Observation planning support device, method and program

Abstract

To provide a technique capable of optimizing an observation point at a high speed when observation values can be acquired at a plurality of depths at the observation point.SOLUTION: An observation planning support device for supporting observation planning for acquiring values of target items at a plurality of depths at an observation point, the observation point provided on a horizontal plane of a target region, includes a memory and a processor. The memory stores time-series data representing an item value at each time of a target item at a grid point represented by a two-dimensional position and a depth. The processor acquires a spatial mode, which is the mode of a space represented by the two-dimensional position and the depth, from the time-series data, constructs a deformed spatial mode matrix formed by arranging element values of the spatial mode, with two-dimensional position taken in a row direction, and the time and depth taken in a column direction based on the spatial mode, selects a prescribed number of rows from the deformed spatial mode matrix by a prescribed optimization method, and determines a two-dimensional position represented by the selected row as the observation point.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present disclosure relates to a technique for assisting in creating an observation plan for a moving object to acquire data at an observation point. [Background technology]

[0002] In order to predict environmental changes in each ocean area, environmental data is observed at each location at each time. Examples of environmental items to be observed and predicted include seawater temperature and salinity. Since it is not possible to install measurement equipment fixedly on the sea or underwater like on land, it is not easy to set up fixed observation locations. Therefore, it is common to go to observation points by observation vessel and observe the desired items. In this case, increasing the number of observation points improves the accuracy of environmental prediction, but on the other hand, the cost required for observation increases. Therefore, it is necessary to identify observation points as few as possible so that data that can predict the environment of each ocean area with high accuracy can be obtained, and to develop an efficient observation plan.

[0003] Patent Document 1 discloses a technology for calculating an optimal route using measurement data on weather, marine environment, ocean currents, etc., measured by a ship sailing in a target sea area. In the method of Patent Document 1, the measurement data is assimilated into forecast data obtained by numerical analysis, improving the accuracy of forecasts of weather, marine environment, and ocean currents. This makes it possible to calculate an optimal route that takes into account sailing time, fuel consumption, and ship motion.

[0004] Non-Patent Document 1 also describes a method for calculating positions with high observation value using time-series data on sea surface water temperature. This method applies proper orthogonal decomposition (POD) to the time-series data on sea surface water temperature, and identifies optimal observation points by applying QR decomposition to a sensor candidate matrix in which the obtained spatial POD modes are arranged. This method is said to be able to obtain optimal observation point positions from the perspective of reconstructing the temperature field using a low-dimensional model with POD modes.

[0005] Furthermore, Non-Patent Document 2 describes an observation point optimization method for cases where multiple physical quantities can be measured at an observation point, i.e., when vector observation is possible. This method makes it possible to identify the optimal observation point position when multiple physical quantities can be measured. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] JP 2014-13145 A [Non-patent literature]

[0007] [Non-Patent Document 1] Manohar, K., Brunton, BW, Kutz, JN, & Brunton, SL (2018). Data-driven sparse sensor placement for reconstruction: Demonstrating the benefits of exploiting known patterns. IEEE Control Systems Magazine, 38(3), 63-86. [Non-Patent Document 2] Saito, Y., Nonomura, T., Nankai, K., Yamada, K., Asai, K., Sasaki, Y., & Tsubakino, D. (2020). Data-driven vector-measurement-sensor selection based on greedy algorithm. IEEE Sensors Letters, 4(7), 1-4. Summary of the Invention [Problem to be solved by the invention]

[0008] According to the method described in the above-mentioned Patent Document 1, it is possible to determine an optimal route by utilizing measurement data obtained within a target sea area, taking into consideration sailing time, fuel consumption, ship motion, etc. However, this method does not improve the accuracy of prediction of the marine environment, because it does not evaluate the quality of the measurement point positions from the viewpoint of prediction accuracy of the marine environment.

[0009] The method described in Non-Patent Document 1 above has been shown to be effective in observing seawater temperature data at the sea surface. However, in addition to observing seawater temperature at the sea surface, there are cases where the distribution of water temperature in the depth direction is observed at one observation point. For example, an expandable bathythermograph (hereinafter referred to as XBT) is a device that is dropped from a sailing observation ship and continuously records seawater temperature up to a depth of about 300 m while free falling. In this way, an effective method is required when observing data at each position in the depth direction. However, although the method in Non-Patent Document 1 is effective in observing seawater temperature data at the sea surface, it cannot be said to be an effective observation point optimization method when the water temperature distribution in the depth direction is observed.

