.

First, the configuration of the individuals (parameter sets) generated by the initial group generation unit 305 will be described.
In the present embodiment, the seafloor sedimentary layer is modeled as an n-layer multilayer structure as shown in FIG. The parameters of each layer are
Thickness h _i, the density of the layer [rho _i, longitudinal wave velocity alpha _i, shear wave velocity beta _i, is set to five transverse wave attenuation rate gamma _i. Note that i (i = 0,
1, 2,... N) indicate layer numbers, and the layer furthest from the sea bottom 105 is the 0th layer, and the layer having the sea bottom 105 as the upper surface is the nth layer. When the x- and y-axes are set in the horizontal plane of each layer and the z-axis is set in the vertical direction in the model of the multi-layered sedimentary layer 106 in FIG. 7, the value of each parameter (gene) is a function of only z, and Constant, that is, a constant value in the same layer. In the present embodiment, all five types of parameters for each of the 0th to nth layers are genes, and a set of candidate values for these parameters is the individual described above. Therefore, one individual, as shown in FIG.
(5 types) × (n layers) = 5 · n parameters (genes) are indicated by combinations of values.

The model shown in FIG. 7 is effective for obtaining the dispersion characteristics of an elastic wave propagating in a seabed sedimentary layer. Saito: The theoretical explanation of this model is Saito: "Calculation of reflectance and surface wave dispersion curves for stratified structures" Geophysical Exploration Vol. 46 No. 4 (1
993) pages 283-298.

Next, how the initial population generation unit 305 generates individuals of the initial generation will be described. In this embodiment, as shown in FIG. 9, each gene (layer thickness h _i , layer density ρ) other than h ₀ of each individual (parameter set)
Regarding _i , longitudinal wave velocity α _i , shear wave velocity β _i , and shear wave attenuation rate γ _i (where i = 1 to n), numerical ranges (upper and lower limits) are determined in advance from a practical viewpoint. Also, the unit amount within the above range is determined. The initial group generation unit 305 sets the value of the predetermined unit amount in the predetermined range as a candidate value of the numerical value of the parameter, and randomly selects one value from the candidate values, thereby Are determined one by one. h ₀
Only the fixed value of ∞. As the value of n in the nth layer, the value received by the set value receiving unit 207 is used. Note that the unit amount is an amount corresponding to one bit at the time of calculation by the computer.

When the range and unit amount of each parameter are determined as shown in FIG. 9, the total number of combinations of five parameter candidate values for one layer is 512 × 512 × 512.
× 512 × 512 ≒ 3.5 × 10 ¹³ ways. Therefore, the combination of the candidate values that constitute one individual is 3.5 × 10 ¹³ times the number of n layers,
That is, there are n × 3.5 × 10 ¹³ patterns. In the above-described initial group generation unit 305, the number of individuals (group size) set by the set value reception unit 207 from combinations of these candidate values.
The individuals of the initial generation are generated by randomly selecting the individuals. The generated individuals are in the pool 30
4 and stored in the storage area of the storage device.

The estimated dispersion curve calculation unit 301 calculates the dispersion characteristics of each individual (parameter set) stored in the group pool unit 304 when the parameter value of each layer of the marine sedimentary layer 106 is that value. Determined by theoretical calculation. In the present embodiment, when the dispersion curve C ₀ of the group velocity is obtained by the observation dispersion curve calculation unit 306 at least for the Rayleigh wave that is the 0th mode among the elastic waves transmitted through the seabed sedimentary layer 106, Also finds the dispersion curve D ₀ of the phase velocity, so for comparison,
The estimated dispersion curve calculation unit 301 theoretically calculates the dispersion curve C of the group velocity and the dispersion curve D of the phase velocity for the Rayleigh wave from the parameter set in the group pooling unit 304. In this embodiment, the calculation method is SAFA.
RI (Seismo-Acoustic Fast field Algorithm for Rang
e Independent environments) model was used. This model is widely known as a mathematical model of seismic sound propagation in a layered medium, and is currently published on the Internet by OASES (Ocean Acoustic and Seismic Explorat).
This is the model on which ion synthesis is based. SAFARI
Is composed of a plurality of programs. In the present embodiment, the SAFARI-FIPP module is used. In this SAFARI-FIPP module, five layer structure parameters (layer thickness, density, longitudinal wave velocity, shear wave velocity, shear wave attenuation rate), longitudinal wave attenuation rate,
By inputting the values of the wave number integration range (the number of integration points, etc.) and the dispersion curve calculation range (the upper and lower limits of the frequency, etc.),
A dispersion curve C of the group velocity and a dispersion curve D of the phase velocity can be calculated. Here, the numerical values of the parameters of the individual whose dispersion curve is to be calculated are input as the five layer structure parameters. Further, a fixed value (here, 0) input as a longitudinal wave attenuation factor is set in advance. As the wave number integration range and the dispersion curve calculation range, predetermined ranges were input. The dispersion curves C and D obtained by this calculation are passed to the fitness calculator 302.

