JP2016224512A - Decision support system and decision making support method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To naturally model expert knowledge and make future prediction.SOLUTION: A decision support system having an action chain model, a reaction model, a decision model, an evaluation model, and an action selection unit, derives a next action possibly taken by a player from actions of the player using the action chain model; derives next index values from the action of the player and index values using the reaction model; calculates a selection probability of the action of each player from the derived next index values using the decision model; and calculates an evaluation value representing a desirable degree of the derived next index values for each player using the evaluation model, and the action selection unit selects and outputs the action of each player using the action derived using the action chain model, the selection probability calculated using the decision model, and the evaluation value calculated using the evaluation model.SELECTED DRAWING: Figure 2

Description

  The present invention relates to a decision support system.

  Computer systems have been proposed that use the knowledge of experts to predict future situations and behaviors of related parties.

  For example, Patent Document 1 (Japanese Patent Laid-Open No. 5-204991) discloses a step of registering a plurality of patterns in a system including a computer, a time series database, a registration pattern database, and a terminal device, and a time series data from the time series database. From the step of performing the matching with a plurality of patterns already registered for each pattern and every predetermined period, the step of analyzing the causal relationship regarding the appearance between the registered patterns, and the step of displaying the analysis result A time-series data search system is described. The time-series data search system described in Patent Literature 1 predicts a future trend (action) based on a result of comparison with a registered pattern (on a rule basis).

JP-A-5-204991

  However, in a rule-based prediction system as described in Patent Document 1, a model including many rules is created using a chain model, and what happens next is simulated. For this reason, it is difficult to create a model with a rule-based prediction system. In other words, in this model, all events and the actions of related parties must be taken into account, and in order to organize the knowledge of experts and create a model that integrates knowledge, the knowledge of experts is classified. Creating a model is difficult. For this reason, there is a need for a system that models expert knowledge without difficulty and predicts the future.

  A typical example of the invention disclosed in the present application is as follows. That is, a decision support system comprising a computer having a processor and a memory, wherein the memory stores a plurality of index values obtained by quantifying a plurality of situations necessary for decision making, and the decision support system Is an action chain model for the processor to derive the next action from the player's action, and a reaction model for the processor to derive the next index value from the player's action and the index value; An intention model for the player to derive, for each player, a selection probability representing the intention of the action from the index value; an evaluation model for the processor to derive an evaluation value for each player from the index value; An action selection unit for selecting an action of the player, and The system derives the next action that the player can take from the action of the player using the action chain model, and uses the reaction model to determine the next index value from the action of the player and the index value. And using the intention model to calculate the selection probability of each player's action from the derived next index value, and using the evaluation model, the derived next index value is An evaluation value representing a degree desired for the player is calculated, and the action selection unit uses the action derived using the action chain model, the selection probability calculated using the intention model, and the evaluation model. Using the calculated evaluation value, the action of each player is selected, and the selected action is output.

  According to the representative embodiment of the present invention, expert knowledge can be easily organized, and the situation and actions that will occur in the future can be predicted based on the expert knowledge. Problems, configurations, and effects other than those described above will become apparent from the description of the following embodiments.

It is a block diagram which shows the physical structure of the decision support system of a 1st Example. It is a block diagram which shows the logical structure of the decision support system of a 1st Example. It is a figure explaining the action chain model of the 1st example. It is a figure explaining the reaction model of the 1st example. It is a figure explaining the intention model of a 1st Example. It is a figure explaining the evaluation model of a 1st Example. It is a flowchart of the process by the decision support system of a 1st Example. It is a figure which shows the example of the simulation result output screen of 1st Example. It is a flowchart of the process by the decision support system of the modification of a 1st Example. It is a figure which shows the example of the star chart output screen of a 1st Example. It is a figure explaining the Monte Carlo tree search for comprising the star chart of a 1st Example. It is a block diagram which shows the logical structure of the decision support system of a 2nd Example. It is a figure explaining the mediation model of a 2nd Example.

  FIG. 1 is a block diagram showing the physical configuration of the decision support system of the first embodiment.

  The decision support system according to this embodiment includes a plurality of computers (CALC_NODE) and a communication switch (COM_SW) that connects the plurality of computers.