[0010] In addition, the method described in Non-Patent Document 2 uses a spatial POD mode like the method in Non-Patent Document 1. However, since the method uses a spatial POD mode calculated from a plurality of physical quantity distributions, a high-speed method such as QR decomposition cannot be applied, and it is necessary to individually evaluate the row vector norm of a sensor candidate matrix corresponding to a specific coordinate. This may make the method less practical in situations where rapid decision-making is required.

[0011] One objective included in the present disclosure is to provide a technique that can quickly optimize an observation point when observation values ​​can be obtained at multiple depths at the observation point. [Means for solving the problem]

[0012] An observation plan creation support device according to one aspect of the present disclosure is an observation plan creation support device that sets up an observation point on a horizontal plane in a target area and supports the creation of an observation plan for acquiring values ​​of target items at multiple depths at the observation point, the observation plan creation support device having a memory and a processor, wherein the memory stores time series data representing item values ​​of target items at each time at lattice points represented by two-dimensional positions and depths, and the processor acquires a spatial mode, which is a spatial mode represented by the two-dimensional positions and depths, from the time series data, and constructs a modified spatial mode matrix based on the spatial mode, with two-dimensional positions in the row direction and times and depths in the column direction, arranging element values ​​of the spatial mode, selects a predetermined number of rows from the modified spatial mode matrix using a predetermined optimization method, and determines the two-dimensional positions indicated by the selected rows as the observation points.

[0013] An observation plan creation support method according to one aspect of the present disclosure is a method for supporting the creation of an observation plan that sets an observation point on a horizontal plane in a target area and supports the creation of an observation plan for acquiring values ​​of target items at the observation point at multiple depths, the method being executed by a computer to record in a memory time series data representing item values ​​of the target item at each time at a lattice point represented by a two-dimensional position and depth, obtain from the time series data a spatial mode which is a spatial mode represented by the two-dimensional position and depth, construct a modified spatial mode matrix based on the spatial mode, with two-dimensional positions in the row direction and times and depths in the column direction, and arrange the element values ​​of the spatial mode, select a predetermined number of rows from the modified spatial mode matrix using a predetermined optimization method, and determine the two-dimensional positions indicated by the selected rows as the observation points.

[0014] An observation plan creation support program according to one embodiment of the present disclosure is an observation plan creation support program that sets up an observation point on a horizontal plane in a target area and supports the creation of an observation plan for acquiring values ​​of target items at multiple depths at the observation point, and causes a computer to perform the following: record in a memory time series data representing item values ​​of the target item at each time at a lattice point represented by a two-dimensional position and depth, acquire from the time series data a spatial mode which is a spatial mode represented by the two-dimensional position and depth, construct a modified spatial mode matrix based on the spatial mode, with two-dimensional positions in the row direction and times and depths in the column direction, and arrange the element values ​​of the spatial mode, select a predetermined number of rows from the modified spatial mode matrix using a predetermined optimization method, and determine the two-dimensional position indicated by the selected rows as the observation point. Effect of the Invention

[0015] According to the present disclosure, observation points can be quickly optimized when observation values ​​can be obtained at multiple depths at the observation points. [Brief description of the drawings]

[0016] [Figure 1] 1 is a block diagram showing an example of the configuration of an observation plan formulation support device according to a first embodiment. [Diagram 2] 5 is a flowchart showing an example of observation point optimization processing according to the first embodiment. [Diagram 3] The following shows Equation 1 according to the first embodiment. [Figure 4] Equation 2 according to the first embodiment is shown below. [Diagram 5] Equation 3 according to the first embodiment is shown below. [Figure 6] Equation 4 according to the first embodiment is shown below. [Figure 7] Formula 5 according to the first embodiment is shown below. [Figure 8] 3 shows an example of tabular data according to the first embodiment. [Figure 9] FIG. 4 is a diagram showing an example display of an observation point position optimization result in the first embodiment. [Figure 10]13 is a flowchart showing an example of observation point optimization processing according to the second embodiment. [Figure 11] 13 is a flowchart showing an example of an observation point reduction process according to the second embodiment. [Figure 12] Formula 6 according to the second embodiment is shown below. [Figure 13] Formula 7 according to the second embodiment is shown below. [Figure 14] Formula 8 according to the second embodiment is shown below. [Figure 15] Formula 9 according to the second embodiment is shown below. [Figure 16] FIG. 11 is a diagram showing a display example of trade-off visualization according to the second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0017] (Embodiment 1) FIG. 1 is a block diagram showing an example of the configuration of an observation plan formulation supporting device 1 according to the first embodiment.

[0018] The observation plan creation support device 1 according to this embodiment supports creation of an observation plan in which an observation point is set on a horizontal plane of a target area and values ​​of target items are acquired at multiple depths at the observation point. The observation plan creation support device 1 includes a control device 100, an input / output unit 110, and a storage device 120. The input / output unit 110 and the storage device 120 are connected to the control device 100 so as to be able to communicate with each other.