Next, the operation of the fitness calculating section 302 will be described in detail. The fitness calculation unit 302 determines whether or not the dispersion curve C of an individual calculated by the estimated dispersion curve calculation unit 301 matches the dispersion curve C ₀ obtained from the received data by the observed dispersion curve calculation unit 306. The fitness Φ ₀ , which is an index indicating, is calculated. Specifically, the fitness calculating unit 302 calculates the group velocity C ₀ (f _i ) of the dispersion curve C _{0 and} the group velocity of the dispersion curve C for each predetermined frequency f _i as in the following equation (1). The reciprocal of the average of the square of the difference from C (f _i ) is defined as fitness Φ ₀ . The fitness Φ given by equation (1) increases as the degree of coincidence between the dispersion curve C ₀ and the dispersion curve C increases.

[0030]

(Equation 1)

When the dispersion curve D ₀ of the phase velocity is obtained by the observation dispersion curve calculation section 306, the fitness calculation section 302 also determines the degree of coincidence between the dispersion curve D ₀ and the dispersion curve D by the fitness calculation. In order to reflect this, the fitness Φ is calculated by the following equation (2).
Seek _1. Then, the fitness Φ of the individual is an arithmetic mean of the fitness Φ ₀ and the fitness Φ _{1 as in} the following equation (3). Also, when the higher-order mode dispersion curve is obtained, the fitness is similarly obtained, and the fitness Φ ₀ or Φ _{1 of the} zero-order mode is obtained.
The arithmetic mean of the above is taken as fitness Φ.

[0032]

(Equation 2)

[0033]

(Equation 3)

In _equation (1), the frequency f _i used for calculating the fitness Φ ₀ is determined in a geometric series. In particular,
Equation 1 is a case where the frequency f _i is set to 20 in a geometric progression from 1 Hz to 100 Hz, assuming that m = 20 is set in the set value receiving unit 207 by the operator as the number m of samples of the frequency f _i. Is shown. However, how to determine the frequency f _i is not limited to this. For example, when the rate of change of the dispersion curves C ₀ and C is significantly different depending on the frequency, the frequency sampling interval (f
It is also possible to improve the analysis performance by reducing the distance d) between _{i + 1} and f _i and increasing the frequency sampling interval in the portion where the change in the dispersion characteristic is relatively small.

The fitness Φ ₀ or fitness Φ calculated by the fitness calculation unit 3 is stored in the data storage area in correspondence with the individual (parameter set) stored in the group pool unit 304.

Next, the operation of the gene manipulation unit 303 will be described in detail. The gene manipulation unit 303 includes a selection unit 401, a crossover unit 402, and a mutation unit 403, and
A group of individuals (parameter sets) stored in the group 04 is subjected to three kinds of genetic operations of selection, crossover, and mutation by a genetic algorithm. Thereby, it evolves into a group of individuals having higher fitness Φ ₀ (or fitness Φ). This genetic algorithm is a mathematical model that simplifies the mechanism of natural selection and genetic phenomena, and makes it possible for a group of individuals who are candidates for solving the target problem to adapt to the external environment of fitness Φ ₀ (or fitness Φ). Then, a group configuration based on the following rules is generated for each generation.

(Rule 1) The higher the fitness, the higher the survival probability.

(Rule 2) A new individual is generated based on an old individual.

