Estimation device, estimation method, and program
The estimation device uses a Koopman operator to estimate phase models for multiple elements, addressing the limitation of existing methods by accurately analyzing collective vibrations and interactions, thereby enhancing system understanding.
Patent Information
- Application Number
- JP2023546600
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2021-09-07
- Publication Date
- 2024-11-13
- Estimated Expiration
- 2041-09-07
AI Technical Summary
Existing methods struggle to estimate phase models for multiple elements using the Koopman operator, as they require only one element.
An estimation device and method that utilizes a Koopman operator to estimate a phase model for a plurality of elements by analyzing time series data, employing a Koopman operator estimator and phase model estimation unit to derive collective vibrations and interactions among elements.
Accurately estimates a topological model for multiple elements, enabling analysis of interactions and collective vibrations, reflecting the estimated phase model closely matches the actual system dynamics.
Smart Images

Figure 0007582494000022 
Figure 0007582494000023 
Figure 0007582494000024
Abstract
Description
[Technical field]
[0001] The present invention relates to an estimation device, an estimation method, and a program. [Background technology]
[0002] Investigating interactions between data composed of multiple elements is a common problem in various fields such as statistics, machine learning, physics, and molecular dynamics. In physics and molecular dynamics, a method has been proposed to describe collective vibrations of multiple elements and information on interactions between them using a model called a phase model (Non-Patent Document 1). In addition, a method has been proposed to estimate a phase model from given data using a Fourier series or a Hilbert transform (Non-Patent Document 2). In the phase model, data is converted into phase information using a function called a phase function, and then the relationship between the phases is described using a function called a phase coupling function.
[0003] Here, when there is one element, the phase model describes only the vibration of one element, but in this case, a method of estimating a phase model using a Koopman operator has been proposed (Non-Patent Document 3). The Koopman operator is a linear operator that describes the time evolution of time series data, and a method has been proposed for estimating the Koopman operator that the given time series data follows from the data using RKHS (Reproducing Kernel Hilbert Space) and its generalization vvRKHS (Vector-Valued RKHS, Non-Patent Document 4) (Non-Patent Document 5). Therefore, by using these methods, when the number of elements is one, it is possible to estimate a phase model from given data. [Prior art documents] [Non-patent literature]
[0004] [Non-Patent Document 1] H. Nakao, S. Yasui, M. Ota, K. Arai, and Y. Kawamura, “Phase reduction and synchronization of a network of coupled dynamical elements exhibiting collective oscillations,” Chaos 28, 045103 (2018).
Outdoor Tool2
Outdoor Tools3
Outdoor Tools 4
Direct Environment 5
[0005] While it is difficult to estimate the entire phase function using the Fourier series or Hilbert transform, it is possible to estimate the entire phase function using the Koopman operator. However, the method of estimating the phase model using the Koopman operator requires only one element.
[0006] An embodiment of the present invention has been made in consideration of the above points, and has an object to estimate a phase model for a plurality of elements using a Koopman operator. [Means for solving the problem]
[0007] In order to achieve the above object, an estimation device according to one embodiment includes an operator estimator that receives time-series data composed of a plurality of elements as an input and estimates a Koopman operator from the time-series data, and a phase model estimator that uses the Koopman operator to estimate a phase model that represents collective oscillations of the plurality of elements and interactions between the elements. Effect of the Invention
[0008] The Koopman operator can be used to estimate a topological model for multiple elements. [Brief description of the drawings]
[0009] [Figure 1] FIG. 2 is a diagram illustrating an example of a hardware configuration of an estimation device according to the present embodiment. [Diagram 2] FIG. 2 is a diagram illustrating an example of a functional configuration of an estimation device according to the present embodiment. [Diagram 3] 10 is a flowchart illustrating an example of a phase model estimation process according to the present embodiment. [Figure 4] This is an example of a scatter plot of the eigenvalues of the Koopman operator on the complex plane (part 1). [Diagram 5] FIG. 13 is a diagram showing an example of a phase function estimated from observed data. [Figure 6] FIG. 1 shows an example of a transformation calculated from an FHN model. [Figure 7] FIG. 13 is a diagram showing an example of values of the antisymmetric part of the interaction. [Figure 8] This is a second example of a scatter plot of the eigenvalues of the Koopman operator on the complex plane. [Figure 9] FIG. 13 is a diagram showing an example of a heat map showing the strength of interactions. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0010] An embodiment of the present invention will be described below. In this embodiment, an estimation device 10 capable of estimating a phase model for a plurality of elements using the Koopman operator will be described.