[0019] The control device 100 is a computer device (e.g., a PC (Personal Computer)) and includes a memory 101 and a processor 102. The memory 101 may be configured with a volatile storage medium and / or a non-volatile storage medium. The processor 102 may be, for example, a CPU (Central Processing Unit). An observation plan planning support program 103 may be stored in the memory 101. The processor 102 reads out and executes the observation plan planning support program 103 from the memory 101, thereby realizing functions related to marine planning support described in this embodiment.

[0020] The input / output unit 110 includes a keyboard 111, a mouse 112, and a display 113. The keyboard 111 and the mouse 112 are examples of input devices. The display 113 is an example of an output device.

[0021] A user can input commands through a keyboard 111 and a mouse 112, which are examples of input devices. The user can visually grasp the results of processing by the control device 100 from a display 113, which is an example of an output device.

[0022] The storage device 120 is a non-volatile storage medium configured, for example, by a hard disk drive (HDD) or a solid state drive (SSD). The storage device 120 stores ocean analysis data 121, map data 122, observation point data 123, and the like. The data stored in the storage device 120 may be read into the memory 101 as necessary and used by the processor 102.

[0023] The ocean analysis data 121 is two-dimensional or three-dimensional time-series data on the state of the ocean obtained by using an ocean simulation. The ocean simulation involves arranging grid points in a space represented by three axes of latitude, longitude, and depth, and obtaining temporal changes in physical quantities such as sea level altitude, water temperature, salinity, and flow velocity by numerical analysis. The ocean analysis data 121 has values ​​at the grid points of each variable. The ocean analysis data 121 may be created by an ocean research institute or the like and made public. The ocean analysis data 121 may be interpreted as time-series data. In other words, the time-series data represents item values ​​at each time of target items (e.g., sea level altitude, water temperature, salinity, flow velocity, etc.) at grid points represented by two-dimensional positions (latitude, longitude) and depth.

[0024] The map data 122 is data showing a map relating to the ocean. The map data 122 may be publicly available.

[0025] Observation point data 123 is data that indicates the correspondence between the position of a lattice point corresponding to an observation point (hereinafter referred to as a lattice position) and the actual latitude and longitude. Details of observation point data 123 will be described later (see FIG. 8).

[0026] Fig. 2 is a flowchart showing an example of observation point optimization processing according to Embodiment 1. Next, with reference to Fig. 2, a process in which the observation plan development supporting device 1 determines and presents optimal positions and numbers of observation points will be described.

[0027] The processor 102 acquires the ocean analysis data 121 from the storage device 120 and expands it in the memory 101 (step 201). This process may be executed in response to an instruction from a user via the input / output unit 110. The processor 102 may also determine which of the latitude, longitude, depth (water depth), time range, and physical quantity included in the ocean analysis data 121 to acquire based on an instruction from a user. In this embodiment, an example of performing observation point optimization processing using the ocean analysis data 121 for the analysis of environmental changes in a sea area will be described, but it is also possible to perform observation point optimization processing using atmospheric meteorological analysis data instead of the ocean analysis data 121 for the analysis of meteorological changes.

[0028] The processor 102 performs singular value decomposition (SVD) on the oceanographic analysis data 121 acquired in step 201 to perform mode decomposition and construct a low-dimensional model (step 202). Next, the process of step 202 will be described in detail.

[0029] First, the processor 102 transforms the acquired oceanographic analysis data 121 according to Equation 1 shown in Fig. 3 to construct a data matrix X. Here, the number of latitude elements is nlat, the number of longitude elements is nlon, the number of depth elements is ndepth, and the number of time range elements is ntime.

[0030] As shown in Equation 1, a plurality of data corresponding to a specific time in the ocean analysis data 121 are arranged one-dimensionally to form a state vector. That is, a state vector corresponding to a specific time corresponds to one column of the data matrix X. Therefore, as shown in Equation 1, the number of columns of the data matrix X is the number of elements in the time range, ntime, and one column of the data matrix X is a state vector corresponding to the time of that column.

[0031] When there is only one type of physical quantity (target item), for example, temperature, the length of the state vector is (ndepth×nlat×nlon). When there are multiple physical quantities to be reduced-dimensionally modeled, the state vector can be extended in the row direction (i.e., the number of rows can be increased) according to the arrangement in Equation 1. Note that, although the present embodiment describes a case where there is only one type of physical quantity, when there are, for example, two types of physical quantities, the length (number of elements) of the state vector can be (2×ndepth×nlat×nlon).

[0032] Let the final data matrix X have n rows and m columns, where n is the number of elements in the state vector and m is the number of elements in the time range, ntime.