  Each computer (CALC_NODE) is a communication device (CPU) that executes a program, a temporary storage device (RAM) that stores data and a program, an auxiliary storage device (STOR), and a communication device (COM_SW) that is connected to a communication switch (COM_SW). COM_DEV). The processor (CPU), the temporary storage device (RAM), the auxiliary storage device (STOR), and the communication device (COM_DEV) are connected by a bus (BUS).

  The processor (CPU) executes a program stored in a temporary storage device (RAM). The temporary storage device (RAM) includes a ROM that is a nonvolatile storage element and a RAM that is a volatile storage element. The ROM stores an immutable program (for example, BIOS). (RAM) is a high-speed and volatile storage element such as D (RAM) (Dynamic Random Access Memory), and temporarily stores a program executed by the processor (CPU) and data used when the program is executed. To do.

  The auxiliary storage device (STOR) is a large-capacity non-volatile storage device such as a magnetic storage device (HDD) or a flash memory (SSD), for example, and is used when executing a program executed by a processor (CPU) and a program. Data to be stored. That is, the program is read from the auxiliary storage device (STOR), loaded into the storage device (RAM), and executed by the processor (CPU).

  The communication device (COM_DEV) is a network interface device that controls communication with other devices via a communication switch (COM_SW) according to a predetermined protocol.

  Each computer (CALC_NODE) may have an input interface and an output interface. The input interface is an interface that receives input from an operator, and specifically includes a mouse, a keyboard, a touch panel, a microphone, and the like. The output interface is an interface that outputs the execution result of the program in a format that can be visually recognized by the operator, such as a display device or a printer.

  A terminal computer having an input interface and an output interface may be connected to the communication switch (COM_SW).

  A program executed by the processor (CPU) is provided to each computer (CALC_NODE) via a removable medium (CD-ROM, flash memory, etc.) or a network, and is a nonvolatile auxiliary storage device (STOR) that is a non-temporary storage medium. ). Therefore, each computer (CALC_NODE) may have an interface for reading data from the removable medium.

  Each computer (CALC_NODE) is a computer system that is configured on a single computer or a plurality of computers that are logically or physically configured, and operates on separate threads on the same computer. Alternatively, it may operate on a virtual machine built on a plurality of physical computer resources. Each functional unit of each computer (CALC_NODE) may be realized on a different computer.

  FIG. 2 is a block diagram illustrating a logical configuration of the decision support system according to the first embodiment.

  The decision support system according to the present embodiment includes four models: an action chain model 1, a reaction model 2, a intention model 4, and an evaluation model 5. Specifically, the decision support system shown in FIG. 2 includes an action chain model 1, a reaction model 2, and a plurality of action determination units 3. The action determination unit 3 includes an intention model 4, an evaluation model 5, and an action selection unit 6, and is provided for each player.

  As shown in FIG. 3, the action chain model 1 is a rule-based simulator, and derives an action that is likely to be executed next from the current action of the player. A player is a subject who makes decisions and acts (executes actions) in the world, and includes, for example, the public, administrative bodies (each ministry and agency), the Diet, the Cabinet, foreign governments, and the media.

  As shown in FIG. 4, the reaction model 2 is a rule-based simulator, and derives the next index value from the current action and the current index value of each player. The index value is, for example, an index that quantifies events occurring in the world (changes in circumstances), such as economic indicators (GDP, stock price, exchange rate, etc.), and public opinion survey results (Cabinet support rate, etc.).

  The action determination unit 3 is provided for each player, and derives the next action of each player.

  As shown in FIG. 5, the intention model 4 is a simulator that derives an action intention from the next index value. The action intention is a numerical value indicating the strength of the will to execute the action adopted in a certain situation (represented by a combination of index values). That is, each player has a high probability (expected value) of selecting an action having a large action intention value. As shown in FIG. 6, the evaluation model 5 is a simulator that derives an evaluation value from the next index value. The evaluation value is a numerical value representing the degree to which a certain situation is desirable for the player.

  The action selection unit 6 is a selector that derives the next action of the player from the probability of the next action of the player, the action intention, and the evaluation value. The action selection unit 6 weights the action output from the action chain model 1 with the action intention output from the intention model 4 and the evaluation value output from the evaluation model 5, for example. select.