When a genetic algorithm is used as a solution to the optimization problem, (Rule 1) can be regarded as a stochastic search method and (Rule 2) can be regarded as an empirical search method. It can be said that it has both aspects of the stochastic search method and the empirical search method. The specific operation of the gene manipulation unit 303 will be further described. Selector 401 genetic manipulation unit 303, the adaptive calculation unit 302 has calculated fitness [Phi ₀ (or fitness [Phi) reading for each individual (parameter set), this fitness [Phi ₀ (or fitness [Phi) The individual stored in the pool unit 304 is “selected” with a proportional probability. The population of the population shall not be changed. Therefore, an individual with a higher fitness Φ ₀ (or fitness Φ) has a higher probability of being selected, and a fitness Φ ₀
Some individuals with high ₀ (or fitness Φ) are selected multiple times. As a result, individuals with high fitness Φ ₀ (or fitness Φ) leave more offspring to the next generation, and individuals with low fitness Φ ₀ (or fitness Φ) are eliminated. Through the operation of the selection unit 401, an individual who is to leave a gene in the next generation is selected from the population of individuals in the initial generation.

Next, the crossover section 40 of the gene manipulation section 303
2 performs a crossover process on the individual selected by the selection unit 401. As a method of the crossover processing, the simplest stochastic averaging crossover method is used as a crossover method for a numerically expressed gene. That is, all the selected individuals are randomly assigned to 2
Assuming that the corresponding genes (parameter values) of the combined two individuals (parameter sets) are g ₁ and g ₂ , these are expressed by the following formula (4):
Substituting in (5), two new genes g ₁ ′ and g ₂ ′
Generate. Then, the gene g ₁ that caused 'replacement gene g _1, gene g _2' replacing the gene g _2. This crossover process is performed for all genes of all sets of individuals.

G ₁ ′ = g ₁ + k (g ₁ −g ₂ ) R (4) g ₂ ′ = g ₂ + k (g ₁ −g ₂ ) R (5) where R is − Uniform random number generated in the range of 1-1, k
Is the crossover coefficient received by the set value receiving unit 207 from the operator. When the values of the new genes g ₁ ′ and g ₂ ′ become negative or exceed a predetermined upper limit as a result of the calculations of the formulas (4) and (5), recalculate the values within the range. Use only. FIG. 10 conceptually shows the range of new genes g ₁ ′ and g ₂ ′ generated by the stochastic averaging crossover method of equations (4) and (5). By this crossover process, new genes g ₁ ′ and g ₂ ′ can be generated while inheriting the characteristics of the genes g ₁ and g ₂ of the two individuals.

Next, the mutation unit 403 of the gene manipulation unit 303 causes a mutation in the gene of the individual subjected to the crossover process in the crossover unit 402. That is, assuming that the number of all individuals constituting a population of one generation is N and the number of genes per individual is M, the total number of genes of one generation is NM, and thus the set value reception unit 207 receives the number of genes. Using the mutation rate a, aNM genes are randomly selected. Then, the selected gene (parameter value)
Replace with a randomly generated number. Generating a mutation in this way can generate a gene having characteristics that do not appear at crossover, thereby maintaining the diversity of the population. Therefore, it is possible to avoid the danger that the solution falls into the local optimum.

As described above, the selection unit 4 of the gene manipulation unit 303
01, the group of individuals generated by the operation of the crossover unit 402, and the mutation unit 403 are stored as the next-generation group, replacing the group stored in the group pool unit 304.
The estimated dispersion curve calculation unit 301 newly adds the group pool unit 30
The dispersion curve C and the dispersion curve D are obtained for each individual of the next generation population stored in No. 4 as described above. The fitness calculation unit 302 calculates the fitness Φ ₀ (or fitness Φ) for each individual in the next generation population based on the variance curve C (variance curve D) as described above. In the genetic operation unit 303, the selection unit 401, the crossover unit 402, and the mutation unit 403 perform operations on the next generation population based on the fitness Φ ₀ (or fitness Φ), and generate the next generation population. Let it. By repeating this series of operations by the number of generations set in the set value receiving unit 207, a group of individuals having a higher fitness Φ ₀ (or fitness Φ) is gradually generated. Group pool part 3
The gene (parameter value) of the individual (parameter set) having the highest fitness Φ ₀ (or fitness Φ) among the individuals of the last generation stored in 04 is displayed on the display 206. In addition, the genes (parameter values) of individuals (parameter sets) having the highest fitness Φ ₀ (or fitness Φ) in the respective generations are displayed on the display 206 to indicate the progress of each generation before the last generation. Will be displayed.