[0011] <Theoretical structure> First, a theoretical configuration when the estimation device 10 according to this embodiment estimates a phase model for a plurality of elements using the Koopman operator will be described.
[0012] 1. Settings Consider analyzing time series data generated as follows:
[0013]
number
[0014] Let X1, . . . , X be the N elements described by the network (network model of a coupled dynamical system) shown in the following equation (1). N Let ∈Xset.
[0015]
number
[0016] In this case, X=[X1, ,X N ]. The model shown in equation (1) above generates data at regular time intervals Δt, that is, x i,l =X i Let us assume that time series data (Δt·l) is generated. This time series data is observable data and is also called observed data. Note that l is a lowercase L.
[0017] We assume that there is a common frequency ω that governs the N elements, and that each element weakly interacts with the others. In other words, we assume that the model shown in the above formula (1) is reduced to the phase model shown in the following formula (2).
[0018]
number
[0019] 2. Estimation of Topological Models Using Koopman Operators Xset N From C N (N-dimensional complex space) where v(x)∈C N The i-th component of x∈Xset N Let H be the Hilbert space consisting of functions v that depend only on the i-th component of N ×Xset N →C N×N It is possible to set the vvRKHS generated from as H.
[0020] [Φ(x1,x2)] i,j =k((x 1,i ,i),(x 2,j ,j)) where k is
[0021]
number
[0022] For v∈H, define the Koopman operator K on H as a linear operator that satisfies Kv(X(t))=v(X(t+Δt)). For a natural number l (l is lowercase L), let t0=Δt·l. Then, consider the optimization problem shown in the following equation (3).
[0023]
number
[0024] For some ω∈[0,2π],
[0025]
number
[0026]
number
[0027] Eigenvalue λ j,i Let us assume that there exists an eigenvector for v j,i Then, for each j=2, ,M, v j,1 (X(t0)), ,v j,N Assuming that (X(t0)) are linearly independent,
[0028]
number
[0029] At this time, for j = 2, ···, M, consider the optimization problem P shown in the following formula (4) j inductively.
[0030]
Number
[0031]
Number
[0032]
Number
[0033] The optimization problem P shown in the above formula (4) j is solved inductively up to j = M, and the following is obtained. Since the optimization problem shown in the above formula (4) is a linear problem, it can be solved analytically.
[0034]
Number
[0035] θ i,j (t)=arg(u i,j (X i(t))), r i,j (t)=|u i,j (X i (t))|far away. u j From the definition and equation (4),
[0036]
number
[0037]
number
[0038]
number
[0039]
number
[0040]
number
[0041]
number
[0042]
number
[0043] ≪3. Estimation of the Koopman operator≫ If the Koopman operator K can be estimated from the observed data, it is possible to estimate the topological model that the observed data follows by the method described in "2. Estimation of a Topological Model Using the Koopman Operator." As a method for estimating the Koopman operator K from the observed data, for example, the method described in Non-Patent Document 5 may be used.
[0044] <Hardware Configuration of Estimation Device 10> Next, the hardware configuration of the estimation device 10 according to this embodiment will be described with reference to Fig. 1. As shown in Fig. 1, the estimation device 10 according to this embodiment is realized by the hardware configuration of a general computer or computer system, and has an input device 101, a display device 102, an external I / F 103, a communication I / F 104, a processor 105, and a memory device 106. Each of these pieces of hardware are connected to each other via a bus 107 so as to be able to communicate with each other.
[0045] The input device 101 is, for example, a keyboard, a mouse, a touch panel, various physical buttons, etc. The display device 102 is, for example, a display, a display panel, etc. Note that the estimation device 10 does not necessarily have to include at least one of the input device 101 and the display device 102, for example.
[0046] The external I / F 103 is an interface with an external device such as a recording medium 103a. The estimation device 10 can read and write data from and to the recording medium 103a via the external I / F 103. Examples of the recording medium 103a include a compact disc (CD), a digital versatile disc (DVD), a secure digital memory card (SD memory card), and a universal serial bus (USB) memory card.