[0033] Next, the processor 102 performs singular value decomposition (SVD) on the data matrix X as shown in Equation 2 in FIG. 4 to obtain a spatial POD mode U, a singular value matrix Σ, and a temporal POD mode V T POD is an abbreviation for Proper Orthogonal Decomposition. POD is an example of principal component analysis.

[0034] If the data matrix X can be sufficiently approximated by r spatial POD modes, the data matrix X can be approximated in a reduced dimension as shown in Equation 3 in Fig. 5. The subscripts a:b at the bottom right of the matrix in Equation 3 mean that columns a to b of the matrix are extracted.

[0035] Finally, the reduced spatial POD mode U is calculated according to Equation 4 shown in FIG. 4 and Equation 5 shown in FIG. 1:r Here, the reduced spatial POD mode U1:r has n rows and r columns. Also, the spatial POD mode U of some order s s , i.e. U 1:r The vector obtained by extracting the s-th column of stores the mode values ​​corresponding to the arrangement of state vectors, that is, the (ndepth×nlat×nlon) arrangement.

[0036] According to the above, the processor 102 regards the time series data as a time series data matrix (e.g., matrix data X) in which two-dimensional positions and depths are taken in the row direction and time is taken in the column direction, and decomposes the time series data matrix into a spatial mode matrix, a singular value matrix, and a time mode matrix by singular value decomposition (SVD) as shown in Equation 2 in FIG. 4, thereby obtaining a spatial mode (e.g., matrix U 1:r ) is obtained. In this way, the spatial mode can be obtained by singular value decomposition (SVD).

[0037] Next, the processor 102 calculates the reduced spatial POD mode U of Equation 4 shown in FIG. 1:r , and U in Equation 5 shown in FIG. mod That is, U 1:r The spatial POD modes in the depth direction are stored in the row direction, but U mod To construct U, the spatial POD modes in the depth direction are arranged in the column direction. mod is a matrix with (nlat×nlon) rows and (r×ndepth) columns. mod is an example of a low-dimensional model, and is called a sensor candidate matrix in this embodiment. By rearranging in this way, the spatial POD modes at a certain latitude and longitude coordinate are arranged in one row while taking into account the depth direction, and a high-speed observation point optimization method, such as QR decomposition, can be applied.

[0038] When the spatial POD mode is obtained by Equation 2 shown in FIG. 4, the number of modes is the number of time direction data of the data matrix X, that is, m. When constructing a reduced-dimensional model by Equation 3 shown in FIG. 5, the processor 102 may terminate at r modes. However, if the computer resources are sufficient, the processor 102 does not terminate and performs subsequent processing using m spatial POD modes to obtain a transformed sensor candidate matrix U mod may be created.

[0039] According to the above, the processor 102 extracts a spatial mode (e.g., U 1:r Then, based on the spatial mode, the processor 102 obtains a modified spatial mode matrix (for example, a matrix U mod ) is being constructed.

[0040] Also, as described above, the processor 102 extracts a predetermined number of columns of a spatial mode matrix (e.g., a matrix U before reduction) that represents a spatial mode with two-dimensional position and depth in the row direction and time in the column direction, thereby generating a reduced-dimensional spatial mode matrix (e.g., a reduced-dimensional matrix U 1:r The processor 102 then transforms the low-dimensional spatial mode matrix so that the two-dimensional position is in the row direction and the time and depth are in the column direction to construct a transformed spatial mode matrix (e.g., matrix U mod In this way, by reducing the dimensionality of the spatial mode, it is possible to speed up the determination of observation points using optimization techniques.

[0041] Returning to the explanation of Figure 2.

[0042] Processor 102 sets conditions for performing observation point optimization (hereinafter referred to as optimization conditions) (step 203). The optimization conditions may be set by the user via input / output unit 110. For example, the number of observation points and an optimization target area are set as the optimization conditions. The number of observation points is set as the number of observation points. The range of latitude, longitude, depth, etc. for performing observation point optimization is set as the optimization target area.

[0043] The processor 102 performs the observation point position optimization process by calculating the sensor candidate matrix U mod For example, an observation point optimization method using QR decomposition is applied to determine the observation point position (step 204). The observation point position indicates the position of the observation point that is optimal for observing the ocean state among multiple observation points.

[0044] The observation point optimization method using QR decomposition is to optimize the sensor candidate matrix U mod The observation point optimization method is not limited to QR decomposition. For example, when the number of observation points is equal to or less than the number of low-dimensional modes, it is equivalent to QR decomposition. However, even when the number of observation points is greater than the number of low-dimensional modes, a determinant-based greedy method that can determine the optimal observation point or a determinant-based greedy method that takes into account the correlation of observation noise may be applied.

