River data predictor variable selection device, river data predictor variable selection method and program

Abstract

To provide a river data predictor variable selection device with which variables for predicting a river state can be properly selected.SOLUTION: A river data predictor variable selection device 100 includes: a spot relevance calculation unit 120 which, according to river information D11 which indicates connecting states of multiple rivers, observation point information D12 which indicates positional information of multiple observation points provided in the multiple rivers, and observation information D20 which is time-series data of observation target quantities at the multiple observation points, calculates relevance data D40 which indicates mutual relations between multiple spots including any of the observation points; and a predictor variable selection unit 140 which selects a predictor variable(s) corresponding to a prediction point Y and a prediction time T which have been specified on the basis of the relevance data D40.SELECTED DRAWING: Figure 4

Description

[Technical field]

[0001] The present invention relates to a river data predictor variable selection device, a river data predictor variable selection method, and a program. [Background technology]

[0002] As background art in this technical field, the abstract of the following Patent Document 1 states, "The present invention is a data analysis method targeted at data related to variables having temporal precedence, and includes an extraction step of sparse conditional independent relationships suitable for constructing a causal relationship related to a specific variable that exists between variables by applying a regression method that can select variables while taking into account the precedence between variables. The extraction step can be performed by a selection step in which an explanatory variable is selected from among the variables that precede a set objective variable, and the selected explanatory variable is reset as a new objective variable and a new explanatory variable is reselected from among the variables that precede the new objective variable. In this way, it is possible to clarify accurate conditional independent relationships and, therefore, causal relationships that exist between each variable." [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2019-191634 A Summary of the Invention [Problem to be solved by the invention]

[0004] Incidentally, Patent Document 1 does not specifically mention how to appropriately select variables for predicting river conditions. This invention has been made in consideration of the above-mentioned circumstances, and aims to provide a river data prediction variable selection device, a river data prediction variable selection method, and a program that can appropriately select variables for predicting river conditions. [Means for solving the problem]

[0005] In order to solve the above problems, the river data prediction variable selection device of the present invention is characterized in that it comprises a point relevance calculation unit that calculates association data that represents the relationship between multiple points including any of the observation points based on river information that represents the connection state of multiple rivers, observation point information that represents the location information of multiple observation points installed on multiple rivers, and observation information that is time series data of the observed quantities at the multiple observation points, and a prediction variable selection unit that selects a prediction variable corresponding to a specified prediction point and prediction time based on the association data. Effect of the Invention

[0006] According to the present invention, variables for predicting river conditions can be appropriately selected. [Brief description of the drawings]

[0007] [Figure 1] FIG. 2 is a diagram showing an example of river network information. [Diagram 2] FIG. 2 is a diagram showing an example of the arrangement of observation points. [Diagram 3] FIG. 11 is a diagram illustrating an example of observation information. [Figure 4] FIG. 1 is a block diagram of a river data predictor variable selection device according to a first embodiment. [Diagram 5] FIG. 1 is a block diagram of a computer. [Figure 6] 13 is a flowchart of a location association processing routine. [Figure 7] 13 is a flowchart of a predictor variable selection process routine. [Figure 8] 11A and 11B are schematic diagrams showing a specific example of a process for calculating branch point data. [Figure 9] FIG. 11 is a schematic diagram showing a specific example of a process for calculating association data. [Figure 10] FIG. 10 is a diagram illustrating a process for determining a predictor variable. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] [Overview of the embodiment] In river and dam management, the flow rate or water level at various points is predicted. To this end, data from each observation point, such as a water level gauge or rain gauge, is used, and variables are set in consideration of the flow delay time between the point where the flow rate or water level is to be obtained and each observation point, and the relationship between rivers, to predict the flow rate or water level at the desired point. As a variable selection method for data with a time sequence, for example, the contents of the above-mentioned Patent Document 1 can be applied. In other words, by applying Patent Document 1, it is considered possible to target data with a time sequence and clarify causal relationships using a regression method to select explanatory variables.

[0009] However, in the above-mentioned technology applying Patent Document 1, when selecting variables for predicting river data (flow rate, water level, etc.), simply using the time series data of the river may result in the selection of variables with a weak causal relationship. In addition, when simply comparing the time series data of a certain river, for example, a problem may occur in that data in a tributary is not selected as a variable with no correlation to a prediction point in the main river. Therefore, in the embodiment described later, the river network information and observation information, the position information of the prediction point, and the prediction target time are used to automatically select variables to be used in the river data prediction model. As a result, according to the embodiment described later, it is possible to automatically select variables to be used in the river data prediction model while taking into account the network relationship of the river.

[0010] [First embodiment] The first embodiment will be described below based on the accompanying drawings. The following description and drawings are examples for explaining the first embodiment, and are omitted and simplified as appropriate for clarity of explanation. The present invention can be implemented in various other forms. Unless otherwise specified, each component may be singular or plural. In order to facilitate understanding of the embodiment, the position, size, shape, range, etc. of each component shown in the drawings may not represent the actual position, size, shape, range, etc. Therefore, the present embodiment is not necessarily limited to the position, size, shape, range, etc. disclosed in the drawings. In addition, when there are multiple components having the same or similar functions, they may be described by using the same reference numerals with different subscripts, but when there is no need to distinguish between these multiple components, the subscripts may be omitted.