  The next action derived by the action selection unit 6 is input to the action chain model 1 and used for simulation of the next action. The next index value derived by the reaction model 2 is input to the reaction model 2 and used for the simulation of the next index value.

  FIG. 3 is a diagram for explaining the action chain model 1 of the first embodiment. The action chain model 1 is represented by a Markov decision process model in which each player's action is a node. Each node is associated with a probability that the player selects an action, and an edge between the nodes is associated with a probability that the state transitions between the nodes.

In the action chain model 1 shown in FIG. 3, the probabilities that the player 1 selects the action 1, the action 2 and the action 3 are 0.3: 0.5: 0.2, respectively. When player 1 selects action 1, player 2 selects action 2 with a probability of 0.8. That is, the probability that the player 2 selects the action 2 can be expressed by Equation 1.
1-((1-0.3) × (1-0.8)) (Formula 1)

  The action chain model 1 can derive one or more next actions that are likely from the player's current action.

  FIG. 4 is a diagram for explaining the reaction model 2 of the first embodiment. The reaction model 2 is a model that represents the correlation between each index by a causal loop diagram.

  For example, as illustrated in FIG. 4, the reaction model 2 can be represented by a graphical model in which a plurality of indices are nodes and the nodes are connected by edges. A solid line arrow at each edge indicates a positive correlation, and a broken line arrow indicates a negative correlation. Furthermore, by defining a coefficient for each edge, a system dynamics model representing the behavior of each index can be obtained. The coefficient of each edge is defined by the ratio with the increase / decrease amount of the index. For example, if the index 3 and the index 2 have a positive correlation and the edge coefficient is 0.5, when the index 3 increases by 1, the index 2 increases by 0.5. In addition, the index 3 and the index 5 have a negative correlation, and if the edge coefficient is 1.2, when the index 3 increases by 1, the index 5 decreases by 1.2.

  When performing a discrete system simulation, the reaction model 2 can be represented by a recurrence formula in the computer. Moreover, when performing a continuous system simulation, the reaction model 2 can be represented by a primary differential equation in the computer.

  Indicators include stock elements and flow elements. The stock element indicates an amount at a certain point in time, for example, a stock of crude oil. The flow element indicates a flow of variables in a time zone such as the import amount (production amount) and consumption amount of crude oil. Whether an index is a stock element or a flow element may be determined depending on whether the index is generally used as a stock quantity or a flow quantity in the world. Moreover, you may use the quantity (for example, risk impact, nationalism) which cannot be actually measured numerically as a parameter | index. In this way, by configuring the reaction model 2 by mixing the stock elements and the flow elements, it is possible to model the thinking of the expert without any restrictions.

  FIG. 5 is a diagram for explaining the intention model 4 of the first embodiment. The intention model 4 is a model that expresses a correlation between each index by a causal loop diagram, and further represents a correlation between each player's action and each index.

  The intention model 4 can be expressed by a model similar to the reaction model 2 described above. That is, the intention model 4 can be represented by, for example, a graphical model in which a plurality of indices are nodes and nodes are connected by edges. A solid line arrow at each edge indicates a positive correlation, and a broken line arrow indicates a negative correlation. Furthermore, by defining a coefficient for each edge, a system dynamics model representing the behavior of each index can be obtained. The coefficient of each edge is defined by the ratio with the increase / decrease amount of the index.

  In the intention model 4, the indexes have a correlation with each other. However, between the player's action and the index, only the edge from each player's action toward the index is defined, and the opposite edge is not defined. Also, the edge between each action is not defined. The intention model 4 can model the influence of each player's action on the index.

  When performing discrete system simulation, the intention model 4 can be expressed by a recurrence formula in the computer. Moreover, when performing a continuous system simulation, the intention model 4 can be represented by a primary differential equation in the computer. In the intention model 4, as in the reaction model 2, the indicators of the intention model 4 include a stock element and a flow element.

  FIG. 6 is a diagram for explaining the evaluation model 5 of the first embodiment. The evaluation model 5 represents a correlation between each indicator by a corusal loop diagram, represents a correlation between each player's action and each indicator, represents a correlation between each indicator and each player's evaluation, This model represents the correlation between each indicator and each player's intention of action. The evaluation is a numerical value indicating whether each player prefers the situation represented by the combination of a plurality of indicators. Note that the evaluation may have a different evaluation value for each combination of numerical ranges of the index. The action intention is represented by a set of actions that each player can take and the probability of selecting each action.