In the above description, the genetic operation unit 30
In the third example, selection, crossover, and mutation operations are performed on all genes (parameter values) of all individuals in the current generation population based on the fitness Φ ₀ (or fitness Φ). However, other genetic manipulation methods can be used to improve the estimation efficiency. For example, the genetic operation unit 303 can perform operations in three stages as follows. In the first stage, among the five parameters (layer thickness, density, longitudinal wave velocity, shear wave velocity, and shear wave attenuation rate) of genes constituting an individual, predetermined specific parameters (for example, shear wave velocity and longitudinal wave The selection, crossover, and mutation operations are performed only on the gene of (speed), and the operations of the genes with other parameters are not performed. This operation is repeated until the fitness Φ ₀ (or the fitness Φ) calculated by the fitness calculation unit 302 is saturated. When the fitness Φ ₀ (or the fitness Φ) is saturated, the process proceeds to the second stage. Second
In the step, the values of the specific parameters (shear wave velocity and longitudinal wave velocity) of each individual are fixed to the values at the time of saturation, and other parameters (layer thickness, Genetic operation is applied to the values of the genes of density and shear wave attenuation rate), and this genetic operation is repeated until the fitness is saturated. When the fitness is saturated again, the process proceeds to the third stage, in which a genetic operation is simultaneously applied to all parameters, and an individual having the maximum fitness is determined as a solution.

In this genetic operation method, by selecting a parameter that has the greatest influence on the dispersion curve as a specific parameter to be subjected to the genetic operation in the first stage, the genetic operation is performed on the genes of all the parameters. It is more likely that a solution can be reached with a smaller number of generations than in the case of applying. This method is particularly effective for an apparatus configured to compare the observed dispersion curve and the estimated dispersion curve to obtain a solution as in the present embodiment.

Further, in the present embodiment, the estimated dispersion curve calculation unit 301 uses the set values of the crossover coefficient k and the mutation rate a initially set in the set value reception unit 207, and the like, from the initial generation to the last generation. Although the configuration is such that the calculation is performed, it is also possible to adopt a configuration in which these set values can be changed during the evolution.

In the above description of the mutation processing, a method of replacing a gene (parameter value) selected at a mutation rate a with a numerical value selected at random is used. Instead of this method, a numerical creep method is used. Can be used. The idea of numerical creep is based on the assumption that the genes of individuals that have undergone genetic processing and that have undergone generation alternation will be located in much better locations than the earlier ones. Since the parameter to be optimized here is considered to be a continuous function with many peaks and valleys, if the reached peak (parameter value) is the peak of the largest peak (optimal parameter value), If you are in a good location that is fairly close, you may be able to move closer to the top by jumping from that location to somewhere nearby. The numerical creep method is based on this idea. The behavior of mutation by numerical creep method is
This is performed by moving the gene g selected according to the mutation rate a by a certain small random number value above or below the location (gene value) where the gene g is located according to a certain probability. Specifically, the gene (parameter value) is mutated according to the following equation (6).

G ′ = g + bg _max R (6) where g ′ is the value of the new gene,
g is the current gene value, g _max is the upper limit value defined as the range of the parameter of g as shown in FIG. 9, b is the creep degree, and R is the uniformity generated in the range of −1 to 1. It is a random number.

In the present embodiment, the pool unit 304 includes a gene correction unit 314 that corrects a gene (parameter value) to a reference value. Gene correction unit 3
Numeral 14 performs gene correction for each generation number set by the operator in the set value receiving unit 207. Specifically, first, the set value receiving unit 20 is selected from all individuals of the corresponding generation.
7, the number of individuals set in advance is selected. As a selection method, a method of selecting at random, a method of selecting from the one with the highest fitness, a method of selecting from the one with the lowest fitness, etc. are prepared, and the setting unit receiving unit 207 Use the method selected by the operator. Then, for each selected individual, a reference value is determined for each of the five parameters of the layer thickness h, density ρ, longitudinal wave velocity α, shear wave velocity β, and shear wave attenuation rate γ. That is, the genes of each layer (parameter values) for each parameter, for example, the genes α ₀ , α ₁ , α ₂ , α ₃ ... Α _n of the longitudinal wave velocity are plotted as shown in FIG. And a line connecting the n-th layer gene and the values α _1R , α _2R , α _3R , α
_4R ... Are the reference values of the genes in each layer. Then, the gene farthest from the reference value (α _{3 in} FIG. 11) is defined as α ₀ to
One is selected from α _n , and correction for replacing it with the reference value α _3R is performed. All parameters of the selected individual (layer thickness h, density ρ, longitudinal wave velocity α, shear wave velocity β,
This is performed for the shear wave attenuation rate γ). However, the layer thickness h is the value of the zeroth layer is fixed to ∞, draw a straight line connecting the gene h _n of the n-layer and the first layer gene h _1, based on the values on the straight line Value. Thus, the gene correction unit 3
By correcting the gene value for some individuals to the reference value, the efficiency of estimation by the genetic algorithm can be improved.