[0047] The communication I / F 104 is an interface for connecting the estimation device 10 to a communication network. The processor 105 is, for example, various types of arithmetic devices such as a central processing unit (CPU) and a graphics processing unit (GPU). The memory device 106 is, for example, various types of storage devices such as a hard disk drive (HDD), a solid state drive (SSD), a flash memory, a random access memory (RAM), and a read only memory (ROM).
[0048] The estimation device 10 according to this embodiment can realize various processes described below by having the hardware configuration shown in Fig. 1. Note that the hardware configuration shown in Fig. 1 is only an example, and the estimation device 10 may have, for example, multiple processors 105, multiple memory devices 106, or various other hardware besides the hardware shown in the figure.
[0049] <Functional configuration of the estimation device 10> Next, the functional configuration of the estimation device 10 according to this embodiment will be described with reference to Fig. 2. As shown in Fig. 2, the estimation device 10 according to this embodiment includes a Koopman operator estimation unit 201, a phase model estimation unit 202, and a storage unit 203. The Koopman operator estimation unit 201 and the phase model estimation unit 202 are realized, for example, by a process in which one or more programs installed in the estimation device 10 are executed by the processor 105. The storage unit 203 is realized, for example, by the memory device 106. The storage unit 203 may be realized by a storage device (for example, a database server, etc.) connected to the estimation device 10 via a communication network.
[0050] The Koopman operator estimation unit 201 estimates the Koopman operator K from observed data. Here, the Koopman operator estimation unit 201 includes a data acquisition unit 211 and an operator estimation unit 212. The data acquisition unit 211 acquires observed data from the storage unit 203. The operator estimation unit 212 estimates the Koopman operator K from the observed data acquired by the data acquisition unit 211, for example, by using the method described in Non-Patent Document 5.
[0051] The phase model estimation unit 202 uses the Koopman operator K estimated by the Koopman operator estimation unit 201 to estimate a phase model to which the observation data conforms, by the method described in “2. Estimation of a phase model using a Koopman operator”.
[0052] The storage unit 203 stores the observation data. The storage unit 203 also stores the phase model (phase function, frequency ω, interaction Γ) estimated by the phase model estimation unit 202. i ) may be stored.
[0053] <Phase model estimation process> Next, the phase model estimation process according to this embodiment will be described with reference to FIG.
[0054] The data acquisition unit 211 of the Koopman operator estimation unit 201 acquires observed data from the storage unit 203 (step S101).
[0055] The operator estimation unit 212 of the Koopmann operator estimation unit 201 estimates the Koopmann operator K from the observation data acquired in the above step S101 (step S102). As described above, the operator estimation unit 212 may estimate the Koopmann operator K from the observation data by using, for example, the method described in Non-Patent Document 5.
[0056] The phase model estimation unit 202 solves the optimization problem shown in the above formula (3) using the Koopman operator K estimated in the above step S102 (step S103). Note that the phase model estimation unit 202 can solve the optimization problem shown in the above formula (3) by, for example, the gradient method or the like.
[0057] The phase model estimation unit 202 sets j←2 (step S104).
[0058] The phase model estimation unit 202 determines whether j<M + 1 (step S105). Note that M is a natural number of 2 or more determined in advance.
[0059] If it is determined in the above step S105 that j<M + 1, the phase model estimation unit 202 solves the optimization problem shown in the above formula (4) using the Koopman operator K estimated in the above step S102 (step S106). Note that since the optimization problem shown in the above formula (4) is a linear problem, the phase model estimation unit 202 can analytically solve the optimization problem shown in this formula (4).
[0060] The phase model estimation unit 202 sets j←j + 1 (step S107) and returns to the above step S105. As a result, for j = 2, ···, M, the optimization problem shown in the above formula (4) is recursively solved.
[0061] If it is not determined in the above step S105 that j<M + 1, the phase model estimation unit 202 estimates the phase model by the above formula (5) using the solution of the optimization problem shown in the above formula (3) and the solution of the optimization problem shown in the above formula (4) (step S108). As a result, it is considered that the phase model followed by the observation data acquired in the above step S101 is estimated.
[0062] <Evaluation> Hereinafter, the evaluation of the phase model estimated by the estimation device 10 according to the present embodiment will be described.