[0011] Data Structure First, the data structure applied in the first embodiment will be described. FIG. 1 is a diagram showing an example of the river network information D10. The river network information D10 includes river information D11 for a plurality of rivers to be predicted, and observation point information D12. River information D11 indicates the connection state of multiple rivers belonging to either of the river systems 30, 40. In the following description, a "river system" refers to a group of rivers that do not intersect with each other. In the illustrated example, the river systems 30, 40 do not intersect with each other. The rivers in the river system 30 include a main river 32 (river) and a tributary 34 (river). In addition, the rivers in the river system 40 include a main river 42 (river) and tributaries 44 to 49 (rivers).

[0012] The dashed arrows along each river indicate the direction of flow of each river, which is from left to right in the figure. In the river systems 30 and 40, there are points where multiple rivers branch off or merge. These points are collectively called branch points. In FIG. 1, the river system 30 has branch points a m1 (point). In addition, the river system 40 includes the branch point b n1 ~b n4 In this way, the river information D11 is information that indicates the shape of each river, the river system to which it belongs, the flow direction, whether it is a main river or a tributary, the location of the branch point, and the like.

[0013] In the example of FIG. 2, which will be described later, the water system 30 has an observation point a 1 (Point), Observation point b in river system 40 1 ~b 8Each observation point is provided with measuring instruments (not shown) such as a water level meter and a flow meter. As shown in FIG. 1, the observation point information D12 includes 1 ,b 1 ~b 8 For each of the observation points, the map includes the river system (30 or 40) to which the observation point belongs and the location information of the installation location of the observation point. Here, the location information includes the latitude, longitude, and altitude of each observation point. However, the location information is not limited to that shown in the figure, as long as it can identify the location of each observation point.

[0014] FIG. 2 is a diagram showing an example of the arrangement of observation points. In FIG. 2, the water system 30 includes an observation point a 1 In addition, the water system 40 has an observation point b 1 ~b 8 As mentioned above, each observation point a 1 ,b 1 ~b 8 Measuring instruments such as water level gauges and flow meters (not shown) are installed in the observation points. As a result, water level H(t) and flow rate Q(t), which are time-series information on the water level H and flow rate Q, are measured at each observation point. The water level H(t) and flow rate Q(t) at these observation points are collectively referred to as observation information D20.

[0015] FIG. 3 is a diagram showing an example of the observation information D20. Observation information D20 is for all observation points a 1 ,b 1 ~b 8 (See Fig. 2) 1 Observation information D20-a 1 Only the above is illustrated. In FIG. 3, observation information D20-a 1 includes observation information for the observation date and time. The observation information includes the water level H and the flow rate Q. However, the observation information D20 is not limited to that shown in the figure, and may include other information.

[0016] Configuration of the First Embodiment FIG. 4 is a block diagram of a river data predictor variable selection device 100 according to the first embodiment. In Figure 4, the river data predictive variable selection device 100 (hereinafter sometimes simply referred to as device 100) includes a data memory unit 110, a point relevance calculation unit 120 (point relevance calculation step, point relevance calculation means), an input unit 130, and a predictive variable selection unit 140 (predictive variable selection means, predictive variable selection step).

[0017] The input unit 130 inputs a prediction point Y and a prediction time T to the prediction variable selection unit 140 based on an operation by a user or the like. Here, the prediction point Y is a point to be predicted for the flow rate Q and / or water level H. Furthermore, the prediction time T indicates how many hours from the present time the flow rate Q and / or water level H at the prediction point Y is predicted to be. The data storage unit 110 stores river network information D10 and the above-mentioned observation information D20. As shown in FIG. 1, the river network information D10 includes river information D11 and observation point information D12.

[0018] The point relevance calculation unit 120 calculates branch point data D30 and association data D40, and supplies them to the predictive variable selection unit 140 together with the observation information D20. Here, the branch point data D30 is data calculated for a branch point that is not an observation point, and is time-series data having items corresponding to the observation information D20 (see FIG. 3). The association data D40 is data representing the relationship between each observation point and each branch point. The predictive variable selection unit 140 selects a predictive variable used to predict the flow rate Q and / or water level H at the prediction point Y at the prediction time T. Here, the predictive variable is a part of the observation information D20 or the branch point data D30. Other specific functions of each unit in the device 100 will be described later.

[0019] 5 is a block diagram of a computer 900. The apparatus 100 shown in FIG. 4 includes a computer 900 shown in FIG. 5, the computer 900 includes a CPU 901, a RAM 902, a ROM 903, a HDD 904, a communication I / F 905, an input / output I / F 906, and a media I / F 907. The communication I / F 905 is connected to a communication circuit 915. The input / output I / F 906 is connected to an input / output device 916.

[0020] A media I / F 907 reads and writes data from a recording medium 917. A ROM 903 stores a control program executed by the CPU, various data, etc. The CPU 901 executes an application program loaded into the RAM 902 to realize various functions. The inside of the device 100 shown in Fig. 4 above is a block diagram showing the functions realized by the application program, etc.