  The evaluation model 5 can be represented by a model similar to the reaction model 2 described above. That is, the evaluation model 5 can be represented by, for example, a graphical model in which a plurality of indices are nodes and nodes are connected by edges. A solid line arrow at each edge indicates a positive correlation, and a broken line arrow indicates a negative correlation. Furthermore, by defining a coefficient for each edge, a system dynamics model representing the behavior of each index can be obtained. The coefficient of each edge is defined by the ratio with the increase / decrease amount of the index.

  In the evaluation model 5, the indexes have a correlation with each other, but only an edge from each player's action toward the index is defined, and no opposite edge is defined between the player's action and the index. Also, the edge between each action is not defined. Also, between each index and each player's evaluation, only the edge from each index toward the evaluation is defined, and the opposite edge is not defined. Also, between each index and each player's action intention, only the edge from each index toward the action intention is defined, and the opposite edge is not defined. In addition, the edges between intentions of action are not defined. The evaluation model 5 can model the effect of each index on the strength of the intention of each player's action, and can determine the evaluation value of each player's action.

  When performing discrete system simulation, the intention model 4 can be expressed by a recurrence formula in the computer. Moreover, when performing a continuous system simulation, the intention model 4 can be represented by a primary differential equation in the computer. Similar to the reaction model 2, the evaluation model 5 includes stock elements and flow elements as indices of the evaluation model 5.

  The evaluation model 5 includes the intention model 4, and the intention model 4 includes the reaction model 2. Therefore, the reaction model 2, the intention model 4, and the evaluation model 5 may be configured by logically dividing one model.

  In the reaction model 2, the intention model 4, and the evaluation model 5, the events represented by each node are somewhat related, so that an edge can be defined between almost all nodes. However, if the edges are defined between all the nodes, the model becomes complicated. Therefore, the model may be configured with edges having high correlation (for example, edges whose coefficients are larger than a predetermined threshold).

  FIG. 7 is a flowchart of processing by the decision support system of the first embodiment.

  First, when the current situation and simulation period are input to the input interface (S101), the simulation start time is set as the initial value of the repetitive control parameter t, and the simulation end time t_end is set. The current state entered includes each player's current action and each current index value.

  Next, it is determined whether t is smaller than t_end (S102). If t is smaller than t_end, the process proceeds to steps S103 and S105. If t is equal to or greater than t_end, the simulation result for the specified period is obtained, so the processing is terminated and the simulation result output screen (FIG. 8) is output.

  In step S103, the action chain model 1 is driven, and the next action having a probability is derived from the current action of the player, and stored in the auxiliary storage device (STOR) (S104). In step S105, the reaction model 2 is driven, and the next index value is derived from the current action and the current index value of each player, and stored in the auxiliary storage device (STOR) (S106).

  The processes of steps S103 to S106 can be executed in parallel, but the process of starting the action chain model 1 (S103) and the process of driving the reaction model 2 (S105) may be executed in order.

  Next, the intention model 4 and the evaluation model 5 are driven for all players (S107, S109). The processes of steps S107 to S110 can be executed in parallel, but the process of starting the intention model 4 (S107) and the process of driving the evaluation model 5 (S109) may be executed in order.

  In step S107, the intention model 4 is driven, the action intention of each player is derived from the next index value, and stored in the auxiliary storage device (STOR) (S108). In step S109, the evaluation model 5 is driven, an evaluation value is derived from the next index value, and stored in the auxiliary storage device (STOR) (S110).

  Thereafter, the action selection unit 6 determines the next action of the player in consideration of the probability of the next action of the player, the action intention, and the evaluation value, and stores them in the auxiliary storage device (STOR) (S111).

  After the next action of all players is determined, the next index value output from the reaction model 2 is set to the current index value, and the time of the reaction model 2 is advanced by one (S112). Then, 1 is added to the repetitive control parameter t (S113), and the process returns to step S102. In addition, 1 added to t shows the time interval which performs simulation, and it is good for an operator to preset (for example, 1 day).