[0045] Row vectors are associated with coordinates on a two-dimensional map indicated by latitude and longitude. When a state vector is generated by looping through latitude and longitude in the order of equation 4 shown in Fig. 6, the latitudinal grid position lon can be calculated as i%nlon (% indicates a modulus operation) and the longitudinal grid position lat can be calculated as i / / nlon ( / / indicates an integer quotient) using the row position i of the row vector in equation 5 shown in Fig. 7.

[0046] Furthermore, as an observation point position optimization process in step 204, processor 102 finds the number of observation points equal to the number of observation points in the optimization conditions set in step 203, and converts each observation point into a grid position in the latitude and longitude directions. Processor 102 then associates the actual latitude and longitude information at the grid positions of the observation points to create tabular data 301 as shown in FIG. 8, and stores this in storage device 120 as observation point data 123.

[0047] As shown in FIG. 8, tabular data 301 has the following data items: ID, latitudinal element position, longitudinal element position, longitude, and latitude. ID is identification information of an observation point position. Latitude element position indicates the latitudinal grid position of an observation point position corresponding to an ID. Longitude element position indicates the longitudinal grid position of an observation point position corresponding to an ID. Latitude indicates the actual latitude of an observation point position corresponding to an ID (i.e., the actual latitude corresponding to the latitudinal element position). Longitude indicates the actual longitude of an observation point position corresponding to an ID (i.e., the actual longitude corresponding to the longitudinal element position).

[0048] According to the above, the processor 102 optimizes the deformation space mode matrix (e.g., matrix U mod ), a predetermined number of rows (e.g., r) are selected from the set of rows, and the two-dimensional position indicated by the selected rows is determined as the observation point. In this way, by acquiring spatial modes from time-series data, constructing a spatial mode matrix in which the depth direction is arranged in the column direction, and determining the observation point by selecting rows from the spatial mode matrix, it is possible to optimize the observation point using a high-speed method when observation values ​​can be acquired at multiple depths at the observation point.

[0049] Also, as described above, the processor 102 selects a predetermined number of rows (e.g., r) from the modified space mode matrix in descending order of row vector norm to set them as observation points. In this way, by selecting rows in descending order of row vector norm to set them as observation points, it is possible to quickly determine the observation points.

[0050] Processor 102 displays the observation point position optimization results on display 113 of input / output unit 110 as observation point position visualization processing (step 205).

[0051] FIG. 9 is a diagram showing an example of how a result of optimization of observation point positions according to the first embodiment is displayed.

[0052] A GUI 401 for displaying the observation point position optimization results displays map data, the observation point position optimization results, and setting items such as the display range and depth of the map data. For example, the GUI 401 includes a map display area 402, a display range setting area 403, an observation point position display area 404, and an observation point position 405.

[0053] At least a part of the map data 122 stored in the storage device 120 is displayed in the map display area 402. The range of the map data to be displayed in the map display area 402 may be adjusted by dragging the map in the map display area 402 with the mouse 112, or may be adjusted by inputting a value in the display range setting area 403.

[0054] Observation point position display area 404 displays tabular data 301 (see FIG. 8) corresponding to the observation point positions calculated in the observation point position optimization process in step 204 .

[0055] The observation point position 405 is displayed superimposed on the map data displayed in the map display area 402 .

[0056] The map data displayed in the map display area 402 may display, in addition to the observation point positions 405, the distribution of physical quantities of the oceanographic analysis data 121, and the results of statistical processing of the distribution of physical quantities. In addition, while Fig. 9 shows an example in which the depth to be visualized is specified and then visualized using two-dimensional map data, the observation point positions and the map data may be visualized in three dimensions.

[0057] (Embodiment 2) In the second embodiment, an observation plan planning support device 1 that visualizes the trade-off between observation cost and approximation error of ocean analysis data 121 and supports observation plan planning will be described.

[0058] Fig. 10 is a flowchart showing an example of observation point optimization processing according to embodiment 2. Next, with reference to Fig. 10, a process in which the observation plan development support device 1 calculates and presents a trade-off between observation cost and approximation error of the ocean analysis data 121 will be described.

[0059] The processor 102 performs the same processes as steps 201, 202, 203, and 204 shown in Fig. 2. This process has been described in the first embodiment, so a description thereof will be omitted here.

[0060] Next, as an observation point reduction process, in order to reduce the observation cost, processor 102 identifies observation point positions from the candidate observation point positions determined in the observation point position optimization process of step 204 that will have a small contribution to the prediction error when reduced (i.e., the prediction error is unlikely to become large), and removes those observation point positions that are determined to be so from the candidate observation point positions (step 501).