[0021] Operation of the First Embodiment 6 is a flowchart of a location relevance processing routine. This routine is executed in the location relevance calculation unit 120 (see FIG. 4). When the process proceeds to step S12 in FIG. 6, the point relevance calculation unit 120 calculates branch point data D30. Here, the branch point data D30 has the same data structure as the observation point information D12 (see FIG. 3), and the branch point a m1 ,b n1 ~b n4 (See Figure 1) This is time series data listing estimated values ​​of flow rate and water level.

[0022] The branch point data D30 can be estimated, for example, as follows. First, observation information D20 at one or more observation points located upstream of the branch point is synthesized while shifting the time. Next, this synthesis result is compared with observation information D20 at an observation point located downstream of the branch point to determine the degree of correlation between the two. In this manner, the shift time at which the degree of correlation is highest is determined. Next, the shift time at the branch point is estimated based on the position information of the upstream observation point, the downstream observation point, and the branch point. The branch point data D30 can be estimated based on the result of shifting the upstream observation information D20 by the estimated shift time.

[0023] In addition, among the branching points, for example, branching point b shown in Fig. 1 n2 There are also branching points where a river branches off, as shown in the figure. For this type of branching point, the point relevance calculation unit 120 also estimates the flow rate Q(t) and water level H(t) of each branch immediately after the branching. That is, the flow rate Q(t) and water level H(t) of each branch immediately after the branching are also included in the branching point data D30. The flow rate Q(t) and water level H(t) of each branch immediately after the branching can be estimated, for example, according to the observation information D20 of an observation point upstream of the branching point and a plurality of observation points downstream of the branching point, and the distance between these observation points and the branching point. However, the estimation method of the branching point data D30 is not limited to the above. A specific example of the process of estimating the branching point data D30 will be described later.

[0024] Next, when the process proceeds to step S14, the location relevance calculation unit 120 calculates the relevance data D40. Here, the relevance data D40 is 1 ,b 1 ~b 8 And branch point a m1 ,b n1 ~b n4The association data D40 includes a "downstream delay time" that is the time it takes for water to flow from a certain point upstream (e.g., an observation point) to another point downstream (e.g., a branch point), and a "degree of mutual association" that is the relationship between time-series information between points. Note that the "degree of mutual association" can be, for example, a correlation coefficient, but is not limited to this. The "downstream delay time" is, for example, the time when the "degree of mutual association" is maximum.

[0025] FIG. 7 is a flowchart of a predictor variable selection process routine executed by the predictor variable selection unit 140. 7, when the process proceeds to step S22, the predictor variable selection unit 140 acquires the branch point data D30 and the association data D40 from the point relevance calculation unit 120. Furthermore, the predictor variable selection unit 140 acquires the predicted time T and the predicted point Y from the input unit 130.

[0026] Next, when the process proceeds to step S24, the predictor variable selection unit 140 selects a group of points that are in the same river system as the predicted point Y from the river network information D10 based on the position information of the predicted point Y input from the input unit 130. The selected group of points is called a same river system point group GA. For example, if the predicted point Y belongs to the river system 40 (see FIG. 2), the same river system point group GA includes the observation point b 1 ~b 8 and branch point b n1 ~b n4 The following will be included:

[0027] Next, when the process proceeds to step S26, the predictor variable selection unit 140 judges whether the predicted point Y is included in the same river system point group GA. For example, if the predicted point Y is included in the river system 40, the predictor variable selection unit 140 judges whether the predicted point Y is included in the observation point b 1 ~b 8 Or branch point b n1 ~b n4 It is determined whether it matches any of the following.

[0028] If the result of step S26 is "No", the process proceeds to step S28. Here, the predictor variable selection unit 140 calculates the prediction point information D50 based on the position information of the prediction point Y and each position information of the same river system point group GA. The prediction point information D50 includes the downstream delay time from the nearest upstream point included in the same river system point group GA to the prediction point Y, and the mutual association degree. The calculation method can adopt the method described in step S12 (see FIG. 6). In addition, for example, the mutual association degree between other observation points based on an adjacent upstream observation point can be used as the mutual association degree at the prediction point Y. However, the method of calculating the mutual association degree at the prediction point Y is not limited to these methods.

[0029] Next, when the process proceeds to step S30, the prediction variable selection unit 140 selects from the group of points in the same river system GA points that have usable data for the flow rate Q and / or water level H at the prediction point Y when the prediction time T has passed. A set of one or more selected points is called the group of related points GB. Here, "usable" means that data for the water level H and / or flow rate Q exists and can be used as variables even if the time is shifted back by the downstream delay time from the prediction time T.

[0030] Next, when the process proceeds to step S32, the predictor variable selection unit 140 selects an application point group GC from the associated point group GB. The application point group GC is selected as follows. First, the predictor variable selection unit 140 identifies branches (main river and / or tributaries) within the range connecting each point of the associated point group GB and the prediction point Y. Then, it selects the point group (observation point and / or branch point) most downstream in each branch. This selected point group is the application point group GC.