[0063] <Estimation of phase function and phase coupling function> Xset=R 2 By changing the initial value from the FHN (FitzHugh-Nagumo) model above, 10 time series data (observation data) x0 i ,x 2000 i (i=1, ,10), estimate the Koopman operator K, and then use the method explained in "2. Estimation of a topological model using the Koopman operator" to estimate X i (t) to θ i (t) is the phase function η i (η i (x) = arg(u i,1 (x)), x∈R 2 ) and the interaction Γ in the above equation (5) i (ψ1, ,ψ N ) was calculated.
[0064] The FHN model is a network model in which X i (t)=[y i (t),z i (t)]∈R 2 For F1(X i )=F(X i )=[y i (y i -c)(1-y i )-z i ,μ -1 (y i -dz i )], G i,j (X i ,X j )=G(X i ,X j )=[0.01(z i -z j ),0], c=-0.1, d=0.5, and μ=100.
[0065] 4 shows a scatter diagram in which the eigenvalues of the Koopmann operator K estimated by the Koopmann operator estimation unit 201 are plotted on the complex plane. Note that the horizontal axis represents the real part, and the vertical axis represents the imaginary part.
[0066] As shown in Figure 4, there are three places where two eigenvalues with almost equal values overlap (Q 11 , Q 12 , Q 13 ) is the closest one of these three locations (Q 13 ) are the two eigenvalues in λ 1,1 , λ 1,2 In addition, the eigenvalue λ 1,1 The eigenvector for v 1,1 , eigenvalue λ 1,2 The eigenvector for v 1,2 In this case, s=1899, c 1,1 Σ i=1 10 v 1,1 (x s i ) / 10+c 1,2 Σ i=1 10 v 1,2 (x s i ) / 10=1 1,1 ,c 1,2 For ∈C, u1 0 =c 1,1 v 1,1 +c 1,2 v 1,2 Let λ = λ 1,1 , u=u1 0 , a 1,2 =a 2,1 = 0.002, a 1,1 =a 2,2 = 0 as the initial value. Furthermore, M = 3, and among the points where two eigenvalues with almost equal values overlap, λ 1,1 and λ 1,2 Each eigenvalue that exists in a place other than the place where 2,i (i=1,2), λ 3,i (i=1,2) and solved the optimization problem shown in equation (4) above.
[0067] In this example, N=2, and in the above formula (1), F1=F2, G 1,2 =G 2,1Therefore, η1 should be equal to η2. Therefore, only η1 was calculated by the phase model estimation unit 202. The result is shown in FIG. 5. On the other hand, X i (t) to θ i The results of the conversion to (t) calculated directly from the FHN model are shown in Figure 6. Comparing Figures 5 and 6, it can be seen that the results estimated from the observational data (Figure 5) are close to the results calculated directly from the FHN model (Figure 6).
[0068] In addition, the interaction Γ i The antisymmetric part of Γ α The results of calculating (ψ) = Γ1(ψ) - Γ2(-ψ) are shown in Figure 7. As shown in Figure 7, in this case too, it can be seen that the value estimated from the observational data is close to the value calculated directly from the FHN model.
[0069] <Estimation of interaction strength> Xset=R 2 By changing the initial value from the SL model above, 10 time series data (observation data) x0 i ,x 2000 i (i=1, , 10), estimate the Koopman operator K, and solve the optimization problem shown in equation (3) to obtain the strength of interaction a i,k was estimated.
[0070] The SL model is a network model in which X i (t)=[y i (t),z i (t)]∈R 2 For F1(X i )=F(X i )=[y i -a-z i -(y i 2 +z i 2 )(by i +z i ),z i -(y i 2 +z i 2 )(by i +zi )], G(X i ,X j )=[0.01(z j -z i ),0],G 1,2 (X i ,X j )=G 2,1 (X i ,X j )=G(X i ,X j ), G 1,3 (X i ,X j )=G 2,3 (X i ,X j )=G 3,1 (X i ,X j )=G 3,2 (X i ,X j )=0.1G(X i ,X j ), a=2, b=1.
[0071] 8 shows a scatter diagram in which the eigenvalues of the Koopmann operator K estimated by the Koopmann operator estimation unit 201 are plotted on the complex plane. Note that the horizontal axis represents the real part, and the vertical axis represents the imaginary part.