  Through the above processing, each player's action during the simulation period can be derived.

  FIG. 8 is a diagram illustrating an example of a simulation result output screen 1000 according to the first embodiment. The simulation result output screen 1000 represents an action selected by each player as time passes, and is displayed on an output interface (display device). For example, as shown in FIG. 8, the actions at each time of each player are displayed in a table format in which players are listed in the vertical direction and times of simulation results are listed in the horizontal direction. The simulation result output screen 1000 allows the user to know the actions taken by each player in time series.

  Next, a modification of the first embodiment will be described. In the modification described below, the action of each player is selected using a star chart.

  FIG. 9 is a flowchart of processing by the decision support system according to the modification of the first embodiment.

  First, when the current situation and simulation period are input to the input interface (S121), the initial value of the repetitive control parameter t is set at the start of simulation, and t_end at the end of simulation is set. The input current state includes the current action of each player and the current index values.

  Next, it is determined whether t is smaller than t_end. If t is smaller than t_end, the process proceeds to steps S103 and S105 (S122). If t is equal to or greater than t_end, the simulation result for the specified period is obtained, so the processing is terminated and the simulation result output screen (FIG. 8) is output.

  Next, all combinations of action options that all players can take are listed (S123), and for all the listed combinations, the action chain model 1 and the reaction model 2 are driven to t + α, and each player's next combination is selected. The action and the next index value are derived and stored in the auxiliary storage device (STOR) (S124). α is a future time at which a prediction is considered at time t in calculating the score of each action described in the star chart.

  Thereafter, the intention model 4 and the evaluation model 5 are driven for all the players, and the action intention and evaluation value of each player are derived and stored in the auxiliary storage device (STOR) (S125).

  Thereafter, a star chart is created based on the evaluation value at time t + α (S126), the action of the player at time t is determined using the star chart, and stored in the auxiliary storage device (STOR) (S127). Specifically, the star chart is a table in which one's own evaluation and other player's evaluation are described, and is used to select an appropriate action as will be described later.

  After the next action of all players is determined, the next index value output from the reaction model 2 is set as the current index value, and the time of the reaction model 2 is advanced by one (S128). Then, 1 is added to the repetition control parameter t (S129), and the process returns to step S122.

  Through the above processing, each player's action during the simulation period can be simulated in consideration of the actions of others.

  FIG. 10 is a diagram illustrating an example of the star chart output screen 1100 according to the first embodiment. The star chart output screen is displayed on the output interface (display device) by selecting the action column in the simulation result output screen 1000 (FIG. 8). Since the star chart represents the relationship between the actions of the two players, the operator selects the opponent player after selecting the action column on the simulation result output screen 1000.

  Note that the star chart is preferably composed of a Monte Carlo tree shown in FIG. 11 inside the computer.

  First, the contents of the screen will be described. The star chart output screen 1100 shown in FIG. 10 shows the relationship between two players, and includes a star chart 1110 in which the actions of the player 1 are listed in the vertical direction and the actions of the player 2 are listed in the horizontal direction. A “return” button 1120 is provided at the bottom of the screen. The operator can return to the simulation result output screen 1000 by operating the “return” button 1120.

  Next, the contents of the star chart will be described. The star chart shown in FIG. 10 is recorded with a combination of the evaluation of the player 1 and the evaluation of the player 2 in the combination of the action of the player 1 and the action of the player 2. The evaluation may be expressed by a symbol as shown in FIG. 10 or may be expressed by a numerical value. By using the star chart, it is possible to determine an action that evaluates the evaluation values of a plurality of players.

  Next, a method for selecting an action using the star chart will be described in the case where the player is the player 1 and the opponent is the player 2. The action is determined using the MinMax method so that the largest possible damage is minimized.

  For example, if there are two players, paying attention to each row of the star chart (your own action is the same), the evaluation value (MinMax evaluation value) of player 1 in the action with the best evaluation value of the opponent (player 2) is The best action is determined as the next action of the player. In the illustrated case, the action of the player 1 is determined as action 2.

  If there are three or more players, the actions of other players are fixed in order for the actions of the player (player 1) (assuming that the player with the highest evaluation value of the player is selected), Decide on the highest action. This will be specifically described below.