[0031] Next, when the process proceeds to step S34, the predictor variable selection unit 140 determines predictor variables. The procedure will be described. First, the predictor variable selection unit 140 selects one time from the observation information D20 or the branch point data D30 for the application point group GC that is usable in response to the prediction time T and has the maximum degree of correlation. Here, the meaning of "usable" is the same as that explained in step S30 above, and indicates that even if the prediction time T is shifted back by the downstream delay time, data on the water level H and / or the flow rate Q exists and can be used as a variable. For example, assume that the prediction time T is "3 hours" and the downstream delay time between the prediction point Y and the upstream observation point is "4 hours". In this case, data from one hour before the current time at the upstream observation point is required, but since the data exists, it is "usable".

[0032] Also, assume that the application point group GC includes a branch point upstream of the prediction point Y. As described above, the branch point data D30 at the branch point is estimated based on the observation information D20 at the observation point located further upstream. Therefore, the branch point data D30 also includes data corresponding to the downstream delay time from these upstream observation points to the branch point. As a result, even if the prediction time T is longer than the downstream delay time from the branch point to the prediction point Y, the branch point data D30 may be "usable". In this way, in step S34 of FIG. 7, a flow rate Q and / or a water level H at one time t is selected in one or more observation information D20 and / or branch point data D30 upstream of the prediction point Y. The prediction variable selection unit 140 outputs the selected flow rate Q and / or water level H as a prediction variable.

[0033] FIG. 8 is a schematic diagram showing a specific example of the process of calculating the branch point data D30 (step S12 in FIG. 6). In FIG. 8, the branch point b of the water system 40 n1 The procedure for calculating the flow rate at branch point b is explained below. n1 is not an observation point, so branch point b n1The flow rate at the upstream observation point b 1 ,b 2 Flow rate Q b1 ,Q b2 Based on this, the branch point b n1 Predicted flow rate Q bn1 More specifically, the point relevance calculation unit 120 calculates a predicted flow rate Q, which is time-series data, based on the following formula (1): bn1 Calculate (t).

[0034] Q bn1 (t)=k 1 Q b1 (tS b1,bn1 )+k 2 Q b2 (tS b2,bn1 ) …Formula (1)

[0035] In equation (1), Q b1 ,Q b2 is the observation point b 1 ,b 2 is the flow rate at S b1,bn1 is observation point b 1 From branch point b n1 is the flow delay time until b2,bn1 is observation point b 2 From branch point b n1 The delay time from the flow to the 1 ,k 2 is observation point b 1 ,b 2 is the weighting factor for Predicted flow rate Q bn1 The flow delay time S b1,bn1 ,S b2,bn1 For example, it can be assumed that only the short side of "S b1,bn1 >S b2,bn1 If " is true, the predicted flow rate Q bn1 is the current time t now "t" in the future now +S b2,bn1 " can be considered to have been obtained.

[0036] By the way, in the main river 42 shown in Figure 8, the branch point b n1 Downstream of observation point b 4 The observation point b 4 Flow rate Q b4 This allows the flow rate Q b1 ,Q b2 ,Q b4 and observation point b 1 ,b 2 ,b 4 Location information of branch point b n1 Based on the position information, the flow delay time S in the above formula (1) is calculated. b1,bn1 ,S b2,bn1 The procedure is explained below.

[0037] First, the weighting coefficient k in Eq. (1) 1 ,k 2 Assume that is "1". Next, at observation point b 1 ,b 2 Flow rate Q b1 ,Q b2 Next, we compare the synthetic data with the flow rate Q b4 Then, the degree of correlation between the observation points and the composite pattern having the maximum degree of correlation is calculated. Then, the downstream delay time S in Equation (1) is calculated based on the determined composite pattern and the position information of each observation point. b1,bn1 ,S b2,bn1 etc. can be estimated.

[0038] For example, in the "composite pattern with the maximum degree of correlation", the flow rate Q b1 ,Q b2 There is a 2 hour difference between Q b1 (t-2)+Q b2 Flow rate Q for (t) b4 Assume that the flow delay time of (t) is 7 hours. Also, at observation point b 1 and branch point b n1 The distance between b1,bn1 (not shown), and observation point b 2 and branch point b n1 The distance between b2,bn1(not shown), and branch point b n1 and observation point b 4 The distance between bn1,b4 (not shown) when "L b1,bn1 / L bn1,b4 =2", "L b2,bn1 / L bn1,b4 = 4 / 3". If we assume that the flow speed of each river is constant, the ratio of the downstream delay times will be equal to the ratio of the distances.

[0039] Therefore, the flow delay time S b1,bn1 is "6 hours", and the flow delay time S b2,bn1 It can be estimated that the time is "4 hours". Also, at the branch point b n1 From observation point b 4 Flow delay time S bn1,b4 In this way, the flow delay time S b1,bn1 ,S b2,bn1 Once is estimated, the branch point b is determined by the above formula (1). n1 Predicted flow rate Q bn1 (t) can be calculated. Also, the predicted flow rate Q bn1 As in (t), branch point b n1 Predicted water level H bn1 (t) can also be calculated. However, the predicted flow rate Q bn1 (t) and predicted water level H bn1 The method of estimating (t) is not limited to the above.