As described above, the marine sediment layer parameter estimating apparatus according to the present embodiment relates to the problem of obtaining the parameters of the marine sedimentary layer based on the measured data including many errors peculiar to the marine sedimentary layer. However, if the two kinds of dispersion characteristics can be obtained by any method, it is possible to evaluate the difference (the fitness) between the two relatively easily. Made it possible to ask. As the two kinds of dispersion characteristics, a dispersion characteristic of the elastic wave group velocity theoretically calculated based on the assumed parameter set and a dispersion characteristic of the elastic wave group velocity based on the observed elastic wave were used. Therefore, the parameter estimating apparatus of the present embodiment prevents the solution from falling into a local minimum or diverging as in the conventional numerical analysis method, and measures the measured value of the marine sedimentary layer containing many errors. The parameter value of the optimal solution or the quasi-optimal solution that can be practically satisfied can be stably estimated from.

A brute-force method can be considered as a method of obtaining a solution from the fitness of the two dispersion characteristics. However, if the brute-force method is used, the adaptiveness is obtained for all parameter sets in a predetermined range. The degree Φ ₀ (or fitness Φ) is calculated, and the parameter set having the largest fitness Φ ₀ (or fitness Φ) is the solution. If the predetermined range is, for example, the condition shown in Table 1 above, n × 3.5 × 10 ¹³
Since there are combinations (parameter sets) of candidate values, if the brute force method is used, all of the fitness values Φ
_It is necessary to calculate ₀ (or fitness Φ), and the amount of calculation becomes enormous. For this reason, the brute force method is impractical from the viewpoint of real-time properties. On the other hand, the estimating apparatus according to the present embodiment is configured to generate individuals (parameter sets) with high fitness by the genetic algorithm as described above, and thus the number of individuals to be treated at one time is predetermined. Only the population size (eg, 50 individuals). Therefore, in the apparatus according to the present embodiment, the amount of calculation can be significantly reduced, and the parameter set of the solution can be efficiently and stably obtained.

In the apparatus according to the present embodiment, a genetic algorithm is introduced to obtain a solution to a target problem, and the inherent parallelism of simultaneously treating a large number of solution candidates, which is a feature of the genetic algorithm. , Avoid local convergence of the solution. Also, since the genetic algorithm is a search process within a finite number of solution candidates, the phenomenon of solution divergence does not occur essentially.
In addition, the genetic algorithm can maintain stability even when the number of variables increases. Further, even when the model is refined, for example, by increasing the number of parameters, the increase in the amount of calculation in the genetic algorithm is caused by the estimated variance characteristic calculation unit 301.
However, there is an advantage that it only affects the calculation process of the dispersion characteristics in the above. When using a genetic algorithm,
At the time of model construction, it is only necessary to pay attention to the method of calculating fitness and the method of expressing genes, and the means for finding the solution itself is a genetic method that mimics several processes seen in biological evolution such as selection, crossover, and mutation described above. There is also an advantage that it is only necessary to perform a manual operation.

Further, in this embodiment, since the estimated dispersion curve calculation unit 301 uses the above-mentioned SAFARI to calculate the dispersion curve from the parameter set, the dispersion of the group velocity is relatively short. The curve and the dispersion curve of the phase velocity can be obtained simultaneously.