  If there are three players, the action is decided by the following steps.

  Step 1: Set own (player 1) action to action 1.

In that state, consider a two-player game between player 2 and player 3.
Step 1-1: With the player 2 as the player, the above player determines the action of the player 2 by the method of two people (MinMax according to the star chart).
Step 1-2: With the player 3 as its own player, the player determines the action of the player 3 by the method of two people (MinMax according to the star chart).

  Evaluation of the player (player 1) in the combination of the actions of each player determined as described above (player 1 = action 1, player 2 = action determined in step 1-1, player 3 = action determined in step 1-2) The value is set to the evaluation value in action 1 of player 1.

  Step 2: Set own (player 1) action to action 2.

  Further, the evaluation value in the action 2 of the player 1 is determined by the same method as described above.

  Step 3: The calculation of step 1 is performed for the number of actions of the player (player 1), and the action having the best evaluation value for the player (player 1) is determined.

  Furthermore, when there are four players, the action is determined by the following steps.

Step 4: Fix yourself (player 1) 's action. As a result, the players 2 to 4 become three players.
Step 4-1: In the game of three players, the player 2 is determined as the player, and the player determines the action of the player 2 by the method of three players.
Step 4-2: In the three-player game, the player 3 is the player, and the player determines the action of the player 3 by the method of three players.
Step 4-3: In the three-player game, the player 4 is determined to be the player, and the player determines the action of the player 4 by three methods.

In the combination of the actions of each player determined as described above (player 1 = action 1, players 2-4 = actions determined in steps 4-1 to 4-3), the evaluation value of the player (player 1) Set to the evaluation value in action 1.
Step 5: Step 4 is calculated for the number of actions of the player (player 1), and the action having the best evaluation value for the player (player 1) is determined.

  FIG. 11 is a diagram for explaining a Monte Carlo tree search for configuring the star chart of the first embodiment.

  That is, in the method described above with reference to FIG. 10, since all combinations of actions of all players are calculated, the calculation amount is large. For this reason, the same processing can be approximately executed with a small amount of calculation by using Monte-Carlo Tree Search. If the Monte Carlo tree is calculated infinitely, the same result as when all combinations of actions of all players are calculated can be obtained.

  In the Monte Carlo tree search shown in FIG. 11, when each player can select one of actions 1 and 2 in the game of four players 1 to 4, the evaluation value in action 1 of himself (player 1) is calculated. The processing to be performed is shown.

  First, the player (player 1) selects action 1. Next, one of the players 2 to 4 is randomly selected (equal probability). Hereinafter, the case where the player 2 is selected will be described. The player 2 selects an action having a high average evaluation value for the player 2 at the child node.

  Next, one player 3 or 4 is selected at random (equal probability). Hereinafter, the case where the player 3 is selected will be described. The player 3 selects an action having a high average evaluation value for the player 3 at the child node. Finally, the player 4 selects an action having a high evaluation value for the player 4 at the child node.

  With the above processing, a set of actions of each player at time X + 0 is determined.

  Further, when prefetching is performed, the following processing is executed.

  First, one of the players 1 to 4 is selected at random (equal probability), and an action having a high evaluation value for the selected player is selected at the child node. Thereafter, for the remaining players, the player is randomly selected and the action is determined in the same manner as described above. After expanding the tree to some extent, all players randomly select an action and advance the time by random playout. Then, the evaluation value of each player when the predetermined number of pre-reads is reached is calculated. Finally, each node of the tree that has passed so far is traced in reverse, the evaluation value at the time of prefetching is added to the evaluation value attached to the node, the average value is obtained, and the evaluation value is updated.

  After performing the above processing about several hundred times, the evaluation value of the player 1 attached to the root node becomes the evaluation value of the action 1 of the player 1.

  The above processing is performed for each action that the player (player 1) can take, and the action with the best evaluation value for the player (player 1) is selected.

  Thus, the star chart allows the user to know the reason why the action is derived.

  As described above, according to the first embodiment of the present invention, since the decision support system is configured using the action chain model 1, the reaction model 2, the intention model 4, and the evaluation model 5, the knowledge of experts Can be easily organized and modeled. Therefore, it is possible to predict the situation and actions that will occur in the future based on the knowledge of experts. In particular, since the intention model 4 and the evaluation model 5 are separately modeled, the intention factor and the suppression factor can be separated, and expert knowledge can be incorporated into the model without being processed.