[0040] FIG. 9 is a schematic diagram showing a specific example of the process of calculating the relationship data D40 (step S14 in FIG. 6). The table 202 shown in FIG. 9 is a table that stores the downstream delay time between each point in the river system 40. The point relevance calculation unit 120 stores the downstream delay time S b1,bn1 ,S b2,bn1 Similarly, the other downstream delay times are calculated and a table 202 is created. Furthermore, the point relevance calculation unit 120 also calculates the mutual relevance between the points for which the downstream delay times have been calculated. As an example, the graph 204 shown in FIG. 9 shows n1 and observation point b 4The correlation between the two is shown. For the downstream delay time (3 hours in the illustrated example), the delay time at the maximum correlation (max) is used.

[0041] Here, the correlation coefficient obtained for the observation information D20 of two points within a certain period of time can be used as the correlation degree. Here, the correlation degree may be constant regardless of the flow rate, but it is more preferable to determine the correlation degree according to the flow rate. In other words, it is preferable to determine the correlation degree so that the larger the flow rate, the shorter the downstream delay time. For example, it is preferable to use several flow rate thresholds to divide the observation information D20 by the magnitude of the flow rate, and calculate the correlation degree at each flow rate according to the divided observation information D20. The same applies to the water level, and it is preferable to determine the correlation degree so that the higher the water level, the shorter the downstream delay time. This makes it possible to apply a correlation degree suited to the state of the river, and to determine the predictive variables more appropriately.

[0042] FIG. 10 is a diagram showing a schematic diagram of the process for determining the predictor variables. In the river system 40 of FIG. 10, the predicted point Y 1 ,Y 2 Here, the predicted location Y 1 is the observation point b 5 It is consistent with the predicted point Y 2 is a position where there is no observation point. In the example described below, the flow rate Q is used as the prediction target and the prediction variable. However, other river data such as the water level H may be used as the prediction target and the prediction variable.

[0043] In FIG. 10, table 10Y1 is a prediction point Y 1 10Y1 is a table showing the prediction variables for the prediction time T of 5 hours, 6 hours, 10 hours, and 13 hours. 2 10Y1 and 10Y2 are tables showing predictive variables when the prediction time T is 1 hour, 2 hours, 6 hours, and 9 hours. The reason why the predictive variables are determined as shown in Tables 10Y1 and 10Y2 will be explained below.

[0044] (Y 1 (Prediction for 5 hours later) First, the predicted point Y after 5 hours 1 In step S24 (see FIG. 7) described above, all points belonging to the river system 40 are selected as the same river system point group GA. 1 ~b 8 and branch point b n1 ~b n4 In the example shown in FIG. 1 is observation point b 5 7, step S28 is skipped, and the related point group GB is selected in the next step S30. In the above example, when the prediction time T of 5 hours has elapsed, the points that have data that can be used in consideration of the downstream delay time are the observation point b 1 ~b 4 and branch point b n1 Therefore, these points are selected as the relevant point group GB.

[0045] Next, in step S32 of FIG. 7, the application point group GC is selected. In the above example, the application point group GC is the observation point b 4 Next, in step S34 of FIG. 7, the predictor variables are determined. In the water system 40 shown in FIG. 10, the observation point b 4 From predicted point Y 1 (i.e., observation point b 5 ) The delay time to the forecast point Y is 5 hours. 1 Predicted flow rate Q after 5 hours Y1 The predictor variables selected to calculate (t+5) are the current observation point b 4 Flow rate Q b4 It becomes (t).

[0046] (Y 1 (Prediction for 6 hours after) Next, the predicted location Y after 6 hours 1 We will consider the following: In this case, the same river system point group GA as in the case of "5 hours later" is selected in step S24 of FIG. 7. However, when the prediction time T of 6 hours has passed, the points that have usable data in consideration of the downstream delay time are observation point b 1 ~b 3 and branch point b n1 and observation point b 4 Therefore, in step S30, the data of observation point b 1 ~b 3 and branch point b n1 is selected as the relevant point group GB.

[0047] Next, in step S32 of FIG. 7, the observation point b located at the most downstream side among the branches to which the related point group GB belongs is 3 and branch point b n1 is selected as the application point group GC. In the water system 40 shown in FIG. 10, the observation point b 3 From predicted point Y 1 The flow delay time to the branch point b is 8 + 5 = 13 hours. n1 From predicted point Y 1 The flow delay time to observation point b is 3 + 5 = 8 hours. 3 Flow rate Q 7 hours before b3 (t-7) and branch point b n1 Predicted flow rate Q 2 hours ahead bn1 (t-2) and is the predicted point Y 1 Predicted flow rate Q after 6 hours Y1 It is selected as a predictor variable for calculating (t+6).