[0061] Next, the observation point reduction processing in step 501 will be described in detail with reference to Fig. 11. Fig. 11 is a flowchart showing an example of the observation point reduction processing according to the second embodiment.

[0062] Processor 102 initializes observation point set S to be subject to observation point reduction (step 601). As the initial set, for example, the observation point positions determined in the observation point position optimization process in step 204 are used. However, the initial set may also be the observation point positions that the user manually adds or removes from the observation point positions determined in the observation point position optimization process.

[0063] The processor 102 calculates the approximation error E, the observation cost, and the observation route of the observation point set S (step 602). The approximation error E is the error when the low-dimensional model constructed in step 202 is used to approximate the oceanographic analysis data 121 on which the low-dimensional model is created.

[0064] For the approximation, we use Equation 6 shown in FIG. 12, which is a modification of Equation 3 shown in FIG. 5. X hat (X^) is the approximation result of the data matrix, i.e., the state vector that represents the state of the ocean. 1:r is a matrix that arranges the reduced spatial POD modes, and z is a coefficient vector. Therefore, Equation 6 shown in Figure 12 is an equation that approximates the ocean state data by a linear combination of the spatial POD modes. The matrix U 1:r If the spatial POD modes of are known, the data matrix X can be approximated by appropriately determining z.

[0065] In order to obtain the coefficient vector z, if we assume that the original ocean analysis data 121 is observed at the observation point position obtained by the observation point position optimization process, the relationship between them is expressed by Equation 7 shown in Figure 13. Here, y is an observation vector that arranges the ocean analysis data 121 at the observation point position obtained by the observation point position optimization process, and H is an observation matrix that indicates the observation point position (a matrix that is 1 at the observation point position and 0 elsewhere). The approximate solution z of z hat (ẑ) can be calculated by Equation 8 shown in Fig. 14, assuming that the matrix C has full row rank or full column rank. Here, p is the number of observation points, r is the number of spatial POD modes of the reduced model, and C is the observation matrix H and the spatial POD mode U 1:r From these relationships, once the observation matrix H is determined, an approximate solution for the coefficient vector z can be obtained, and the matrix X that approximates the original ocean data matrix X can be obtained based on this. hat (X^) is found.

[0066] Therefore, the original ocean data matrix X and the approximate matrix X hat The error (approximation error) with respect to (X̂) can be evaluated. For example, the processor 102 calculates the error using Equation 9 shown in FIG. 15. Here, e is the approximation error, || F denotes the Frobenius norm.

[0067] The observation cost may be calculated based on, for example, time, travel distance, fuel cost, or observation equipment cost, or may be calculated by combining these. In this embodiment, the observation cost will be described as travel distance as an example. If the shortest travel distance required to visit all observation points obtained by one observation ship is taken as the observation cost, the shortest travel distance can be found by solving the shortest Hamilton path problem using dynamic programming or the like. Also, by adding the start point and end point of the route as observation points, a more practical observation route and observation cost can be calculated.

[0068] The processor 102 stores the approximation error E and the observation cost of the observation point set S calculated in step 602 (step 603).

[0069] Processor 102 assigns IDs to the observation points included in temporary observation point set S (step 604). The IDs are consecutive integer numbers starting from 1, with the number of observation points included in observation point set S as the maximum value. Also, the observation point with ID=i is designated as p[i].

[0070] The processor 102 initializes an internal counter i for reducing the number of observation points to 1 (step 605).

[0071] The processor 102 copies the contents of S to a temporary observation point set S' (step 606).

[0072] The processor 102 removes p[i] from the temporary observation point set S' (step 607).

[0073] The processor 102 assumes that observations are made at observation points included in the temporary observation point set S', and obtains the approximation error e[i] using Equation 9 shown in FIG. 15 from Equation 6 shown in FIG. 12 (step 608).

[0074] The processor 102 increments the internal counter i (step 609).

[0075] The processor 102 determines whether the internal counter i has exceeded the number of observation points in the temporary observation point set S' (step 610).

[0076] If the processor 102 determines that the internal counter i has exceeded the number of observation points in the temporary observation point set S' (step 610: YES), the processor 102 proceeds to step 611. If the processor 102 determines that the internal counter i has not exceeded the number of observation points in the temporary observation point set S' (step 610: NO), the processor 102 returns the process to step 606, and repeats the reduction of the observation points and the evaluation of the approximation error at that time.

[0077] By the processing from step 604 to step 609, the approximation error e[i] when the observation points are reduced one by one from the observation point set S is found.

[0078] Processor 102 sets i' as the value of i that minimizes the absolute value of the difference in approximation error e[i] when the observation points are reduced one by one from the approximation error E of observation point set S (step 611).