The provision of the correction unit 316 in the observation dispersion curve calculation unit 306 allows the dispersion curves C ₀ and D ₀ to be corrected in response to an instruction for manual correction from the operator.
This makes it possible to correct the dispersion curve disturbance caused by noise or the like, which is difficult to identify by calculation but can be relatively easily found by the operator. Therefore, when calculating the fitness values Φ ₀ , Φ, the dispersion curves C ₀ , D
_{Since the} influence of noise included in ₀ on the fitness Φ ₀ , Φ can be reduced, the level of fitness of the optimal solution increases, and the accuracy of the solution improves.

Further, by arranging the gene correction unit 314 in the group pool unit 304 and periodically correcting the gene (parameter value) to the reference value, the advantage of gene diversity generated by the genetic algorithm is maintained. However, there is an effect that a solution can be efficiently found.

In the above-described embodiment, the configuration in which the entire configuration of the analysis processor 102 is mounted on the observation vessel 101 and all the calculations are performed on the observation vessel 101 has been described. The receiver 103 and the exciter 104 are configured to be detachable, and the data measured by the geophone 103 is temporarily stored in a storage device, carried to the analysis processor 102 arranged on land, and subjected to arithmetic processing on land. Is also possible. Further, the analysis processor 102 may be provided with a ship analysis mode and a land analysis mode in advance. The onboard analysis mode in which the analysis processor 102 is installed on the ship and the analysis is performed on the ship is a simple mode in which the number of individuals, the number of generations, the number of samples m, and the like are set to a small value and the operation is completed in a short time. Is used to determine whether or not the measured data is sufficient to obtain a solution through the processing of the analysis processor 102. The on-land analysis mode, in which the analysis is performed by moving the analysis processor 102 on land while the ship is anchored or on the ground, is based on the measurement data determined to be processable in the on-board analysis mode. The mode can be set to obtain a practically satisfactory sub-optimal solution.

In the above-described embodiment, each unit of the analysis processor 102 is constituted by one computer including an arithmetic unit, a storage unit, a reception unit such as a keyboard and a mouse, and a display device. Of course, it is also possible to configure each unit of the processing device 103 as an independent device.

[0058]

As described above, according to the present invention,
It is possible to provide a seabed sediment layer parameter estimating apparatus capable of stably estimating the value of a parameter indicating the characteristics of the seabed sedimentary layer from measured data.

[Brief description of the drawings]

FIG. 1 is an explanatory diagram showing a configuration of a seabed sediment layer parameter estimation device according to an embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of an analysis processor 102 of the estimating apparatus of FIG.

FIG. 3 is a parameter estimating unit 2 of the analysis processor 102 of FIG. 2;
FIG. 5 is a block diagram showing a detailed configuration of the power supply unit 05.

FIG. 4 is a block diagram showing a more detailed configuration of a gene operation unit 303 of the parameter estimation unit 205 in FIG.

FIG. 5 is an explanatory diagram showing a dispersion curve C ₀ and a dispersion characteristic displayed on a display unit 206 by an observation dispersion curve calculation unit 306 in the estimation device of FIG.

FIG. 6 is an explanatory diagram showing a dispersion curve C ₀ and a dispersion characteristic displayed on a display unit 206 after the correction unit 316 of the observation dispersion curve calculation unit 306 receives and corrects the manual correction in the estimation device of FIG. It is.

FIG. 7 is an explanatory diagram showing a model of a submarine sedimentary layer for which a submarine sedimentary layer parameter estimating apparatus according to one embodiment of the present invention estimates parameters;

FIG. 8 is an explanatory diagram showing genes of one individual used for a genetic algorithm in the apparatus for estimating parameters of a seabed sedimentary layer according to an embodiment of the present invention.

FIG. 9 is an explanatory diagram showing upper and lower limits and unit amounts of respective parameter values of individuals of an initial generation generated by an initial population generating unit 305 in the marine sediment layer parameter estimating apparatus of one embodiment of the present invention.

FIG. 10 is an explanatory diagram showing the concept of diversity of the range of genes generated by the crossover process of the gene manipulation unit 303 in the marine sediment layer parameter estimation device according to one embodiment of the present invention.

FIG. 11 is a graph showing how the gene correction unit 314 of the pool unit 304 calculates a reference value in the marine sediment layer parameter estimation apparatus according to one embodiment of the present invention.