[0048] (Y 1 (Prediction for 10 hours later) Next, the predicted point Y after 10 hours 1 When the 10-hour prediction time T has passed, the only points that have usable data in consideration of the downstream delay time are observation point b 1 ~b 3 and branch point b n1 These are selected as the related point group GB. In other words, the related point group GB is the same as when the forecast time T is 6 hours. n1The reason for the inclusion is that the branch point b n1 The data is from the upstream observation point b 1 ,b 2 The data is synthesized, and the future forecast flow rate Q bn1 This is because

[0049] Next, in step S32 of FIG. 7, the observation point b located at the most downstream side among the branches related to the related point group GB is 3 and branch point b n1 is selected as the application point group GC. In other words, the application point group GC is also the same as when the prediction time T is 6 hours. As described above, the observation point b 3 From predicted point Y 1 The flow delay time to the branch point b is 13 hours. n1 From predicted point Y 1 The flow delay time to observation point b is 8 hours. 3 Flow rate Q 3 hours before b3 (t-3) and branch point b n1 Predicted flow rate Q after 2 hours bn1 (t+2) and is the predicted point Y 1 Predicted flow rate Q after 10 hours Y1 It is selected as a predictor variable for calculating (t+10).

[0050] (Y 1 (Prediction for 13 hours later) Next, the predicted point Y after 13 hours 1 When the 13-hour forecast time T has passed, the only points that have usable data in consideration of the downstream delay time are observation point b 1 ,b 3 These are selected as the related point group GB. Next, in step S32 of FIG. 7, the observation point b 1 ,b 3 is selected as the application point group GC.

[0051] As shown in the figure, observation point b 3 From predicted point Y 1 The flow delay time to observation point b is 13 hours.1 From predicted point Y 1 The flow delay time to observation point b is 14 hours. 3 The current flow rate Q b3 (t) and observation point b 1 Flow rate Q one hour before b1 (t-1) and the predicted point Y 1 Predicted flow rate Q after 13 hours Y1 It is selected as a predictor variable for calculating (t+13).

[0052] (Y 2 (Prediction for one hour later) Next, the predicted location Y one hour later 2 In step S24 (see FIG. 7), all points belonging to the river system 40 are selected as a group of points in the same river system GA. 2 Since "observation point b" is not included in the same river system point group GA, the process proceeds to step S28. 4 From predicted point Y 2 "Distance to" and "Predicted point Y 2 From observation point b 5 The ratio of the distance to observation point b is 1:4. 4 From observation point b 5 Flow delay time S b4,b5 In step S28 of FIG. 7, the predictor variable selection unit 140 selects the downstream delay time S b4,b5 Based on the ratio of distances, observation point b 4 From predicted point Y 2 Flow delay time S b4,Y2 The predictor variable selection unit 140 calculates that the time required for the observation point b 4 The correlation coefficient between the upstream and the predicted point Y is shifted by one hour. 2 This is used as the degree of correlation with

[0053] When the 1-hour prediction time T has passed, the only point that has usable data in consideration of the downstream delay time is observation point b. 1 ~b 4 and branch point b n1Therefore, in the next step S30, these points are selected as the related point group GB. Next, in step S32 of FIG. 7, the application point group GC is selected. In the above example, the application point group GC is selected from the observation point b 4 Next, in step S34 of FIG. 7, the predictor variables are determined. 4 From predicted point Y 2 The flow delay time to the predicted point Y was estimated to be 1 hour. 2 The predicted flow rate Q after 1 hour Y2 The predictor variables selected to calculate (t+1) are the current observation point b 4 Flow rate Q b4 It becomes (t).

[0054] (Y 2 (Prediction for 2 hours later) Next, the predicted location Y after 2 hours 2 When the 2-hour prediction time T has passed, the only points that have usable data in consideration of the downstream delay time are observation point b 1 ,b 2 ,b 3 and branch point b n1 Therefore, in step S30 of FIG. 7, these points are selected as a group of related points GB. Next, in step S32, the observation point b 3 and branch point b n1 is selected as the application point group GC. Next, in step S34 of FIG. 7, the predictor variables are determined. 4 From predicted point Y 2 The flow delay time to the predicted point Y was estimated to be 1 hour. 2 The predicted flow rate Q after 2 hours Y2 The predictor variables selected to calculate (t+2) are the predicted flow rate Q bn1 (t-2) and the flow rate Q 7 hours ago b3 The result is (t-7).

[0055] (Y 2 (Prediction for 6 hours after) Next, the predicted location Y after 6 hours 2 When the 6-hour prediction time T has passed, the only points that have usable data in consideration of the downstream delay time are observation point b 1 ,b 2 ,b 3 and branch point b n1 These are selected as the related point group GB. As mentioned above, the branch point b n1 The data is from the upstream observation point b 1 ,b 2 The data is synthesized, and the future forecast flow rate Q bn1 Next, in step S32 of FIG. 7, the observation point b located at the most downstream side of the related point group GB is obtained. 3 and branch point b n1 is selected as the application point group GC. Next, in step S34 of FIG. 7, the predictor variables are determined. 4 From predicted point Y 2 The flow delay time to the predicted point Y was estimated to be 1 hour. 2 Predicted flow rate Q after 6 hours Y2 The predictor variables selected to calculate (t+6) are the predicted flow rate Q bn1 (t+2) and the flow rate Q three hours ago b3 The result is (t-3).