[0079] The processor 102 removes p[i'] from the observation point set S (step 612).

[0080] Through the processing of steps 611 and 612, the observation points can be reduced one by one from the original observation point set S so as to minimize the increase in approximation error.

[0081] The processor 102 determines whether the observation point set S is an empty set (step 613). If the processor 102 determines that the observation point set S is an empty set (step 613: YES), the processor 102 completes the observation point reduction process of step 501 shown in Fig. 10. If the processor 102 determines that the observation point set S is not an empty set (step 613: NO), the processor 102 returns the process to step 602.

[0082] As described above, for multiple sets with different numbers of observation points, processor 102 calculates the approximation error when approximating item values ​​of each grid point in the target area based on item values ​​at the observation points included in the set, and the cost (e.g., observation cost) required for observation at the observation points included in the set. This makes it possible to calculate the approximation error and cost (observation cost).

[0083] Also, as described above, processor 102 calculates the cost required for observation at an observation point (e.g., observation cost) using a set of observation points determined by selecting a predetermined number of rows, and repeatedly deletes from the set of observation points the observation point that causes the smallest change in the absolute value of the approximation error when deleted, and calculates the cost. This makes it possible to calculate the cost (observation cost) by deleting the observation point that causes the smallest change in the absolute value of the approximation error when deleted.

[0084] Returning to the explanation of Figure 10.

[0085] The processor 102 displays information on the observation point set, approximation error, observation cost, and observation route evaluated in the observation point reduction in step 501 on the display 113 of the input / output unit 110 as trade-off visualization, thereby providing support for observation plan creation (step 502).

[0086] FIG. 16 is a diagram showing a display example of trade-off visualization according to the second embodiment.

[0087] The processor 102 displays a GUI 701 for visualizing the trade-off regarding the approximation error and the observation cost shown in Fig. 16. The GUI 701 displays map data, the observation point position optimization results, the observation route, and a scatter plot of the trade-off between the approximation error and the observation cost. For example, the GUI 701 includes a map and observation route display area 702, an observation point position 703, an observation route 704, a candidate observation route display area 705, and a trade-off visualization area 706.

[0088] In the map and observation route display area 702, the observation point positions 703 and the observation route 704 with the minimum observation cost are displayed, superimposed on the map data 122 stored in the storage device 120. If the start point and end point of the route are set when evaluating the observation cost, the map and observation route display area 702 may also display the start point and end point of the route. The range of the map data displayed in the map and observation route display area 702 may be adjusted by dragging the map in the map and observation route display area 702 with the mouse 112, or by inputting a value into the display range setting area 403 (not shown in FIG. 12) as shown in FIG. 9.

[0089] As described above, the processor 102 displays on a map (e.g., the map data 122) a route (e.g., the observation route 704) that a moving body (e.g., an observation ship) follows to the observation points included in the set in order of the shortest travel distance. This allows the user to visually recognize the route on the map that is followed in order of the shortest travel distance.

[0090] The observation route candidate display area 705 displays the coordinates of the observation point positions 703, etc., in the order in which the observation vessel will patrol. In addition to the observation point positions 703, the observation route candidate display area 705 may also display the coordinates of the start point, end point, etc. of the route.

[0091] In the trade-off visualization area 706, the approximation error and observation cost for each observation point set evaluated in the observation point reduction in step 501 are displayed in the form of a scatter plot as shown in Fig. 16. Each plot in the scatter plot corresponds to a different observation point set. When the user uses the mouse 112 to pick a plot in the trade-off visualization area 706, the observation point positions 703 and observation routes 704 corresponding to the observation point set of that plot are displayed in the map and observation route display area 702, and the display contents of the observation route candidate display area 705 are updated accordingly.

[0092] As described above, the processor 102 changes the number of observation points included in the set and displays a graph (for example, a scatter plot displayed in the trade-off visualization area 706) plotting the approximation error against the cost. This allows the trade-off relationship between the cost and the approximation error to be visualized as a graph.

[0093] By using the observation plan formulation support device 1 according to the second embodiment, it becomes possible to formulate an observation plan that takes into account the approximation error, that is, the trade-off between the observation effect and the observation cost.