<Second embodiment>
Next, a second embodiment of the present invention will be described. The decision support system of the second embodiment is composed of five models: an action chain model 1, a reaction model 2, a intention model 4, an evaluation model 5, and an arbitration model 7. In the second embodiment, description of the same configuration and processing as those of the first embodiment is omitted, and different configuration and processing will be described.

  FIG. 12 is a block diagram illustrating a logical configuration of the decision support system according to the second embodiment.

  The decision support system of the second embodiment includes an action chain model 1, a reaction model 2, a plurality of action determination units 3, and an arbitration model 7. The action determination unit 3 includes an intention model 4, an evaluation model 5, and an action selection unit 6, and is provided for each player.

  The action chain model 1, the reaction model 2, the intention model 4, the evaluation model 5, and the action selection unit 6 are the same as those in the first embodiment described above. The action determination unit 3 of the second embodiment outputs a plurality of actions that can be taken for each player together with the selection rate.

  The mediation model 7 mediates the actions output from the plurality of action determination units 3 and determines the action of each player. For example, there are conflicting actions that each player can take. The arbitration model 7 uses these relationships to select a combination of actions that can be performed at the same time, and determines an action for each player.

  Specifically, the arbitration model 7 calculates a selection rate of a plurality of actions output from the action selection unit 6, selects an action with the highest selection rate for each player, and determines the action of each player.

  FIG. 13 is a diagram illustrating the arbitration model 7 according to the second embodiment. The arbitration model 7 is a model that represents the correlation between each index (stock element, flow element, state element) by a causal loop diagram.

  For example, in the illustrated arbitration model 7, the stock elements 1, 2, 3, and 4 are x1, x2, x3, and x4, respectively, and the state elements 1 and 2 are y1 and y2, respectively. k1 to k9). The state element is a number representing the current state, and can be set to 1 when the player 1 is currently performing action 1, 0 when the player 1 is performing action other than action 1, and the like.

  As shown in the drawing, the amount of flow flowing into and out of the stock element is determined by the stock element and the state element, and the amount of the stock element changes in an empirical manner.

When defined as described above, in the discrete system simulation, the value of each stock element at time t + 1 can be calculated by the following recurrence formula.
x1 (t + 1) = k1 * y1 (t) + k2 * x4 (t) -k3 * x3 (t)
x2 (t + 1) = x1 (t) + k3 * x3 (t) -k4 * x4 (t)
x3 (t + 1) = k5 * x4 (t) -k6 * x1 (t)
x4 (t + 1) = k7 * x1 (t) -k8 * x3 (t) -k9 * y2 (t)

In the continuous system simulation, the value of each stock element can be calculated by the following differential equation.
d [x1 (t)] / dt = k1 * y1 (t) + k2 * x4 (t) -k3 * x3 (t)
d [x2 (t)] / dt = x1 (t) + k3 × x3 (t) −k4 × x4 (t)
d [x3 (t)] / dt = k5 * x4 (t) -k6 * x1 (t)
d [x4 (t)] / dt = k7 * x1 (t) -k8 * x3 (t) -k9 * y2 (t)

  In the arbitration model 7, by controlling the flow amount by the stock element and the state element, a plurality of stock elements are controlled in association with each other, the probability (selection rate) that each player selects an action is determined, and the action of each player is determined. Mediation is possible.

  As described above, according to the second embodiment of the present invention, the arbitration model 7 arbitrates the actions of each player selected by the action selection unit 6 and determines the actions of each player. Models 1, 2, 4, and 5 can be created without considering mediation of actions. That is, it is possible to create a model separately from the mediation of actions.

  The present invention is not limited to the above-described embodiments, and includes various modifications and equivalent configurations within the scope of the appended claims. For example, the above-described embodiments have been described in detail for easy understanding of the present invention, and the present invention is not necessarily limited to those having all the configurations described. A part of the configuration of one embodiment may be replaced with the configuration of another embodiment. Moreover, you may add the structure of another Example to the structure of a certain Example. In addition, for a part of the configuration of each embodiment, another configuration may be added, deleted, or replaced.