[0056] (Y 2 (Prediction for 9 hours later) Next, the predicted location Y after 9 hours 2 When the 9-hour forecast time T has passed, the only points that have usable data in consideration of the downstream delay time are observation point b 1 ,b 3 These are selected as the related point group GB. Next, in step S32 of FIG. 7, the observation point b 1 ,b 3 is selected as the application point group GC. Next, in step S34 of FIG. 7, the prediction variables are determined. 4 From predicted point Y 2The flow delay time to the predicted point Y was estimated to be 1 hour. 2 The predicted flow rate Q after 9 hours Y2 The predictor variables selected to calculate (t+9) are the flow rate Q b1 (t-1) and the current flow rate Q b3 (t) and so on.

[0057] [Effects of the embodiment] As described above, according to the embodiment, the river data predictor variable selection device (100) selects river information (D11) representing the connection state of a plurality of rivers (32, 34, 42, 44 to 49) and a plurality of observation points (a 1 ,b 1 ~b 8 ) and multiple observation points (a 1 ,b 1 ~b 8 Based on the observation information (D20) which is time series data of the observation target quantity (Q(t),H(t)) at any observation point (a 1 ,b 1 ~b 8 ) including multiple points (a 1 ,b 1 ~b 8 ,a m1 ,b n1 ~b n4 ), and a prediction variable selection unit (140) that selects a prediction variable corresponding to a specified prediction location (Y) and prediction time (T) based on the correlation data (D40).

[0058] This makes it possible to predict the predictive variables corresponding to the specified prediction point (Y) and prediction time (T) based on the association data (D40), and therefore to appropriately select predictive variables for predicting the state of the river. More specifically, it is possible to automatically select predictive variables to be used in the river data prediction model while taking into account the network relationships of the river.

[0059] The river information (D11) further includes a river system (30, 40) to which the multiple rivers (32, 34, 42, 44 to 49) belong and the flow directions of the multiple rivers (32, 34, 42, 44 to 49). The observation point information (D12) further includes multiple observation points (a 1 ,b 1 ~b 8 It is more preferable to include the river system (30, 40) to which the river belongs. This allows for a more appropriate selection of predictor variables depending on the river system to which the river belongs and the direction of the river flow.

[0060] In addition, the relational data (D40) is a data set of a plurality of points (a 1 ,b 1 ~b 8 ,a m1 ,b n1 ~b n4 It is more preferable to include the correlation degree of the flow rate (Q(t)) or water level (H(t)) between the above-mentioned three predictors. This allows for more appropriate selection of predictor variables corresponding to the flow rate (Q(t)) or water level (H(t)).

[0061] In addition, the point relevance calculation unit (120) 1 ,b 1 ~b 8 ) other than the branch point (a m1 ,b n1 ~b n4 It is more preferable that the system further has a function of calculating branch point data (D30) which is time series data of the estimated values ​​of the observation target quantities (Q(t),H(t)) at the branch point (a m1 ,b n1 ~b n4 ) state, leading to better selection of predictor variables.

[0062] In addition, the predictor variable selection unit (140) selects a point (a 1 ,b 1 ~b 8 ,a m1 ,b n1 ~b n4It is more preferable to have a function of calculating the degree of correlation between the observed quantities (Q(t),H(t)) between the observation point (t) and the prediction point (Y). This makes it possible to more appropriately select the prediction variables based on the calculated degree of correlation.

[0063] In addition, the predictor variable selection unit (140) selects one or more points (a) having available data corresponding to the prediction time (T). 1 ,b 1 ~b 8 ,a m1 ,b n1 ~b n4 It is even more preferred to have the ability to select a set of relevant locations (GB) that are 100% of the total locations. This allows for a better selection of predictors based on locations that have available data.

[0064] In addition, the predictor variable selection unit (140) selects one or more points (a 1 ,b 1 ~b 8 ,a m1 ,b n1 ~b n4 It is more preferable to further have a function of selecting an application point group (GC) that is a group of application points (GC) that is a prediction point (Y). This allows points close to the prediction point (Y) to be selected as the application point group (GC), thereby making it possible to select prediction variables more appropriately.

[0065] In addition, the predictor variable selection unit (140) selects the predictor variables for each point (a 1 ,b 1 ~b 8 ,a m1 ,b n1 ~b n4 It is more preferable to select, as a predictive variable, the data at the time when the degree of correlation with the prediction point (Y) is maximum among the observation information (D20) at the prediction point (Y). This allows the predictive variable to be selected more appropriately.

[0066] It is more preferable that the location relevance calculation unit (120) classifies the observation information (D20) into a plurality of groups according to the value of the observation information (D20) and calculates the degree of mutual relevance for each group. This makes it possible to calculate the degree of mutual relevance according to the value of the observation information (D20), thereby making it possible to more appropriately select the predictive variables.