[0094] As described above, the processor 102 visually displays relationship information indicating the relationship between the observation points in the set or the number of observation points, the approximation error, and the cost as the GUI 701. This makes it possible to visually see how the approximation error and the cost change due to the deletion of an observation point, facilitating the creation of an observation plan. [Explanation of symbols]

[0095] 1...observation planning support device, 100...control device, 101...memory, 102...processor, 103...observation planning support program, 110...input / output unit, 111...keyboard, 112...mouse, 113...display, 120...storage device, 121...ocean analysis data, 122...map data, 123...observation point data, 401...GUI, 402...map display area, 403...display range setting area, 404...observation point position display area, 405...observation point position, 701...GUI, 702...map and observation route display area, 703...observation point position, 704...observation route, 705...observation route candidate display area, 706...cost trade-off visualization area

Claims

1. An observation plan creation support device that supports creation of an observation plan in which observation points are set on a horizontal plane of a target area and values ​​of target items are acquired at a plurality of depths at the observation points, A memory and a processor, the memory stores time series data representing item values ​​at each time of a target item at a lattice point represented by a two-dimensional position and a depth; The processor, A spatial mode is obtained from the time series data, the spatial mode being a mode of space represented by two-dimensional position and depth; Based on the spatial mode, a modified spatial mode matrix is ​​constructed in which two-dimensional positions are taken in the row direction, time and depth are taken in the column direction, and element values ​​of the spatial mode are arranged; selecting a predetermined number of rows from the modified space mode matrix by a predetermined optimization method, and determining the two-dimensional positions indicated by the selected rows as observation points; Observation planning support device.

2. the processor selects the predetermined number of rows from the modified space mode matrix in descending order of row vector norm, and sets the selected rows as the observation points.

2. The observation planning support device according to claim 1.

3. The processor, The time series data is regarded as a time series data matrix in which the two-dimensional position and the depth are in the row direction and the time is in the column direction, and the time series data matrix is ​​decomposed into the spatial mode matrix, the singular value matrix, and the time mode matrix by singular value decomposition, thereby obtaining the spatial mode.

2. The observation planning support device according to claim 1.

4. The processor, constructing a low-dimensional spatial mode matrix by extracting a predetermined number of columns of a spatial mode matrix representing the spatial modes, the rows being indicative of two-dimensional position and depth and the columns being indicative of time; constructing the transformed spatial mode matrix by transforming the low-dimensional spatial mode matrix so that two-dimensional positions are in the row direction and time and depth are in the column direction; 2. The observation planning support device according to claim 1.

5. The processor, calculating, for a plurality of sets each having a different number of observation points, an approximation error when approximating item values ​​of each grid point in the target region based on item values ​​at the observation points included in the set, and a cost required for observation at the observation points included in the set; visually displaying relationship information that indicates a relationship between the observation points in the set or the number of observation points in the set, the approximation error, and the cost; 2. The observation planning support device according to claim 1.

6. The processor, calculating costs required for observation at the observation points based on the set of observation points determined by selecting the predetermined number of rows; repeatedly removing an observation point from the set of observation points that causes the smallest change in the absolute value of the approximation error when the observation point is removed, and calculating the cost; generating the relationship information based on information obtained by repeating the calculation of the cost; 6. The observation planning support device according to claim 5.

7. The processor, displaying a graph plotting the approximation error against the cost while varying the number of observation points included in the set; 6. The observation planning support device according to claim 5.

8. The processor further comprises: displaying on a map a route that the mobile object follows through the observation points included in the set in order of the shortest travel distance; 6. The observation planning support device according to claim 5.

9. 1. A method for supporting the creation of an observation plan for providing an observation point on a horizontal plane of a target area and acquiring values ​​of a target item at a plurality of depths at the observation point, comprising: The computer Recording in memory time series data representing item values ​​at each time of the target item at a lattice point represented by the two-dimensional position and depth; A spatial mode is obtained from the time series data, the spatial mode being a mode of space represented by two-dimensional position and depth; Based on the spatial mode, a modified spatial mode matrix is ​​constructed in which two-dimensional positions are taken in the row direction, time and depth are taken in the column direction, and element values ​​of the spatial mode are arranged; selecting a predetermined number of rows from the modified space mode matrix by a predetermined optimization method, and determining the two-dimensional positions indicated by the selected rows as observation points; A method for supporting the creation of observation plans to achieve this.

10. An observation plan creation support program for supporting creation of an observation plan for setting observation points on a horizontal plane of a target area and acquiring values ​​of target items at a plurality of depths at the observation points, the program comprising: On the computer, Recording in memory time series data representing item values ​​at each time of the target item at a lattice point represented by the two-dimensional position and depth; A spatial mode is obtained from the time series data, the spatial mode being a mode of space represented by two-dimensional position and depth; Based on the spatial mode, a modified spatial mode matrix is ​​constructed in which two-dimensional positions are taken in the row direction, time and depth are taken in the column direction, and element values ​​of the spatial mode are arranged; selecting a predetermined number of rows from the modified space mode matrix by a predetermined optimization method, and determining the two-dimensional positions indicated by the selected rows as observation points; This is an observation planning support program to help achieve this.

Images