[0072] As shown in Figure 8, there are three places where three eigenvalues with almost equal values overlap (Q 21 , Q 22 , Q 23 ) There are three Q 21 , Q 22 , Q 23 Let the eigenvalues in be λ1, λ2, and λ3, respectively. Also, let the eigenvector for eigenvalue λ1 be v1, the eigenvector for eigenvalue λ2 be v2, and the eigenvector for eigenvalue λ3 be v3. In this case, let s = 1899, and c1Σ i=1 10 v1(x s i ) / 10+c2Σ i=1 10 v2(x s i ) / 10+c3Σ i=1 10v3(x s i For c1, c2, c3∈C such that u1 0 =c1v1+c2v2+c3v3.
[0073] In this example, i,k Since the only purpose is to find the magnitude of 0 and solved the optimization problem shown in equation (3) as a linear problem. i,k A heat map showing the magnitude of is shown in Fig. 9. As shown in Fig. 9, the interaction between element X1 and element X2 is estimated to be large, which reflects the large interaction between element X1 and element X2 in the original model (SL model).
[0074] <Summary> As described above, the estimation device 10 according to this embodiment can accurately estimate a topological model for a plurality of elements using the Koopman operator. Therefore, it is possible to perform analysis, such as extracting interactions between a plurality of elements in relation to data composed of these elements, using the topological model estimated by the estimation device 10 according to this embodiment. Note that the estimation device 10 according to this embodiment may perform the above analysis using the estimated topological model, or may use the analysis result to control a system (or devices constituting the system) represented by a network model of a coupled dynamics system.
[0075] The present invention is not limited to the above-described embodiments specifically disclosed, and various modifications, changes, combinations with known technologies, etc. are possible without departing from the scope of the claims. [Explanation of symbols]
[0076] 10 Estimation device 101 Input Device 102 Display device 103 External I / F 103a Recording media 104 Communication I / F 105 processors 106 Memory Device 107 Bus 201 Koopman operator estimator 202 Phase model estimation unit 203 Storage section 211 Data Acquisition Department 212 Operator estimation part
Claims
1. an operator estimator that receives time series data composed of a plurality of elements as input and estimates a Koopman operator from the time series data; a phase model estimating unit that estimates a phase model representing a collective vibration of the plurality of elements and an interaction between the elements by using the Koopman operator; An estimation device having the following:
2. The phase model estimation unit using the Koopman operator to solve the first optimization problem by a gradient method; recursively solving a second optimization problem a predetermined number of times using the solution of the first optimization problem and the Koopman operator; The estimating apparatus of claim 1 , further comprising: an estimation unit configured to estimate the phase model using a solution of the first optimization problem and a solution of the second optimization problem.
3. The time series data is expressed as X(t) = [X 1 (t), ..., X N (t)], the Koopman operator is K, B i,k u=u k e i Let B be a linear operator on the Hilbert space defined by i,k , u k is the k-th component of the vector-valued function u, e i is an N-dimensional vector whose i-th element is 1 and the other elements are 0. Let t be t 0 As, The first optimization problem is: [0020] The estimation device according to claim 2 , wherein:
4. The data interval of the time series data is Δt, a certain frequency is ω∈[0, 2π], and the solution of the first optimization problem is λ=e^((√(-1))Δtω), a i,k 1 , u 1 , the eigenvalues of the Koopman operator K are λ j,i = e^((√(-1))Δtjω) (where j = 2, ..., M, i = 1, ..., N, and M is a predetermined integer of 2 or more), The second optimization problem is: ##EQU00021## It is expressed as The phase model estimation unit 4. The estimating device of claim 3, further comprising: a step of recursively solving the second optimization problem for j=2, . . . , M.
5. The phase model estimation unit The solution λ, a of the first optimization problem i,k 1 , u 1 and the solution a of the second optimization problem i,k 2 , ..., a i,k M The estimation device according to claim 4 , wherein the phase model constituted by the frequency ω and the phase coupling function is estimated by approximating a phase coupling function using
6. an operator estimation step of estimating a Koopman operator from time series data composed of a plurality of elements as input; a phase model estimation step of estimating a phase model representing collective vibration of the plurality of elements and interactions between the elements using the Koopman operator; The estimation method is performed by a computer.
7. A program causing a computer to function as the estimation device according to any one of claims 1 to 5.