[0067] In addition, the location relevance calculation unit (120) m1 ,b n1 ~b n4 ) where splitting occurs (b n2 For the branch point (a), it is more preferable to estimate the observable quantity (Q(t),H(t)) for each branch immediately after the branch. m1 ,b n1 ~b n4 ) the predictor variables can be more appropriately selected for the prediction point (Y) downstream of the

[0068] [Variations] The present invention is not limited to the above-mentioned embodiment, and various modifications are possible. The above-mentioned embodiment is exemplified to explain the present invention in an easy-to-understand manner, and is not necessarily limited to those having all the configurations described. In addition, other configurations may be added to the configurations of the above-mentioned embodiment, and it is also possible to replace some of the configurations with other configurations. In addition, the control lines and information lines shown in the figure show those that are considered necessary for explanation, and do not necessarily show all the control lines and information lines necessary on the product. In reality, it may be considered that almost all the configurations are connected to each other. Possible modifications of the above-mentioned embodiment are, for example, as follows.

[0069] (1) Since the hardware of the river data prediction variable selection device 100 in the above embodiment can be realized by a general-purpose computer, the flowcharts shown in Figures 6 and 7 and other programs for executing the various processes described above may be stored on a storage medium or distributed via a transmission line.

[0070] (2) In the above embodiment, the processes shown in FIGS. 6 and 7 and the other processes described above have been described as software-based processes using a program. However, some or all of the processes may be replaced with hardware-based processes using an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array), etc.

[0071] (3) The various processes executed in the above-described embodiments may be executed by a server computer via a network (not shown), and the various data stored in the above-described embodiments may also be stored in the server computer. [Explanation of symbols]

[0072] 30,40 Water-based 32 Honkawa (river) 34, 42, 44~49 Tributaries (rivers) 100 River Data Predictor Variable Selection Device 120 Point relevance calculation unit (point relevance calculation step, point relevance calculation means) 140 Predictor variable selection section (predictor variable selection means, predictor variable selection step) 900 Computers T Prediction time Y Prediction point GB related sites GC applicable points group D11 River Information D12 Observation point information D20 Observation Information D30 Branching Point Data D40 Related Data a 1 ,b 1 ~b 8 Observation point (location) a m1 ,b n1 ~b n4 Branch point (point)

Claims

1. a point relevance calculation unit that calculates relevance data representing relationships between a plurality of points including any of the observation points based on river information representing a connection state of a plurality of rivers, observation point information representing position information of a plurality of observation points installed in the plurality of rivers, and observation information being time-series data of quantities to be observed at the plurality of observation points; A prediction variable selection unit that selects a prediction variable corresponding to a specified prediction point and prediction time based on the association data. A river data predictor variable selection device characterized by:

2. The river information further includes a river system to which the plurality of rivers belong and a flow direction of the plurality of rivers; The observation point information further includes a river system to which the observation points belong.

2. The river data predictor variable selection device according to claim 1 .

3. The association data includes: Including the degree of correlation of flow rate or water level between the plurality of said points 2. The river data predictor variable selection device according to claim 1 .

4. The point relevance calculation unit further has a function of calculating branch point data, which is time-series data of an estimated value of the observation target quantity at a branch point of the river that is a point other than the observation point.

2. The river data predictor variable selection device according to claim 1 .

5. The prediction variable selection unit further has a function of calculating a degree of correlation between the observation target quantity and any one of the points and the prediction point.

2. The river data predictor variable selection device according to claim 1 .

6. The predictor variable selection unit further has a function of selecting a group of related locations, which is one or more of the locations having data available corresponding to the prediction time.

2. The river data predictor variable selection device according to claim 1 .

7. The predictor variable selection unit further has a function of selecting an application point group, which is one or more of the points located downstream, from the relevant point group.

7. The river data predictor variable selection device according to claim 6.

8. The predictor variable selection unit selects, from the observation information at each of the points belonging to the application point group, data at a time when the degree of correlation with the prediction point is maximum, as the predictor variable.

8. The river data predictor variable selection device according to claim 7.

9. The location relevance calculation unit classifies the observation information into a plurality of groups according to values ​​of the observation information, and calculates a degree of mutual relevance for each of the groups.

2. The river data predictor variable selection device according to claim 1 .

10. The location relevance calculation unit, For the branch points where a branch occurs, the observation target quantity for each branch immediately after the branch is estimated.

5. The river data predictor variable selection device according to claim 4.

11. a point relevance calculation step of calculating, based on river information indicating a connection state of a plurality of rivers, observation point information indicating position information of a plurality of observation points provided on the plurality of rivers, and observation information being time series data of the quantities to be observed at the plurality of observation points, a point relevance calculation step of calculating relevance data indicating the interrelationships between a plurality of points including any of the observation points; A prediction variable selection step of selecting a prediction variable corresponding to a specified prediction location and prediction time based on the association data. A method for selecting predictive variables for river data.

12. Computer, a point relevance calculation means for calculating, based on river information indicating a connection state of a plurality of rivers, observation point information indicating position information of a plurality of observation points provided on the plurality of rivers, and observation information being time-series data of quantities to be observed at the plurality of observation points, relevance data indicating relationships between a plurality of points including any of the observation points; a predictor variable selection means for selecting a predictor variable corresponding to a specified predictor location and predictor time based on the association data; A program to function as a

Images