  In addition, each of the above-described configurations, functions, processing units, processing means, etc. may be realized in hardware by designing a part or all of them, for example, with an integrated circuit, and the processor realizes each function. It may be realized by software by interpreting and executing the program to be executed.

  Information such as programs, tables, and files that realize each function can be stored in a storage device such as a memory, a hard disk, or an SSD (Solid State Drive), or a recording medium such as an IC card, an SD card, or a DVD.

  Further, the control lines and the information lines are those that are considered necessary for the explanation, and not all the control lines and the information lines that are necessary for the mounting are shown. In practice, it can be considered that almost all the components are connected to each other.

CALC_NODE computer CPU processor RAM temporary storage device STOR auxiliary storage device COM_SW communication switch COM_DEV communication device 1 action chain model 2 reaction model 3 action decision unit 4 intention model 5 evaluation model 6 action selection unit 7 arbitration model

Claims (8)

A decision support system comprising a computer having a processor and a memory,
The memory stores a plurality of index values obtained by quantifying a plurality of situations necessary for decision making,
The decision support system includes:
An action chain model for the processor to derive a next action from a player action;
A reaction model for the processor to derive a next index value from a player's action and the index value;
An intention model for the processor to derive, for each player, a selection probability representing the intention of the action from the index value;
An evaluation model for the processor to derive an evaluation value from the index value for each player;
The processor has an action selection unit for selecting an action of a player;
The decision support system includes:
Using the action chain model, deriving the next action that the player can take from the action of the player,
The reaction model is used to derive the next index value from the player's action and the index value,
Using the intention model, calculate the selection probability of each player's action from the derived next index value,
Using the evaluation model, calculating an evaluation value representing the degree to which the derived next index value is desirable for each player;
The action selection unit uses the action derived using the action chain model, the selection probability calculated using the intention model, and the evaluation value calculated using the evaluation model. A decision support system, characterized by selecting a player action and outputting the selected action. The decision support system according to claim 1,
The processor has an arbitration model for arbitrating the actions of each selected player to determine the actions of each player;
The action selection unit selects a plurality of actions that each player can take, and outputs probabilities of the selected plurality of actions.
The decision support system uses the arbitration model to arbitrate the actions of the players selected by the action selection unit to determine the actions of the players. The decision support system according to claim 1,
The action selection unit outputs screen data for displaying the actions of the selected players in time series. The decision support system according to claim 1,
The decision selection support system, wherein the action selection unit outputs screen data for displaying an evaluation of each player for a set of actions of the plurality of players. A decision support method executed by a computer having a processor and a memory,
The memory stores a plurality of index values obtained by quantifying a plurality of situations necessary for decision making,
The calculator includes an action chain model for deriving a next action from a player action, a reaction model for deriving a next index value from the player action and the index value, and an intention of the action from the index value. An intention model for deriving a selection probability for each player, and an evaluation model for deriving an evaluation value for each player from the index value,
The method
The processor derives the next action that the player can take from the action of the player using the action chain model, and stores it in the memory;
The processor uses the reaction model to derive the next index value from the player action and the index value, and stores the next index value in the memory,
The processor calculates the selection probability of each player's action from the derived next index value using the intention model, and stores it in the memory;
The processor uses the evaluation model to calculate an evaluation value representing the degree to which the derived next index value is desirable for each player, and stores the evaluation value in the memory;
The processor uses the action derived using the action chain model, the selection probability calculated using the intention model, and the evaluation value calculated using the evaluation model to A decision support method, wherein an action is selected and stored in the memory. The decision support method according to claim 5,
The calculator has an arbitration model for arbitrating the actions of each selected player to determine the actions of each player;
The processor selects a plurality of actions that each player can take, and outputs probabilities of the selected plurality of actions.
A decision support method, wherein the processor arbitrates the action of each selected player using the arbitration model, and determines the action of each player. The decision support method according to claim 5,
The said processor outputs the screen data for displaying the action of each said selected player in time series, The decision support method characterized by the above-mentioned. The decision support method according to claim 5,
A decision support method, wherein the processor outputs screen data for displaying an evaluation of each player for a set of actions of the plurality of players.

Images