[Explanation of symbols]

101: Observation boat, 102: Analysis processor, 103: Geophone, 104: Exciter, 105: Sea bottom, 106:
Seabed sedimentary layer, 201: distance calculating unit, 202: waveform accumulating unit, 203: frequency analyzing unit, 204: observation dispersion characteristic calculating unit, 205: sedimentary layer parameter estimating unit, 206 ··· Display unit, 301: estimated dispersion curve calculation processing unit, 302: fitness operation unit, 303: gene operation unit, 304: population pool unit, 305: initial population generation unit, 306: observation dispersion curve calculation unit, 314: gene correction unit, 316: correction unit, 401: selection unit, 402: crossover unit, 403: mutation unit.

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Masanori Oiwa 216 Totsuka-cho, Totsuka-ku, Yokohama-shi, Kanagawa Prefecture Inside Hitachi Advanced Systems, Ltd. (72) Kazuo Sato 216 Totsuka-cho, Totsuka-ku, Yokohama-shi, Kanagawa Prefecture Hitachi Advanced Systems, Inc. (72) Kenichi Asano, Inventor Kenichi Asano 216 Totsuka-cho, Totsuka-ku, Yokohama-shi, Kanagawa F-term (reference) 2G047 AA00 BC02 BC03 BC04 BC15 BC18 CB01 EA11 GG17 GG19 GG37 GG41 GH04

Claims (3)

[Claims] 1. An observation characteristic value calculation unit for calculating characteristic values of an elastic wave propagating through a seabed sedimentary layer from data observed on a seabed sedimentary layer to be estimated, A candidate value of the parameter as a gene, an individual obtained by combining the candidate values as an individual, an initial population generator for generating the individual by a predetermined number of individuals, and performing a genetic operation on the individual by a genetic algorithm. A genetic operation unit for evolving the individual into a next generation individual, an estimated characteristic value calculator for theoretically estimating the characteristic value of the elastic wave for the individual, and an estimated characteristic value calculator for the individual. The characteristic value, determine the fitness indicating the degree of coincidence with the characteristic value determined by the observation characteristic value calculation unit, to configure the individual with the highest fitness among the individuals. The fitness value is a solution of the value of the parameter; and the characteristic value is at least a dispersion characteristic of a frequency of a group velocity of the elastic wave, and the parameter is the sea floor. For each of the one or more layers comprising the deposition layer,
An apparatus for estimating parameters of a seabed sediment, including layer thickness, density, longitudinal wave velocity, shear wave velocity, and shear wave attenuation factor. 2. The marine sediment layer parameter estimating device according to claim 1, wherein the observation characteristic value calculating unit displays a dispersion curve from the dispersion characteristic, and displays the dispersion curve on an external display device. Correction means for receiving an instruction from an operator and correcting the dispersion curve. 3. An observation characteristic value calculation unit for calculating characteristic values of an elastic wave propagating through the seabed sedimentary layer from data observed on the seabed sedimentary layer to be estimated, and a plurality of data indicating physical characteristics of the seabed sedimentary layer. A candidate value of the parameter as a gene, an individual obtained by combining the candidate values as an individual, an initial population generator for generating the individual by a predetermined number of individuals, and performing a genetic operation on the individual by a genetic algorithm. A genetic operation unit for evolving the individual into a next generation individual, an estimated characteristic value calculator for theoretically estimating the characteristic value of the elastic wave for the individual, and an estimated characteristic value calculator for the individual. The characteristic value and the fitness value indicating the degree of coincidence with the characteristic value obtained by the observation characteristic value calculation unit are obtained, and the indication constituting the individual having the highest fitness among the individuals is obtained. A complement value, a fitness calculating unit that is a solution of the value of the parameter, the genetic operation unit, in the genetic operation, the candidate value of the plurality of parameters constituting the individual at the time of the genetic operation Among them, the genetic operation is given only to a predetermined candidate value of a specific parameter, and the genetic operation is repeated until the fitness calculated by the fitness calculation unit is saturated, and when the fitness is saturated, Fixes the value of the predetermined parameter to the value at the time when the fitness is saturated, gives the genetic operation to other parameter candidate values, and repeats the genetic operation until the fitness is saturated. A method of estimating parameters of a seabed sedimentary layer, wherein when the fitness is saturated again, the genetic operation is applied to all parameters simultaneously.